Intergenerational conflict and publicly provided goods

MISSING IMAGE

Material Information

Title:
Intergenerational conflict and publicly provided goods
Physical Description:
viii, 64 leaves : ; 29 cm.
Language:
English
Creator:
Fletcher, Deborah
Publication Date:

Subjects

Subjects / Keywords:
Economics thesis, Ph.D   ( lcsh )
Dissertations, Academic -- Economics -- UF   ( lcsh )
Genre:
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Thesis:
Thesis (Ph.D.)--University of Florida, 2003.
Bibliography:
Includes bibliographical references.
Statement of Responsibility:
by Deborah Fletcher.
General Note:
Printout.
General Note:
Vita.

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 030441769
oclc - 83559451
System ID:
AA00013583:00001


This item is only available as the following downloads:


Full Text










INTERGENERATIONAL CONFLICT AND PUBLICLY PROVIDED GOODS


By

DEBORAH FLETCHER

















A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA



















This dissertation is dedicated to my husband, Peter Thompson.














ACKNOWLEDGMENTS

I thank the members of my dissertation committee, David Figlio, Lawrence Kenny

and Steven Slutsky for their constant guidance and patience. I also thank my husband,

Peter Thompson, for unending moral support and frequent reality checks. Finally, many

thanks go to my parents for their pride in my professional and personal accomplishments.















TABLE OF CONTENTS

page

ACKNOW LEDGEMENTS................ .................................................................. iii

LIST OF TABLES ....................................................... ................................................ vi

ABSTRACT.............................................................................. vii

CHAPTER

1 INTRODUCTION ........................................................................... ........................1......

2 OUT WITH THE OLD, IN WITH THE NEW? CONGRESSIONAL VOTING
PATTERNS, INTERGENERATIONAL CONFLICT AND DEMOGRAPHIC
CHANGE............................... ................................................... ............................ 6

Introduction..................................................................................................................6......
C ensus D ata ................................................................................... ................................ 9
The Voting Index ........................................................................................................1...... 1
The Model ............................................................................................. ....................13
Hypotheses.......................................................................................... ............................ 14
R esu lts ................................................................................................................................ 16
P airs of B ills............................................................................ ..................................... 17
Children's Issues......................................................... ................................................. 23
C conclusion ........................................................................................ ..................... 27

3 IT TAKES A VILLAGE? INTERGENERATIONAL CONFLICT AND
COOPERATION IN EDUCATION EXPENDITURES .......................................... 29

Introduction .................................................................................... .............................. 29
Empirical Strategy and Data Collection ............................................... ........................ 32
Results............................................................... ........... .......................... 39
Controlling for the Effects of Tiebout Sorting...................................... ........................ 45
R esults............................................... ........................................................................... 4 9
C conclusion ................................................................................................................. 56









APPENDIX

VOTES ON AGING ISSUES WITH 20% OR GREATER
CONTENTION, 1987 TO 1998 ...................................................................................58

R EFER EN C E S .......................................................... .................................................. 61

BIOGRAPHICAL SKETCH...................................................64














LIST OF TABLES


Table page

2-1 Summary statistics: changes in demographic variables.....................................10

2-2 Changes in demographic variables for Florida's Representatives......................10

2-3 Sum m ary statistics: voting index .............................................. ...................... 12

2-4 Changes in elderly voting index .............................................. ......................... 16

2-5 Summary statistics: fraction changes in constituent demographic
characteristics for representatives voting on antipoverty measures
for the elderly ......................................................................... ........................... 19

2-6 Summary statistics: fraction changes in constituent demographic
characteristics for representatives voting on age discrimination issues..............20

2-7 Factors influencing congressional votes on elderly issues ................................21

2-8 Summary statistics: fraction changes in constituent demographic
characteristics for representatives voting on antipoverty measures
for children ............................................................................ .... ........................ 24

2-9 Summary statistics: fraction changes in constituent demographic
characteristics for representatives voting on education improvements ..............25

2-10 Factors influencing congressional votes on children's issues.............................26

3-1 Sam ple sum m ary statistics...................................................... .......................... 37

3-2 Determinants of per pupil state and local revenue for education........................40

3-3 Summary statistics for the determinants of elderly migration............................48

3-4 Determinants of elderly migration from out-of-state..........................................50

3-5 Determinants of per pupil state and local revenue for education........................53













Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

INTERGENERATIONAL CONFLICT AND PUBLICLY PROVIDED GOODS

By

Deborah Fletcher

August 2003

Chair: David Figlio
Major Department: Economics

This dissertation explores the ways in which intergenerational conflict can affect

the provision of publicly provided goods that are consumed by people in specific age

groups, such as public education and Medicare. Many factors can influence the relative

redistribution of government resources toward the various age groups. I examine this

redistribution at both the local and national levels.

