Citation
Economic Base Analysis Using Panel Data Regression

Material Information

Title:
Economic Base Analysis Using Panel Data Regression a Case Study of the Florida Regional Economy
Creator:
Li, Fan
Place of Publication:
[Gainesville, Fla.]
Publisher:
University of Florida
Publication Date:
Language:
english
Physical Description:
1 online resource (109 p.)

Thesis/Dissertation Information

Degree:
Master's ( M.S.)
Degree Grantor:
University of Florida
Degree Disciplines:
Geography
Committee Chair:
Fik, Timothy J.
Committee Members:
Thrall, Grant I.
Qiu, Youliang
Graduation Date:
8/7/2010

Subjects

Subjects / Keywords:
Counties ( jstor )
Economic models ( jstor )
Economic regions ( jstor )
Economics ( jstor )
Employment ( jstor )
Estimation methods ( jstor )
Linear regression ( jstor )
Modeling ( jstor )
Regression analysis ( jstor )
Scalars ( jstor )
Geography -- Dissertations, Academic -- UF
base, economic, panal, regional, regression
City of Miami ( local )
Genre:
Electronic Thesis or Dissertation
bibliography ( marcgt )
theses ( marcgt )
Geography thesis, M.S.

Notes

Abstract:
The main purpose of this research is to investigate and estimate the increase in total regional employment as it relates to employment increases in specific economic sectors, using economic base theory and a regional economic base model. Past empirical research in economic base modeling typical relies on single year data and original least square (OLS) regression, with unstable results that are possibly explained by the fact that the regression analysis is influenced by abnormal value changes in one or more explanatory variables. As an alternative, this thesis utilizes panel data regression with data for multiple years in an economic base analysis of Florida employment patterns. Sector-specific employment multipliers are estimated and compared using OLS and the panel data regression. Of the other widely used methodologies: the survey method, location quotient (LQ) method, and minimum requirements (MR), regression analysis is the only one which can estimate impact multipliers. Regression models to estimate employment multipliers are typically divided into disaggregate and aggregate models. The former can obtain the multiplier estimates for specific industries, while the latter one independently cannot distinguish between different industries. Hence, disaggregate regression models are widely used in economic base research. Unable to account for economic spillover effects, such as cash flows or trans-boundary monetary transfers, disaggregate regression models tend to have upwardly biased multipliers. It is necessary, therefore, to apply scaling factors and utilize independent variables that capture the influence of monetary transfers to account for such factors as unemployment and retirement benefits and other state income sources in an effort to decrease bias in the modeling process. Transfer payments, dividends, interest and rents, and social security benefits are considered in this analysis, and prove to be effective in reducing the upward bias of multiplier estimates. This research also examines the assumptions of original least square and alternatives, to test for the potential violation of assumptions. In order to obtain reasonable and stable estimates of multiplier related coefficients and avoid potential problems associated with endogenous error terms, panel data regression is preferred to original least squares regression when data for more than one sample time period is available. The estimation results suggest that employment increases in wholesale trade can create most total employment benefits in Florida's regional economy, while job increases in the transportation and public utility sector leads to relatively small job returns to Florida's overall regional economy. Comparison of results between ordinary least square regression based on a single year sample and panel data regression using a two-year sample suggest that the latter can avoid the impact of sudden and discontinue change in a regional economic trend. In short, the panel regression approach is deemed as preferable over ordinary least square regression in the estimation of employment multipliers. ( en )
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis:
Thesis (M.S.)--University of Florida, 2010.
Local:
Adviser: Fik, Timothy J.
Statement of Responsibility:
by Fan Li.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Applicable rights reserved.
Embargo Date:
10/8/2010
Resource Identifier:
004979610 ( ALEPH )
769016412 ( OCLC )
Classification:
LD1780 2010 ( lcc )

Downloads

This item has the following downloads:

li_f.pdf

li_f_Page_009.txt

EGSTUITN8_W6SMAJ_xml.txt

li_f_Page_098.txt

li_f_Page_096.txt

li_f_Page_053.txt

li_f_Page_084.txt

li_f_Page_012.txt

li_f_Page_091.txt

li_f_Page_051.txt

li_f_Page_097.txt

li_f_Page_037.txt

li_f_Page_052.txt

li_f_Page_044.txt

li_f_Page_109.txt

li_f_Page_013.txt

li_f_Page_047.txt

li_f_Page_006.txt

li_f_Page_080.txt

li_f_Page_099.txt

li_f_Page_019.txt

li_f_Page_066.txt

li_f_Page_011.txt

li_f_Page_027.txt

li_f_Page_042.txt

li_f_Page_064.txt

li_f_Page_014.txt

li_f_Page_048.txt

li_f_Page_069.txt

li_f_Page_017.txt

li_f_Page_022.txt

li_f_Page_002.txt

li_f_Page_067.txt

li_f_Page_107.txt

li_f_Page_092.txt

li_f_Page_008.txt

li_f_Page_075.txt

li_f_Page_041.txt

li_f_pdf.txt

li_f_Page_087.txt

li_f_Page_018.txt

li_f_Page_074.txt

li_f_Page_057.txt

li_f_Page_004.txt

li_f_Page_015.txt

li_f_Page_020.txt

li_f_Page_039.txt

li_f_Page_089.txt

li_f_Page_065.txt

li_f_Page_101.txt

li_f_Page_035.txt

li_f_Page_028.txt

li_f_Page_036.txt

li_f_Page_100.txt

li_f_Page_070.txt

li_f_Page_045.txt

li_f_Page_072.txt

li_f_Page_023.txt

li_f_Page_061.txt

li_f_Page_026.txt

li_f_Page_031.txt

li_f_Page_056.txt

li_f_Page_081.txt

li_f_Page_029.txt

li_f_Page_093.txt

li_f_Page_086.txt

li_f_Page_001.txt

li_f_Page_106.txt

li_f_Page_083.txt

li_f_Page_007.txt

li_f_Page_005.txt

li_f_Page_077.txt

li_f_Page_055.txt

li_f_Page_038.txt

li_f_Page_032.txt

li_f_Page_049.txt

li_f_Page_062.txt

li_f_Page_104.txt

li_f_Page_030.txt

li_f_Page_054.txt

li_f_Page_088.txt

li_f_Page_016.txt

li_f_Page_034.txt

li_f_Page_043.txt

li_f_Page_079.txt

li_f_Page_003.txt

li_f_Page_073.txt

li_f_Page_046.txt

li_f_Page_071.txt

li_f_Page_085.txt

li_f_Page_095.txt

li_f_Page_033.txt

li_f_Page_082.txt

li_f_Page_025.txt

li_f_Page_090.txt

li_f_Page_050.txt

li_f_Page_105.txt

li_f_Page_103.txt

li_f_Page_060.txt

li_f_Page_021.txt

li_f_Page_094.txt

li_f_Page_010.txt

li_f_Page_076.txt

li_f_Page_068.txt

li_f_Page_102.txt

li_f_Page_059.txt

li_f_Page_024.txt

li_f_Page_108.txt

li_f_Page_040.txt

li_f_Page_063.txt

li_f_Page_078.txt

li_f_Page_058.txt


Full Text


















S s w ies


Bay -Slbwanrn.eColmbaBakerI
tLIberty' W ,Rla 'II I
Gulf Tayor Lafaefte- Clay Joh
G f ranklin J '1nd... t t. -
K G'ncrl'rin'ait i.'f I
SDixie i Putnam'

Levy Marion

Citrus n e
Hemando .
SPaso


Legend

Non-Basic Employment

for Services in 1993
56- 5130

5130-17100

17100 30930

30930 54670

54670 277201


0 62.5 125
I I I I I


Indian River
1 Manatee HarCee Okeenhobee t
Highlands t. Lucia
DesToro Martin
Charlotte lades
Hendry









Monroe
^7 Q^- ,--^~5


250 Kilometers
I N


Figure 4-9. Non-basic employment for services sector of Florida in different years by county
A)1993 B)1997




















L | ,' 'Jack'on

l -Calhoun Ma disorna
Bay Libertt Wakulla SuwaneeC nn
SGuI I- TaylorLafayette.
S Jranklln '-
\/---./ "// G' h"^^^a,,,
Dixie I
SLevy Ma

Citrus
Hemando-
Pasco

Pin


Legend
Total Employment in 1997

|__ 764-18670
18670 54780
54780- 102520
102520 150280
150280 8162200


0 62.5 125 250 Kilometers
I l I I L I I I


jdilan Rive
Manatee Hsroee OkechobeeL Lu
Highlan Lucle
Deoto Martin
Charlotte Glades
Hendry









)-
6, -'Monr e







N


Figure 4-1. Continued









Table 4-1. Continued
County scalar 1 scalar 2 nonbasic in 1993 nonbasic in 1997
Nassau 0.67 0.8529 8982 11287
Okaloosa 0.63 0.8529 40977 49660
Okeechobee 0.55 0.8529 5001 5344
Orange 0.7 0.8529 369316 453648
Osceola 0.67 0.8529 29503 36068
Palm Beach 0.46 0.8529 328170 389281
Pasco 0.48 0.8529 48966 56339
Pinellas 0.55 0.8529 306103 348737
Polk 0.61 0.8529 120189 133398
Putnam 0.54 0.8529 9134 10624
St.Johns 0.6 0.8529 23046 30035
St.Lucie 0.47 0.8529 32541 37475
Santa Rosa 0.69 0.8529 13465 15891
Sarasota 0.42 0.8529 95709 119964
Seminole 0.73 0.8529 87836 107294
Sumter 0.45 0.8529 3614 3703
Suwannee 0.58 0.8529 4769 5141
Taylor 0.61 0.8529 3082 3086
Union 0.69 0.8529 1291 1115
Volusia 0.5 0.8529 104654 118695
Wakulla 0.71 0.8529 1443 1686
Walton 0.57 0.8529 6119 7871
Washington 0.57 0.8529 2554 3272
U.S., Florida has a relatively large number of in-migrants and retirees that relocated to Florida

from other states. Subsequently, Florida has a large ratio of non-employment related income

when compared to other states, and conceivably an inordinate share and influx of external

income from sources that are varied categorically and geographically. Note that the ratio of net

earnings to total personal income in Florida is approximately 58%, while the ratio of net earnings

to total personal income of United States is 68%. Thus, the ratio of 58% to 68% (or .8529) is the

second scalar that can be used to deflate estimated non-basic employment. This second scalar

denotes the influence of state-level non-employment related income transfers that filter down on

average to county level economies. Therefore, the dependent variable Fy in regression model

can be re-sated, adjusting for the impact of the two scalars:

EN = EN (4-2)





















Sn a' Holmes
S,-an -"-' I I "oJack orr
SE 'a tWaiQ a i.o Gads 4son lanso Hariuto
'y u .Uibe Wakulla/ SwanneeColn rn BKr











Non-Basic Employment. TalLafate Unn Clay
Gr C strctin i. L: ord St Johns
S franklt ."
Dir FlagI
L vy 1 2 rloK

Citrus ke
~mFigurer
Hemando .
) Pasc .
Osceola

I dian Rive
allaS Harde, Okee,,hobee
Highland St. Luc
DeStol Mamn
ChadotteI




Legend
Non-Basic Employment

for Construction in 1997
44-920
920 2720 _._ sOr
2720- 4880
4880 8820
S8820 33105


0 62,5 125 250 Kilometers N


B

Figure 4-3.Continued









Figure 4-18 shows social security payment of year 1993 and 1997 by county, which

shows same spatial distribution by county across years.

Regression Analysis

The disaggregate regression model which shows non-basic employment as a function of

basic employment by sector can be expressed as

EN = a + blEBprm + b2EBcons + b3EBman + b4EBtran + b5EB whole + b6EBeta + b7EBfina +
b8,EA e + b9Soci (4-1)

where _y denotes total non-basic employment in a j-th regional economic or county,

independent variables are the estimated basic employment of each industry/sector: agricultural

services, forestry, fishing and mining (prim); construction (cons); manufacturing (manu);

transportation and public utilities (tran); wholesale trade (whol); retail trade (reta); finance,

insurance, and real estate (fina); and services(serv). Note that the model is also expanded to

include social security benefits (soci), in thousands of dollars; a variable that is used as a

surrogate for income that creates a cash flow to regional economy and an expansion in sales and

services. The use of this surrogate helps to account for a source of state income that can

augment the non-basic sector. This especially important in the state of Florida were such a large

proportion of the population is 65 years of age or older, many of whom are retired and drawing

Social Security payments.

The sample database consists of data for n=67 Florida counties and data for two distinct

time periods (1993 and 1997). If one-year samples (for either year 1993 or year 1997) are

applied, the OLS regression model can be used. However, to obtain more stable and consistent

estimation of sector-specific economic base multipliers, data for these two time periods can be

combined using the panel data regression approach. Before this can be accomplished, the

dependent variable (county level non-basic employment) must be scaled back to account for










TABLE OF CONTENTS

page

A C K N O W L E D G M E N T S ......... .. ............................................................................................. 4

L IST O F T A B L E S ... .. .................... ..... ..................................................................... .............. ....... 8

L IST O F F IG U R E S ................................................................. 9

A B S T R A C T .................................................................................... ............. ...... 1 1

CHAPTER

1 IN TR O D U C TIO N ...................................................... 13

Overview and Problem Statem ent ...................................................... 13
B ack g rou n d........................................................................................................... ............... ...... 14
R research O bjectiv es ....................................................................................... 15

2 LITERATURE REVIEW ......................................... 17

E conom ic B ase Theory ..................................................... 17
Tw o V versions of Econom ic B ase M odel ........................................................... ............... 21
The Average Multiplier of the Economic Base Model ......................................................22
The Marginal Multiplier of Economic Base Model ..........................................25
The Procedure of Estimating Multipliers of Economic Base Model ..................................... 27
Survey M ethod .................. .............. ......... ................................... 27
The Location Quotient Approach ........... ........ ......... ...... 28
Minimum Requirements........................................................ 31
Regression-based Approaches .......... ................... ................ 32

3 M E T H O D O L O G Y ....................................................................... 35

The Location Quotient Method ........... .............................35
Ordinary Least Squares (OLS) ........................................ ........ 36
Linear Unobserved Effects Panel Data Models .................................... 38

4 EXPLORATORY ECONOMIC BASE REGRESSION MODEL ......................................42

Initial Data Collection ............. ..... .... 42........... .........
Regression Analysis ............. ............................................. 44
C choosing a Scalar ............. ....... .. ......... ............................................ 81
R results and D iscu ssion .................................................................. 84

5 CONCLUSIONS ........................................ ..................... 102









where a is the ratio of net earnings to total personal income for each county and f is the scalar

of influence of the state level non-employment income. Note that the first scalar is different

across the counties, while the second scalar (without more detailed information) is assumed to be

constant across counties. Scaling back non-basic employment allows for more accurate sector-

specific multiplier estimates. Because of limitations in data sources, the scalars used in this study

were based on information for the closest year 1998.

Results and Discussion

The results of OLS estimation of the model illustrated in equation 4-1 using the modified

dependent variable from equation 4-2 for the year 1993 are shown Table 4-2. Except for the

primary sector, estimated coefficients are significant at 90% confidence level or higher. Basic

employment in wholesale trade is estimated to have the largest return to non-basic employment

in regional economy. One job increase in wholesale trade leads to approximately fourteen jobs in

non-basic sectors. Basic employment in construction; finance, insurance, and real estate; and

services also have large returns to non-basic employment, while basic employment of primary

sector industry and transportation have small returns to non-basic employment relative to other

sectors. For example, it is estimated that a one job increase in the primary sector leads to

approximately 2 jobs in non-basic sectors. The estimated coefficient associated with "soci"

implies that every additional $3,336 in social security benefits contributes roughly, on average, a

1 job increase in the non-basic sector of the regional economy.

The results of the OLS regression for the year 1997 are shown in Table 4-3. Note the

consistency in both the estimated values and the significance levels, with the exception of the

coefficient associated with primary sector industry (as it is now significant at the 95% confidence

level) and the increase in the coefficient associated with wholesale trade. The change in the









ET EN
MB = = 1+- (2-6)
EB EB

which implies that each unit change in basic employment will lead to an MB employment

EN
change in total employment and a change in non-basic employment. It is assumed that the
EB

EN
non-base/basic employment ratio remains the same and that the impact trend is embodied in
EB

the multiplierMB.

Practitioners frequently apply this multiplier to forecast the total employment impact of the

new establishment or the expansion of existing facilities. The multiplier is also used to predict

the total employment changes following an increase or decrease of export activity (and hence.

Describes the expansion or contraction potential of an employment change associated with a

change in exports, respectively). Large multiplier values indicate that export employment plays

a predominant role in the regional economy in terms of its ability to generate (or lose) jobs in the

non-basic sectorss. Once an estimate of an economic base multiplier is available, it is possible

to predict the future impact of a change in export-oriented employment on the total employment

within a regional economy (and the overall change in the employment in non-basic sectors).

More specifically, the change in total employment is equal to the multiplier times the change in

basic employment:

AET = MBAEB (2-7)

For example, if the current economy has 2,000 full-time workers employed in basic

(export-oriented) activities and 1,000 workers in non-basic basic the multiplier should be 1.5.

Then 50 new export jobs would create 25 non-basic jobs and 75 jobs in total. Equation (2-7)

supposes that the current MB can reflect the return of new basic employment change to total









BIOGRAPHICAL SKETCH

Fan Li was born in Xiangfan, China. She received a Bachelor of Engineering degree in

Information Engineering from Jinan University in 2007 and a master's degree in industrial

economics from Jinan University in 2009.

Her interest in economic geography brought her to the University of Florida to pursue a

second master's degree. She plans to continue her education at the University of Florida and

enter into a doctorate program in economics.










Hence, it is highly possible that the value of(xT x) is small but the value of "ut -u is

relatively large. If so, it would lead to a bias in estimation, something that would be especially

pronounced when the individual or sampling unit does not change much across time periods. To

illustrate the effectiveness of these methods, a two-period exploratory economic base model is

constructed and estimated using Florida employment data at the county level for the year 1993

and year 1997.









Viewed in this light, it is apparent that OLS regression cannot obtain "consistent" estimators. It

is necessary, therefore, to find an alternative method to OLS regression. Two commonly applied

alternatives are random-effects methods and fixed-effect methods. Both of these methods


assume thatE( The random effects method assumes thatE(ci x) = while the

fixed-effect method does not retain any such assumption.

Random effects methods belong to the family of "general least squares" (GLS) estimators;

techniques which utilize an unrestricted variance estimator as a transformer as a way to

overcome problems related to serial autocorrelation. The key issue is whether the random-

effects method should be applied in the panel data model given the assumption that

E(ul x,c) =0 (Baltagi 1995). In an economic base analysis using say two-years of data, c, can

be considered as part of various regional location factors such as the "weather" a random effect

that is largely uncorrelated with industrial employment or export orientation. As such, the

random-effects methods can be used within a PD context.

Fixed-effects methods apply an internal transformation, where

yt -y, = (x, -x )b +u, u (2-21)


y~ = T1 x, = T1 T x
and where "t T I tx and T is the number of total periods. Note that the

component c, is deleted in the transformation process. This transformation is necessary to

ensure that the error term satisfies the assumptions of OLS regression; allowing OLS to obtain

consistent estimators.


Fixed-effects methods do not assume that E(u, x, ) = Note, however, that the

transformation depends on the difference between equation from one period and the average.









factor would ultimately result in biased and unreasonable estimates for economic base

multipliers and the degree to which specific industries or sectors contribute to the overall

employment base. Social security benefits are another non-employment revenue source. Social

security payments bring in money, stimulate demand for goods and services, and contribute to

overall employment growth. Therefore, a critical assessment of how key economic sectors

expand a regional economy must also address the importance of transfer incomes and social

security benefits.

In the estimation of economic base multipliers, which reveal the extent to which a sector

contributes to the expansion of the regional economy at large, several modeling and estimation

techniques will be compared with controls put in place to lessen bias and/or overstatement of the

impact. The research objectives of this thesis are two-fold: (1) to develop a regional economic

base model that incorporates the transfer income factor; and (2) using data for more than one

time period, demonstrate the effectiveness of panel data regression as a techniques that sidesteps

some of the limitations imposed by classic regression in the estimation of economic base

multipliers which describe the extent to which employment in key economic sectors contribute to

a region's overall total employment (ceterisparibus).









As with the methods discussed earlier, the minimum requirements approach can only

estimate the average multiplier. The classification of the similar range regional economy makes

a significant influence on the accuracy of final result, so it is critical that care be taken to find

regional economies of similar size, though the same criticism can be levied as there is no reason

to believe that similar sized economies should have a similar industry mix. Some regions have

an inherent advantage based on location, natural resources, and/or other production advantages,

and even if similar in size will not necessarily share similar sector-specific production levels or

similar sector-specific demand. Researchers continue to discuss standards by which to evaluate

economies of similar size economy, but no agreement has been reach on exactly how to go about

justifying the minimum requirements benchmark. Nonetheless, tremendous strides have been

made in the estimation of multipliers using regression models of various kinds.

Regression-based Approaches

Regression analysis has been used effectively in the estimation of regional economic base

multipliers, yet is known that the type and quality of estimates depend on the specific regression

approach being applied regression model. The aggregate and disaggregate models represented

by equation (2-9), (2-11) and (2-13) can all be estimated using ordinary least square (OLS)

regression. This requires a survey and the gathering of a sample data, and then the use of OLS to

provide estimates of the coefficients of the model. Multipliers are then calculate from the

estimated coefficients using equations (2-10), (2-12) and (2-14). Typically, multiplier estimates

will be dramatically different from the method described above. Nevertheless, the regression-

based approaches have become quite popular in the economic base theory literature. The

regression-based approaches are also expandable, as illustrated by the seminal work of Mulligan

and Gibson (1987). Mulligan and Gibson emphasized the importance of including transfer

payments and have shown that the employment-derived estimates of economic base multipliers




















( I -tolmes' o
Escam.' a WialonWlshn1gon w a.J .-jeffetason -- H il- o a
r 'Cailho Leon MadisO.N-R a 9lIt d
By Wak-ula / swrrnneeCou i u Banrkc
ut j ayloria la et U n Clay Jon s
S Frarilin .frd .Jo
'- 4 l' r r/- ll, Gi hristA'ai.a ar. i
Dixie ".. Fl aglet,
Levy Marion -Iusr
Volusia
'Citris Lake ,
mter minor
Hernando I L
i Pasoo o os revard
*i ----b h Osceola
Polk
PRneilas isboFM h
SI ..... dian Riv r
SManatee Hardee Okeechobeet Luc
Hi gninds I Luce
S rasc.- DeSclo Martin
SCharlotte Glades ,,: -.
I ... Palm Beach j
Lee Hendry
Legend '- .
Collier Broward
Basic Employment
for Transportation in 1993 .
--0 70
70-180
180-1150 Monroe
1150-7700
7700 35570



0 62.5 125 250 Kilometers
I I I I i i a I N

A

Figure 4-13. Basic employment for transportation and public utilities sector of florida in different
years by county A)1993 B)1997






















1 -- -i
.ainp osao al a 'Jackson .
Esc saOa l .lrorWahnglor Gads de-JeffeIIson -. assalun
r-30sden^e ---N--,- rN s
I "- .q'o l Hanift)
ScCahot Leon, Mad or
Bay
Liberrj Wakulla Sa...lumb-aB -
T Unr~n Clay
Gulf "- TaYlorLafayette Bradfrd ST Johnl
IG l frs'n n.l \
G'&hrFs'AlachUa *pinarn -
D ixie J- agier

Levy Ma don
ii S __.Volusia\
Citrus Lahe
Surrnter SFneinroI
Hemandb' Orange
SPasco E-revaid
Osceola
Polk
Pinelas ^ h h1


Legend

Basic Employment

for Wholesale in 1993

0-225

225-1079

L | 1079-2742

2742-10413

10413 21806




0 62.5 125
II j i i I i i


In:llan R ,er
Manatee Hardee Oke Ihobeeu
'an highland t Lucie
arasola DeSo I Martin
Charlotte Glades

Lee Hendry alm Beach I


Colifer







Monroe


250 Kilometers A'
i- N


Figure 4-14. Basic employment for wholesale trade sector of Florida in different years by county
A)1993 B)1997


IH 1 I I I


\


I









The Procedure of Estimating Multipliers of Economic Base Model

In order to estimate a regional economic base multiplier, it is important to determine how

much total employment in a given region is basic employment and how much is non-basic

employment. There are numerous methods to arrive at numerical values or estimates for basic

and non-basic employment.

Survey Method

The Survey Method relies on a comprehensive survey of employers of all firms in a region

to identify approximately how much of each firm's income is obtained from external sales versus

local sales. The proportion of sales revenue from export sales is assumed to be equal to the

proportion of basic employment. Suppose a firm generates 30% of its revenues from exports

sales, compared to 70% revenues from local sales, and 150 "full-time equivalent" (FTE)

employees (that is, a number of employees that total to an equivalent of 150 full-time

employees). Under such a scenario, it is assumed that there are 45 basic-sector employees and

105 non-basic sector employees (or 30% basic and 70% non-basic employees). Note that

standardized employment data must be transformed into FTEs as employment is typically

divided into three types: full-time, part-time and seasonal workers. All employment data should

be converted to a full-time equivalent (FTE) employment scale (Gibson and Worden, 1981).

Part-time employment is converted to FTE employment by summing the hours each employee

works and then dividing by 40 (assuming a standard workweek of 5 days at eight hours per day

or 40 hours). The number of weeks that seasonal employees work must be summed and divided

by 52 to estimate its contribution to FTE employment. When the data are in FTE format,

formula (2-6) can be applied to estimate the economic base multiplier. As mentioned before, this

approach can only be sued to estimate an average multiplier. The survey method is a least

preferred method as it requires a comprehensive survey of all companies in a study region; so it









other non-employment related income and transfers which tend to bloat the non-basic sectors.

The choice of a scalar, while not obvious, is the focus of the next section.

Choosing a Scalar

If the regression model in previous section is applied directly, the estimation of sector-

specific coefficients would be very large. While the independent variables of the model may do

well in explaining employment related patterns, they lack in their ability to account for other

sources of personal or transfer payment income (e.g., investment income, dividends, interest and

rents, unemployment benefits, income from personal or corporate retirement accounts, money

spent by visitors to the state). Note that the dependent variable embodies both the impact of

employment income and non-employment income, and as such is typically thought of as

exaggerating the impact of basic employment. As such, there is a need to scale back the

estimates for non-basic employment to account for the impact of transfer payments.

Total personal income includes net earnings by place of residence; dividends, interest, and

rent; and personal current transfer receipts. The first part is from employment, while the second

and the third are from non-employment sources. Nontraditional sources of personal income and

transfers have become important in the economic base literature, especially in regions with high

population growth rates (Nelson, Beyers 1998); a feature that certainly applies to Sunshine State.

The ratio of net earnings to total personal income by county is applied as a scalar to the

dependent variable to account for the non-employment part of non-basic employment.

Numerical values for the scalar are presented in Table 4.1 for each of Florida's 67 counties. This

scalar alone, however, is not sufficient in capturing the entire effect of income transfers as there

are also inherent via multiplier effects associated with income transfers that must be account for.

Furthermore, the scalar only decreases the initial non-basic employment figures, and ignores the

influence of state level non-employment income sources. Compared with other states in the




















,'' ,- home i
SaW h Nassau
r____Ia y Suw amv tr:-o r -ilB k
3 I I _- "a i n o -n ^ o M ad i so n ar m | r
.Bayl Liberty Wakulla/ SuwvaneeColumr
Gulf .. T alrLalavett$ B ordClaSt. Johs
Frankii n chua ipr
SDixie Pu a
Lew ai Flglq.

Citrus Lake
Summer 41
Herna-rd,

Osceola
Pi2ne 82
Ir din Rrver
\Manatee Hardee Okee~:hobee t Lucl
-Hghlandst Lucie
D t Martin
Charloctte Glades
--- -L- Hendry

Legend
Collier
Non-Basic Employment "
for Wholesale in 1993
......................... 10 7 9 0

790-2220
2220-8260 ; onroe
8260 29670
29670 48200



0 62.5 125 250 Kilometers
I I I I I i I N

A

Figure 4-6. Non-basic employment for wholesale trade sector of Florida in different years by
county A)1993 B)1997










models are typically preferred in cases involving complex inter-regional economic inter-

dependencies. Unfortunately, there is no clear boundary between two kinds of methods in terms

of which would be preferred as one increases the geographic scale of the analysis (Mulligan,

2009). It is safe to say that for short-run analyses, a traditional economic base analysis should

suffice if one operates under the assumption that factors such as Consumption, Investment,

Government expenditures, and Imports remain fairly constant over the period in question.

Two Versions of Economic Base Model

Economic base models can be divided into two varieties: aggregate models and

disaggregate models. The distinction between them is that the aggregate model considers the

regional economy as a whole, while the disaggregate model identifies sectors or industries of

interest in a regional economy (Vias, Mulligan 1997). In additional to the model definition given

in equations (2-1) and (2-2), the following definitions are required for a disaggregate version of

the regional economic base model :

n
EB = EBI
1-1 (2-4a)
n
EN = Y Em
1-1 (2-4b)

ET =Y ET,
1 1 (2-4c)


Er, = EN, + EB, (2-5)


as defined for i = 1,..., n sectors, where the various relative industry-specific (i-th sector) figures

are summed to arrive at economy-wide figures for non-basic employment, basic employment,

and total employment.


































-50000


0
Residuals


I I I 1 1
-60000 -40000 -20000 0 20000
Residuals


50000


40000


Figure 4-19. Normality test for different regression A)Year 1993 OLS regression B)Year 1997
OLS regression C)Both year 1993 and 1997 OLS regression D) Both year 1993 and
1997 panel data regression


89









weak as it is asserted that different economic regions adhere to the same benchmark; that is, have

the same minimum demand for a goods and services from a given industry or sector. In short, it

assumes that sector demand, in a relative sense, is spatially invariable across sub-regions and is

similar to that of the larger benchmark economy. The choice of a benchmark economy can also

impact the accuracy of the estimates; and a question arises as to whether it is advisable to use a

regional or national economy as a benchmark for a sub-region; especially as that region may be

highly specialized in terms of its industry mix and/or have a comparative or competitive

advantage in production that allows it achieve say economies of scale or exploit its geographic

location. When a region's economy is compared with two benchmark economies, say at the state

and national levels, estimated basic and non-basic economic activity usually varies, and there is

no direct indicator for which estimate is more reflective of the region's economic position.

Unfortunately, no wildly-applied standard is available by which to evaluate the benchmarks.

Consider the urban economy of Miami, Florida a node that is recognized as a global city

because of its importance in several sectors including finance, commerce, media and

entertainment, and international trade. Miami is radically different in terms of its industry mix

than say the capital of Florida -- the city of Tallahassee, which focuses on education, government

administration, and agriculture. If the two cities' economies are compared with the state level,

they will share the same benchmark to distinguish basic and non-basic economic activity in all

sectors. Yet obviously, Miami has a higher local minimum demand for finance, insurance, real

estate, retail trade and service, while Tallahassee has higher demand for public administration.

This inherent flaw can lead to multiplier estimates that are exceedingly large and over-estimated.

As an alternative benchmark by which to compare these two very different regional economies,

one could utilize information on sector activity for northern Florida and southern Florida and









25 new non-basic jobs. Empirical evidence and common sense suggest that external demand of

two different industries will most likely produce dissimilar impacts in the regional economy.

Average economic base multipliers that depend on aggregate models do not fully express

the various influences of external demand on the local economy for goods and services produced

by different industries nor do they explain the different impacts of industry-specific export shifts

on a region's economic/industry mix. By contrast, average multiplier obtained from the

disaggregate models provide better overall explanatory ability in terms of their ability to estimate

industry-specific differences and change. Average economic base multipliers, however, cannot

describe or forecast the complex dynamics of regional economic change as they relate to specific

sectors for changes that may be occurring simultaneously, and therefore, it is necessary to turn to

"marginal economic base multipliers" as a way to express regional adjustment and related

impacts in association with a given change in a region's employment base.

The Marginal Multiplier of Economic Base Model

Assume that a direct linear relationship exists between export-oriented activity and locally-

oriented activities, then following:

EN = a + bEB (2-9)

where a is the autonomous component of non-basic employment and b is the marginal returns

of basic employment to non-basic employment. Hence, the marginal multiplier is:

aET
M =( )=l+b
aEB (2-10)

The major difference between multipliers MB and Mi is that the former represents the

average ratio while the latter one is marginal one. As a result, MB may be applied in static

economy, whereas Mi may be applied in a dynamic economic setting. The marginal multiplier

is more valuable as a forecasting tool for economic impacts in comparison to the average


















Holmes
S H r Jackson
"H W oashirIlon (I S rlNI





Levyu MbiaBak
Liberty Warlau s ke


_Sun b Clay'r
'"G,A l yee .Srj ,j -St Johns
,, Jrankl!n,-- -...G. .

Le' "' 1 1o.

Citrus Lakeuir
Hernando '
SPasoU


Legend

Non-Basic Employment
for Retail in 1993
S98- 3730
3730 9530
9530- 18290
18290 34600
S34600- 163785



0 62.5 125 250 Kilometers
I I I I I I I I


Indian River
Manatee Hardee Okee6hobeei
Highlands. St Lucle
DeSolo
S'/ Martin
Charlotte Glades
Hendry








Monro.e
C? 3 -^ '


A
N


Figure 4-7. Non-basic employment for retail trade sector of Florida in different years by county
A)1993 B)1997

















Jackson


NassaJ
-J./ Duval


Hendry


Legend


Basic Employment
for Construction in 1993 Miani-Dade
10-94
94 316
I I 316-545
545 1628
1628- 4158


0 62.5 125 250 Kilometers

A

Figure 4-11. Basic employment for construction sector of Florida in different years by county
A)1993 B)1997


I Manatee



'




















H, nolmes' ,-
f7 ,. -ljackson, r-i^.
Sti Okaloosaw cnWa g -_ Nasu
f, r L I alho ',-y Leon Madicri'n D ,
Bay 7 Wa kua Suwarr Colmb 'Baker
Lib:"rs ,v v W akulla .-J ,
ul TyIrLaaye Ur lo Clay Jis
'yGuN 4'l-- Lmye Bladf.rd St Jri s
FR nRaln i ri ,,,,,
SO Gi h.rt Alacha Putnam
Dixie lale r,
SLevy Marion I
V-olusia
Citrus \ Lake
S Summer Soemino
lemrand" Orange

(-^\ Osceola
Polk l
Ir.dran Rive
SManatee 'Hardee Okeechobees
H bgeland St. Lucis
Hgriland,
Sara r ta DeSucI I ..........
F&S-a Martin \
Charlotte Glades -
I L HeIndry Palm Beach
Lee He"dry

L Collier
Legend..
Basic Employment

for Wholesale in 1997
S 0-220
220 770 Monroe
770- 3650
3650 9730
9730 25800

0 62.5 125 250 Kilometers N



Figure 4-14. Continued
Figure 4-14. Continued









Krikelas, A.C. 1992. Why Regions Grow: A Review of Research on the Economic Base Model.
Economic Review of the Federal Reserve Bank ofAtlanta 1992:16-29.

Lego, B., Gebremedhin, T. and Cushing, B. 2000 A Multi-Sector Export Base Model of Long-
Run Regional Employment Growth. Agricultural and Resource Economics Review 29:192-
97.

Leichenko, R.M. 2000. Exports, Employment, and Production: A Causal Assessment of U.S.
States and Regions. Economic Geography 76:303-25.

Lesage, J. and Reed, D.J. 1989. The Dynamic Relationship between Export, Local, and Total
Area Employment. Regional Science and Urban Economics 19:615-36.

Loveridge, S. 2004. A Typology and Assessment of Multi-Sector Regional Economic Impact
Models. Regional Studies 38:305-17.

Malecki, E.J. 1991. Technology and Economic Development: The Dynamics ofLocal, Regional,
and National Change. New York: John Wiley and Sons.

Moore, C.L. 1975. A New Look at the Minimum Requirements Approach to Regional Economic
Analysis. Economic Geography 51:350-56.

Mulligan, G.F. and Gibson, L.J. 1984. Regression Estimates of Economic Base Multipliers for
Small Communities. Economic Geography 60: 225-37.

Mulligan, G.F. 2009. Industry-Specific Employment Multipliers in U.S. Nonmetropolitan
Economies. Studies in Regional Science 39:681-98.

Mulligan, G.F. 1987. Employment Multipliers and Functional Types of Communities: Effects of
Public Transfer Payments. Gi ,,ii thl and Change 18:1-11.

Mulligan, G. and Fik, T. 1994. Using Dummy Variables to Estimate Economic Base Multipliers.
The Professional Geographer. 46:368-78.

Nelson, P.B. and Beyers, W.B. 1998. Using Economic Base Models to Explain New Trends in
Rural Income. GI ni th and Change 29:295-318.

Nishiyama, Y. 1997. Exports Contribution to Economic Growth: Empirical Evidence for
California, Massachusetts, and Texas, Using Employment Data. Journal of Regional
Science 37:99-125.

North, D.C. 1955. Location Theory and Regional Economic Growth. The Journal ofPolitical
Economy 63:243-58.

Roberts, D. 2003. The Economic Base of Rural Areas: A SAM-Based Analysis of the Western
Isles, 1997. Environment and Planning A 3 5:95-111.









tend to be biased upward unless the transfer payments are directly specified in the regression

model. Mulligan and Fik (1994) extended previous research using the Arizona Community Data

set, and distinguished between multipliers for different types of communities by introducing

dummy variables into the regression models. The estimation results have also been shown to be

dependent on Meta-analysis (Vollet, Bousset 2002). In this approach, a host of variables are

used to reflect characteristics of study regions and account for variability in population size and

the magnitude of the employment base, as well as the physical size of the study regions; and in

particular, how the economic base is determined, how it is expressed, and which statistic method

is applied, and how it might impact the accuracy of the model based on a multitude of variables.

As mentioned previously, the fundamental assumptions of economic base theory are at odds with

the Keynesian equation, something that researchers have made great efforts to correct in time-

series econometric applications. One very influential work was the use of bi-variate vector

autoregression (VAR). In this approach, sector employment levels were analyzed in the state of

Ohio, and its economic base was derived and evaluated using Granger causality tests and

interpreted dynamic base multipliers by impulse-response functions (Lesage, Reed 1989).

Although Krikelas (1992) argued that the identifying restrictions required to derive multiplier

estimates become arbitrary and the estimation procedure becomes unstable as the number of

sectors included in a VAR model is increased. Nevertheless, additional research in this area

supports the vector auto-regression approach. An interregional trade model from which testable

parameter restrictions for economic base theory were shown to be equivalent when subject to

Granger causality tests, suggest that economic base theory holds strongly for the crudest

definition of the base, in a study where the data set included the states of California,

Massachusetts and Texas (Nishiyama 1997).


















SHolmes
S aJackson -Nassau I'
SE WaltonWlashe t Gn -Gadse .sdon .. i l
1Ealoun i BMado i4on.
y ,iberly Wa ulla Suw nneeClumb .k r
Gulf TayorUatayette ...o S. ohns
,Franklin -~ j h- -r "a
G ch" a Putnam
Dixie L
1 Levy Marlon

Citrus ke
Lemandor

Osceola

Irdlan Riv
Manatee Hardee Okeecnohobee ui
Highlands. $t i

Charlotte Glades
HeIndry


Legend

Total Employment in 1993
398 15000
S15000-38000
38000 74000 Mo n.oe
74000- 144000
144000 781000


Figure 4-1. Total employment of Florida in different years by county A) 1993 B)1997









Finally, I would like to express my special gratitude to my family in China my loving

parents and friends for all their encouragement and support; something that I feel is very

strong, even though I am thousands of miles away from my home.









to the proportion of employment associated with that sector as reported for the Florida economy

at large, e. The LQ method is then used to estimate the amount of basic and non-basic
e
employment for each sector and each county. These estimates were then used in a disaggregated

model to estimate sector-specific economic base multipliers using various regression-based

models. Total employment by county of year 1993 and 1997 show in Figure 4-1. Estimated

non-basic employment for each sector by county of year 1993 and 1997 show from Figure 4-2

to Figure 4-9. Estimated basic employment for each sector by county of year 1993 and 1997

show from Figure 4-10 to Figure 4-17.

According to Figure 4-1a and Figure 4-1b, Orlando, Jaxsonville, Tampa, Miami,

St.Petersburg, West Beach and Boca Raton areas have relative large total employment compared

with other regions. Non-basic employment and basic employment for sector from Figure 4-2 to

Figure 4-17 base on the location quotient method. From Figure 4-2 to Figure 4-9 show that non-

basic employment of different sectors have disimilar spatial distribution by county, which is

probably not consistent with the spatial distribution of total employment. In Figure 4-2A,

Broward, Miami-Dade, Palm Beach, Orange, Pinellas,Hillsborough, Brevard and Duval counties

have largest total employment in 1993, however, on on hand, Brevard has medium non-basic

transportation employment in 1993 from Figure 4-5A, on the other hand, largest non-baic series

employment in 1993 from Figure 4-9A is consistent with total employment.Therefore, the change

of non-basice employment of different sectors should have disimilar return to total employment.

Some of the same sectors of year 1993 and 1997 shows disimilar employment distribution

by county, which implies that employment distribution by county of single year cannot

characterise that in long term. Therefore, panel data regression of mutliple years should instead

regression of single year to estimate the steady trend ofregionl employment.





















Holrrme s \
',. Jackson r-
i -' .' aloosaWal nW -n"h'-n u sonM -- -rJ
Ch' lhourl MaLso n Had I C/ Duval

a Liberty WakulIa Su, r ,ee I-c..e al
f-- Tayi:r a Union Clay
GuO Lf't-/ -- L l:'t" or, S Jo0ns
S rarklir -"
SDixie Flagle
Levy Marion

CilruI Lake
SumTer d
Hernandbr.
/ Pas co .. ---........
jPasco.j


Legend

Basic Employment

for Services in 1993
0 200

S200 760

760 3610

3610-9350

9350-30403


m_ _lr)lari River
Hardee Oke hobee L
Highland Luc
-Desoto I --- ----

Charlo'Te Glares "

> Lee Hendry Pam Beach
I __.
Collier Broward


Miami-Dade)


Monroe


0 62.5 125 250 Kilometers
I I I I I I I I I N

A

Figure 4-17. Basic employment for services sector of Florida in different years by county
A)1993 B)1997

















koineakson
Sant RosaOacks
Esambis W shn- Gadsde son Nassau
a- hcun Md. MISHm m il1- Duval
L-er-ty" ~ k ua SuwarrieeColunmbiaBake
"^ H ..Libert'^ Ula lr U'nl ) l-
Gulf -, lorLafayette. or JohUs











Marialee Haraee Oke hobee
DHe Put hlandam tu





SarastaDeSoto "





Legend
Basic Employment

for Retail in 1997
0 340
340 1380
1380 2440
2440 4790
4790-7245

0 62.5 125 250 Kilometers N



Figure 4-15. Continued
Figure 4-15. Continued









less expensive in comparison to input-output models, a feature that explains why they are so

popular in the analysis of regional economic growth (Malecki 1991).

According to economic base framework, total economic activity, ET, is assumed to be

divided into non-basic economic activity, EN and basic economic activity EB, where EN is a

function of EB More formally,

ET= EN + EB (2-1)
EN = f(E) (2-2)

Note the narrow focus on exports and basic economic sectors as the engine of local

economy makes economic base theory lack the complexity to offer an adequate framework for

analyzing regional and sub-regional economic issues. The focus on exports only consider the

demand side, and excludes important supply-side factors (Blumenfeld 1955). Hence, the

assumption that basic economic activity is the predominant driver of regional economic change

minimizes the contribution that non-basic economic activity makes to regional growth and

development (Tiebout 1956). In an effort to expand the economic base model, other factors such

as transfer payments have been added (Mulligan 1987), as well as internal population growth

(Leichenko 2000). Discussions of the counter-veiling force of "leakage" lost income due to

purchases made outside a regional economy by consumers who live in that region (Fik, 2000).

Many researches argued that economic base models are limited in their ability to forecast

long-run regional economic impacts as they focus on short-run expansion effects tied to export-

oriented income generation, while ignoring other important growth factors. Hildebrand and

Mace (1950) apply the familiar Keynesian equation:

Y = C + I + G + (X -M) (2-3)
































2010 Fan Li









where Y denotes total regional income, which is divided into four parts: C, consumption; I,

investment; G, government expenditures; and (X -M) exports minus imports. The Keynesian

equation suggests that regional export activity is only one of four other factors C, I, G and M

that explain regional growth and development; and hence, exports are not the only thing driving

economic growth. When relative factors C, I, G and Mare held constant, economic base theory

and the Keynesian equation are congruent. Note that the short-run economy always assumes that

those four factors remain constant, so economic base models can be applied rather effectively to

estimate short-run changes, but should not be used to produce long-run forecasts. The additional

factors considered by Keynesian equation are related to both demand and supply. Thus, a more

detailed regional economic model would need to incorporate supply-side factors to anticipate

long-term changes; thus, making it non-traditional in the sense that it would also rely on su-pply

side patterns and trends. Theoretically, the most important determinate of a region's long-run

development is the ability to attract capital and labor from outside into region. Such supply

accumulation would, in turn, stimulate export-oriented production sectors, thereby augmenting

export activity and bringing in exported-related revenues and income (North 1955). All in all,

the connection between the economic base model and the Keynesian equation has to do with

differences in social accounting techniques and the degree to which each model may be used

depending on the time horizon (short run versus long run), the geographic scale (the areal

coverage), and the type of impact assessment or forecast required. Roberts (2003) applied an

accounting model to quantify the relative importance of traditional and non-traditional elements

of economic base theory for local and regional economies in rural areas.

Due to its inherent simplicity of theoretical foundations, economic base theory,

nevertheless, does well to explain regional economic growth over the short-run. Input-output









demand and were labeled as basic to the survival of the region, with dollars generated from

export activities. By contrast, there were industries that were region serving in the sense that

they were associated with the sale of goods and services from one company in the region to

others inside the region. Hence, city-forming activities serviced the demand that was internal to

the region (demand generated by the local population), with money that was circulated and re-

circulated within the regions. To sum, basic or export-oriented activities were regarded as the

city forming and a direct result of external demand, while city-serving activities were there to

meet a region's internal demand for non-exportable goods and services. The latter set of

activities and associated employment were in place for the maintenance and well-being of the

people inside the region to satisfying its internal demand. The distinction between city-

forming and city serving activities is important in the sense that the regional employment base of

a region could then be broken down into basic employment (employment associated with export-

oriented activity) and non-basic employment (employment associated with meeting internal

demand). There is a fundamental assumption, central to the notion of the duality of regional

economic activity, that non-basic economic activity relies on basic economic activity (Andrews

1953) and that basic employment supports employment in non-basic sectors. This modeling

framework focuses on regional export activity as the primary determinant of local economy

growth. The theory also assumes that the money-generated by basic industry expands as it

circulates and re-circulates, creating non-basic jobs as money and income change hands an

effect known as the multiplier effect (Fik, 2000). Note that the economic base models and input-

output models are different. While both modeling approaches can estimate the impact of a

change in employment, the former yields an aggregate overall impact, while the latter yields

sector-specific impacts (Billings 1969). Moreover, economic base models are computationally



















*Holmes
S- sa loIosa IA lton"asin'on -an- Na-sau
Es iem ia I it Gase E or,
S .. 1 -- Calho2n L .
LietyBay Walkulla 2 SuwanneeCOlumbiSBaker
Bay. al Liberty^ W I_ o a b '^
SGu -f Ta ylorLafaye, t ,a or'Clay SI Johns
.F ran li -
"-- -/ r tAlaPchua 'IPutnam--
Dixie -Flagler
SLevy Marion
S |Le Volusla
Citrus Lake
nMter MiO
Hernandr-
Pasco
Isceola
PiMartin
Indian Ri
Marate Hardee OketchobeeL Luci

Charlotte G




Legend
Non-Basic Employment

for Services in 1997
156-7830
7830-28910 r Monroe
I 128910- 57590
S57590-218830
S218830-305425 /


0 62.5 125 250 Kilometers



Figure 4-9. Continued
Figure 4-9. Continued














variables very well. However, figure 4-21D shows that fitted value of non-basic employment is



less than actual value of non-basic employment in the middle of plots.


o
0
0
0
0



o
o
o
o-
CO






0)



0
C0




u-










0 -
Co
C)
C
O

O



O
O
O






0L
O







0)
O


O




C _
O
O
O
O
O
O

CO

U)
a)
CO


O


o
Co





O
O
O
0)

0)


0 -


S


0 100000 200000 300000
NEB


Figure 4-21. Plots of fitted versus actual non-basic employment values for different regression

A)Year 1993 OLS regression B)Year 1997 OLS regression C)Both year 1993 and

1997 OLS regression D) Both year 1993 and 1997 panel data regression


0 100000 200000 300000
NEB


400000

A


400000

B









Time-series procedures have also been applied in the estimation of economic base

multipliers, with moderate success. Separating the employment base into at least two sectors, a

shortcut method has been developed that utilizes both a structural adjustment procedure and a

time-adjustment procedure (Was and Mulligan, 1999; Mulligan, 2009). The first procedure

recognizes that each industry has a sector-specific assignment to the local employment base that

relies not only on the size of the regional economy but also its industrial mix and the degree of

industry concentration. The second procedure applies a time variable to control for dramatic

growth or depression of an economy in a particular year, using a variant of the VAR method.

These regression-based approaches represent the current trend in economic base research. They

represent flexible and expandable strategies for the estimation of coefficients and marginal

economic base multipliers and are methodologies that have potential forecasting abilities.

Regression-based procedures also offer the capacity to added more explanatory variables, such

as supply-side factors, regional characteristics, transfer payments and spillover incomes, leakage,

structural adjustment considerations, and time.









obtain far more reasonable results as there would be less of a contrast in economic activity levels

across various sectors. Another disadvantage of the LQ method is that it fails to estimate the

marginal economic base multiplier. Recently, there has been an attempt to improve on

traditional location quotients estimates through the specification of the dynamic location quotient

(DLQ) method. This approach was developed to decompose regional employment into base and

non-basic components across multiple sectors using spatial and temporal data (Lego,

Gebremedhin & Cushing 2000).

Minimum Requirements

The minimum requirements (MR) method was popularized in 1960s and 1970s (Moore

1975). It is similar to the location quotient method in terms of distinguish sector activity by

comparing activity levels in the study area to a predetermined benchmark. In the minimum

requirements method, groups of regional economies or sub-regions with approximately similar

sizes are compared, and a sector's minimum requirement is assumed to be the minimum level of

activity in those regions. It is assumed that this minimum level of employment observed within

the group is a level that satisfies internal demand. Furthermore, it is assumed that sub-regions

with greater activity than the observed minimum for a given economic sector must be exporting

goods or services. Excess sector employment (above the sector minimum), within a given size

class, is assumed to be basic employment. The minimum local benchmark for a given economic

sector, for a given size class, represents minimum non-basic employment E,(mn),. Once these

values are determined for all sectors, the economic base multiplier is found from the following

expression:


ET,
M-= "0
n
Y (ET- EN(mmn),)
-o0 (2-18)



















-, --- _--- --
I. / -ackso' il--_ _.. .
S1n s OkaIoOsaWao nr acon sder Msn -

Cral in C b
V Lioerrt NaKulla Su' iareree eCOI' n R e*
'Gulf TylcrLar9yette g aford ySt Johnc





\i a Ia'ae Rkeve




1 Higlnlrnds,.
rrnl DeSoo Martin


LevLegend











Non-Basic Employment
for Wholesale in 1997
Hernando
75016Paso80

















Osc1680-eola4960
Indian Rive

















H 4960 92500
I 92500 502420
0 625 125 250 Kilometers Ok


Figure 4-6. ContinuedMart
Charlotte Glades
Hendry


Legend
Non-Basic Employment

for Wholesale in 1997
100- 750
750 1680Mn
1680-4960
4960 92500
92500 502420

0 62.5 125 250 Kilometers





Figure 4-6. Continued









The other obvious change from year 1993 to 1997 is that the sign of coefficient associated

with the transportation and public utility sector becomes negative (literally interpreted as

meaning a contraction of the non-basic sector for each additional job added in that sector). This

counterintuitive and illogical result means that a 1 job increase in the transportation and public

utility sector would lead to a decrease of two jobs overall in the regional economy. The results

for 1993 and 1997 show, nonetheless, that wholesale trade has the largest multiplier effect on

non-basic employment; something that is consistent in the estimation of both models. This result

is no surprise as wholesale trade represents an important industry as Florida maintains its

position as a key player in the global economy with its vibrant international shipping ports

located in Miami, Tampa, and Jacksonville. Employment in construction; finance, insurance,

and real estate; and services also have large return to non-basic employment. The estimated

coefficient for "soci" means that for every $1,896 dollar increase in social security benefits, one

additional non-basic job is added to the regional economy.

The regression analysis using the single-year samples cannot capture trends or the

adjustment of regional economies accurately, as the results are easily influenced by sudden

shocks or discontinuous changes that occur from one year to the next. This could explain why

the estimated coefficient associated with the primary industry in 1993 is not significant at even a

80% confidence level, even though it is represents one of the most important sectors in the state

of Florida (given that agriculture and mining activities generate a fairly large percentage of

income). The variability in single-year estimation can be overcome by applying a regression

platform that utilizes information for more than just one year. Table 4-4 lists the result of

ordinary least regression estimation for the years 1993 and 1997 combined. In this analysis, the

two sample years were combined and the model regressed without any classification. No









surprise that the estimated coefficients are located between the two single year estimates.