At the national level, I examine how demographic change affects Congressional

voting on aging issues. I used demographic data from the 1990 Census and vote records

of individual members of the House of Representatives from 1986 to 1998 to examine

how demographic changes in the constituencies of Representatives, and particularly

changes in the age distribution of the constituencies, affected how the Representatives

voted on issues specific to age. Redistricting is used as a natural experiment, and allowed

me to separate the effects of changes in demographics from the individual inclinations of

the various members of Congress. I offer evidence that politicians are indeed

Representatives, responding to changes in the demographic makeup of their districts.








At the local level, I explored the effects of elderly migration on the public provision

of education. Previous research offered mixed evidence about the effect of the elderly on

education spending, but consistently showed that racial differences between the older and

younger cohorts had a strong negative effect on education spending. The elderly

preferred higher education spending if they felt a sense of connection to their

communities, particularly to the youth in those communities. As a result, long-time

elderly residents or those who migrated only a short distance are more likely to support

education.

I used 5-year migration data from the U.S. Census to explore this idea, in a model

controlling for the possible endogeneity of elderly migration with respect to education

spending. I found that education expenditures are higher when a larger fraction of the

elderly has lived in the same county for at least 5 years, but are lower when the elderly

who do migrate are from farther away. I also found cross-cohort racial heterogeneity to

have no significant effect on education spending.














CHAPTER 1
INTRODUCTION

The government provides goods and services for many different groups in society.

These groups are not taxed differentially by their preferences on public spending, but

their preferences vary. Therefore, these various groups must compete for government

resources. A conflict exists between generations because different services exist for

different age groups; examples include education and welfare and nutrition programs

aimed specifically to aid children, and Medicare and other programs for the elderly. The

mix of programs chosen reflects the characteristics of the population. Therefore, changes

in the demographic composition of the population may have far-reaching effects on the

mix of services provided by the public sector.

According to the 2000 Census, just over 12% of the American population is

currently aged 65 or higher. By the year 2030, this proportion is expected to rise to

nearly 20%. As America continues to age, it is likely that interest in issues affecting

older citizens will increase. Two of these important issues are how the voting behavior of

legislators in a representative democracy changes when constituent demographics

change; and how the increasing elderly population will affect spending on public

education.

Most of the literature to date on intergenerational conflict has addressed Social

Security and education finance. The Social Security literature has concentrated on the

viability of the system in the face of the booming retiree population and possible reforms

(Baker and Weisbrot 1999 and Schultz 1995).






A growing literature has concentrated on education finance issues. The literature

on intergenerational conflict and education expenditures gives mixed hypotheses and

evidence on the effects of change in the proportion of older people on education

expenditures. If the two cohorts, children and the elderly, are competing for resources, it

would seem that a higher proportion of older constituents would cause fewer votes for

services for children. However, altruism may mitigate this tendency, as Logan and Spitze

(1995) point out. Poterba (1998) argues that increased spending on education may

decrease crime, which increases the utility of the elderly; other cross-generational effects

could include the improved level of services, such as health care, the elderly enjoy with a

better-educated younger cohort. It is feasible that cross-generational effects could hold

for children's issues other than education. This would be particularly true where altruism

exists. However, antipoverty and other programs for children could act to reduce crime

and improve services for the elderly in much the same way as improved education, so

altruism need not be present. Kemnitz (2000) models Social Security and public

education expenditures, and finds that population aging may increase the Social Security

contribution rate as well as educational expansion.

The empirical evidence is also mixed. Poterba (1997) finds that older voters do not

support increases in state education expenditures. Furthermore, he finds that, if the

elderly and children in a state are of different races, the elderly do not support education

spending. Ladd and Murray (2001) perform a similar analysis at the county level and

find the fraction of elderly to have no effect on education expenditures. Harris, Evans

and Schwab (2001) get a similar result in a district-level analysis, and suggest that this

may be because local education expenditures are capitalized into housing prices.

However, the latter two sets of authors find the same strong negative effect of cross-






cohort racial heterogeneity as Poterba. It is possible that this is due not to a racial bias

in the elderly, but that the race variables are capturing a "connectedness" between the

generations.

This sense of connection forms the basis of the analysis of Chapter 3. The elderly

do not enjoy the direct benefits of public education, unlike families with children. Other

researchers including Poterba (1998) and Kemnitz (2000) have discussed the external

benefits of education that might make the elderly more willing to improve it. These

include decreased crime, a higher level of services such as health care, and increased

Social Security contributions from the better educated younger cohort. In the real world,

though, the elderly will not live to enjoy these external benefits of paying to improve

education now. For example, improving a child's education now may make him less

likely to commit crimes now, but the effect presumably (and hopefully) extends to the

child's full lifetime. If this is the case, the elderly will not live long enough to benefit

from the cumulative reduction in crime. The immediate gains to the welfare of the

elderly will be small; thus, the elderly should be unwilling to fund education.