Nevertheless, the results are suspect and a test for exogeneity is necessary. Table 4-5 highlights

the results of exogeneity test on the error terms of the combined OLS model. The test depends

on the regression of error terms from the 1997 model (uhat) on error terms from the 1993 model

(uhat 1). The test results suggest that the coefficient is significant at 95% confident level, and

therefore, panel data (PD) regression should be applied instead of OLS regression on the two

year samples.

Table 4-4. Ordinary least regression estimation result of year 1993 and 1997
Coefficient Std.Err t-test P>|t| 95% Conf. Interval
prim 5.282 2.611 2.02 0.045 .114 10.449
cons 5.189 1.851 2.80 0.006 1.526 8.853
manu 2.911*** 0.750 3.88 0.000 -1.426 4.395
tran 0.082*** 0.676 0.12 0.903 -1.256 1.421
whol 14.850*** 1.078 13.77 0.000 12.716 16.985
reta 3.806*** 0.993 3.81 0.000 1.828 5.784
fina 4.767*** 0.462 10.32 0.000 3.853 5.681
serv 5.984*** 0.315 19.00 0.000 5.360 6.607
soci 0.205*** 0.040 5.10 0.000 0.126 0.285
constant -4881.392 2110.40 -2.31 0.022 -9058.46 -704.316
Note: *** indicates significant at the 0.01 level, ** indicates significant at the 0.05 level, indicates significant at the 0.1
level. Adjust R2=0.946

Table 4-5. Test result of exogeneity of error term
uhat Coefficient Std.Err t-test P>|t|
uhat 1 0.563 0.101 5.59 0.000
constant -3365.202 3544.38 -0.9 0.346
The results for the panel data regression are shown in Table 4-6. Note that the estimated

coefficients for the primary and construction sectors are significant at 80% confidence level, and

the coefficients of all other industries are significant at 95% confidence level. Once again, the

largest estimated coefficient is for wholesale trade, which implies that one additional job in that

sector of wholesale trade will lead to approximately 11 additional non-basic jobs in the regional

economy at large. Retail trade, the service sector, and manufacturing are also important sectors,



















Holmes
1acvr ^ on -
Es6 oa 08 0 1 altonVVashngton -.dsdeeon
Jefielsor I-rnill
1alho" "Leon Medison.Ham il U
iBy ,Lbert ula SiJwanneeColumoaBaker
Gulf T 'Lata Joh
-rar ,kinr ,
Gi chr',. purnam-
Diie t6L .- 'Ftag
Levy Marion

CiTrus
_sjrnth,
hernando
Pasco


Pine


Legend


Non-Basic Employment
for Manufacturing in 1993
20-1330
1330-3170
S3170-6750
6750 -14360
14360-77530


0 62.5 125
I Ia a I I I


Indian River
Hardee Okeecrnobee
Hlghlands
p Martin
Charlotte Glades
L.Le Hendry

Collier



/oY

i 'Monroe /
"""i- -^


A
N


250 Kilometers
a I


Figure 4-4. Non-basic employment for manufacturing sector of Florida in different years by
county A)1993 B)1997









the panel data approach for these years, though validation of the multiplier estimates represents

the next step in the scientific process. Further research on this topic could also consider

expanding the analysis to include more than two years to improve accuracy in the estimation of

sector-specific multipliers. Panel data regression could also be used to estimate economic base

multipliers for other states in the Southeast or the U.S. (and the results compared to estimates for

Florida). While this study applied transfer income related scalars and social security benefits as

a surrogate independent variable to weaken the influence of non-employment revenue sources,

further steps could be taken to scale down non-basic employment figures in places were there is

a secondary or overlapping multiplier effect associated with the indirect expansion related to

non-employment income (as might be in the case of Florida). Further inquires could also look at

various data transformations and alternative functional forms for the regression model. The

panel regression model could be further expanded to incorporate spatial or geographical effects

as suggested by Hanink (2007), who estimated multipliers as elasticities using regression

analyses to uncover the of the associations between retail trade earnings by place of work and

total income by place of residence across New England counties in 2002. As an extension,

further modeling efforts could consider spill-over and spill-inward effects across political

jurisdictions that are both part of the study region and part of the regions) juxtaposed or adjacent

to the study region. Spatial autocorrelation problem of panel data model with dummy variable

"coastal urban" should be addressed. Lastly, a larger battery of regression diagnostics could be

performed to more fully inspect the error terms and to test model assumptions.










According to the normality tests figures, error terms of both OLS and panel data regression

show normally distributed. Peak density of error terms for both single year regression is located

in residual with value above 0, while peak density of error terms of panel data regression is

located in residual with value more closed to 0, so error terms of panel data regression show

more normally distributed.

Residual plots( fitted values versus residuals) for year 1993 OLS regression, year 1997 OLS

regression, both year 1993 and year 1997 OLS regression and panel data regression are showed

in Figure 4-20a, 4-20b, 4-20c and 4-20d respectively. Residual plots are applied to test

homoscedasticity assumption of regression.



o

-o
0




LO











oC


0 100000 200000 300000 400000

Fitted values
A

Figure 4-20. Residual plots for different regression A)Year 1993 OLS regression B)Year 1997
OLS regression C)Both year 1993 and 1997 OLS regression D) Both year 1993 and

1997 panel data regression
0-*









locate above the line of"fitted value equal actual value", while others plots locate below the line,

which implies that estimation problem during range 100000 to 300000 is solved effectively.

Therefore, panel data model obtains better "goodness of fit" after dummy variable is added.


* **


100000 200000 300000
NBE


400000


Figure 4-23. Plots of fitted versus actual non-basic employment values for both year 1993 and
year 1997 panel data regression with dummy variable

In order to further compare the estimation results of two panel data models, spatial

autocorrelation of error terms are considered. According to Moran's I statistic test applying

"inverse distance" as conceptualized spatial relation, the spatial pattern of error terms of original

panel data model is random, while the spatial pattern of error terms of panel data model with

dummy variable is clustered in 10% significant level. Therefore, on one hand, the panel data

model with dummy variable "urban" has better estimation compared with original panel data

model, on the other hand, the panel data model with dummy variable suffer from spatial









Linear Unobserved Effects Panel Data Models

This section highlights the advantages and problems of using "panel data", and examines

two common methods for estimating linear unobserved effects in panel data (PD) models.

Before proceeding to review the panel data modeling approach, a short review of various

definitions as they pertain to regression data is necessary to panel data in its proper perspective.

A data set containing observations on a single phenomenon observed over multiple time

periods is called a "time series". A data set containing observations on multiple phenomena

observed at a single point in time is called "cross-sectional" (where each individual, sampling

unit, data point or polygon is observed only once). A data set containing observations on

multiple phenomena observed over multiple time periods is called "panel data" (which implies

that each individual, sampling unit, data point or polygon is observed in more than one time

period). Panel data analyses do not require that the time period in which different individuals or

sampling units are observed are exactly the same.

The major factor that distinguishes time-series data and cross-sectional data from panel data

is that both time-series and cross-sectional data are one-dimensional, while panel data is two-

dimensional (given that it is both spatial and temporal, and unrestricted in the temporal domain).

Time-series data suffers from its lack of spatial sensitivity and requires that observations be

equally spaced in time. Cross-sectional data analyses do not adequately capture the dynamics of

inter-relationships amongst variables as they occur across time and largely ignore the possibility

of consistency of phenomenon across the temporal domain. Cross-sectional data for a single

time period cannot adequately capture the aggregate effects inherent amongst variables or

phenomena and the inter-relationships that spans across multiple time periods. Panel data offers

an alternative that is widely used in economic analysis and can be viewed as offering information

that is spatial-temporal.









CHAPTER 2
LITERATURE REVIEW

Economic base theory remains a respected field of study in geography and regional

economics. Economic base analysis offers an inexpensive yet reasonably accurate method to

assess small regional economic and employment impacts. The objective of economic base

analysis is to calculate regional multipliers that describe the extent to which employment or

income will grow as a function of new jobs added to a regional economy. The following section

provides a brief literature review of economic base theory, including related assumptions,

applicability and scale, and data requirements. Aggregate and disaggregate models are

discussed, as well as the derivation of multiplies; focusing on sector-specific application and

distinguishing between average multipliers and impact multipliers. The overview will highlight

the four most-commonly used methods of estimating economic base multipliers; namely, the

survey method, the location-quotient (LQ) method, minimum requirement (MR) method, and

regression-based approaches. It is noted that regression-based models tend to produce impact

multipliers, while the other methods are used to estimate average multipliers.

Economic Base Theory

The history of the economics base concept dates back to the early 1900s. Early regional

economic analysts observed the duality of urban and regional economic activity, which includes

city-forming activities (characterized as the basic sector) and city-serving activities

(characterized as the non-basic sector) (Hewings 1985). The city-forming activities were said to

provide the reasons for the city's existence; typically associated with export-oriented industries

that provided goods and services for markets both inside and outside the region. As these

industries satisfied demand outside the region, they were responsible for generating income from

sales related to exports. As such, these industries were viewed as responding to exogenous

















s HolmesJacon
ESa a Okalsa ( Nassau
Es .a' .. ..Wa.llon.VVshnrgion Ga-dsdew-l s S on ihon' .
-- nCaihoU h o adisooHan l bn kj 4
Bay SuwanneeCclumb'dBaE
:Lloeir,"y' au Tulla -I io i
Gulfjln ,_ TaylorLafe Union Clayt. Johns
Gulf L -. 1,- rao.rd .St. Johns
SDii", L 1' ach u a utnmi -

l"'^^
Levy Marion
Cilrus Lake Z
Sumtr -
Hernanso
Pasco


-In Rive
, Manatee Harreel Okeechobes Le
-Highland St Luce
S DeSoto i Martin
Chadrotte Ga[ de
Hendry


Legend
Non-Basic Employment
for Retail in 1997
91-4410
4410- 12950
12950- 28190
28190 68930
68930 166538

0 62.5 125 250 Kilometers
II I I I I I


Monroe
0 ^ r^


A
N


Figure 4-7. Continued









primary sector coefficient implies that a 1 job increase in the basic sector of Florida's primary

industries leads to an estimated 9 job increase in non-basic sector employment. Note that this

coefficient is radically different form that which was estimated for 1993, which showed only a 2-

to-1 increase in non-basic employment for an additional basicjob added to the economy. The

results also suggest the volatility that exists in year-to-year estimates. Note the jump in the

estimated coefficient for whole trade, from 12 to approximately 17. All in all, OLS estimation is

suspect as the sector-specific multipliers appear to remain somewhat exaggerated in spite of the

use of the scaling factors applied to the dependent variable.

Table 4-2. Ordinary least regression estimation result of years 1993
Coefficient Std.Err t- P>|t| 95% Conf. Interval
test


prim 2.253 3.534 0.64 0.526
cons 8.085** 3.122 2.59 0.012
manu 4.243*** 1.015 4.18 0.000
tran 1.901** 0.882 2.16 0.035
whol 12.076*** 1.400 8.62 0.000
reta 3.182** 1.541 2.06 0.044
fina 5.399*** 0.735 7.34 0.000
serv 5.955*** 0.500 11.90 0.000
soci 0.237*** 0.062 3.85 0.000
constant -5704.684* 2732.42 -1.72 0.091
Table 4-3. Ordinary least regression estimation result of years 1
Coefficient Std.Err t- P>|t|
test
prim 9.621** 3.736 2.58 0.013
cons 4.184* 2.334 1.79 0.078
manu 1.593** 1.092 1.46 0.150
tran -1.747** 1.028 -1.70 0.095
whol 17.513*** 1.635 10.71 0.000
reta 4.151*** 1.345 3.09 0.003
fina 4.691*** 0.595 7.88 0.000
serv 6.295*** 0.412 15.28 0.000


soci
constant


0.135**


-4.824
1.835
2.210
0.135
9.272
0.095
3.926
4.953
0.113
-10176
993
95% Conf.

2.141
-.490
-.595
-3.806
14.238
1.458
3.450
5.470


0.058 2.34 0.023 0.020


-5011.103 3157.99 -1.59 0.118 -11334 1312.663


Note: *** indicates significant at the 0.01 level, ** indicates significant at the 0.05 level, indicates
significant at the 0.1 level. Adjust R2=0.948


9.330
14.336
6.276
3.667
14.880
6.269
6.871
6.957
0.360
766.907


Interval

17.102
8.858
3.780
0.312
20.788
6.844
5.883
7.120
0.250









relocated (permanently or seasonally) into the state. The influx of the elderly and retired brings a

large number of non-employment incomes to Florida in the form of transfer payments tied to

retirement accounts, social security, and interest, dividends and rents paid on investments. Non-

employment income in the form of transfer payments has had an enormous historical and

geographic impact on Florida's regional economy. This impact sometimes clouds the nature of

how sector-specific industries affect total employment in any given sub-region. The nature of the

relationship between industrial employment and regional economic development in Florida is

investigated in this study, with acknowledgment that that transfer payments play a critical role in

influencing employment patterns. Hence, analysis of employment patterns and the identification

of a region's economic base is understood as something that can be affected by transfer income.

Adjusting for transfer income, this study will focus on the estimation of economic base

multipliers coefficients that embody the expansion effects of key sectors or industries that

contribute greatly to the state's overall employment. More specifically, this study will seek to

compare several different methods and techniques for estimating multipliers and the inter-

relationships between the state's economic sectors using county level data.

Background

Warm weather and hundreds of miles of beaches attract about 60 million visitors to the

Sunshine State every year, so tourism replaces agriculture to makes up the largest economic

sector in the state economy. As such, much of Florida's economy is tied to retail trade and

services. The second largest industry is agriculture. Citrus fruit, especially oranges, and various

winter vegetables play major roles in supporting the regional economy. Florida produces the

majority of citrus fruit grown in the U.S.; for example, in 2006, 67% of all citrus, 74% of

oranges, 58% of tangerines, and 54% of grapefruit came from growers in Florida. Hence,

employment activities in agricultural sectors are known to affect regional growth directly. The









This study introduced economic base theory and methods used to distinguish between basic

and non-basic employment in a regional economy. It also gave an overview of the mechanics of

estimating employment multipliers using traditional and contemporary approaches. Discussion

focused on estimation techniques, assumptions, and the shortcomings of the traditional OLS

approach to estimating economic base multipliers. In addition, this study discussed the

difference between aggregate and disaggregate modeling. It also distinguished between average

multipliers and impact multipliers. Four commonly used methods of estimating economic base

multipliers were reviewed: the survey method, the location-quotient method, the minimum

requirements method, and linear regression-based alternatives. While the survey method is a

preferred approach, in that it is based on a comprehensive survey of all firms in all industries, it

is viewed as problematic and costly for large regions. The location-quotient (LQ) method and

minimum requirements methods, while easier in terms of data collection requirements, depend

on the choice of a benchmark (whose accuracy can be debated on several levels). The

regression-based approach is the most popular method by far, largely due to its accuracy and

forecasting ability. The largest difference between the regression-based approach and the other

three widely used methods is that regression approach allows for estimation of sector-specific

impact multipliers, while the other three index-driven approaches only produce average

multipliers.

In the methodology chapter, this study provided a comparison of two regression methods

that are used in the estimation of sector-specific multipliers namely, original least square

regression and panel data regression. Non-basic employment figures were initially obtained

using the LQ method at the county level, with state-level employment figures as a benchmark.

These estimates were adjusted to account for the overstatement problem that is commonly









associated with the LQ method. Furthermore, this thesis offered a unique attempt as scaling

down estimates of non-basic employment (through the use of two scalars). This was done to

reduce the problem of overstating multipliers when non-basic employment levels are overstated

due to the effects of non-employment income or transfer payments. The two scalars, used to

adjust non-basic employment figures, were defined as (1) the ratio of net earnings to total

personal income of each county within the state, and (2) the ratio of the percentage of net

earnings to total personal income for Florida versus that of the U.S. at large. Non-basic

employment figures, as the dependent variable, were simply multiplied by the product of these

two scalars. The adjusted non-basic employment figures were then used as dependent variables

to estimate sector-specific multipliers. Apart from LQ-generated basic employment figures that

were calculated for each industry/sector (the explanatory variables of the model), social security

benefits were added as an additional surrogate variable on the right-hand side of the regression

equation given the state's heavy reliance on this income source. These corrections were

necessary to account for the various factors related to non-employment income that typically

leads to overestimation of both non-basic employment and sector-specific multipliers.

It was shown that the panel data regression model for a two years sample yielded reasonable

more sensible results than single-year estimation using OLS regression. OLS residuals were also

shown to be suspect. Exogeneity tests on the residuals suggested that it was necessary to apply

panel data regression rather in the estimation of the disaggregated economic base model. It was

concluded that panel data regression offered a more even-handed and consistent estimation of

multipliers than those produced by OLS regression.

This study focused exclusively on county level employment patterns in the state of Florida

for the years 1993 and 1997. Reasonable economic base multiplier estimates were obtained using































To my beloved husband, Wei









ACKNOWLEDGMENTS

I would like to express my most sincere appreciation to Dr. Timothy Fik, Dr. Grant I.

Thrall and Dr. Youliang Qiu, members of my advisory committee, for their academic support

and guidance through the completion of this study. I would like to give my special gratitude to

my advisor Dr. Timothy Fik, who is always ready, available and patient to help with academic

problems and discussion. I am also grateful for the help and encouragement from other

committee members, including Dr. Grant Thrall and Dr. Youliang Qiu. They provide insight,

suggestions and valuable feedback to my thesis.

I am thankful to all members of the Geography Department, faculty, staff and students for

creating a friendly atmosphere that makes every day of study so enjoyable. As an international

student, all members make me feel as if I am living and working in my home country. I also

express my special thanks to Dr. Peter Waylen and Dr. Youliang Qiu for their support and help

to my computer lab job. My dear friend, Anna Maria Szynizewska, work with me together to

find the data source. Yang Yang helps me with regression programming.

I also want to thank Dr. Chunrong Ai from Department of Economics and Dr. Alfonso

Flores-Lagunes from Department of Food and Resources Economics. Their Econometric

courses enable me to apply the regression models used in my research.

There are no words that could describe my gratitude to my husband Wei. If it wasn't him

to encourage me to apply for Geography Department, I wouldn't pursue a graduate study at such

an excellent American university. After I began my study in University of Florida, it is his

immense love and unconditional supports that make me get accustomed to life and study abroad.

Mostly I would like to thanks to his help in my thesis writing. He teaches me how to correct

thesis format that saved me many months of work on this thesis.









LIST OF REFERENCES

Andrews, R.B. 1953.Mechanics of the Urban Economic Base: A Classification of Base Types.
Land Economics 29:343-50.

Baltagi, B.H. 1995. Econometric Analysis of PanelData. New York: John Wiley and Sons.

Billings, R.B. 1969. The Mathematical Identity of the Multipliers Derived from the Economic
Base Model and the Input-Output Model. Journal ofRegional Science 9:471-73.

Blumenfeld, H. 1955. The Economic Base of the Metropolis: Critical Remarks on the "Basic -
Nonbasic" Concept. Journal of the American Planning Association 21:114-32.

Chiang, S. 2009. Location Quotient and Trade. The Annals ofRegional Science 43:399-414.

Fik, T.J., Amey, R.G, and Malecki, E.J. 1991. Changing Employment Profiles and Growth: An
Economic Base Study of Florida Counties. GI ,, ith and Change 22:86-104.

Fik, T.J., Malecki, E.J., and Amey, R.G. 1993. Trouble in Paradise? Employment Trends and
Forecasts for a Service-Oriented Economy. Economic Development Quarterly 7:358-72.

Fik, T.J. 2000. The Geography ofEconomic Development: Regional Changes, Global
Challenges. New York: McGraw-Hill.

Gibson, L.J. and Worden, M.A. 1981. Estimating the Economic Base Multiplier: A Test of
Alternative Procedures. Economic Geography 57:146-59.

Greene, W.N. 2000. Econometric Analysis. Englewood Cliffs, NJ: Prentice-Hall.

Hanink, D.M. 2007. Spatial and Geographical Effects in Regional Multiplier Analysis.
Environment and Planning A 39:748-62.

Harris, T.R., Ebai, G.E. and Shonkwiler, J.S. 1998. A Multidimensional Estimation of Export
Base. The Journal ofRegionalAnalysis & Policy 28:3-17.

Hayashi, F. 2000, Econometrics. New Jersey: Princeton University Press.

Hewings, G. 1985, Regional nput-output Analysis. Beverly Hills: Sage Publications.

Hildebrand, G.H. and Mace, A. 1950. The Employment Multiplier in an Expanding Industrial
Market: Los Angeles County, 1940-47. The Review of Economics and Statistics 32:241-49.

Isserman, A.M.J. 1977a. A Bracketing Approach for Estimating Regional Economic Impact
Multiplers and a Procedure for Assessing their Accuracy. Environment and Planning 9:
1003-11.

Isserman, A. 1977b. The Location Quotient Approach to Estimating Regional Economic Impacts.
Journal of the American Planning Association 43:33-41.




















Lioime

* Cal nou '-Lern ,Madl / val i
Bay 1 L k /% SUii mb
SSatranklin o '
,y I / L-e. GllchjrlsAlAMdno Duva
Gua St Joh
..e D "lagler
Levy I j


Legend

Basic Employment
for Manufacturing in 1993
0 220
220 760
S760 1670
1670 3840
3840 12827


Collier


Monroe


V---


0 62.5 125
I i 1 1 1 1


250 Kilometers
, N


Figure 4-12. Basic employment for manufacturing sector of Florida in different years by county
A)1993 B)1997


Broward


















[( I holme& S
9 .___kIos 4 Jackson
EsJa bi i alton wa in l'gLo -i Le e
Bay a
Liberty' Wakulla
Gulf
Franlin -


















Legend
Basic Employment

for Primary in 1997
0-55


w


56-161
162-545
546-1589
1590-3767


0 62.5 125 250 Kilometers
I I I I I I I I I


dson e
__ -'-,// { Du 5sli
/ SuwnneeColumoiaBake
Union tin
'lorfaye. Johqs
S G'Ichslachua Putnam.
1 Diu,.i--. ,"g'
SL Flagler
SLevy Mrton

Citrus Ake

PanC Oirange
Pasco


L ee Okeechobes-
ighland.s
Sarase / -.
Charlotte Glades
-I
L Hendry

S Collier Broward

!Miami-Dade



zo
..--"- "--
Monroe -






A
N


Figure 4-10. Continued













0







0
0



0 13
N
o






co .
> 0
o




o
U0-


o
o
o










cNE
(O
o
0 *
0o
0 *

0




o -


0 100000 200000 300000 400000

C



NEB





0
0
o C
o
o
0




0
0
0







0
C




















Figure 4-21. Continued




















S -- Holmes
HGsac Jackson)X

1daihoi 'een t
Esarn o I WaltonI Wasn / .Gad.e.eJe. scr HarrIon
S.. -- alhor, Leon Madri-.. .a -N ,
"Bay ,b Wakulla su/ar-neColimbtaBake
Libe y Un aon tJ
V JGulf TatlnIrnLat5or Johns
uf nalora ettke Bradford lyt Johs
I _F ra n kl-n
lI chriAlachua --
pDl 1'- Flagie
SLevy Marion \

Citrus Lake
Su_ 18F
Hernande
Pasco r o l
L Osceola
Polk ..
Iq.ar, River
Manatee Hardee OkeachobeLtu
S Hghlands L
SCDeSoto Marin
SCharlotte Glades
Lee Hendry
Legend

Non-Basic Employment Collier
for Transportation in 1993
F-24- 1070
1070-3330

| | 3330 7300 Monroe
7300 19700
19700 47600



0 62.5 125 250 Kilometers A


A
I, i i i I N




Figure 4-5. Non-basic employment for transportation and public utilities sector of Florida in
different years by county A)1993 B)1997









LIST OF TABLES


Table page

3-1 OLS assum options and possible solution ........................................ .......................... 37

4-1 Scalars and non-basic employment in year 1993 and 1997 by county .............................. 82

4-2 Ordinary least regression estimation result of years 1993................................................ 85

4-3 Ordinary least regression estimation result of years 1993................................................ 85

4-4 Ordinary least regression estimation result of year 1993 and 1997................................ 87

4-5 Test result of exogeneity of error term ........................................... .......................... 87

4-6 Panel data regression estimation result of year 1993 and 1997 without dummy
v a ria b le ........................ ................................................................................... ..... ...... 8 8

4-7 Panel data regression estimation result of year 1993 and 1997 with dummy variable......97

4-8 Multipliers of OLS regression and panel data regression............................. 101




















-- -/ "--- ._u ....
Sarta R.osa)kaloos3 -ako. N
Escarnbia l. .WallonWashirton Gad e.=-re-I.so0 _'I
S- alhoh ,Leon M,,I"- ,Hamlol a
Bay -f SuwanneeColUeBk rI
-.. .... .__ ,Union i
S ulf aylrLarav e S rafori ClS Johns
I ranklin- i
\ 2' )"ln'st''' 'Ja Putamr. t, F
SDxie FP.. agl
,1 Levy Marion

Citrus Lake
-- imter rmn \nhe
Hernandc.
Pasco evard
Osceola
Polk I \
Pinelias illsborou __,
Indian Rive
\ Manatee Hadee OHar kehobee, Luci
--+.ghland St -Luci
Sarasota DeSoo Martin
Charlotte Giades -
Lee Hendry Pal Beach
1 r --r Broward i
Legend \Collier
Basic Employment

for Transportation in 1997
0 300
300 2990 Monroe
S 2990- 5380 --L
5380 8530
8530- 44340 /


0 62 5 125 250 Kilometers N


B
Figure 4-13. Continued









ECONOMIC BASE ANALYSIS USING PANEL DATA REGRESSION:
A CASE STUDY OF THE FLORIDA REGIONAL ECONOMY



















By

FAN LI



















A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE


UNIVERSITY OF FLORIDA

2010









Tiebout, C.M. 1956. Exports and Regional Economic Growth. The Journal of Political Economy
64:160-164.

Vias, A.C. and Mulligan, G.F. 1997. Disaggregate Economic Base Multipliers in Small
Communities. Environment and Planning A 29: 955-74.

Vollet, D. and Bousset, J. 2002. Use of Meta-Analysis for the Comparison and Transfer of
Economic Base Multipliers. Regional Studies 36: 481-94.

Was, A.C. and Mulligan, G.F. 1999. Integrating Economic Base Theory with Regional
Adjustment Models: The Nonmetropolitan Rocky Mountain West. Gi 1, thi1 and Change 30:
507-25.

Wooldridge, J.M. 2002. Econometric Analysis of Cross Section and PanelData. Cambridge, MA:
MIT Press.











0
0

0 -
0




S0-
C 0
O-

0


S.

0o
0
0
O_
o S
*


0_


0 100000 200000 300000 400000
Linear prediction D


Figure 4-20. Continued


While residual plots of OLS regression show obvious triangle-shade, residual plots of


panel data regression show probably random distributed if the upper and below outliers are

deleted. Therefore, all OLS regression estimates here suffer from heteroscedasticity problem.


Panel data regression also cannot avoid this problem, but alleviate to some extent according to


residual plots.

Plots of fitted versus actual values of the scaled non-basic employment for year 1993 OLS


regression, year 1997 OLS regression, both year 1993 and year 1997 OLS regression and panel


data regression are showed in Figure 4-21A, 4-21B, 4-21C and 4-21D respectively. All plots of

fitted versus actual non-basic employment values shows satisfied "goodness of fit" of the


economic base models, which implies that independent variables can explain scaled dependent









Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Science

ECONOMIC BASE ANALYSIS USING PANEL DATA REGRESSION:
A CASE STUDY OF THE FLORIDA REGIONAL ECONOMY

By

Fan Li

August 2010

Chair: Timothy Fik
Major: Geography

The main purpose of this research is to investigate and estimate the increase in total regional

employment as it relates to employment increases in specific economic sectors, using economic

base theory and a regional economic base model. Past empirical research in economic base

modeling typical relies on single year data and original least square (OLS) regression, with

unstable results that are possibly explained by the fact that the regression analysis is influenced

by abnormal value changes in one or more explanatory variables. As an alternative, this thesis

utilizes panel data regression with data for multiple years in an economic base analysis of Florida

employment patterns. Sector-specific employment multipliers are estimated and compared using

OLS and the panel data regression. Of the other widely used methodologies: the survey method,

location quotient (LQ) method, and minimum requirements (MR), regression analysis is the only

one which can estimate impact multipliers.

Regression models to estimate employment multipliers are typically divided into

disaggregate and aggregate models. The former can obtain the multiplier estimates for specific

industries, while the latter one independently cannot distinguish between different industries.

Hence, disaggregate regression models are widely used in economic base research. Unable to

account for economic spillover effects, such as cash flows or trans-boundary monetary transfers,































200000
Fitted values


**


St.


200000
Fitted values


Figure 4-20. Continued


qt*

().


100000


300000


400000


100000


300000


400000









Compared with Figures 4-1A and 4-1B, these counties are coincident with the counties that

have largest total employment except Miami-Dade, which actually is the plot above

NEB=400000 in Figure 4-21D. The common geographic characteristic of these 6 counties is that

all of them are coastal urban area. As tourist and wholesale trade are highly developing, coastal

and waterway resource boom the relative counties' employment. Therefore, it is reasonable to

argue that the geography characteristic of coastal urban area should obtain more attention in the

model.

In order to eliminate the bias of estimates for these 6 counties, the original panel data

model should be added a new dummy variable "urban" defined as coastal urban counties,

including Broward, Palm Beach, Orange, Duval and Pinellas and Hillsborough. The result of

panel data regression estimation with dummy variable list in Table 4-7. Dummy variable is

positive significant at 1% level, which coincide with expected impact that coastal urban property

can increase the employment of county.

Table 4-7. Panel data regression estimation result of year 1993 and 1997 with dummy variable
Coefficient Std.Err Z-test P>|IZ 95% Conf. Interval
prim 2.220 2.320 0.95 0.340 -2.345 6.787
cons 3.987** 1.391 2.87 0.004 1.260 6.715
manu 2.354*** 0.708 3.33 0.001 -0.997 3.741
tran 4.163*** 0.638 6.52 0.000 2.912 5.413
whol 8.889*** 1.017 8.52 0.000 6.675 10.662
reta 4.071*** 0.764 5.33 0.000 2.573 5.569
fina -.087 0.496 -0.17 0.862 -1.060 .887
serv 2.752*** 0.311 8.85 0.000 2.143 3.361
soci 0.202* 0.031 2.72 0.006 0.024 0.146
urban 109119.4*** 8501.87 12.83 0.000 92456.1 125782.8
constant -532.287 1978.03 -0.27 0.788 -4409.16 3344.584
Note: *** indicates significant at the 0.01 level, ** indicates significant at the 0.05 level, *
indicates significant at the 0.1 level. Wald Chi 2(10)=2558.00, Prob> Chi 2=0.000
Plots of fitted versus actual non-basic employment values for both year 1993 and year

1997 panel data regression with dummy variable show in Figure 4-23. In the middle, some plots









CHAPTER 1
INTRODUCTION

Overview and Problem Statement

Employment patterns by industry or economic sector for sub-regions of a regional

economy are important indicators of regional economic growth and development. For example,

county employment patterns and industry mix are useful in assessing the degree to which a

county is reliant on specific industries when compared to an employment mix at the state or

national levels. The more employment in a given sector, relatively speaking, the more important

that sector is to the economic vitality of a regional economy. Furthermore, it is known that

overall employment levels may be a by product of supporting or linked industries that are known

to generate spin-off employment.

County-level economies typically have very different industry employment patterns than a

state-level economy. These patterns can offer insights as to the importance of various industries

or sectors to a county's overall total employment and the extent to which key sectors have an

impact on regional economic development and the expansion of a region's overall employment

base. Key sectors are generally referred to as the "economic base" of a regional economy. They

are the sectors that are most responsible for generating the majority of linked or spin-off

employment. The net benefits of a region's economic base are generally distributed unevenly

over space, and it is useful to gather information from many sub-regions as to the overall

importance of various sectors to the regional economy at large.

Florida, nicknamed the "Sunshine State" because of its generally warm climate, is ranked

as the fourth-most populous state in the U.S. according to the United States Census Bureau in

2008. Florida's regional economy focuses on three leading industries; namely, tourism,

agriculture and mining. In the last fifty years, more and more retired people have moved and


















Holm.es \
S t osaOkl I rO | /I 'Jacksonm
Escarnb a /IMlTosi o nWasitn -Gadsdr fle n )Nassau
Esca'. rr,,
SCail. .- Leon Madison i /Duval
-Bay 7 L ibet)Wakulla SuwanneeCo 0 bIn Bakr
"\" Liberty-" lla^ T t no Icla, \
Ga V- lorLa ayete B rad lay S Jo
SFranHli, L r \ Brafd S J
S / Glchrist Alachua P -
Dixie 1 ... -- Fiagler
\ i ", .... I 'l


Legend
Basic Employment

for Services in 1997
0- 730
I 730 2870
2870- 14590
14590-22130
22130-35410


0 62.5 125 250 Kilometers
I I I I I I I I


revard
Osceola
Polk
I. hdian Rive
gte Hardee Okeewhobee, .
Highlands St. Lucie
I DeSo" _1-" / Martin
Charlotte Glades
-Pan Beach
Lee I nHenciry

Collier Broward
........... .. .


Monro


N
N


Figure 4-17. Continued


,-3


















Holmnes
S so Ja cksoon a N ass a u
E ia vWalIcnW.;rlrn sn N sau
Slhouh jMadis | 0
------ L B.~.~.y' d;&'a 'so "-'
Bay W SuvvanneCCol mbt, keFr
S I "Llerly W kw llua I Union I I
J G Taylor a ete o Cla' St. Johns
,I ranklin- \ Dx.i hris P u "
Gir r PuTriam -
Dixl- Fla gle
L ^ M onL
Citrus e
temando
Pacoa


ian River
Manatee Hardee Okeec obee Lu
highland s

Cha G.Mdrbn
Hendry


Legend

Non-Basic Employment
for Finance in 1993
31 550
S1 550 2010
1 2010 4160
4160 9400
9400 61030



0 62.5 125 250 Kilometers
I I I I I i I


Monroe







N


Figure 4-8. Non-basic employment for finance, insurance and real estate sector of Florida in
different years by county A)1993 B)1997









autocorrelation of error term, while the original panel data model gets rid of the problem. It is

highly possible that dummy variable lead to such a problems, because some of coastal urban

counties are neighborhood.

According to test of exogeneity of error term and three comparison between OLS and

Panel data regression, estimates results depending on both OLS and panel data regression fit the

actual scaled non-basic employment to large extent, while compared with OLS regression, panel

data regression shows better results on test of normality, homoscedasticity and exogeneity.

Multipliers of each industry/sector, from Tables 4-2, 4-3, 4-6 and 4-7 are summarized in

Table 4-8. While the panel data results appear promising, the OLS estimates are most certainly

suspect given the dramatic differences that are observed between 1993 and 1997 in the

magnitude of the estimated coefficients. In 1993, the multiplier associated with the primary

sector is estimated at 3.253, smaller than that of the multiplier for the transportation and public

utility sector. This result does not make sense, because it is known that the agriculture industry

in Florida creates more employment opportunities than does public transportation. Moreover,

single equation estimation of multipliers for the year 1997 produced negative multipliers. This

result is somewhat of an absurdity as a job added to an economy cannot take away more than that

which it adds (as a generalization, any increase in income via a new job will undoubtedly

produce additional demand). By contrast, the panel data regression results seem to fall more in

line with expectations, producing positive multipliers across the board and coefficients that are of

a reasonable magnitude. Compared with regression estimates for a single year, panel data

regression provides a more consistent and even-handed estimation of multipliers across sectors,

avoiding the pitfall of abnormal disturbances associated with single year samples and illogical

results such as negative coefficients/multipliers.



















-- < i Holme"
S rig t I Jackson
.i(I.ua:Okaloosa,,^la.n ... .. w _Y
'2S sa l' l ar, rgto -.- Gadsde r son H .. ilo
-_ 1 al-o Le Madson rt
Bay Libert Wakulla Sda-nneeCColUmba
-Gulf Taylorafavette.
Franklin GilC',s
I Dixie ]
SLevy

Ctru
Hern.
Pa!

Pin 9








Legend
Non-Basic Employment

for Primary in 1997
8-189
190 729 i
S730- 1817
1818 3290
3291 5354

0 62.5 125 250 Kilometers
I I I I I I I I


Figure 4-2. Continued


.J

r
aI


I


0


Giades
Hendry


Monroe


A









4-16 Basic employment for finance, insurance and real estate sector of Florida in different
year by county A )1993 B )1997 ......... ............... ............... .................. ............... 75

4-17 Basic employment for services sector of Florida in different years by county A)1993
B )1997.............. ................................................................ 77

4-18 Social security payment by county of Florida in different years by county A)1993
B )1997................ ....................................................................... ...... 79

4-19 Normality test for different regression A)Year 1993 OLS regression B)Year 1997
OLS regression C)Both year 1993 and 1997 OLS regression D) Both year 1993 and
1997 panel data regression ............................................................ ............. ............ 89

4-20 Residual plots for different regression A)Year 1993 OLS regression B)Year 1997
OLS regression C)Both year 1993 and 1997 OLS regression D) Both year 1993 and
19 9 7 p an el data regression .......................................................... .................................... 9 1

4-21 Plots of fitted versus actual non-basic employment values for different regression
A)Year 1993 OLS regression B)Year 1997 OLS regression C)Both year 1993 and
1997 OLS regression D) Both year 1993 and 1997 panel data regression.........................94

4-22 M ap of coastal urban counties.................................................................. ...................... 96

4-23 Plots of fitted versus actual non-basic employment values for both year 1993 and
year 1997 panel data regression with dummy variable ...............................................98




















SanHolmes
Esi kaloosI'tonWashiglton -GsddeJeesn Hamton -
Calnow Leon Maoiso R.amic. i a-
Bay bertWakulla SUnrleeCllmbiaBaker
..yGu. T 1aylorla e nion' ClayS. Jo
Gu-r..ii- Bradford r Johs
F ranrli"n -- -
G ,.i-GIchristAachua i in
Dixie P, lnam-
/ -- .- --, Flagler,
Levy Marion -lu" \

Citrus Lake
Sumter i
Hernan.Jo I\
Pasco
LLOsceola
Polk
\ Ildian Ri r
Manatee Hardee Okeecnobee t Lucie
,H Highlands
DeSotI -1-
ra DeStr Martin\
'' Charlotte Glades
Lee4 Hendry

Legend --I
Collier
Non-Basic Employment
for Primary in 1993
0 286
287 782

783- 1623 ii Monroe
1624-2791
2792-4376



0 62.5 125 250 Kilometers A
I i I I I I I I N
A


Figure 4-2. Non-basic employment for primary sector of Florida in different years by county
A)1993 B)1997





















Sar a oaO 0 Jackson
Espambla WallnWasri., F- GMsde adlsn -,Ham Nassau
.. I ao rcI ,cn ...i. so- Duval
berty ,Waku ll SuwanneeC ohmbiaBae e
Gull r .Iin1 OFLalIvette Joh s_
Franklin J
Glicnr. UU Putnam
D le Flagle
Levy Maron

C rus r

Orange
HL rear

PineliasHrlisboroughi i _
dian River
Manatee Haraeel Okeehobee
Hlghlandri
D e S o l o _I I
Glades


Legend

Basic Employment
for Retail in 1993 Miami-dea
0-310
S310 1179
1179- 2618
2618 4044
4044 6572



0 62 5 125 250 Kilometers
I I I I I I N

A
Figure 4-15. Basic employment for retail sector trade of Florida in different years by county
A)1993 B)1997


















I Ho m-s
Holmes
.. L / |J a c k s o n) / -
S ait9 oSaOI loosawa tonwt.-ng .r., H- -o nssg
.__l I | Calh Leon Ma..iso y. -L'
-Bay- L Wakil SuwanneeColumbIraBa, .

Gu lf TylOrL ayette I Unn Clayt Johns
Franklin, --..a. -
nrStAlachua Putnam-"
l- Dixie ,




P ein Mo"ioe




Monet Harde- Okei-hocee^^'










Social Security Payment in 1993

478 4975
S4975 20037 Manroe
20037 40512 ._^
40512 83647
S83647 220434

0 62.5 125 250 Kilometers K
F g r I I i i i II II



Figure 4-18. Continued









CHAPTER 3
METHODOLOGY

This chapter discusses the methodology utilized to fulfill the research objectives stated in

Chapter. In the first three sections of this chapter, employment data are used estimate economic

base multipliers using the location quotient (LQ) method to distinguish between basic and non-

basic employment, and a comparison of multiplier estimates from ordinary least squares (OLS)

regression and a linear unobserved effects panel data (PD) model. At the conclusion of this

chapter, alternative research models are proposed, with elaboration on the selection of variables

and potential data sources.

The Location Quotient Method

The Location Quotient (LQ) method is commonly used to calculate an average economic

base multiplier. For our purposes here, it represents preliminary stage, as typically applied when

no industry-specific data regarding basic versus non-basic employment is available. In short, the

LQ method is a tool that provides sector-specific estimates for basic and non-basic employment,

under the usual assumptions. Given the lack of survey method data for counties in the state of

Florida, the LQ estimates will serve as important input for a regression-based approach that

compares the results of alternative estimators. The focus is on economic activity levels at the

county level, with the state of Florida used as a benchmark economy. It is well known that LQ

multipliers typically give large and bloated estimates when run at the regional level given that

much of sector-specific demand is satisfied by imports or supply coming from outside the region

and that some sectors are exporting across the regional boundary to satisfy non-internal demand

(Chiang 2009). Hence, it may be prudent not to use the actual location ratio as a benchmark, as

the LQ method would undoubtedly produce overstated multipliers. In order to make the

multiplier estimates more reasonable, a "scaling factor" could be employed. Any attempt,









Consider the following linear panel data (PD) model, expressed as

yt = a + bilx,1 + b2tX,2 +... + btxt + ',t (2-20)

where xlt is a 1 K vector that contains observable variables that change across time t but not i,

variables that change across i rather than t, and variables that change across i and t (Wooldridge,

2002). The error terms can be denoted as

et = c, + ut (2-20a)

where ci is the part of error term that changes across the individual rather than time, while ult is

the part of error term that change across both individual and time. Wooldridge (2002) states that

the component CI could be treated as a random variable or as a parameter that is to be estimated.

Accordingly, ci is divided into a "random effect" and a "fixed effect", where the former means

zero correlation between the observed explanatory variables and the unobserved effect; that is,

Cov(x,, c,) = ; while the latter means that correlation between the observed explanatory


variables and the unobserved effect is non-zero and allowed: Cov(x,,,c) 0 In short, two

types of regression possibilities apply here: "random-effect methods" and "fixed-effect

methods".

Recall that the fifth assumption of OLS regression states that no serial auto-correlation is


found in the error terms of the model, orE(E'1'2 x) = 0 For the two time period panel data

model, because ci is unchanged across two time periods

E(E,,,2 x)= E[(c, + u, )(c, + u,)x]= E[(c,c, + c,u, +u,,c, +u,lu,)x]
= E(c, c, Ix) +E(c,u,2 x)+E(c,u,, x)+E(u,,u,2 x)= E(c, c, x) 0 (2-21)


E(c,u,, x) + E(c, u1, x) + E(uu,, 11 x) = E(c, c, x) = 0


where the assumption is that









extent to which a given industry affects a region's total economy can vary dramatically

depending on that region's industrial mix, and its overall impact on urban and regional

development is something that be estimated having knowledge of the inter-relationships or inter-

dependencies between industries and sectors. As a regional economy is highly dependent on

revenue from both industry mix and external sources, it is vital to understand the nature of the

influence of key economic sectors or industries as well as the effect of things like income

transfers. It is also important to determine the extent to which overall employment levels are

affected by key sectors or industries while accounting for the effects of income transfers.

Research Objectives

The goal of this study is to identify the extent to which key sectors or industries affect

overall employment levels in the Florida economy, controlling for the effects of income transfers

and geographic variability in the inter-relationships between economic sectors. The analysis

takes into account both industrial employment patterns and underlying transfer payment impact

using county level data which links total employment to employment by sector. Total and

sector-specific employment data were chosen as they are widely used to measure the

development level of regional economy. A major task of this large interdisciplinary project is to

distinguish between import employment and export employment -- employment related to

income transfers versus that which is generated from a region's economic base (also called an

export base), taking into account the transfer income factor (i.e., income generated from the

inflow of retirement dollars, investments, and other sources of personal revenue). The

importance of this factor can not be understated given the large proportion of elderly and retired

in Florida. Thus, it is recognized that the expansion of a region's employment base is both tied

to industry mix and non-employment or transfer income. Transfer income is a crucial part of

state income, and cannot be ignored in the model. Failure to account for the transfer income












From Figure 4-21D, there are 11 plots below the line of "fitted value equal actual value" from

100000 to 300000, while only one plot is above the line, which implies that fitted value is

underestimated during this range. After projecting the underestimated plots to the total

employment map shown in Figure 4-22, these plots are standard for Broward, Palm Beach,

Orange, Duval and Pinellas and Hillsborough.










































Figure 4-22. Map of coastal urban counties
P i i





SBroward








0 60,000 120,000 240,000 Kilometers





Figure 4-22. Map of coastal urban counties









CHAPTER 5
CONCLUSIONS

Job creation is a characteristic of basic industry. Activity and employment in basic sectors

are vital for economic growth, and basic employment plays a crucial role in contributing to the

growth of a regional economy by bringing in dollars from external sources. Economic base

multipliers estimate the impact that an influx of dollars has on the creation of jobs for each

additional job added to a given sector. Basic employment, in response to external demand for

goods and services, is responsible for creating additional jobs as income circulates and re-

circulates throughout a regional economy, and subsequently, as the employment base expands.

Quantitative models that attempt to estimate those multipliers through the disaggregation of

employment or income data must reckon with problems that are tied inherently to the nature of

sample data and to problems that arise from various estimation procedures. Economic base

models can be a quick, effective, and inexpensive way in which to ascertain the expansion

potential of various sectors or industries and their relative impact on a region's economic growth.

This information is useful to policymakers and legislators in their attempt to construct economic

and development policies that could conceivably yield the biggest bang for the buck. Moreover,

economic base analysis and estimated multipliers are useful in explaining variability in regional

economic and employment growth rates (Malecki, 1991), factors that can be attributed to a

region's economic or employment mix. As such, critical assessments of multipliers are needed

to ensure accurate and reasonable estimates. As part of this assessment, the techniques by which

multipliers are estimated must also be evaluated to lessen the chance of bias or misleading

results. This thesis has demonstrated that multiple-year panel data regression models offer a

useful alternative to OLS regression model as an estimation technique.









CHAPTER 4
EXPLORATORY ECONOMIC BASE REGRESSION MODEL

This chapter presents the statistical results of an exploratory economic base analysis of

the Florida economy using county level data (for n=67 counties). The first section provides an

overview of the employment data used in this study and the estimation of basic and non-basic

employment. Section two discusses the potential overestimation problem that is inherent to

economic base studies, and offers a potential solution the use of various scaling factors to

deflate non-basic employment estimates. Lastly, the results of the regression analysis are

highlighted; comparing those obtained from ordinary least squares (OLS) regression and panel

data (PD) regression.

Initial Data Collection

Florida county-level industrial employment data were collect from the U.S. Census Bureau

website. Standard Industrial Classification of employment figures were obtained for nine

separate industries/sectors: (1) agricultural services, forestry, and fishing; (2) mining; (3)

construction; (4) manufacturing; (5) transportation & public utilities; (6) wholesale trade; (7)

retail trade; (8) finance, insurance and real estate; and (9) services, as well as employment

figures for unclassified establishments. Previous research, such as the economic base research of

communities in Arizona, combines employment in agricultural services, forestry and fishing, and

mining; defining it is as primary sector employment. This study follows this convention.

Using the reported employment data, estimates for basic and non-basic employment were

obtained using location quotient (LQ) method. A location quotient was calculated for each

county and each industrial sector using state level data as a benchmark. Recall that the LQ

method relies on the formulas listed in equations (2-15) to (2-17), where the ratio E- is the
s
proportion of a j-th county's industry employment in a given i-th sector; a value that is compare









Table 4-1. Scalars and non-basic employment in year 1993 and 1997 by county
County scalar 1 scalar 2 nonbasic in 1993 nonbasic in 1997
Alachua 0.65 0.8529 57107 66567
Baker 0.69 0.8529 2974 2956
Bay 0.63 0.8529 39872 45597
Bradford 0.68 0.8529 3068 3463
Brevard 0.59 0.8529 128870 137691
Broward 0.6 0.8529 463034 541136
Calhoun 0.59 0.8529 1365 1241
Charlotte 0.35 0.8529 21692 24698
Citrus 0.38 0.8529 17659 18615
Clay 0.73 0.8529 19526 22320
Collier 0.4 0.8529 56239 72960
Columbia 0.61 0.8529 10740 14044
Dade 0.62 0.8529 719150 740137
De Soto 0.55 0.8529 3778 3976
Dixie 0.52 0.8529 850 841
Duval 0.71 0.8529 291661 317162
Escambia 0.64 0.8529 88129 96929
Flagler 0.45 0.8529 5232 7281
Franklin 0.48 0.8529 1610 1800
Gadsden 0.64 0.8529 7610 7453
Gilchrist 0.66 0.8529 868 963
Glades 0.54 0.8529 269 1242
Gulf 0.56 0.8529 1583 1999
Hamilton 0.6 0.8529 1535 1453
Hardee 0.62 0.8529 2735 3109
Hendry 0.67 0.8529 4004 3793
Hernando 0.41 0.8529 16452 24306
Highlands 0.39 0.8529 15409 16688
Hillsborough 0.68 0.8529 366108 434392
Holmes 0.56 0.8529 1709 1903
Indian River 0.34 0.8529 25096 30632
Jackson 0.57 0.8529 7016 7083
Jefferson 0.63 0.8529 1633 1622
Lafayette 0.68 0.8529 418 487
Lake 0.48 0.8529 36262 41728
Lee 0.46 0.8529 103366 124288
Leon 0.72 0.8529 65815 74493
Levy 0.54 0.8529 3306 4037
Liberty 0.66 0.8529 578 568
Madison 0.58 0.8529 2178 2589
Manatee 0.5 0.8529 59985 79526
Marion 0.51 0.8529 48426 54307
Martin 0.37 0.8529 31768 36130
Monroe 0.48 0.8529 24340 26256


















SHolmes
s tosa o WalTona Jackson "
I v V ITo n a s ih o u'r i'o n -T,- -r S a n e..I... -N a s: a u
l[," Madr0o ar-m lllu
^ ,_ ----^ialr, o d- ^.B.-. I ---,
Liberly Wakulla S .neeCl- Unionb a er
JGI Tylorf^l^e YrSt Johns
,G. ".. a,,Putnam'-








Mllana HHardee Okeecobee. '.








Non-Basic Employment
for Finance in 1997
26- 970
970 2370 -Monroe
2370 5690
5690 9440
I 9440 62810 A

0 62.5 125 250 Klometers N
I r I | I I I i I

B

Figure 4-8. Continued




















I Holmes
oa' *.)-Ja ckson,

9 alriouL Leon Maois'~ i B e


F rarnklin -' \
Sf ran "--ri IA G inhrAlachuia n .
D .xie -- Fagler
'< Levy Marion L sia

CtrJus e ke_
Hernardo I
Pasco ;-la


Indan RIe
Manatee Hardee' Okeech:.beet Luc'
Highlands
Srasola DeSoto Martin
Ch~harIotte Glades
SHendry


Legend collier
Non-Basic Employment

for Transportation in 1997
10- 1400
1400-4600 Monroe-
4600- 11680 -J "
S11680-21890
21890-49443


0 62.5 125 250 Kilometers N


B

Figure 4-5. Continued









however, to scale down the LQ multiplier estimates would run the risk of being viewed as

subjective as the choice of a scalar is not readily clear and for the simple fact that the scalar may

differ across locations. As such, we will turn to OLS estimators which rely on various statistical

properties in the estimation of multipliers.

Ordinary Least Squares (OLS)

This section introduces the single-equation linear model and the ordinary least squares

(OLS) estimators. It is the most basic estimation procedure in econometrics, yet it is known that

accurate application of OLS depends on meeting a series of strict assumptions (Hayshi, 2000).

Once one or more assumptions are violated, OLS method cannot obtain reasonable estimation

results. Consider a multivariate regression model of the variety

y, = a + blXl + b2X?2 +...+ bkXik + (2-19)


where b 's are unknown parameters to be estimated and 6, are unobserved error terms with

certain properties (as stated below). Note that the b 's represent the marginal and separate

effects of regressors (the independent variables of the model). The error term represents the part

of the dependent variable left unexplained by the regressors. It can consist of a variety of things,

including omitted variables and measurement error. An explanatory variable xkis said to be

endogenous in the regression equation if is it correlated with E. It is assumed that Xkis exogenous

in regression equation. Endogenous variables are a common problem in OLS regression, and can

lead to violation of the assumptions of OLS regression.

Five assumptions of OLS may be summarized. First, the relationship between the

dependent variable and the regressors is linear; i.e., the model is linear in its parameters (or can

be specified in a linear form such as the model shown above). Second, variable matrix is full

rank, which implies that each column of regressors is linearly independent (assuming that multi-









with empirical results that are consistent with those reported by Fik et al (1991, 1993). The

smallest coefficients are associated with transportation and finance; not surprising, as Florida

does not actively engage in the building of public transportation (e.g., limited railway and

subway). All in all, the sector-specific multiplier estimates provide a glimpse of Florida's

orientation toward a sales- and service-oriented economy, as discussed by Fik et al (1991). The

size of the multiplier estimates indicate that may still be slightly overstated, yet one could argue

Table 4-6. Panel data regression estimation result of year 1993 and 1997 without dummy
variable
Coefficient Std.Err Z-test P>|Z| 95% Conf. Interval
prim 5.483 4.072 1.35 0.178 -2.499 13.464
cons 3.311 2.137 1.55 0.121 -.877 7.500
manu 4.019*** 1.307 3.07 0.002 -1.457 6.580
tran 2.768*** 1.059 2.61 0.009 -.693 4.843
whol 10.832*** 1.702 6.36 0.000 7.496 14.168
reta 4.306*** 1.201 3.58 0.000 1.950 6.659
fina 2.736*** 0.685 3.99 0.000 1.393 4.079
serv 4.454*** 0.396 11.25 0.000 3.678 5.230
soci 0.202*** 0.046 4.39 0.000 0.112 0.292
constant -1336.749 3636.65 -0.37 0.713 -8464.45 5790.9
Note: *** indicates significant at the 0.01 level, ** indicates significant at the 0.05 level, indicates
significant at the 0.1 level. Wald Chi 2(9)=0629.42, Prob> Chi 2=0.000

that their relative proportion is fairly representative of Florida's economic reality (with an

estimation bias that has been reduced), given that the transfer income effect has been largely

accounted for by scaling back non-basic employment levels.