Altruism would mitigate this tendency. Following Becker's (1981) work on the

economics of the family, grandparents are very likely to support those children who mean

the most to them: their grandchildren. If grandparents living in the same locale as their

grandparents shift the position of the median voter, this will lead to higher spending on

education. Once we are outside the confines of the family, however, the literature does

not explain why we might be more altruistic toward some people than others.

I propose that the extent to which we are altruistic depends on how connected we

are to the beneficiaries of our good will. Charity begins at home, and then extends to

others. That is, if the elderly subscribe to the idea that "it takes a village to raise a child,"






they will also act to help other children who are part of their village --- close to them in

a geographic sense or with whom they identify in some other way.

It may also take time for people to identify with their communities; if this is the

case, they will become more altruistic with longer residence. If recent residents of a

community maintain some sense of connection to their previous residence, support for

education among the elderly should be highest where more of the elderly are long-time

residents (i.e. moved a distance of zero); and lowest where more of the elderly have

moved in from far away.

Chapter 2 examines how demographic change affects Congressional voting on

aging issues. In a simple majority-voting scheme bills would be passed which would

reflect the demographic makeup of the area. Abstracting from differential voting

participation rates between age groups, we would expect to see more elderly-friendly

bills in places where more elderly people reside, and fewer bills passed which would

provide goods and services to children; the opposite should hold true where there are

relatively more families with children. However, in the United States we have a

representative government to make most decisions about public expenditures. Do

politicians respond to varying demographic characteristics of their constituent bases, or

do they vote according to their own preferences?

I use demographic data from the 1990 Census and vote records of individual

members of the House of Representatives from 1986 to 1998 to examine how

demographic changes in the constituencies of Representatives affect how the

Representatives vote on issues specific to age. Of particular interest is how changes in

the age distribution of the constituencies affect voting behavior. The 1992 redistricting of







the House of Representatives is used as a natural experiment to isolate and examine

these effects.

Redistricting provides us with multiple observations on demographic variables for

individual lawmakers. We can examine how a Representative's voting patterns might

change on issues related to demographic characteristics, such as age, when his constituent

demographics change. This allows me to separate the effects of changes in demographics

from the individual inclinations of the various members of Congress.

The analyses in Chapter 2 offer evidence that Representatives do indeed respond to

changes in the demographics of their constituencies. When a larger proportion of the

constituency is made up elderly people, legislators are more likely to vote to support

redistribution toward the elderly. Similarly, when a larger proportion is made up of

children, legislators are more likely to vote to support redistribution toward children.

However, cross-effects do not hold, where having a large population of one age group

would make a Representative less likely to vote for redistribution toward the competing

group.

One possible explanation for this is cross-generational altruism. The results of

Chapter 3 show that cross-generational altruism can have important effects on the

provision of age-specific, publicly provided goods. In particular, if a larger fraction of a

county's elderly has lived in that same county for a period of 5 years or more, education

spending is higher in the county. The elderly who migrate from another state or county,

however, are less willing to support education. These results indicate that connection to

the community, and members of the competing cohort who reside in the community, can

affect support for redistribution to the competing cohort.













CHAPTER 2
OUT WITH THE OLD, IN WITH THE NEW? CONGRESSIONAL VOTING
PATTERNS, INTERGENERATIONAL CONFLICT AND DEMOGRAPHIC CHANGE

To my knowledge, there are no analyses of the determinants of Congressional

action on aging issues. This chapter employs exogenous demographic change at the

appropriate level of aggregation to investigate changes in Congressional voting behavior.

Inman (1987) summarizes the models addressing the competition of various

groups for resources and resulting representative voting behavior. Among these models,

Downs' (1957) median voter model is perhaps the best known. Demographic change will

not necessarily change the position of the median voter, but if it does, the model predicts

that changes in voting behavior should result. Other models concentrate on the relative

sizes of the competing groups, positing that shifts in the relative group sizes should

translate to changes in voting. For example, Preston (1984) shows that government

transfers depend on the political power of the cohort, with a larger cohort getting higher

transfers. He suggests that this partially explains the increased well-being of the elderly

and decreased well-being of children from 1960 to 1980.

The House of Representatives is made up of 435 members. Each state's districts

must be of equal size, so population shifts result in the various states redrawing their

district boundaries. After each Decennial Census, each state redraws its Congressional

districts in accordance with these Federal guidelines and the state's own constitution. For

the 1990 Census, redistricting occurred after the 1992 session.1


A small amount of additional redistricting occurred in 1994; this analysis does not consider any effects
this might have on voting behavior.