In order to test if the above OLS regression and panel data regression is valid, error term

diagnostics for both OLS regression and panel data regression are applied. Normality test of

error term for residuals of year 1993 OLS regression, year 1997 OLS regression, both year 1993

and year 1997 OLS regression and panel data regression are showed in Figure 4-19a, 4-19b, 4-

19c and 4-19d respectively.









is extremely time-consuming and costly. The quality of estimates obtained from this approach

depends greatly on the quality and accuracy of answers provided by survey respondents. Many

companies may be reluctant to reveal their sales data, and hence, there is a potential for under-

reporting or for firms to be uncooperative (Harris, Ebai & Shonkwiler 1998). Nevertheless, the

survey method is the most direct way to gather information on non-basic and basic employment

and does not rely on complex statistical techniques or mathematical transformations. It is a

technique that is effective in small communities or those settlements with a small number of

firms.

The Location Quotient Approach

The location quotient (LQ) method of estimating economic base multipliers is more

elaborate in its design. The approach begins by assuming that that a given sub-region requires a

level of economic activity in an particular sector or industry that is directly proportional to the

national or regional economy at large (Isserman 1977a). Specifically, the method compares a

region's observed industry employment or income concentration with that of a benchmark;

typically the concentration of employment or income that industry in the state or nation in which

the region is located in. The location quotient is a ratio of ratios, and may be defined as the ratio

of employment in sector i for region j divided by the ratio of employment in sector i for an

economy at a larger geographic scale. Specifically,

SE, / e,
E1 e (2-15)
EI e,
where Ej and e are the ratios of employment in sector i to the total employment in sub-region

j and the larger geographic region, respectively. Once the value ofLQ' is estimated, non-basic

employment be easily identified. IfLjQ > 1, it means the proportion of i th sector employment to









LIST O F R E FER EN CE S ............... .............................................................................................. 106

B IO G R A PH IC A L SK E T CH ........................................................................................................... 109





















































7




















S~k~r---- ------- -
Holmes7 \
SJ b acson
ESc a'a 0' 'mNassau
Esa al"ooVaVltonMShirgtnon GadsdenJe fson H-m- ; N 1 1 .
I ._.. alhoih y Leonr Macisor op D uval
Bay Libertv Wakulla / SuanneeColumbB
'-. L(l- Lfa'yt4 ..., -
Gulf r ,hTavlrBradyor y Joh ns
G hri... / "'Alachu IpUPnam "
ID xie P i- aglr
Levy Marion
LVolusia
CiItrus Lake\
umtt~r ninIoe
Hernando. Oranget(
SPasco aBrevard
U Osceota
Pineblas Hilsborou
I
Manatee Hardee Okeechobee
Highlands.
Saraso- DeSolo _- Mitrn
Cnarlotte Glades

Legendndr
Broward i
Basic Employment
for Primary in 1993 \ Miami-Dadeo
0-101
102-383
S384- 909 Monroe
910-1836
1837-3317



0 62.5 125 250 Kilometers J
I I I I I I N

A

Figure 4-10. Basic employment for primary sector of Florida in different years by county A)1993
B)1997




















Hormes
SSar os t Jackson '-
osaol^ W al T ^n1". Nassau
W v-Il.as rigwr G s S .n -L_ a o
airi.L om .Madison Hamit at
I.Y ilb q SllwarneeColt b13 Bak.er _I
-' Gul.f TaylorLafyeft U ClaY J s
I Franklin 'r
Putnam
D"ie DFlagI
Levy Maion

Crtrus ke
Suril, r-
Hernando
'Pasc


Legend

Non-Basic Employment
for Construction in 1993
I 23- 900
900 2300
2300 4420
4420 7720
7720 38977


0 62.5 125
I I a I I I I


Irdian River
MarInae Hardee Okeachobee
Highlands Luc
et e r1 V Martin
Charlotte Glades
Hendry









Monroe
C7 r^'-o ^-


250 Kilometers A
1 1 N


Figure 4-3. Non-basic employment for construction sector of Florida in different years by
county A)1993 B)199



















Santa RosaOk-ioosa, a -" -.- M'-?
Esambria o I va3In""vrvn'asdnsI, -~"f7Y,6 ,Jef-rsn-,-
Ih ,-n-1 C .aloun r 'i Leer, Duval
Bay f o~ mb Baker
L"- Wakulla. i'- Ban [" ei
-Gu lf br Bradord aS. Johbs
i ranJk lin I GIchr stAlchus
Dixie
S Levy
j, IVolusia
Citrus ake
m, r Senlinol
Herrand Orange
Pasco -
Osceol
Pin illsboro

Hardee OReech
Hignlands
Ssraso DeSoTo
r -T
"- Charlotte Glades

Lee Hendr


Legend Collier
Basic Employment -"

for Manufacturing in 1997
0 250
S250- 690 Monroe
690- 1640
1640-4600
4600-12300

0 62.5 125 250 Kilometers
I i i I l i I I l


Figure 4-12. Continued


Miami-Dade>

/ '





,N


NV





















Holmerl \
TI I re 'Jackson
Snta ~ S OKalocsaWlonwashnrgtn GadsJe erson H n Nassau
Esrambs 0 nashgol.
S Calho Lecn M s rH n illon
SBay i .bertyW.ulia Su.warneeComoaB er ,
I -T LinVz Clay L
-"Gulf orLfaelT Br r .l JO ilS
/ fGra n r, ..., .....-- ,
S Icl, tAlachua 'Putnam -
Dixie lPutnam
Levyr- Marion -
SVI olusia
SCitrus \ Lake
r 1emand- I
n o range
Pasco I reva,



Pin Marting


Charlotte Glades

M Lee Hendcry


Legend \Collier





4--
Basic Employment

for Finance in 1997C


S50 320 ZMonroe
320-1660 ,
1660 5750
5750- 31106


0 625 125 250 Kilometers I
I I | I I I I l

B

Figure 4-16. Continued









multiplier. Practitioners typically apply the marginal multipliers to forecast which industry

obtain the additional local employment as driven by the entire export-oriented employment shift

and to estimate the new local job creation that is linked to changes in industry-specific export

employment. The calculation of both MB and M1 require estimation of the disaggregate model.

Let's specify a liner relationship between total export activity and industry-specific local

activity. This can be formally expressed as

EN1 = a, + bEB (2-11)
where b, is the marginal returns of entire basic employment to industry-specific non-basic

employment. Following this logic, the marginal multiplier is

cET
M= (E) = 1 + bi
SEB (2-12)

and represents the local employment change in an i th industry resulting from the each unit shift

in the total number of basic jobs.

Alternatively, a linear relationship between industry-specific export employment and total

local employment may be specified as

EN = a + biEBl + b2EB2 + b3EB3 +... + bEBN (2-13)
where b, is the marginal returns of industry-specific basic employment to total non-basic

employment as defined for i = 1,.., N industries. The marginal multiplier from this equation is

defined as

aET
M = ( ) = 1+ bi
OEB (2-14)

and represents which the total local employment change associated with each unit shift of basic

employment in an i th industry. In short, either of these two methods could be used to estimate

the marginal multiplier.









The aggregate version of this model concentrates on the entire regional economy, and as

such fails to distinguish between the various impacts of basic economic activity change on

specific sectors or different industries that fall under the rubric of non-basic economic activity.

The aggregate model does identify dissimilarities of influence or the external industry-specific

demand on local economy (Mulligan 2009). Input-output models have clearly demonstrated that

regional changes in different export industries are likely to have different effects on a regional

economy, should one be able to trace the flow of income associated with exports and external

demand to inside economic activity.

Disaggregate regional economic base models represent an acceptable compromise between

aggregate models and the more computationally intensive and information-dense input-output

model. It extends the traditional economic base framework to include industry- or sector-

specific levels, a feature that has allowed these models to gain a broader acceptance (Loveridge

2004). Income data or employment data are widely used by researchers to measure the level of

regional economic activity by sector, thereby allowing the estimation of sector-specific

multipliers which describe the expected expansion in total income or employment for a given

change in income or employment in a given sector. Although it is recognized that employment

data fails to account adequately for productivity or wage/earnings differences between workers

employed in different industries (or even differences within the same industry or sector, across

various firms), employment data are widely used given their availability.

The Average Multiplier of the Economic Base Model

After distinguishing basic and non-basic economic activity/employment, the economic base

model is then used to calculate an economic base multiplier. The average regional economic

base multiplier MB for the entire economy is calculated as:




















I iolmes
- To I jacrk. on -
Estl kaloo VVa' l shngtn -GadsdeiJferon as
,_ lnou, Leon MedisorlH"a'1 i"
SBay Wberty a SuanrIeeClumb
L e Gilt Ta orL af- aete uClay T. Johns
Franklin -
GIIclrIAe 'IacU PuTnam -
SDlle I- FlagI
Levy ^ l

SMio
mrr M
Oemrd,


Legend
Social Security Payment in 1993

S478 4975
4975 20037
| 20037- 40512
40512-83647
83647 220434


0 62.5 125 250 Kilometers
I I I i I Ia I


odlan RIvE
Manatee Hardee Oke h.d-ee R
Highlands, Luc
DeSolo
Charlotle Glaees
Hendry








, Monroe







N


Figure 4-18. Social security payment by county of Florida in different years by county A)1993
B)1997









LIST OF FIGURES


Figure page

4-1 Total employment of Florida in different years by county A) 1993 B)1997 ...................45

4-2 Non-basic employment for primary sector of Florida in different years by county
A )19 9 3 B )19 9 7 .......................................................................... ................ 4 7

4-3 Non-basic employment for construction sector of Florida in different years by
cou n ty A )19 9 3 B )19 9 ........................................................ .............................................. 4 9

4-4 Non-basic employment for manufacturing sector of Florida in different years by
cou nty A )19 9 3 B )19 9 7 ............................................. ....................................................... 5 1

4-5 Non-basic employment for transportation and public utilities sector of Florida in
different years by county A )1993 B)1997..................................... ................... ............ 53

4-6 Non-basic employment for wholesale trade sector of Florida in different years by
cou nty A )19 9 3 B )19 9 7 ............................................. ....................................................... 5 5

4-7 Non-basic employment for retail trade sector of Florida in different years by county
A )19 9 3 B )19 9 7 ......................................................................................................................... 5 7

4-8 Non-basic employment for finance, insurance and real estate sector of Florida in
different years by county A )1993 B)1997..................................... ................... ............ 59

4-9 Non-basic employment for services sector of Florida in different years by county
A )1993 B )1997 ......................................................... ................ .. 6 1

4-10 Basic employment for primary sector of Florida in different years by county A)1993
B )1997................ ............................................................................. 63

4-11 Basic employment for construction sector of Florida in different years by county
A )19 9 3 B )19 9 7 ......................................................................................................................... 6 5

4-12 Basic employment for manufacturing sector of Florida in different years by county
A )19 9 3 B )19 9 7 ......................................................................................................................... 6 7

4-13 Basic employment for transportation and public utilities sector of florida in different
years by county A )1993 B )1997 ........................... ..... ........................ ............... 69

4-14 Basic employment for wholesale trade sector of Florida in different years by county
A )19 9 3 B )19 9 7 ........................................................... ................ 7 1

4-15 Basic employment for retail sector trade of Florida in different years by county
A )19 9 3 B )19 9 7 ......................................................................................................................... 7 3




Full Text
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID EGSTUITN8_W6SMAJ INGEST_TIME 2012-12-07T19:21:27Z PACKAGE UFE0041842_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES



PAGE 1

1 ECONOMIC BASE ANALYSIS USING PANEL DATA REGRESSION: A CASE STUDY OF THE FLORIDA REGIONAL ECONOMY By FAN LI A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIR EMENT S FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2010

PAGE 2

2 20 10 Fan Li

PAGE 3

3 To my beloved husband Wei

PAGE 4

4 5BACKNOWLEDGMENTS I would like to express my most sincere appreciation to Dr. Timothy Fik Dr. Gra nt I. Thrall and Dr. Y ouliang Qiu members of my advisory committee, for their academic support and guidance through the completion of this study. I would like to give my special gratitude to my advisor Dr. Timothy Fik who is always ready, available and patient to help with academic problems and discussion. I am also grateful for the help and encouragement from other committee members, including Dr. Grant Thrall and Dr. Youliang Qiu. They provide insight, suggestions and valuable feedback to my thesis. I am thankful to all members of the G eography D epartment, faculty staff and students for creating a friendly atmosphere that makes every day of study so enjoyable As an international student all member s make me feel as if I am living and working in my ho me country. I also express my special thank s to Dr. Peter Wayl e n and Dr. Youliang Qiu for their support and help to my computer lab job. My dear friend, Anna Maria Szynizewska, work with me together to find the data source. Yang Yang helps me with regress ion programming. I also want to thank Dr. Chunrong Ai from Department of Economics and Dr. Alfonso Flores -Lagunes from Department of Food and Resources Economics. Th eir E conometric courses enable me to apply the regression model s used in my research The re are no words that could describe my gratitude to my husband Wei If it wasn t him to encourage me to apply for Geography Department, I wouldn t pursue a graduate study at such an excellent American university. After I began my study in University of Fl orida, it is his immense love and unconditional supports that make me get accustomed to life and study abroad. Mostly I would like to thanks to his help in my thesis writing. He teaches me how to correct thesis format that saved me many months of work on this thesis.

PAGE 5

5 Finally, I would like to express my special gratitude to my family in China my loving parents and friend s for all their encouragement and support; something that I feel is very strong, even though I am thousands of mile s away from my home.

PAGE 6

6 TABLE OF CONTENTS Upage ACKNOWLEDGMENTS .................................................................................................................... 4 LIST OF TABLES ................................................................................................................................ 8 LIST OF FIGURES .............................................................................................................................. 9 ABSTRACT ........................................................................................................................................ 11 CHAPTER 1 INTRODUCTION ....................................................................................................................... 13 Overview and Problem Statement .............................................................................................. 13 Background .................................................................................................................................. 14 Research Objectives .................................................................................................................... 15 2 LITERATURE REVIEW ........................................................................................................... 17 Economic Base Theory ............................................................................................................... 17 Two Versions of Economic Base Model ................................................................................... 21 The Averag e Multiplier of the Economic Base Model ............................................................. 22 The Marginal Multiplier of Economic Base Model .................................................................. 25 The Procedure of Estimating Multipl iers of Economic Base Model ....................................... 27 Survey Method ..................................................................................................................... 27 The Location Quotient Approach ....................................................................................... 28 Minimum Requirements ...................................................................................................... 31 Regression -based Approaches ............................................................................................ 32 3 METHODOLOGY ...................................................................................................................... 35 The Location Quotient Method .................................................................................................. 35 Ordinary Least Squares (OLS) ................................................................................................... 36 Linear Unobserved Effects Panel Data M odels ........................................................................ 38 4 EXPLORATORY ECONOMIC BASE REGRESSION MODEL .......................................... 42 Initial Data Collection ................................................................................................................. 42 Regression Analysis .................................................................................................................... 44 Choosing a Scalar ........................................................................................................................ 81 Results and Discussion ............................................................................................................... 84 5 CONCLUSIONS ....................................................................................................................... 102

PAGE 7

7 LIST OF REFERENCES ................................................................................................................. 106 BIOGRAPHICAL SKETCH ........................................................................................................... 109

PAGE 8

8 6BL IST OF TABLES U Table U Upage 3 1 OLS a ssumptions and possible solution ............................................................................... 37 4 1 Scalars and non -basic employment in year 1993 and 1997 by county ............................... 82 4 2 Ordinary least regression estimation result of years 1993 ................................................... 85 4 3 Ordinary least regression estimat ion result of years 1993 ................................................... 85 4 4 Ordinary least regression estimation result of year 1993 and 1997 .................................... 87 4 5 Test result of exogenei ty of error term ................................................................................. 87 4 6 Panel data regression estimation result of year 1993 and 1997 without dummy variable .................................................................................................................................... 88 4 7 Panel data r egression estimation result of year 1993 and 1997 with dummy variable ...... 97 4 8 Multipliers of OLS regression and panel data regression .................................................. 101

PAGE 9

9 7BLIST OF FIGURES U Fi gure U Upage 4 1 Total employment of Florida in different years by county A) 1993 B)1997 ..................... 45 4 2 Non -basic employment for primary sector of Florida in different years by county A)1993 B)1997 ....................................................................................................................... 47 4 3 Non -basic employment for construction sector of Florida in different years by county A)1993 B)199 ............................................................................................................ 49 4 4 Non -basic employment for manufacturing sector of Florida in different years by county A)1993 B)1997 .......................................................................................................... 51 4 5 Non -basic employment for tran sportation and public utilities sector of Florida in different years by county A)1993 B)1997 ............................................................................ 53 4 6 Non -basic employment for wholesale trade sector of Florida in different years by county A)19 93 B)1997 .......................................................................................................... 55 4 7 Non -basic employment for retail trade sector of Florida in different years by county A)1993 B)1997 ....................................................................................................................... 57 4 8 Non -b asic employment for finance, insurance and real estate sector of Florida in different years by county A)1993 B)1997 ............................................................................ 59 4 9 Non -basic employment for services sector of Florida in different year s by county A)1993 B)1997 ....................................................................................................................... 61 4 10 Basic employment for primary sector of Florida in different years by county A)1993 B)1997 ..................................................................................................................................... 63 4 11 Basic employment for construction sector of Florida in different years by county A)1993 B)1997 ....................................................................................................................... 65 4 12 Basic employment for manufacturing sector of Florida in different years by county A)1 993 B)1997 ....................................................................................................................... 67 4 13 Basic employment for transportation and public utilites sector of florida in different years by county A)1993 B)1997 ........................................................................................... 69 4 14 Basic employment for wholesale trade sector of Florida in different years by county A)1993 B)1997 ....................................................................................................................... 71 4 15 Basic employment for retail sector trade of Florida in different years by county A)1993 B)1997 ....................................................................................................................... 73

PAGE 10

10 4 16 Basic employment for finance, insurance and real estate sector of Florida in different year by county A)1993 B)1997 ............................................................................................. 75 4 17 Basic employment for services sector of Florida in different years by county A)1993 B)1997 ..................................................................................................................................... 77 4 18 Social security payment by county of Florida in different years by county A)1993 B)1997 ..................................................................................................................................... 79 4 19 Normality test for different regression A)Year 1993 OLS regression B)Year 1997 OLS regression C)Both year 1993 and 1997 OLS regression D) Both year 1993 and 1997 panel data regression ..................................................................................................... 89 4 20 Residual plots for different regression A)Year 1993 OLS regression B)Year 1997 OLS regression C)Both year 1993 and 1997 OLS regression D) Both year 1993 and 1997 panel data regression ..................................................................................................... 91 4 21 P lots of fitted versus actual non-basic employment values for different regression A)Year 1993 OLS regression B)Year 1997 OLS regression C)Both year 1993 and 1997 OLS regression D) Both year 1993 and 1997 panel data regression ......................... 94 4 22 Map of coastal urban counties ............................................................................................... 96 4 23 Plots of fitted versus actual non-basic employment values for both year 1993 and year 1997 panel data regression with dummy variable ........................................................ 98

PAGE 11

11 Abstract of Thesis Presented to the Graduate School of the University of Florida in Pa rtial Fulfillment of the Requir ements for the Degree of Master of Science ECO NOMIC BASE ANALYSIS USING PANEL DATA REGRESSION: A CASE STUDY OF THE FLORIDA REGIONAL ECONOMY By Fan Li August 2010 Chair: Timothy Fik Major: Geography The main purpose of t his research is to investigate and estimate the increase in total regional employment as it relates to employment increases in specific economic sectors using economic base theory and a regional economic base model P ast empirical research in economic ba se modeling typical relies on single year data and original least square (OLS) regression, with unstable results that are possibly explained by the fact that the regression analysis is influenced by abnormal value change s in one or more explanatory variabl es As an alternative, this thesis utilizes panel data regression with data for multiple years in an economic base analysis of Florida employment patterns. Sector -specific employment multipliers are estimated and compared using OLS and the panel data reg ression. Of the other widely used methodologies : the survey method, location quotient (LQ) method and minimum requirement s (MR), regression analysis is the only one which can estimate impact multipliers. Regression models to estimate employment multipl iers are typically divided into disaggregate and aggregate model s The form er can obtain the multiplier estimates for specific industr ies while the latter one independently c annot distinguish between different industries. Hence, disaggregate regression model s are widely used in economic base research Unable to account for economic s pillover effect s such as cash flow s or trans -boundary monetary transfers

PAGE 12

12 disaggregate regression model s tend to have upward ly biased multipliers. It is necessary therefo re, to apply scal ing factor s and utilize independent variable s that capture the influence of monetary transfers to account for such factors as unemployment and retirement benefits and other state income sources in an effort to decrease bias in the modeling process T ransfer payment s, dividends interest and rents, and social security benefits are considered in this analysis, and prove to be effective in reducing the upward bias of multiplier estimates. Th is research also examines the assumptions of origina l least square and alternatives to test for the potential violation of assumptions. In order to obtain reasonable and st able estimat es of multiplier related coefficients and avoid potential problems associated with endogenous error terms, panel data regr ession is preferred to original least square s regression when data for more than one sample time period is available. The estimation results su ggest that employment increases in wholesale trade can create most total employment benefits in Floridas regiona l economy, while job increases in the transportation and public utility sector lead s to relatively small job returns to Floridas overall regional economy. Comparison of results between ordinary least square regression based on a single year sample and pa nel data regression using a two -year sample suggest that the latter can avoid the impact of sudden and discontinue change in a regional econom ic trend In short, the panel regression approach is deemed as preferable over ordinary least square regression i n the estimation of employment multipliers.

PAGE 13

13 0BCHAPTER 1 8BINTRODUCTION 15BOverview and Problem S tatement E mployment patterns by industry or economic sector for sub regions of a regional economy are important indicators of regional economic growth and dev elopment. For example, county employment patterns and industry mix are useful in assessing the degree to which a county is reliant on specific industries when compared to an employment mix at the state or national level s The more employment in a given s ector, relatively speaking, the more important that sector is to the economic vitality of a regional economy Furthermore, it is known that overall employment levels may be a by product of supporting or linked industries that are known to generate spin -of f employment County -level economies typically have very different industr y employment patterns than a state lev el economy. T hese patterns can offer insights as to the importance of various industries or sectors to a countys overall total employment and the extent to which key sectors have an impact on regional economic development and the expansion of a regions overall employment base Key sectors are generally referred to as the economic base of a regional economy. They are the sectors that are mos t responsible for generating the majority of linked or spin off employment. The net benefits of a regions economic base are generally distributed unevenly over space, and it is useful to gather information from many subregions as to the overall importan ce of various sectors to the regional economy at large. Florida, nicknamed the "Sunshine State" because of its generally warm HclimateH, is ranked as the fourth -most populous state in the U.S. according to t he HUnited States Census BureauH in 2008. Floridas regional economy focus es on three leading industries; namely, tourism, agriculture an d mining. In the last fifty years, more and more retired people have move d and

PAGE 14

14 relocated (permanently or seasonally) into the state. The influx of the elderly and retired brings a large number of non-employment incomes to Florida in the form of transfer payments tied to retirement accounts, social security, and interest, dividends and rents paid on investments. Nonemployment income in the form of transfer payments has had an enormous historical and geographic impact on Florida s regional economy. This impact sometimes clouds the nature of how sector -specific industries affect total employment in any given sub-region. The nature of the relationship between industrial employment and regional econom ic development in Florida is investigated in this study w ith acknowledgment that that transfer payments play a critical role in influencing employment patterns. Hence, analysis of employment patterns and the identification of a regions economic base is understood as something that can be affected by transfer i ncome. Adjusting for transfer income, this study will focus on the estimation of economic base multipliers coefficients that embody the expansion effects of key sectors or industries that contribute greatly to the states overall employment. More spec ifically, this study will seek to compare several different methods and techniques for estimating multipliers and the inter relationships between the states economic sectors u sing county level data. 16BBackground Warm weather and hundreds of miles of be aches attract about 60 million visitors to the Sunshine State every year, so tourism replaces agriculture to makes up the largest economic sector in the state economy. As such, much of Floridas economy is tied to retail trade and service s The second la rgest industry is HagricultureH. HCitrusH HfruitH, especially HorangesH, a nd various winter vegetables play major roles in supporting the regional economy. Florida produces the majority of citrus fruit grown in the U.S ; for example, in 2006, 67% of all citrus, 74% of oranges, 58% of H tangerinesH, and 54% of HgrapefruitH came from growers in Florida Hence, employment activities in agricultural sectors are known to affect regional growth directly. T he

PAGE 15

15 extent to which a given industr y affects a regions total economy can vary dramatically depending on that regions industrial mix, and its overall impact on urban and regional development is something that be estimated having knowledge of the inter relationships or inter dependencies between industries and sectors. As a regional economy is highl y dependent on revenue from both industry mix and external sources it is vital to understand the nature of the influence of key economic sectors or industries as well as the effect of things like income transfers It is also important to determine the extent to which overall employment levels are affected by key sectors or industries while accounting for the effects of income transfers. 17BResearch O bjectives The goal of this study is to identify the extent to which key sectors or industries affect overall employment levels in the Florida economy, controlling for the effects of income transfers and geographic variability in the inter relationships between economic sectors. The analysis takes into account both industrial employment patterns and underlying transfer payment impact using county level data which links total employment to employment by sector Total and sector -specific employment data were chosen as they are widely used to measure the development level of regional economy. A m ajor task of this large interdisciplinary project is to distinguish between import employment and export employment -employment related to income transfers versus that which is generated from a regions economic base (also called an export base), taking into account the transfer income factor (i.e., income generated from the in flow of retirement dollars, investments, and other sources of personal revenue) The importance of this factor can not be understated given the large proportion of elderly and retired in Florida. Thus, it is recognized that the expansion of a regions em ployment base is both tied to industry mix and non -employment or transfer income. Transfer income is a crucial part of state income, and cannot be ignored in the model. Failure to account for the transfer income

PAGE 16

16 factor would ultimately result in biased a nd unreasonable estimates for economic base multipliers and the degree to which specific industries or sectors contribute to the overall employment base. S ocial security benefits are another non-employment r evenue source. Social security payments bring i n money, stimulate demand for goods and services, and contribute to overall employment growth. Therefore, a critical assessment of how key economic sectors expand a regional economy must also address the importance of transfer incomes and social security benefits In the estimation of economic base multipliers, which reveal the extent to which a sector contributes to the expansion of the regional economy at large, several modeling and estimation techniques will be compared with controls put in place to les sen bias and/or overstatement of the impact. T he research objectives of this thesis are two -fold: (1) to develop a regional economic base model that incorporates the transfer income factor; and (2) using data for more than one time period, demonstrate the effectiveness of panel data regression as a techniques that sidesteps some of the limitations imposed by classic regression in the estimation of economic base multipliers which describe the extent to which employment in key economic sectors contribute to a regions overall total employment ( ceteris paribus ).

PAGE 17

17 1BCHAPTER 2 9BLITERATURE REVIEW Economic base theory remains a respected field of study in ge ography and regional economics Economic base analysis offers an inexpensive yet reasonabl y accurate meth od to assess small regional economic and employment impacts. The objective of economic base analysis is to calculate regional multiplier s that describe the extent to which employment or income will grow as a function of new jobs added to a regional econom y. The following section provides a brief literature review of economic base theory, including related assumptions, applicability and scale and data requirements. A ggregate and disaggregate model s are discussed, as well as the derivation of multiplies; focusing on sector -specific application and distinguishing between average multipliers and impact multipliers. The overview will highlight the four most -common ly used method s of estima ting economic base multipliers; namely, the survey method, the l ocation -quotient (LQ) method m inimum requirement (MR) method and regression -based approaches It is noted that regression -based models tend to produce impact multipliers, while the other methods are used to estimate average multipliers. 18BEconomic Base Theory The history of the economics base concept dates back to the early 1900s E arly regional economic analysts observed the duality of urban and regional economic activity which includes city -forming activities (characterized as the basic sector ) and city -se rving activities (characterized as the non -basic sector ) (Hewings 1985) The city -forming activities were said to provide the re asons for the citys existence; typically associated with export -oriented industri es that provided goods and services for markets both inside and outside the region. As these industries satisfied demand outside the region, they were responsible for generating income from sales related to exports. As such, these industries were viewed as responding to exogenous

PAGE 18

18 demand and were labeled as basic to the survival of the region, with dollars generated from export activities. By contrast, there were industries that were region serving in the sense that they were associated with the sale of g oods and service s from one company in the region to others in side the region Hence, c ity -forming activities serv iced the demand that was internal to the region (demand generated by the local population ), with money that was circulated and re circulated w ithin the regions. To sum, basic or export -oriented activities we re regarded as the city forming and a direct result of external demand, while city -serving activities were there to meet a regions internal demand for non -exportable goods and service s. Th e latter set of activities and associated employment were in place for the maintenance and well -being of the people inside the region to satisfying its internal demand. The distinction between city forming and city serving activities is important in the sense that the regional employment base of a region could then be broken down into basic employment (employment associated with export oriented activity) and non-basic employment (employment associated with meeting internal demand). There is a fundamenta l assumption, central to the notion of the duality of regional economic activity, that non-basic economic activity relies on basic economic activity (Andrews 1953) and that basic employment supports employment in non-basic sectors. This modeling framework focuses on regional export activity as the primary determinant of local economy growth. The theory also assumes that the money generated by basic industry expands as it circulates and re -circulates, creating non -basic jobs as money and income change hands an effect known as the multiplier effect (Fik, 2000) Note that the economic base model s and input output model s are different While both modeling approaches can estimat e the impact of a change in employment the former yields an aggregate overall impact, while the latter yields sector -specific impacts (Billings 1969) Moreover, ec onomic base model s are computationally

PAGE 19

19 less expensive in comparison to input -output m odels a feature that explains why they are so popular in the analysis of regional economic growth (Malecki 1991) Acc ording to economic base framework total economic activity, TE is assumed to be divided into non -bas ic economic activity, NE and basic economic activity BE where NE is a function of BE More formally, B N TEE E (2 1) ) (B NE f E (2 -2 ) Note th e nar row focus on exports and basic economic sector s as the engine of local economy makes economic base theory lack the complexity to offer an adequate framework for analyzing regional and subregional economic issues. The focus on e xports only consider the de mand side and excludes important supply -side factors (Blumenfeld 1955) Hence, the assumption that basic economic activity is the predominant driver o f regional economic change minimizes the contribution that non -bas ic e conomic activity makes to regional growth and development (Tiebout 1956) In an effort to expand the economic base model, other factors such as transfer payments have been added (Mulligan 1987), as well as internal population growth (Leichenko 2000) Discussions of the counter -veiling force of leakage lost income due to purchases made outside a regional economy by consumers who live in that region (F ik, 2000). M any researches argued that economic base model s are limited in their ability to forecast long run regional economic impacts as they focus on short run expansion effects tied to export oriented income generation while ignoring other important growth factors Hildebrand and Mace (1950) apply the familiar Keynesian equation: ) ( M X G I C Y (2 3)

PAGE 20

20 where Y denotes total regional income, which is divided into four parts: C consumptions; I, investment; G government expenditures; and ) (M X exports minus imports. The Keynesian equation suggest s that regional export activity is only one of four other factors C, I G and M that explain regional growth and development ; and hence, exports are not the only thing driving economic growth. When r elative factors C, I, G and M are held constant economic base theor y and the Keynesian equation are congruent Note that t he short run economy always assumes that those four factors remain constant, so economic base model s can be applied rather effectively to estimate short run changes, but should not be used to produce long -run forecast s The additional factors considered by Keynesian equation are re lated to both demand and supply. Thus, a more detailed regional economic model would need to incorporate supply-side factor s to anticipate long -term changes; thus, making i t non traditional in the sense that it would also rely on su pply side patterns and trends. Theoretically, the most important determinate of a region s longrun development is the ability to attract capital and labor from outside into region. Such supply accumulation would in turn, stimulate export -oriented production sectors, thereby augmenting export activity and bringing in exported-related revenues and income (North 1955) All in all, t he connection betw een the economic base model and the Keynesian equation has to do with differences in social accounting techniques and the degree to which each model may be used depending on the time horizon (short run versus long run), the geographic scale (the areal cove rage), and the type of impact assessment or forecast required. Roberts (2003) applied a n accounting model to q uantify the relative importance of traditional and non -traditional elements of economic base theory for local and regional economies in rural areas. Due to its inherent simplicity of theoretical foundations, economic base theory nevertheless, does wel l to explain regional economic growth over the short run. Input output

PAGE 21

21 model s are typically preferred in cases involving complex inter regional econom ic inter dependencies Unfortunately there is no clear boundary between two kinds of methods in terms of which would be preferred as one increases the geographic scale of the analysis (Mulligan, 2009). It is safe to say that for short run analyses, a traditional economic base analysis should suffice if one operates under the assumption that factors such as Consumption, Investment, Government expenditures, and Imports remain fairly constant over the period i n question. 19BTwo Versions of Economic Base Model E conomic base model s can be divided into two varieties : aggregate models and disaggregate models The distinction between them is that the aggregate model considers the regional economy as a whole while the disaggregate model identifies sectors or industries of interest in a regional economy (Vias, Mulligan 1997) In additional to the model definition given in equations (2 1) and (2 2 ), the following definitions are re quired for a disaggregate version of the regional economic base model : Bi n i BE E1 (2 4a) Ni n i NE E1 (2 4b) Ti n i TE E1 (2 4c) Bi Ni TiE E E (2 5) as defined for i = 1,, n sectors, where the various relative industry-specific (i th sector) figures are summed to arrive at economy -wide figures for non -b as ic employment, basic employment and total employment.

PAGE 22

22 The a ggregate version of this model concentrates on the entire regional economy, and as such fail s to distinguish between the various impact s of basic economic activity change on specific sectors or different industries that fall under the rubric of non -b as ic economic activity. The aggregate model does identify dissimilar ities of influence or the external industry -specific demand on loc al economy (Mulligan 2009) Input -output model s have clear ly demonstrate d that regional changes in different export industries are likely to have different effects on a regional economy should one be able to trace the flow of income associated with exports and external demand to inside economic activity. Disaggregate regional economic base models represent an acceptable compromise between aggregate models and the more computationally intensive and information -dense input -output model It extends the traditional economic base framework to include industryor sector specific level s a feature that has allowed these models to gain a broader acceptance (Loveridge 2004) Income data or employment data are widely used by researchers to measure the level of regional economic activity by sector, thereby allowing the estimation of sector -specific multipliers which describe the expected expansion in total income or employmen t for a given change in income or employment in a given sector. Although it is recognized that employment data fails to account adequately for productivity or wage/ e arnings differences between workers employed in different industries (or even differences within the same industry or sector, across various firms), employment data are widely used given their availability. 20B The Average Multiplier of the Economic Base Model After distinguishing basic and non -basic economic activity /employment the economic ba se model is then used to calculat e an economic base multiplier. The average regional economic base multiplier BM for the entire economy is calculated as:

PAGE 23

23 B N B T BE E E E M 1 (2 6 ) w hich implies that each unit change in basic employment will lead to an BM employment change in total employment and a B NE E change in non -basic employment It is a ssumed that t he non -base/basic employment ratio B NE E remains the same and that the impact trend is embodied in the multiplier BM Practitioners frequently apply this multiplier to forecast the total employment impact of the new establishment or the expansion of existing facilities. The multiplier is also used to predict the total employment changes following an increase or decrease of export activity (and hence Describes the expansion or contraction potential of an empl oyment change associated with a change in exports, respectively). Large multiplie r values indicate that export employment plays a predominant role in the regional economy in terms of its ability to generate ( or lose) jobs in the non -basic sector(s) Once an estimate of an economic base multiplier is available, it is possible to predict the future impact of a change in export -oriented employment on the total employment within a regional economy (and the overall change in the employment in nonbasic sectors). More specifically, the change in total employment is equal to the multiplier times the change in basic employment: B B TE M E (2 7 ) For example, if the current economy has 2 000 full -time workers employed in basic (export -oriented) activities and 1, 000 workers in non -basic basic the multiplier should be 1 .5. Then 50 new export jobs would create 25 non -basic jobs and 75 jobs i n total. Equation ( 2 7 ) supposes that the current BM can reflect the return of new basic employment change to total

PAGE 24

24 employment. However, this impact might be understated as there is no evidence to support th e contention that the multip lier remains constant during the post -impact adjustment period. In fact, empirical evidence suggests that the calculation of the multiplier in equation ( 2 6) will tend to underestimate the multiplier effect and the overall economic impact. Therefore, th e average multiplier is widely applied in static regional econom ies rather than impact situation s were an economy is experience rapid growth A shortcoming here is that the aggregate analysis cannot identify which specific industries are associate d with a n additional X number of nonbasic jobs though it is generally assumed that the expansion will take place in linked or interdependent sectors. By contrast, disaggregate economic base model s benefit from the fact that they can produce estimates of employ ment change associated with specific industry. If the industry -specific assignment of additional non -basic job is equal with the current employment share of industry to total economy, the expected change in local employment in an i th industry is : N N Ni NiE E E E (2 8 ) So if the i th industry is responsible for 200 of 1, 000 nonbasic jobs in the current economy, the practitioner expects that this industry to be allocated .20 (or 20%) of new X non-basic jobs added to the economy. T h is assumption is also suspect given that shift of export employment would simultaneously result in various industry employment shifts in the mix of non -basic activities. A ggregate economic base models cannot distinguish between various shift s that might take place in different industr ies or sectors as a result of an expansion of the regional employment base associated with different basic sectors If the multiplier is equal to 1.5 a shift of 50 BE export jobs both in 1st industry and in 2nd industry has the same anticipated result of

PAGE 25

25 25 new nonbasic jobs Empirical evidence and common sense suggest that external demand of two different industries will most likely produce dissimilar impacts in the regional econom y. A verage economic base multiplier s that depend on aggregate models do not fully express the various influence s of external demand on the local economy for goods and ser vices produced by different industr ies nor do they explain the different impact s of industry -specific export shift s on a region s economic /industry mix By contrast average multiplier obtained from the disaggregate models provide better overall explanato ry ability in terms of their ability to estimate industry-specific difference s and change A verage economic base multiplier s, however, cannot describe or forecast the complex dynamic s of regional economic change as they relate to specific sectors for chan ges that may be occurring simultaneously, and therefore, it is necessary to turn to marginal economic base multiplier s as a way to express regional adjustment and related impacts in association with a given change in a regions employment base. 21BThe Ma rginal Multiplier of Economic Base Model Assume that a direct linear relationship exists between export -oriented activity and locally oriented activities then following: B NbE a E (2 -9) w here a is the autonomous component of non-basic employment and b is the marginal returns of basic employment to non-basic employment. Hence, the margin al multiplier is: b E E MB T I 1 ) ( (2 10) The major difference between multipliers BM and IM i s that the former represents the average ratio while the latter one is marginal one. As a result, BM may be applied in static economy, whereas IM may be applied in a dynamic econom ic setting The marginal mult iplier is more valuable as a forecast ing tool for economic impacts in comparison to the average

PAGE 26

26 multiplier P ractitioners typically apply th e marginal multiplier s to forecast which industry obtain the additional local employment as driv en by the entire export -oriented e mployment shift and to estimate the new local job creat ion that is linked to changes in industry -specific export employment The calculation of b oth BM and IM require estimation of the disaggregate model. Lets spe cify a liner relationship between total export activity and industry -specific local activity This can be formally expressed as B i i NiE b a E (2 11) where ib is the marginal returns of entire basic employment to industry -specific non -basic employment. Following th is logic the marginal multiplier is bi E E MB T Ii 1 ) ( (2 12) and represents the local employment change in an i th industry result ing from the each unit shift in the total number of basic job s Alternatively, a line a r relationship between industry -specific export employment and total local employment may be specified a s BN N B B B NE b E b E b E b a E ...3 3 2 2 1 1 (2 13) where ib is the marginal returns of industry -spe cific basic employment to total non -basic employment as defined for i = 1,.., N industries The marginal multiplier from this equation is defined as bi E EMB T Ii 1 ) ( (2 14) and represents w hich the total local employment change associated with each unit shift of basic employment in an i th industry In short, either of these two methods could be used to estimate the marginal multi plier.

PAGE 27

27 22BThe Procedure of Estimating Multipliers of Economic Base Model In order to estimate a regional economic base multiplier, it is important to determine how much total employment in a given region is basic employment and how much is nonbasic employment. There are numerous methods to arrive at numerical values or estimates for basic and non -basic employment. 30B Survey M ethod The Survey M ethod relies on a comprehensive survey of employer s of all firm s in a region to identify approximately how much of each firm s income is obtained from external sale s versus local sale s The proportion of sales revenue from export sales is assumed to be equal to the proportion of basic employment. Suppose a firm generates 30% of its revenue s from exports sales compa red to 70% revenue s from local sales and 150 full time equivalent (FTE) employees (that is, a number of employees that total to an equivalent of 150 full -time employees) Under such a scenario, it is assumed that there are 45 basic -sector employees and 105 non-basic sector employees (or 30% basic and 70% non -basic employe es ). Note that s tandardized employment data must be transform ed into FTEs as employment is typically divided into three types: full time, part -time and seasonal workers. All employmen t data should be converted to a full -time equivalent (FTE) employment scale ( Gibson and Worden 1981). P art time employment is converted to FTE employment by summing the hours each employee works and then dividing by 40 (assuming a standard workweek of 5 days at eight hours per day or 40 hours) The number of weeks that s easonal employees work must be summed and divided by 52 to estimate its contribution to FTE employment. When the data are in FTE format formula (2 6) can be applied to estimate the econo mic base multiplier. As mentioned before, this approach can only be sued to estimate an average multiplier The s urvey method is a least preferred method as it requires a comprehensive survey of all companies in a study region ; so it

PAGE 28

28 i s extremely time -co nsuming and costly. Th e quality of estimates obtained from this approach depend s greatly on the quality and accuracy of answers provided by survey respondent s. Many companies may be reluctant to reveal their sales data, and hence, there is a potential fo r under reporting or for firms to be uncooperative (Harris, Ebai & Shonkwiler 1998) Nevertheless, the survey method is the most direct way to gather information on non -basic and basic employment and does not rely on complex statistic al technique s or mathematic al transformation s. It is a technique that is effective in small communities or those settlements with a small number of firms 31B The Location Quotient Approach The l ocation quotient (LQ) method of esti mating economic base multipliers is more elaborate in its design. The approach begins by assum ing that that a given sub-region requires a level of economic activity in an particular sector or industry that is directly proportional to the national or regio nal economy at large (Isserman 1977a) Specifically the method compares a regions observed industr y employment or income concentration with that of a benchmark ; typically the concentration of employment or income that industry in the state or nation in which the region is located in The location quotient is a ratio of ratios, and may be defined as the ratio of employment in sector i for region j divided by the ratio of employment in sector i for an economy at a larger geographic scale. Specifically, e e E E LQi j ij ij (2 15) where j ijE E and e ei are the ratio s of employment in sector i to the total employment in sub region j and the larger geographic region, respectively. Once the value of ijLQ is estimated, non -basic employment be easily identified. If 1 ijLQ it means the proport ion of i th sec t or employment to

PAGE 29

29 the total employment in a j th region surpass es the relative proportion at the economy at the larger geographic scale; indicating that the good or service associated with an i th sector is enough to satisfy the local demand and there is a surplus for export to meet demand outside the sub region. In this case, it is assumed that a portion of that sectors employment can be classified as basic employment. If 1 ijLQ the proportion of employment in a sector is equal in both the subregion and the larger region and it is assumed that the good or service associated with an i th sector just satisf ies local demand and there is no surplu s production for export ; and hence, no basic employment If 1 ijLQ the good or service associated with an i th sector is assumed to be less than what is needed to satisfy local demand (i.e., demand with the subregion). As a result, it is assumed that the sub region needs to import from other regions Under such a scenario, it is assumed that all employment in that sector is non basic (or non export oriented) ; and hence, there is no basic employment in that sector Sum marizing the above analysis estimates for non -basic employment are as follows: Tij ij NijE LQ E*) / 1 ( if 1 ijLQ ; (2 16a) Tij NijE E if 1 ijLQ (2 16b) The economic base multiplier can be estimated from the following equation n i Bij Tij n i TijE E E M11) ( (2 17) T he location quotient method is widely used to estimate regional economic base multipliers especially after the latter -half of the 1940s. Applying this methodology requires only industry em ployment or income data for the region and a similar data set for an appropriate benchmark economy. Compared with survey method, it is more feasible and less expensive (Isserman 1977b) However there are some di sadvantage s The underlying assumption of this method is

PAGE 30

30 weak as it is asserted that different economic region s adhere to the same benchmark ; that is, have the same minimum demand for a goods and service s from a given industry or sector. In short, it ass umes that sector demand, in a relative sense, is spatially invariable across sub regions and is similar to that of the larger benchmark economy. The choice of a benchmark economy can also impact the accuracy of th e estimates; and a question arises as to w hether it is advisable to use a regional or national economy as a benchmark for a sub region; especially as that region may be highly specialized in terms of its industry mix and/or have a comparative or competitive advantage in production that allows it a chieve say economies of scale or exploit its geographic location When a regions economy is compared with two benchmark econom ies s ay at the state and national level s estimated basic and non -basic econom ic activity usually var ies, and there is no direc t indicator for which estimate is more reflective of the regions economic position. Unfortunately, no wildly applied standard is available by which to evaluate the benchmark s Consider the urban economy of Miami Florida a node that is recognized as a global city because of its importance in several sectors including finance, commerce, media and entertainment and international trade. Miami is radically different in terms of its industry mix tha n say the capital of Florida -the city of Tallahassee, which focus es on education, government administration and agriculture. If the two cities econom ies are compared with the state level, they will share the same benchmark to distinguish basic and non -basic economic activity in all sectors Yet o bviously, Miami has a higher local minimum demand for finance, insurance, real estate, retail trade and service, while Tallahassee has higher demand for public administration. This inherent flaw can lead to multiplier estimates that are exceedingly large and over -estimated. As an alternative benchmark by which to compare these two very different regional econom ies one could utilize information on sector activity for northern Florida and southern Florida and

PAGE 31

31 obtain far more reasonable results as there would be les s of a contrast in economic activity levels across various sectors Another disadvantage of th e LQ method is that it fails to estimate the marginal economic base multiplier. Recent ly, there has been an attempt to improve on traditional location quotients estimates through the specification of the dynamic location quotient (DLQ) method This approach was developed to decompose regional employment into base and non -basic components across multiple sectors using spatial and temporal data (Lego, Gebremedhin & Cushing 2000) 32B Minimum Requirements The minimum requirement s (MR) method was popularized in 1 960s and 1970s (Moore 1975) It is similar to the location quotient metho d in terms of distinguish sector activity by compari ng activity levels in the study area to a predetermined benchmark. In the minimum requirement s method groups of regional economies or sub regions with approximately similar sizes are compared, and a sec tor s minimum requirement is assumed to be the minimum level of activity in those regions. It is assumed that this minimum level of employment observed within the group is a level that satisfies internal demand Furthermore, it is assumed that sub regions with greater activity than the observed minimum for a given economic sector must be exporting goods or s ervices Excess sector employment ( above the sector minimum ), within a given size class, is assumed to be basic employment. The minimum local benchm ark for a given economic sector, for a given size class, represents minimum non -basic employment i nEmin) ( Once these values are determined for all sectors, the economic base multiplier is found from the following e x pression: n i i N Ti n i TiE E E M0 (min) 0) (. (2 18)

PAGE 32

32 As with the methods discussed earlier, the minimum requirement s approach can only estimate the average multiplier. The classification of th e similar range regional economy makes a significant influence on the accuracy of final result, so it is critical that care be taken to find regional econom ies of s imilar size though the same criticism can be levied as there is no reason to believe that s imilar sized economies should have a similar industry mix. Some regions have an inherent advantage based on location, natural resources, and/or other production advantages, and even if similar in size will not necessarily share similar sector -specific pro duction levels or similar sector -specific demand R esearcher s continue to discus s standard s by which to evaluate economies of similar size economy, but no agreement has been reach on exactly how to go about justifying the minimum requirements benchmark. Nonetheless, tremendous strides have been made in the estimation of multipliers using regression models of various kinds. 33B Regression -based Approaches Regression analysis has been used effectively in the estimat ion of regional economic base multipliers yet is known that the type and quality of estimates depend on the specific regression approach being applied regression model. The aggregate and disaggregate model s represented by equation (2 9) (2 11) and ( 2 13) can all be estimated using ordinary least square (OLS) regression Th is requires a survey and the gathering of a sample data and the n the use of OLS to provide estimates of the coefficient s of the model. M ultipliers are then calculate from the estimated coefficients using equations (2 10), ( 2 12) and ( 2 1 4 ). T ypically, multiplier estimates will be dramatically different from the method described above. Nevertheless the regression based approaches have become quite popular in the economic base theory literature. The regression -based approach es are also expandable, as illustrated by the seminal work of Mulligan and Gibson (1987) Mulligan and Gibson emphasized the importan ce of including transfer payments and have shown that the employment -deriv ed estimates of economic base multiplier s

PAGE 33

33 tend to be biased upward unless the transfer payments are directly specified in the regression model. Mulligan and Fik (1994) extend ed previous research using the Arizo na Community Data set and distinguished between multipliers for different types of communities by introducing dummy variables into the regression models The estimation result s have also been shown to be depend ent on Meta analysis (Vollet, Bousset 2002) In this approach, a host of variables are used to reflect characteristics of study regions and account for variability in population size and the magnitude of the employment base, a s well as the physical size of the study regions; and in particular, how the economic base is determined, how it is expressed, and which statistic method is applied, and how it might impact the accuracy of the model based on a multitude of variables As mentioned previously, the funda mental assumptions of economic base theory are at odds with the Keynesian equation something that researchers have made great effort s to correct in time series econometric application s One very influential work was the use of bi -variate vector autoregre ssion (VAR). In this approach, sector employment levels were analyzed in the state of Ohio, and its economic base w as derived and evaluated using Granger causality tests and interpreted dynamic base multipliers by impulse response functions (Lesage, Reed 1989) Although Krikelas (1992 ) argued that the identifying restrictions required to derive multiplier estimates become arbitrary and the estimation procedure becomes unstable as the number of sectors included in a VAR model is increased. Nevertheless, a dditional research in this area support s the vector auto -regression approach. A n interregional trade model from which testable parameter restrictions for economic base theory w ere shown to be equivalent when subject to Granger causality test s suggest that economic base theory holds strongly for the crudest definition of the base, in a study where the data set included the states of California, Massachusetts and Texas (Nishiyama 1997)

PAGE 34

34 Time -series procedures have also been applied in the estimation of econ omic base multi p liers, with m oderate success S eparat ing the employment base into at least two sectors a shortcut method has been developed that utilizes both a structur al ad just ment proced ure and a time adjustment procedure (Was and Mulligan, 1999; Mulligan, 2009). The first procedure recognize s that each industry has a sector -spec i fic assignment to the local employment base that rel ies not only on the size of the regional e conomy but also its industrial mix and the degree of industry concentration. The second procedure applie s a time variable to control for dramatic growth or depression of an economy in a particular year, using a variant of the VAR method. These regression -based approaches represent the current trend in economic base research. They represent flexible and expandable strategies for the estimation of coefficients and marginal economic base multipliers and are m ethodolog ies that have potential forecasting abili t ies. Regression -based procedures also offer the capacity to added more explanatory variables, such as supply -side factors, regional characteristics, transfer payment s and spillover incomes, leakage, structural adjustment considerations, and time.

PAGE 35

35 2BCHA PTER 3 10BMETHODOLOGY This chapter discusses the methodology utilized to fulfill the research objectives stated in Chapter1. In the first three sections of this chapter, employment data are used estimate economic base multipliers using the location quoti ent (LQ) method to distinguish between basic and nonbasic employment, and a comparison of multiplier estimates from ordinary least squares (OLS) regression and a linear unobserved effects panel data (PD) model. At the conclusion of this chapter, alternat ive research models are proposed, with elaboration on the selection of variables and potential data source s 23BThe Location Quotient Method The Location Quotient (LQ) method is common ly used to calculate an average economic base multiplier. For our purpo ses here, it represents preliminary stage, as typically applied w hen no industry -specific data regarding basic versus non -basic employment is available In short, the LQ method is a tool t hat provides sector -specific estimates for bas ic and non-basic empl oyment under the usual assumptions. Given the lack of survey method data for counties in the state of Florida, the LQ e stimates will serve as important input f or a regression -based approach that compares the results of alternative estimators The focus i s on economic activity levels at the county level, with the state of Florida used as a benchmark economy. It is well known that LQ multipliers typically give large and bloated estimates when run at the regional level given that much of sector -specific demand is satisfied by imports or supply coming from outside the region and that some sectors are exporting across the regional boundary to satisfy non -internal demand (Chiang 2009) Hence, it may be prudent not to u se the actual location ratio as a benchmark as the LQ method would undoubtedly produce overstated multipliers. In order to make the multiplier estimates more reasonable a scaling factor could be employed. Any attempt,

PAGE 36

36 however, to scale down the LQ multiplier estimates would run the risk of being viewed as subjective as the choice of a scalar is not readily clear and for the simple fact that the scalar may differ across locations. As such, we will turn to OLS estimators which rely on various statistic al properties in the estimation of multipliers. 24BOrdinary Least Squares (OLS) This section introduces the single -equation linear model and the ordinary least squares (OLS) estimato rs It is the most basic estimation procedure in econometrics yet it is known that accurate application of OLS depends on meeting a series of stric t assumptions (Hayshi, 2000) Once one or more assumptions are violated, OLS method cannot obtain reasonable estimation results. Consider a multivariate regression model of the va riety i ik k i i ix b x b x b a y ...2 2 1 1 (2 19) where b s are unknown parameters to be estimated a nd i are unobserved error term s with certain properties (as stated below ) Note that the b s represent the marginal and separate effects of regressors (the independent variables of the model) The error term represents the part of the dependent variable left unexplained by the regressors. It can consist of a variety of things, including omitted variables and measurement error. An explanatory variable kx is said to be endogenous in the regression equation if is it correlated with It is assumed that kx is exogenous in regression equation. Endogenous variables are a co mmon problem in OLS regression, and can lead to violation of the assumptions of OLS regression. Five assumptions of OLS may be summarized. First, the relationship between the dependent variable and the regressors is linear; i.e., the model is linear in its parameters (or can be specified in a linear form such as the model shown above). Second, variable matrix is full rank, which implies that each column o f regressors is linearly independent (assuming that multi -

PAGE 37

37 collinear relations are not a problem) Third, the error term is strictly exogenous, stated as formula 0 ) ,..., (2 1n ix x x E which indicates that regressors do not give information about the expec ted value of the disturbance terms in the model. Fourth, the error term s satis fy conditional homoscedasticity, where 2 2 1 2) ,..., ( n ix x x E (i.e, the error structure is such that it is of constant variance) Fifth, the error term s are not dependent that is, error is not spatially or serial ly auto -correlated; and as such error covariance o x x x En j i ) ,..., (2 1 The fourth and fifth assumptions define the format of the error term covariance matrix with diagonal elements equal to 2 and all nondiagonal elements equal to 0 (Greene 2000) After OLS regression is applied, it is necessary to (a) check for outliers, (b) validate that the error terms are normally distributed, (c) verify that multi -collinearity is not a problem and (d) test for serial or spatial autocorrelation in the residuals When all assumptions of OLS are satisfied the estimators are said to be BLUE -best linear (efficient and/or consistent) unbiased estimators However, it is common that one or more assumptions are violated. When this occurs it is useful to apply a remedial measure or an alte rnative estimation technique. Table 3 -1 provides an overview of O LS assumptions a nd possible solutions and/or remedial measures Tab le 3 1. OLS a ssumptions and p ossible solution Assumption Violation Solution Linearity Nonlinear ity Transform ations (log, square, inverse etc. ) Full rank Multi collinearity Delete one or more independent variable Exogen e ity Endogen eity I nstrumental Var iables or TSLS Homescedasticity Heteroscedasticity GLS(WLS) or Robust Regression No A utocorrelation Temporal and/or Spatial Autocorrelation GLS or spatial/temporal t ransformation Autoregressive models Spatial Econometric model

PAGE 38

38 25BLinear Unobserve d Effects Panel Data Models This section highlights the advantages and problems of using panel data and examines two common m ethods for estimat ing linear unobserved effects in panel data (PD) models. Before proceeding to review the panel data modeling approach, a short review of various definitions as they pertain to regression data is necessary to panel data in its proper perspective. A Hdata setH containing observations on a single phenomenon observed over multiple time periods is called a Htime seriesH A data set containing observations on multiple phenomena observed at a single point in time is called Hcross -sectionalH (where each individual sampling unit, data point or polygon is observed only once) A data set containing observations on multiple phenomena observed ove r multiple time periods is called panel data ( which implies that each individual sampling unit, data point or polygon is observed in more tha n one time period) Panel data analyses do not require that the time period in which different individual s or s ampling units are observed are exact ly the same. The major factor that distinguish es time -series data and cross -sectional data from panel data is that both time -series and cross -sectional data are one-dimension al while panel data is two dimension al (give n that it is bo th spatial and temporal, and unrestricted in the temporal domain). Time -series data suffers from its lack of spatial sensitivity and requires that observations be equally spaced in time. Cross -sectional data analys e s do not adequately capt ure the dynamics of inter relationships amongst variables as they occur across time and largely ignore the possibility of consistency of phenomenon across the temporal domain Cross -sectional data for a single time period cannot adequately capture the agg regate effects inherent amongst variables or phenomena and the inter relationships that spans across m ultiple time periods. P anel data offers an alternative that is widely used in economic analysis and can be viewed as offering information that is spatial temporal.

PAGE 39

39 Consider the following l inear panel data (PD) model expressed as it it it i t i t itx b x b x b a y ...2 2 1 1 (2 20) where itx is a K 1 vector tha t contain s observable variables that change across time t but not i variables that change across i rather than t and variables that change across i a nd t (Wooldridge, 2002). The error term s can be denoted as it i itu c (2 20a) where ic is the part of error term that change s across the individual rather than time, while itu is the part of error term that change across both indi vidual and time. Wooldridge (2002) states that the component i c could be treated as a random variable or as a parameter that is to be estimated. Accordingly, ic is divided into a random effect and a fixed effect, where the former means zero correlation between the observed explanatory variables and the unobserved effect; that is, 0 ) ( i itc x Cov ; wh ile the latter means that correlation between the observed explanatory variables and the unobserved effect is non-zero and allowed: 0 ) ( i itc x Cov In short, two types of regression possibilities apply here: random effect methods and fix ed -effe ct methods Recall th at the fifth assumption of OLS regression states that no serial auto -correlation is found in the error terms of the model, or 0 ) (2 1 x Ei i For the two time period panel data model, because ic is unchange d across two time periods 0 ) ( ) ( ) ( ) ( ) ( ] ) [( ] ))( [( ) (2 1 1 22 1 1 2 2 1 2 1 x c c E x u u E x u c E x uc E x c c E x u u c u u c c c E x u c u cE x Ei i i i i i i i i i i i i ii i i i i i i i i i (2 21) where the assumption is that 0 ) () ( ) ( ) (2 1 1 2 x c c E x u uE x u c E x u c Ei i i i i i i i

PAGE 40

40 Viewed in this light, it is apparent that OLS regression cannot obtain consistent estimators It is necessary therefore, to find an alternative method to OLS regression. Two common ly applied alternatives are random -effects methods and fixed-effect methods. Both of the se methods assume that 0 ) ( i i itc x u E The random effects method assume s that 0 ) ( ix ci E while the fixed -effect method do es no t retain any such assumption. Random effects method s belong to the family of general least square s (GLS ) estimat ors ; techniques which utilize an unrestricted variance estimator as a transformer as a way to overcome p roblems related to serial autocorrelation. The key issue is whether the random effects method should be applied in the panel data model given the assumption that 0 ) ( i i itc x u E (Baltagi 1995) In an econ omic base analysis using say two -year s of data, i c can be considered as part of various regional location factors such as the weather a random effect that is largely uncorrelated with industrial employment or export orientation. As such, the random -effects methods can be used within a PD context Fixed -effects methods apply a n internal transformation, where i it i it i itu u b x x y y ) ( (2 21) and wher e T t it iy T y1 1 T t it ix T x1 1 and T is the number of total periods. Note that the component ic is deleted in the transformation process. This transformation is necessary to ensure that the error te rm satisfies the assumption s of OLS regression; allowing OLS to obtain consistent estimators Fixed -effects methods do not a ssum e that 0 ) ( i i itc x u E Note, however, that the transformation depends on the difference between equation from one period and the average.

PAGE 41

41 Hence, it is highly possible that the value of ) (i itx x is small but the value of i itu u is relatively large. If so, it would lead to a bias in estimation, something that would be especially pronounced when the individual or sampling unit does not change much across time periods. To illustrate the effectiveness of these methods, a two-period exploratory economic base model is constructed and estimated using Florida employment data at the county level for the year 1993 and year 1997.

PAGE 42

42 3BCHAPTER 4 11BEXPLORATORY ECONOMIC BASE REGRESSION MODE L This chapter presents the statistical results of an exploratory economic base analysis of the Florida economy using county level data (for n=67 counties) The first s ecti on provides an overview of the employment data used in this study and the estimation of basic and non-basic employment Section two discusses the potential overestimation problem that is inherent to economic base studies, and offers a potential solution the use of various scal ing factors to deflate non -basic employment estimates Lastly, the results of the regression analysis are highlighted; comparing those obtained from ordinary least squares (OLS) regression and panel data (PD) regression. 26BInitial Data Collection Florida county -level industrial employment data were collect from the U.S. Census Bureau website Standard Industrial Classification of employment figures were obtained for nine separate industries /sectors : (1) agricultural services, fore stry, and fishing; (2) mining; (3) construction; (4) manufacturing; (5) transportation & public utilities; (6) wholesale trade; (7) retail trade; (8) finance, insurance and real estate ; and (9) services, a s well as employment figures for unclassified estab lishments. Previous research, such as the economic base research of communities in Arizona, combines employment in agricultural services, f orestry and fishing, and mining; defining it is as primary sector employment Th is study follows this convention Using the reported employment data estimates for basic and non -basic employment were obtained using location quotient (LQ) method A location quotient was calculated for each county and each i ndustrial sector using state level data as a benchmark Recal l that the LQ method relie s on the formula s list ed in equations (2 15) to (2 17), where the ratio i s the proportion of a j th countys industry employment in a given i th sector; a value that is compare j ijE E

PAGE 43

43 to the proportion of employment associated with tha t sector as reported for the Florida economy at large, The LQ method is then used to estimate the amount of basic and non -basic employment for each sector and each county. These estimates were then used in a disaggregated model to estimate sector -speci fic economic base multipliers using various regression -based models. Total employments by county of year 1993 and 1997 show in Figure 4 1 Estimated non -basic employments for each sector by county o f year 1993 and 1997 show from Figure 42 to F igure 4 9. E stimated basic employment for each sector by county o f year 1993 and 1997 show from Figure 4 10 to F igure 4 17. Accoriding to Figure 4 1a and Figure 4 1b, Orlando, Jaxsonville, Tampa, Miami, St.Petersburg, West Beach and Boca Raton areas have relative lar ge total emploment compared with other regions. Nonbasic employmenet and basic employment for sector from F igure 4 2 to F igure 4 17 base on the location quotient method. From Figure 4 2 to F igure 4 9 show that non basic emploment of different sectors have disimilar spatial distribution by county, which is probably not consistent with the spatial distri bution of total employment. In Figure 4 2A Broward, Miami Dade, Palm Beach, Orange, Pinellas,Hillsborough, Brevard and Duval counties have largest total emp loyment in 1993, however, on on hand, Brevard has medium nonbasic transpor tation employment in 1993 from Figure 4 5A on the other hand, largest non -baic s erives employment in 1993 from F igure 4 9A is consitent with total employment.Therefore, the change of non-basice employment of differents sectors should have disimilar return to total employment. Some of the same sectors of year 1993 and 1997 shows disimilar employment distribution by county, which implies that employment distribution by county of singl e year cannot characterise that in long term. Therefore, panel data regression of mutliple years should instead regression of single year to estimate the steady trend of regionl employment. e ei

PAGE 44

44 Figure 4 18 shows social security payment of year 1993 and 1997 by county, which shows same spatial distribution by county across years. 27BRegression Analysis The disaggregate regression model which shows non-basic employment as a function of basic employment by sector can be expressed as Soci b E b E b E b E b E b E b E b E b a EBserv Bfina Breta whol B tran B Bmanu Bcons Bprim N 9 8 7 6 5 4 3 2 1 (4 1) where denotes total non -basic employment in a j th regional economic or county independent variable s are the estimated bas ic employment of each industry/sector : agricultural s ervices, forestry, fishing and mining (prim) ; const ruction (cons) ; manufacturing (manu) ; transportation and public utilities (tran); wholesale trade (whol) ; retail trade (reta); finance, insurance, and real estate (fina) ; and services (serv) Note that the model is also expanded to include social security benefits (soci) in thousands of dollars; a variable that is used as a surrogate for income that create s a cash flow to regional economy and an expansion in sales and services The use of this surrogate helps to account for a source of state income that c an augment the non -basic sector This especially important in the state of Florida were such a large proportion of the population is 65 years of age or older, many of whom are retired and drawing Social Security payments. T he sample database consists of d ata for n= 6 7 Florida counties and data for two distinct time periods (1993 and 1997) If one -year samples ( for either year 1993 or year 1997) are applied, the OLS regression model can be used. However, to obtain more stable and consistent estimation of s ector -specific economic base multipliers data for these two time periods can be combined using the panel data regression approach Before this can be accomplished, the dependent variable (county level non -basic employment) must be scaled back to account for

PAGE 45

45 A Figure 4 1 Total employment of Florida in different years by county A) 1993 B)1997

PAGE 46

46 B Figure 4 1 Continued

PAGE 47

47 A Figure 4 2 Non -basic employment for primary sector of Florida in different years by county A)1993 B)1997

PAGE 48

48 B Figure 4 2 Continued

PAGE 49

49 A Figure 4 3 Non -basic employment for construction sector of Florida in different years by county A)1993 B)199

PAGE 50

50 B Figure 4 3 .Continued

PAGE 51

51 A Figure 4 4 Non -basic employment for manufacturing sector of Florida in different years by county A)1993 B)1997

PAGE 52

52 B Figure 4 4 Continued

PAGE 53

53 A Figure 4 5 Non -basic employment for transportation and public utilities sector of Florida in different years by county A)1993 B)1997

PAGE 54

54 B Figure 4 5 Continued

PAGE 55

55 A Figure 4 6 Non -basic employment for wholesale trade sector of Flori da in different years by county A)1993 B)1997

PAGE 56

56 B Figure 4 6 Continued

PAGE 57

57 A Figure 4 7 Non -basic employment for retail trade sector of Florida in different years by county A)1993 B)1997

PAGE 58

58 B Figure 4 7 Continued

PAGE 59

59 A Figure 4 8 Non -basic employment for finan ce, insurance and real estate sector of Florida in different years by county A)1993 B)1997

PAGE 60

60 B Figure 4 8 Continued

PAGE 61

61 A Figure 4 9 Non -basic employment for services sector of Florida in different years by county A)1993 B)1997

PAGE 62

62 B Figure 4 9. Continued

PAGE 63

63 A Fi gure 4 10. Basic employment for primary sector of Florida in different years by county A)1993 B)1997

PAGE 64

64 B Figure 4 10. Continued

PAGE 65

65 A Figure 4 11. Basic employment for construction sector of Florida in different years by county A)1993 B)1997

PAGE 66

66 B Figure 4 11. Continued

PAGE 67

67 A Figure 4 12. Basic employment for manufacturing sector of Florida in different years by county A)1993 B)1997

PAGE 68

68 B Figure 4 12. Continued

PAGE 69

69 A Figure 4 13. Basic employment for transportation and public utilites sector of florida in different years b y county A)1993 B)1997

PAGE 70

70 B Figure 4 13. Continued

PAGE 71

71 A Figure 4 14. Basic employment for wholesale trade sector of Florida in different years by county A)1993 B)1997

PAGE 72

72 B Figure 4 14. Continued

PAGE 73

73 A Figure 4 15. Basic employment for retail sector trade of Flori da in different years by county A)1993 B)1997

PAGE 74

74 B Figure 4 15. Continued

PAGE 75

75 A Figure 4 16. Basic employment for finance, insurance and real estate sector of Florida in different year by county A)1993 B)1997

PAGE 76

76 B Figure 4 16. Continued

PAGE 77

77 A Figure 4 17. Basic em ployment for services sector of Florida in different years by county A)1993 B)1997

PAGE 78

78 B Figure 4 17. Continued

PAGE 79

79 A Figure 4 18. Social security payment by county of Florida in different years by county A)1993 B)1997

PAGE 80

80 Figure 4 18. Continued

PAGE 81

81 other non -empl oyment related income and transfers which tend to bloat the non-basic sectors. The choice of a scalar, while not obvious, is the focus of the next section. 28BChoosing a Scalar If the regression model in previous section is applied directly, the estimation of sector specific coefficient s would be very large While the independent variables of the model may do well in explaining employment related patterns, they lack in their ability to account for other sources of personal or transfer payment income (e.g. investment income, dividends, interest and rents, unemployment benefits, income from personal or corporate retirement accounts money spent by visitors to the state ). Note that t he dependent variable embodies both the impact of employment income and non-employment income and as such is typic ally thought of as exaggerating the impact of basic employment. As such, there is a need to scale back the estimates for non basic employment to account for the impact of transfer payments. Total personal income inc ludes net earnings by place of residence; dividends, interest, and rent; and personal current transfer receipts The first part is from employment, while the second and the third are from non -employment sources N ontraditional sources of personal income a nd transfers have become important in t he economic base literature especially in region s with high population growth rate s (Nelson, Beyers 1998) ; a feature that certainly applies to Sunshine State. The ratio of net earnings to total personal income by county is applied as a scalar to the dependent variable to account for the non -employment part of nonbasic employment Numerical values for the scalar are presented in Table 4.1 for each of Florida s 67 counties. Th is scalar alone however, is not sufficient in capturing the entire effect of income transfers as there are also inherent via multiplier effects associated with income transfers that must be account for. Furthermore, the scalar only decrease s the initia l non -basic employment figures and ignor es the influence of state level non -employment income sources Compared with other states in the

PAGE 82

82 Table 4 1 Scalars and n onbasic e mployment in y ear 1993 and 1997 by c ounty County scalar 1 scalar 2 nonbasic in 199 3 nonbasic in 1997 Alachua 0.65 0.8529 57107 66567 Baker 0.69 0.8529 2974 2956 Bay 0.63 0.8529 39872 45597 Bradford 0.68 0.8529 3068 3463 Brevard 0.59 0.8529 128870 137691 Broward 0.6 0.8529 463034 541136 Calhoun 0.59 0.8529 1365 1241 Charlotte 0.3 5 0.8529 21692 24698 Citrus 0.38 0.8529 17659 18615 Clay 0.73 0.8529 19526 22320 Collier 0.4 0.8529 56239 72960 Columbia 0.61 0.8529 10740 14044 Dade 0.62 0.8529 719150 740137 De Soto 0.55 0.8529 3778 3976 Dixie 0.52 0.8529 850 841 Duval 0.71 0.852 9 291661 317162 Escambia 0.64 0.8529 88129 96929 Flagler 0.45 0.8529 5232 7281 Franklin 0.48 0.8529 1610 1800 Gadsden 0.64 0.8529 7610 7453 Gilchrist 0.66 0.8529 868 963 Glades 0.54 0.8529 269 1242 Gulf 0.56 0.8529 1583 1999 Hamilton 0.6 0.8529 153 5 1453 Hardee 0.62 0.8529 2735 3109 Hendry 0.67 0.8529 4004 3793 Hernando 0.41 0.8529 16452 24306 Highlands 0.39 0.8529 15409 16688 Hillsborough 0.68 0.8529 366108 434392 Holmes 0.56 0.8529 1709 1903 Indian River 0.34 0.8529 25096 30632 Jackson 0.5 7 0.8529 7016 7083 Jefferson 0.63 0.8529 1633 1622 Lafayette 0.68 0.8529 418 487 Lake 0.48 0.8529 36262 41728 Lee 0.46 0.8529 103366 124288 Leon 0.72 0.8529 65815 74493 Levy 0.54 0.8529 3306 4037 Liberty 0.66 0.8529 578 568 Madison 0.58 0.8529 2178 2589 Manatee 0.5 0.8529 59985 79526 Marion 0.51 0.8529 48426 54307 Martin 0.37 0.8529 31768 36130 Monroe 0.48 0.8529 24340 26256

PAGE 83

83 Table 4 1. Continued County scalar 1 scalar 2 nonbasic in 1993 nonbasic in 1997 Nassau 0.67 0.8529 8982 11287 Okaloosa 0.63 0.8529 40977 49660 Okeechobee 0.55 0.8529 5001 5344 Orange 0.7 0.8529 369316 453648 Osceola 0.67 0.8529 29503 36068 Palm Beach 0.46 0.8529 328170 389281 Pasco 0.48 0.8529 48966 56339 Pinellas 0.55 0.8529 306103 348737 Polk 0.61 0.8529 120189 1 33398 Putnam 0.54 0.8529 9134 10624 St.Johns 0.6 0.8529 23046 30035 St.Lucie 0.47 0.8529 32541 37475 Santa Rosa 0.69 0.8529 13465 15891 Sarasota 0.42 0.8529 95709 119964 Seminole 0.73 0.8529 87836 107294 Sumter 0.45 0.8529 3614 3703 Suwannee 0.58 0 .8529 4769 5141 Taylor 0.61 0.8529 3082 3086 Union 0.69 0.8529 1291 1115 Volusia 0.5 0.8529 104654 118695 Wakulla 0.71 0.8529 1443 1686 Walton 0.57 0.8529 6119 7871 Washington 0.57 0.8529 2554 3272 U.S. Florida has a relatively la rge number of in -m igra n ts and retirees that relocated to Florida from other states. Subsequently, F lorida has a large ratio of non-employment related income when compared to other states, and conceivably an inordinate share and influx of external income from sources that ar e varied categorically and geographically Note that t he ratio of net earnings to total personal income in Florida is approximately 58%, while the ratio of net earnings to total personal income of Unite d State s is 68%. Thus, the ratio of 58% to 68% (or .85 29) is the second scalar that can be used to deflate estimated non -basic employment. This second scalar denotes the influence of state level non -employment related income transfers that filter dow n on average to county level econom ies Therefore, the depen dent variable in regression model can be re -sated, adjusting for the imp act of the two scalars: *N NE E (4 2)

PAGE 84

84 where is the ratio of net earnings to total personal income f or each county and is the scalar of influence of the state level non -employment income Note that the first scalar is different across the counties, while th e second scalar (without more detailed information) is assumed to be constant across counties. Scaling back non -basic employment allows for more accurate sector specific multiplier estimates. Because of limitations in data sou rces the scalar s used in th is study were based on information for the closest year 1998. 29BResults and Discussion The result s o f OLS estimation of the model illustrated in equation 4 1 using the modified dependent variable from equation 4 2 for the year 1993 are shown Table 4 2 Exce pt for the primary sector estimated coefficients are significant at 90% confidence level or higher Bas ic employment in wholesale trade is estimated to have the largest return to non -basic employment in regional economy. One job increase in wholesale trad e leads to approximately fourteen job s in non -basic sector s Bas ic employment in construct ion ; finance, insurance, and real estate; and services also have large returns to non -basic employment while bas ic employment of primary sector industry and transpor tation have small return s to non -basic employment relative to other sectors. F or example, it is estimated that a one job increase in the primary sector leads to approximately 2 job s in non -basic sector s The estimated coefficient associated with soci imp lies that every additional $ 3 336 in social security benefits contribute s roughly, on average, a 1 job in crease in the nonbasic sector of the regional economy. The result s of the OLS regression f or the year 1997 are shown in Table 4 3 Note the consistenc y in both the estimated values and the significan ce levels with the except ion of the coefficient associated with primary sector industry (as it is now significant at the 95% confiden ce level ) and the increase in the coefficient associated with wholesale t rade Th e change in the

PAGE 85

85 primary sector coefficient implies that a 1 job increase in the bas ic sector of Floridas primary industr ies leads to an estimated 9 job increase in non -basic sector employment. Note that this coefficient is radically different for m that which was estimated for 1993, which showed only a 2 to 1 increase in non-basic employment for an additional b asic job added to the economy. The results also suggest the volatility that exists in year to year estimates. Note the jump in the estimated coefficient for whole trad e, from 12 to approximately 17. All in all, OLS estimation is suspect as the sector -sp e cific multipliers appear to remain somewhat exaggerated in spite of the use of the scaling factors applied to the dependent variable. Table 4 2 Ordinary l east r egression e stimation r esult of years 1993 Coefficient S td.Err t test P>|t| 95% Conf. Interval p rim 2.253 3.534 0.64 0. 526 4.824 9.330 cons manu tran whol reta fina serv soci constant 8.085 ** 4.243*** 1.901** 12.076*** 3.182** 5.399*** 5.955*** 0.237*** 5704.684 3.122 1.015 0.882 1.400 1.541 0.735 0.500 0.062 2732.42 2.59 4.18 2.16 8.62 2.06 7.34 11.90 3.85 1.72 0. 012 0.000 0.035 0.000 0.044 0.000 0.000 0.000 0.091 1.835 2.210 0.135 9.272 0.095 3.926 4.953 0.113 10176 14.336 6. 276 3.667 14.880 6.269 6.871 6.957 0.360 766.907 Table 4 3 Ordinary l east r egression e stimation r esult of years 1993 Coefficient S td.Err t test P>|t| 95% Conf. Interval p rim 9.621 ** 3.736 2.58 0. 013 2.141 17.102 cons manu tran whol reta fina serv so ci constant 4.184 1.593** 1.747** 17.513*** 4.151*** 4.691*** 6.295*** 0.135** 5011.103 2.334 1.092 1.028 1.635 1.345 0.595 0.412 0.058 3157.99 1.79 1.46 1.70 10.71 3.09 7.88 15.28 2.34 1.59 0. 078 0.150 0.095 0.000 0.003 0.000 0.000 0.023 0.118 .490 .595 3.806 14.238 1.458 3.450 5.470 0.020 11334 8.858 3.780 0.312 20.788 6.844 5.883 7.120 0.250 1312.663 Note : *** indicates significant at the 0.0 1 level, ** indicates significant at the 0.05 level, indicates significant at the 0.1 level. Adjust R2= 0.948

PAGE 86

86 The other obvious change from year 1993 to 1997 is that the sign of coefficient associated with the transportation and public utility sector becomes negative (literally interpreted as meaning a contraction of the nonbasic sector for each addition al job added in that sector). This counterintuitive and illogical result means that a 1 job increase in the transportation and public utility sector would lead to a decrease of two jobs overall in the regional economy. The results for 1993 and 1997 show, nonetheless that wholesale trade has the largest multiplier effect on non -basic employment ; something that is consistent in the estimation of both models This result is no surprise as wholesale trade represents an important industry as Florida maintain s its position as a key player in the global economy with its vibrant international shipping ports located in Miami, Tampa, and Jacksonville. E mployment in construct ion ; finance, insurance, and real estate; and services also have large return to nonbasic employment. The estimated c oefficient for soci means that for every $ 1 896 dollar increase in social security benefits one additional non basic job is added to the regional economy. The r egression analysis using the single -year sample s cannot capture trend s or the adjustment of regional economi es accurately, as the results are easily influenced by sudden shocks or discontinuous change s that occur from one year to the next. This could explain why the estimated coefficient associated with the primary in dustry in 1993 is not significant at even a 8 0% confidence level, even though it is represents one of the most important sectors in the state of Florida (given that agriculture and mining activities generate a fairly large percentage of income) The varia bility in single year estimation can be overcome by applying a regression platform that utilizes information for more than just one year. Table 4 4 lists the result of ordinary least regression estimation f or the year s 1993 and 1997 combined In this anal ysis, the two sample years were combined and the model regressed without any classification No

PAGE 87

87 surprise that t he estimate d coefficient s are locate d between the two single year estimates. Nevertheless, the results are suspect and a test for ex o ge n eity is necessary. Table 4 5 highlights the result s of ex o geneity test on the error t erm s of the combi n ed OLS model The test depends on the regression of error term s from the 1997 model (uhat) on error term s from the 1993 model (uhat_1). The test results sugg est that the coefficient is significant at 95% confident level and therefore, panel data (PD) regression should be applied instead of OLS regression on the two year samples. Table 4 4 Ordinary l east r egression e stimation r esult of y ear 1993 and 1997 Co efficient S td.Err t test P>|t| 95% Conf. Interval p rim 5.282 2.611 2.02 0. 045 .114 10.449 cons manu tran whol reta fina serv soci constant 5.189 2.911*** 0.082*** 14.850*** 3.806*** 4.767*** 5.984*** 0.205*** 4881.392 1.851 0.750 0.676 1.078 0.993 0.462 0.315 0.040 2110.40 2.80 3.88 0.12 13.77 3.81 10.32 19.00 5.10 2.31 0. 006 0.000 0.903 0.000 0.000 0.000 0.000 0.000 0.022 1.526 1.426 1.256 12.716 1.828 3.853 5.360 0.126 9058.46 8.853 4.395 1.421 16.985 5.784 5.681 6.607 0.285 704.316 Note : *** in dicates significant at the 0.0 1 level, ** indicates significant at the 0.05 level, indicates significant at the 0.1 level. Adjust R2= 0.946 Table 4 5 Test r esult of e x o geneity of e rror t erm uhat Coefficient S td.Err t test P>|t| uhat_1 0.563 0.101 5.59 0. 000 constant 3365.202 3544.38 0.9 0.346 The result s for the panel data regression are shown in T able 4 6 Note that the estimated c oefficients for the primary and construction sectors are significant at 80% confidence level, and the coefficients of all other industries are significant at 95% confidence level. Once again, the largest estimated coefficient is for wholesale trade, which implies that one additional job in that sector of wholesale trade will lead to approximately 11 additional non -basic job s in the regional economy at large Retail trade, the service sector and manufacturing are also important sectors,

PAGE 88

88 with empirical results that are consistent with those reported by Fik et al (1991, 1993). The smallest coefficients are associated wi th transportation and finance ; not surprising, as Florida does not actively engage in the building of public transportation (e.g., limited railway and subway ). All in all, the sector -specific multiplier estimates provide a glimpse of Floridas orientation toward a sales and service -oriented economy, as discussed by Fik et al (1991) The size of the multiplier estimates indicate that may still be slightly overstated, yet one could argue Table 4 6 Panel d ata r egression e stimation r esult of y ear 1993 and 1997 without d ummy v ariable Coefficient S td.Err Z test P>|Z| 95% Conf. Interval p rim 5.483 4.072 1.35 0. 178 2.499 13.464 cons manu tran whol reta fina serv soci constant 3.311 4.019*** 2.768*** 10.832*** 4.306*** 2.736*** 4.454*** 0.202*** 1336.749 2.137 1.307 1.059 1.702 1.201 0.685 0.396 0.046 3636.65 1.55 3.07 2.61 6.36 3.58 3.99 11.25 4.39 0.37 0. 121 0.002 0.009 0.000 0.000 0.000 0.000 0.000 0.713 .877 1.457 .693 7.496 1.950 1.393 3.678 0.112 8464.45 7.500 6.580 4.843 14.168 6.659 4.079 5.230 0.292 5790.9 Note : *** indicates significant at the 0.0 1 level, ** indicates significant at the 0.05 level, indicates significant at the 0.1 level. Wald Chi 2(9) = 0629.42, Prob> Chi 2=0.000 that their relative proportion is fairly representative of Flo ridas economic reality (with an estimation bias that has been reduced), given that the transfer income effect has been largely accounted for by scaling back non -basic employment levels. In order to test if the above OLS regression and panel data regressio n is valid, error term diagnostics for both OLS regression and panel data regression are applied. Normality test of error term for residuals of year 1993 OLS regression, year 1997 OLS regression, both year 1993 and year 1997 OLS regression and panel data reg ression are showed in Figure 4 19a, 4 1 9 b, 4 1 9 c and 4 1 9 d respectively.

PAGE 89

89 0 5.0e-05 1.0e-04 1.5e-04 Density -50000 0 50000 Residuals A 0 5.0e-05 1.0e-04 1.5e-04 Density -60000 -40000 -20000 0 20000 40000 Residuals B Figure 4 19. Normality test for different regression A)Year 1993 OLS regression B)Year 1997 OLS regression C)Both year 1993 and 1997 OLS regression D) Both year 1993 and 1997 panel data regression

PAGE 90

90 0 5.0e-05 1.0e-04 1.5e-04 Density -100000 -50000 0 50000 OLS residual C 0 1.0e-04 2.0e-04 3.0e-04 Density -40000 -20000 0 20000 40000 Panel residual D Figure 4 19. Continued

PAGE 91

91 According to the normality tests figures, error terms of both OLS and panel data regression show normally distributed. Peak density of error terms for both single year regression is located in residua l with value above 0, while peak density of error terms of panel data regression is located in residual with value more closed to 0, so error terms of panel data regression show more normally distributed. Residual plots( fitted va lues versus residuals) for year 1993 OLS regression, year 1997 OLS regression, both year 1993 and year 1997 OLS regression and panel data regression are showed in Figure 4 2 0 a, 4 2 0 b, 4 2 0 c and 4 2 0 d respectively. Residual plots are applied to test h om o s ce dasticity assumption of re gression. -50000 0 50000 Residuals 0 100000 200000 300000 400000 Fitted values A Figure 4 20. Residual plots for different regression A)Year 1993 OLS regression B)Year 1997 OLS regression C)Both year 1993 and 1997 OLS regression D) Both year 1993 and 1997 panel data regression

PAGE 92

92 -100000 -50000 0 50000 Residuals 0 100000 200000 300000 400000 Fitted values B -100000 -50000 0 50000 Residuals 0 100000 200000 300000 400000 Fitted values C Figure 4 20. Continued

PAGE 93

93 -40000 -20000 0 20000 40000 e[countyid,t] 0 100000 200000 300000 400000 Linear prediction D Figure 4 20. C ontinued While r esidual plots of OLS regression show obvious triangle -shade, residual plots of panel data regression show probably random distributed if the upper and below outliers are deleted. Therefore, all OLS regression estimates here suffer from heter o sc e dastici t y problem. Panel data regression also cannot avoid this problem, but alleviate to some extent according to residual plots. Plots of fitted versus actual values of the scaled non-basic employment for year 1993 OLS regression, year 1997 OLS regr ession, both year 1993 and year 1997 OLS regression and panel data regression are showed in Figure 4 21A 4 21B, 4 21C and 4 21D respectively. All plots of fitted versus actual non-basic employment values shows satisfied goodness of fit of the economic b ase models, which implies that independent variables can explain scaled dependent

PAGE 94

94 variables very well. However, figure 4 21D shows that fitted value of nonbasic employment is less than actual value of non basic employment in the middle of plots. 0 100000 200000 300000 400000 Fitted values 0 100000 200000 300000 400000 NEB A 0 100000 200000 300000 400000 Fitted values 0 100000 200000 300000 400000 NEB B Fi gure 4 21. Plots of fitted versus actual non-basic employment values for different regression A)Year 1993 OLS regression B)Year 1997 OLS regression C)Both year 1993 and 1997 OLS regression D) Both year 1993 and 1997 panel data regression

PAGE 95

95 0 100000 200000 300000 400000 Fitted values 0 100000 200000 300000 400000 NEB C 0 100000 200000 300000 400000 Linear prediction 0 100000 200000 300000 400000 NEB D Figure 4 21. Continued

PAGE 96

96 From Figure 4 21D, there are 11 plots below the line of fitted value equal actual value from 100000 to 300000, while only one plot is above the line, which implies that fitted value is underestimated during this range. After projecting the un derestimated plots to the total employment map shown in F igure 4 22, these plots are standard for Broward, Palm Beach, Orange, Duval and Pinellas a nd Hillsborough. Figure 4 22. Map of coastal urban c ounties

PAGE 97

97 Compared with Figures 4 1A and 4 1B, these co unties are coincident with the counties that have largest total employment except Miami Dade, which actually is the plot above NEB=400000 in Figure 4 21D. The common geographic characteristic of these 6 counties is that all of them are coastal urban area. As tourist and wholesale trade are highly developing, coastal and waterway resource boom the relative counties employment. Therefore, it is reasonable to argue that the geography characteristic of coastal urban area should obtain more attention in the mod el. In order to eliminate the bias of estimates for these 6 counties, the original panel data model should be added a new dummy variable urban defined as coastal urban counties, including Broward, Palm Beach, Orange, Duval and Pinellas and Hillsborough. The result of panel data regression estimation with dummy variable list in Table 4 7 Dummy variable is positive significant at 1% level, which coincide with expected impact that coastal urban property can increase the employment of county. Table 4 7 Pa nel d ata r egression e stimation r esult of y ear 1993 and 1997 with dummy variable Coefficient S td.Err Z test P>|Z| 95% Conf. Interval p rim 2.220 2.320 0.95 0. 340 2. 345 6.787 cons manu tran whol reta fina serv soci urban constant 3. 987 ** 2.354*** 4.163*** 8.889*** 4.071*** .087 2.752*** 0.202* 109119.4*** 532.287 1.391 0.708 0.638 1. 017 0.764 0. 496 0.3 11 0.0 31 8501.87 1978.03 2.87 3. 33 6.52 8.52 5.33 0.17 8.85 2.72 12.83 0. 27 0. 004 0.001 0.000 0.000 0.000 0. 862 0.000 0.006 0.000 0.7 88 1.260 0.997 2.9 12 6.675 2.573 1.060 2.143 0 .024 92456.1 4409.16 6.715 3.741 5.413 1 0.662 5.569 .887 3.361 0. 146 125782.8 3344.584 Note : *** indicates significant at the 0.0 1 level, ** indicates significant at the 0.05 level, indicates significant at the 0.1 level. W ald Chi 2(10)= 2558.00, Prob> Chi 2=0.000 Plots of fitted versus actual non-basic employment values for both year 1993 and year 1997 panel data regression with dummy variable show in F igure 4 23. In the middle, some plots

PAGE 98

98 locate above the line of fitted v alue equal actual value, while others plots locate below the line, which implies that estimation problem during range 100000 to 300000 is solved effectively. Therefore, panel data model obtains better goodness of fit after dummy variable is added. 0 100000 200000 300000 400000 Linear prediction 0 100000 200000 300000 400000 NBE Fig ure 4 23. Plots of fitted versus actual nonbasic employment values for both year 1993 and year 1997 panel data regression with dummy variable In order to further compare the estimation results of two panel data models, spatial autocorrelation of error te rms are considered. According to Morans I statistic test applying inverse distance as conceptualized spatial relation, the spatial pattern of error terms of original panel data model is random, while the spatial pattern of error terms of panel data mode l with dummy variable is clustered in 10% significant level. Therefore, on one hand, the panel data model with dummy variable urban has better estimation compared with original panel data model, on the other hand, the panel data model with dummy variable suffer from spatial

PAGE 99

99 autocorrelation of error term, while the original panel data model gets rid of the problem. It is highly possible that dummy variable lead to such a problems, because some of coastal urban counties are neighborhood. According to test o f e x o geneity of e rror t erm and three comparison between OLS and Panel data regression, estimates results depending on both OLS and panel data regression fit the actual scaled non -basic employment to large extent, while compared with OLS regression, panel data regression shows better re su lts on test of normality, homosce dastic it y and ex o geneity. Multipliers of each industry /sector, from Table s 4 2 4 3 4 6 and 4 7 are summarized in Table 4 8 While the panel data results appear promising, the OLS estimates are most certainly suspect given the dramatic differences that are observed between 1993 and 1997 in the magnitude of the estimate d coefficients I n 1 993, the multiplier associated with the primary sector is estimated at 3.253, smaller than that of the multiplier for the transportation and public utility sector. This result do es not make sense, because it is known that the agriculture industry in Florida creates more employment opportunities than does p ublic transportation. Moreover, single equation est imation of multipliers for the year 1997 produce d negative multipliers. This result is somewhat of an absurdity as a job added to an economy cannot take away more than that which it add s (as a generalization, any increase in income via a new job will undo ubtedly produce additional demand) By contrast, t he panel data regression results seem to fall more in line with expectations, produc ing positive multipliers across the board and coefficients that are of a reasonable magnitude Compared with regression estimates for a si ngle year, panel data regression provides a more consistent and even -handed estimation of multipliers across sectors avoiding the pitfall of abnormal disturbance s associated with single year sample s and illogical results such as negative coefficients /multipliers

PAGE 100

100 From year 1993 to year 1997, wholesale trade, retail trade and services were responsible for generating the greatest employment multiplier impacts. T hese results are undoubtedly tied to Floridas thriving tourist and recreation sectors, centered about the theme parks of Central Florida and its bountiful parks and coastal amenities, not to mention its role as a leader in international trade from import -export activities associated with the cities of Miami, Tampa, and Ja cksonville. These cities represent three of the most important hubs of southeast of U.S. and link the U.S. economy to South America and the remainder of the global economy. T ourism as vital income source for Florida, represents an export -base in and of itself. A lthough no physical products are being exported, tourist and recreation activity are services that provide a source of entertainment and relaxation. Warm weather and hundreds of miles of beaches attract roughly 60 million visitors to Florida every year HAmusement and theme parksH, in the HOrlandoH area, make up a significant portion of that touris m. The HWalt Disney World ResortH is the largest vacation resort in the world, consisting of four Htheme parksH and more than twenty hotels Universal studio also attracts a large number of tourists Many beach towns such as Daytona Beach, Tampa Fort Lauderdale Sarasota and Fort Myers are popular touri st destinations; and attract year ro und visitors and seasonal in -migrants from colder climates to the state during the winter months. Therefore, it is no surprise that the retail and service sectors have relatively large multipliers. While manufacturing in Florida account for less than 8% of the states total employment, it is also very important as an impact industry with an estimated multiplier that is larger than construction, transportation, and financial sectors. T he empirical result s suggest that the regional economic bases multipli ers associated with the Florida economy are substantially larger than th ose estimated for various communities in the state of Arizona (Mulligan, 1987) This is to be expected as Florida is more populated, more

PAGE 101

101 export and tourist oriented, a large sink for retirees over the past several decades, and unique in terms of its amenities and recreational possibilities from its natural, coastal and waterway resources. Table 4 8 M ultipliers of OLS regression and panel data regression Sector Year 1993 Year 1997 Two Years Two Years with dummy prim cons manu tran whol reta fina serv 3.253 9.085 5.243 2.901 13.076 4.182 6.399 6.955 10.621 5.184 2.593 .747 18.513 5.151 5.691 7.295 6.583 4.431 5.019 3.768 11.832 5.406 3.736 5.454 3.220 4.987 3.354 5.163 9.889 5.0 71 0.913 3.852

PAGE 102

102 4BCHAPTER 5 12BCONCLUSIONS Job creation is a characteristic of basic industry. Activity and employment in basic sectors are vital for economic growth, and basic employment plays a crucial role in contribut ing to the growth of a regional e conomy by bringing in dollars from external source s Economic base multipliers estimate the impact that an influx of dollars has on the creation of jobs for each additional job added to a given sector. Basic employment in response to external demand for goods and services, is responsible for cr eating additional jobs as income circulates and re circulates throughout a regional economy, and subsequently, as the employment base expand s. Quantitative model s that attempt to estimate those multipliers throug h the disaggregat ion of employment or income data must reckon with problems that are tied inherently to the nature of sample data and to problems that arise from various estimation procedures. E conomic base models can be a quick effective, and inexpensive way in which to ascertain the expansion potential of various sectors or industries and their relative impact on a regions economic growth. This information is useful to policymakers and legislators in their attempt to construct economic and development policies that c ould conceivably yield th e biggest bang for the buck. Moreover, economic base analysis and estimated multipliers are useful in explaining variability in regional economic and employment growth rates ( Malecki, 1991) factors that can be att ributed to a regions economic or employment mix. As such, c ritical assessment s of multipliers are needed to ensure accurate and reasonable estimates. As part of this assessment, the techniques by which multipliers are estimated must also be evaluated to lessen the chance of bias or misleading results. This thesis has demonstrated that multiple -year panel data regression models offer a useful alternative to OLS regression model as an estimation technique.

PAGE 103

103 This study introduce d economic base theory and me thods used to distinguish between basic and non -basic employment in a regional economy It also gave an overview of the mechanics of estimating employment multipliers using traditional and contemporary approaches Discussion focused on estimation techniq ues, assumptions, and the shortcomings of the traditional OLS approach to estimating economic base multipliers. In addition, th is study discussed the difference between aggregate and disaggregate modeling It also distinguish ed between average multiplier s and impact multipliers. F our common ly used method s of estimating economic base multipliers were reviewed : the survey method, the l ocation-quotient method the m inimum requirement s method and linear regression based a lternatives While the survey method is a preferred approach, in that it is based on a comprehensive survey of all firms in all industries, it is viewed as problematic and costly for large regions. The l ocation-quotient (LQ) method and m inimum requirement s method s, while easier in terms of data collection requirement s depend on t he choice of a benchmark ( whose accuracy can be debated on several levels ). The regression -based approach is the most popular method by far largely due to its accuracy and forecasting ability The largest differ ence between the regression based approach and the other three widely used methods is that regression approach allow s for estimation of sector -specific impact multiplier s while the other three index-driven approaches only produce average multipliers. In t he methodology chapter th is study provided a comparison of two regression method s that are used in the estimati on of sector -spe cific multipliers namely, original least square regression and panel data regression N on -basic employment figures were initially obtained using the LQ method at the county level with state -level employm ent figures as a benchmark These estimates were adjusted to account for the overstatement problem that is common ly

PAGE 104

104 associated with the LQ method. Furthermore, this thesis offered a unique attempt as scal ing down estimates of non -basic employment ( through the use of two scalar s) This was done to reduce the prob lem of overstat ing multipliers when non -basic employment levels are overstated due to the effects of non -employment i ncome or transfer payments T he two scalars used to adjust non -basic employment figures were defined as (1) the ratio of net earnings to total personal income of each county within the state and (2) the ratio of the percentage of net earnings to total personal income f or Florida versus th at of the U.S. at large. Non basic employment figures as the dependent variable, were simply multiplied by the product of the se two scalars The adjusted nonbasic employment figures were then used as dependent varia bles to estimate sector -specific multipliers Apart from LQ -generated basic employment figures that were calculated for each industry/sector (the explanatory variables of the model) social security benefits we re added as a n additional surrogate variable on the right -hand side of the regression equation given the states heavy reliance on this income source These corrections were necessary to account for the various factors related to non -employment income that typically lead s to overestimation of both non-basic employment and sector -specific multipliers It was shown that the panel data regression model for a two years sample yielded reasonable more sensible results than single -year estimation using OLS regression. OLS residuals were also shown to be suspect. E x o geneity tests on the residuals suggest ed that it was necessary to apply panel data regression rather in the estimation of the disaggregate d economic base model It was concluded that panel data regression offer ed a more even handed and consis tent estimation of multipliers than those produced by OLS regression Th is study focused exclusively on county level employment patterns in the state of Florida for the years 1993 and 1997. Reasonable economic base multiplier estimates were obtained using

PAGE 105

105 the panel data approach for these years, though validation of the multiplier estimates represents the next step in the scientific process. Further research on this topic could also consider expanding the analysis to include more than two years to improve a ccura cy in the estimation of sector -specific multipliers Panel data regression c ould also be used to estimate economic base multipliers f or other states in the South east or the U .S (and the results compared to estimates for Florida). While this study a pplie d transfer income related scalar s and social security benefits as a surrogate independent variable to weaken the influence of non -employment revenue sources further steps could be taken to scale down non-basic employment figures in places were there is a secondary or overlapping multiplier effect associated with the indirect expansion related to non -employment income (as might be in the case of Florida). Further inquires could also look at various data transformations and alternative functional forms for the regression model The panel regression model could be further expanded to incorporate spatial or geographical effects as suggested by Hanink (2007) who estimate d m ultipliers as elasticities using regr ession analyses to uncover the of the associations between retail trade earnings by place of work and total income by place of residence across New England counties in 2002. As an e xtension further modeling efforts could consider spill over and spill inw ard effects across political jurisdictions that are both part of the study region and part of the region(s) juxtaposed or adjacent to the study region. Spatial autocorrelation problem of panel data model with dummy variable coastal urban should be addre ssed. Lastly a larger battery of regression diagnostics could be performed to more fully inspect the error terms and to test model assumptions.

PAGE 106

106 13BLIST OF REFERENCES Andrews, R.B. 1953 M echanics of the Urban Economic Base: A Classification of Base Types Land Economics 29: 343 50. Baltagi, B.H 1995. Eco nometric Analysis of Panel Data New York : John Wiley and Sons Billings, R.B. 1969. The Mathematical Identity of the Multipliers Derived from the Economic Base M odel and the Input Output Mo del Journal of Regional Science 9 : 47173. Blumenfeld, H. 1955. The Economic Base of the Metropolis: Critical Remarks on t he "Basic Nonbasic" Concept Journal of the American Planning Association 21: 11432. Chiang, S. 2009. Location Quotient and Trade The Annals of Regional Science 43: 399 414. Fik, T.J., Amey, R.G, and Malecki, E.J. 1991. Changing Employment Profiles and Growth: An Economic B ase Study of Florida Counties Growth and Change 22: 86104. Fik, T.J., Malecki, E.J., and Amey, R.G. 1993. Trouble in Paradise? Employment Trends and Forecasts for a Service -Oriented Economy. Economic Development Quarterly 7 : 35872. Fik, T.J. 2000. The Geography of Economic Development: Regional Changes, Global Challenges New York : McGraw -Hill Gibso n, L.J. and Worden, M.A. 1981. Estimating the Economic Base Multiplier: A Test of Alternative Procedures Economic Geography 57 : 146 59. Greene, W.N. 2000. Econometric A nalysis Englewood Cliffs, NJ: Prentice Hall. Hanink, D.M. 2007. Spatial and Geo graphical Effects in Regional Multiplier Analysis Environment and Planning A 39: 74862. Harris, T.R., Ebai, G .E. and Shonkwiler, J.S. 1998. A Multidimensional Estimation of Export Base The Journal of Regional Analysis & Policy 28: 3 17. Hayashi, F. 2000, Econometrics New Jersey: P rinceton University Press. Hewings, G. 1985, Regional Input -output Analysis Beverly Hills: Sage Publications Hilde brand, G.H. and Mace, A. 1950 The Employment Multiplier in an Expanding Industrial Market : Los Angeles County, 194047. The Revi ew of Economics and Statistics 32: 24149. Isserman, A.M.J. 1977a A Bracketing Approach for Estimating Regional Economic Impact Multiplers and a Procedu re for Assessing their Accuracy Environment and Planning 9 : 100311. Isserman, A. 1977b The Location Quotient Approach to Estimat ing Regional Economic Impacts Journal of the American Planning Association 43: 3341.

PAGE 107

107 Krikelas, A.C. 1992. Why Regions G row: A R eview of R esea rch on the E conomic Base M odel Economic Review of the Federal Re serve Bank of Atlanta 1992: 1629. Lego, B., Gebreme dhin, T. and Cushing, B. 2000 A Multi -Sector Export Base Model of LongRun Regional Employment Growth Agricultural and Resource Economics Review 29: 19297. Leichenko, R.M. 2000. Exports, Employment, and Production: A Causal Assessm ent of U.S. States and Regions Economic Geography 76: 30325. L esage, J. and Reed, D.J. 1989 The Dynamic Relationship between Export, Loc al, and Total Area Employment Region al Science and Urban Economics 19 : 61536. Loveridge S. 2004. A Typology and Assessment of Multi Sector R egional Economic Impact Models Regional Studies 38: 305 17. Malecki, E.J. 1991. Technology and Economic Development: The Dynamics of Local Regional, and National Change New York : John Wiley and Sons Moore, C.L. 1975. A New Look at the Minimum Requirements Approach to Regional Economic Analysis Economic Geography 51: 35056. Mulliga n, G.F. and Gibson, L.J. 1984. Regression Estimates of Economic Base Mul tipliers for Small Communities Economic Geogra phy 60: 225 37. Mulligan, G.F. 2009. Industry Specific Employment Multipliers in U .S. Nonmetropolitan Economies Studies in Regional Science 3 9 : 681 98. Mulligan, G.F. 1987. Employment Multipliers and Functional Types of Communities: Effec ts of Public Tra nsfer Payments Growth and Change 18: 1 11. Mulligan, G. and Fik, T. 1994. Using Dummy Variables to Esti mate Economic Base Multipliers The Professional Geographer 46: 368 78. Nelson P.B. and Beyers, W.B. 1998. Using Economic Base Models to Expla in New T rends in Rural Income Growth and Change 29: 295 318. Nishiyama, Y. 1997 Exports Contribution to Economic Growth: Empirical Evidence for California, Massachusetts, and Texas U sing Employment Data Journal of Regional Science 37 : 99125. North, D.C. 1955. Location Theory and Regional Economic Growth Th e Journal of Political Economy 63: 243 58. Roberts, D. 2003. The Economic Base of Rural Areas: A SAM Based Analys is of the Western Isles, 1997. Environment and Planning A 35: 95111.

PAGE 108

108 Tiebout, C.M. 1956. Exports and Regional Economic Growth Th e Journal of Political Economy 64: 160164. Vias, A.C. and Mulligan, G.F. 1997 Disaggregate Economic Base Mu ltipliers in Small Communities Environment and Planning A 29: 95574. Vo llet, D. and Bousset, J. 2002. Use of Meta -Analysis for the Comparison and Transfer of Economic Ba se Multipliers Regional Studies 36: 48194. Was, A.C. and Mulligan, G.F. 1999. Integrating Economic Base Theory with Regional Adjustment Models: The Nonm etropolitan Rocky Mountain West Growth and Chang e 30: 50725. Wooldridge, J.M. 2002. Econometric A nalysis of Cross S ection and P anel D ata Cambridge, MA: MIT Press.

PAGE 109

109 14BBIOGRAPHICAL SKETCH Fan Li was born in Xiangfan China She received a Bachelor of Engineering degree in Information Engi neering from Jinan University in 2007 and a m aster s d egree in i ndustrial e conomics from Jinan University in 2009. Her interest in economic geography brought her to the University of Florida to pursue a second masters degree. She plans to continue her e ducation at the University of Florida and enter into a doctorate program in e conomics.