Redistricting provides us with multiple observations on demographic variables

for individual lawmakers. We can examine how a legislator's voting patterns change on

issues related to demographic characteristics, such as age, when his constituent

demographics change. As you see below, the 1992 redistricting caused significant

change in the variables of interest for many individual lawmakers, creating a convincing

natural experiment.

There have been a few recent treatments in the political economy literature of the

effects of demographic change on Representatives' voting behavior. Stratmann (2000)

and Rothenberg and Sanders (2000) use redistricting to analyze the effects of exogenous

demographic change. Stratmann finds that, if more than half of the district changed in

the post-1990 redistricting, there was a greater absolute change in the Representative's

Americans for Democratic Action (ADA) index. Stratmann's primary focus is on the

shifting voting records of Representatives over the span of their careers, however, not on

demographic change. His empirical modeling of redistricting is somewhat limited

because of his construction of the explanatory variable for redistricting. Stratmann

visually inspected each district map and recorded redistricting as "yes" if it appeared that

the district boundaries changed by 50% or more, and "no" otherwise. This measure is not

subtle enough to capture changes in voting behavior that result from smaller changes in

district boundaries. Stratmann's measure for the change in constituency preferences due

to redistricting is the percentage of electoral vote for Clinton in 1992 less the percentage

for Dukakis in 1988. This is problematic because there were likely differing preferences

for these 2 candidates that were not due to redistricting, and possibly other changes over

time for which Stratmann does not control.






Stratmann also finds that as the district's median household income increases the

ADA falls; that is, voting becomes more conservative. His income measure is the change

in median household income from 1980 to 1990 in the reapportioned districts relative to

the districts not reapportioned. However, this again does not control sufficiently for

changes which occurred over time, as income changed for everyone over the course of

the decade. Changes in income and voter preferences as measured by the difference in

the district's Democratic vote cannot adequately capture the districts' changing

preferences for specific types of issues, such as those affecting the elderly.

Rothenberg and Sanders find that, if redistricting caused a shift in the liberalness

of the district, a greater ideological shift in Representatives' voting records resulted.

Their measure of liberalness is single-dimensioned: the change in percentage of the

two-party vote for Dukakis in the district before and after redistricting.

The advantage of my approach to redistricting is that it can capture subtle changes

due to minor adjustment of the district boundaries. In addition, my explanatory variables

address many potential dimensions of preference, rather than liberalness alone. My use

of only 1990 Census data, divided up by Congressional district before and after the

boundary changes, means that there are no changes in the variables at the aggregate level

for which I must adjust.

If lawmakers' voting patterns reflect the demography of their constituents and if

intergenerational conflict is an issue, we should observe the relation

vote on elderly issue = f(constituent age, income, fraction children in district, W)

where W is a vector of controls for other demographic variables including gender and

educational attainment.







Census Data

Data for this analysis are taken from the 1990 Census. These data were

aggregated from the census tract level to the Congressional district level for both the

102nd and 103rd Congresses.2 This provides us with a set of demographic data for each

legislator before and after the most recent redistricting.

The variables of interest in this analysis are primarily those related to age and

income, as they should capture any conflict between the generations. 3 The proportion of

the population aged 55 and older is used as the measure of older residents because people

are forward-looking to some extent. While most people do not retire until they reach

their 60s, people start to care about aging related issues somewhat earlier than that. A 55-

year-old probably does not have minor children, so should be less concerned with

children's issues. Fifty-five is also the age at which people are eligible to receive certain

age-related perks, such as senior citizen bank accounts and the ability to live in adult-only

housing communities. Eligibility for membership in the American Association of Retired

Persons (AARP) begins at 50.

The proportion of the population aged 18 or less is used to measure the proportion

of children in a district. Although education programs largely affect children above the

age of 4, the parents of infants are expected to be forward-looking for 4 years. Other

programs for children, such as welfare, begin at birth.

Each district's average per capital income is used as the income variable. Since

per capital income will be lower in areas with a higher proportion of children, ceteris


2 Special thanks are due to Alex Hooper for his most able assistance in constructing the Census dataset.

Proportions (e.g., fraction female and fraction children) are used to normalize districts. Although districts
tend to be rather uniform in size, there are some differences across states. This is particularly the case in
states with small populations, such as Alaska, Vermont and Wyoming, which have only I Congressional
district.







paribus, household or family income might be better measures. This relation could

cause bias in the coefficients for the income and interaction terms. However, per capital

income is highly correlated with family and household income and it is expected that

such bias will be small. Alternative income measures will be investigated in future

studies of this paper. The other variables are intended as controls, and include the

proportion of women and fraction of the constituency with a college degree.

The combination of redistricting and the 1992 elections brought about sweeping

changes in the composition of the House of Representatives, leaving only 322

Representatives who were in both the 102nd and 103rd Congresses. Table 2-1 shows

summary statistics of interest for the constituencies of these 322, including the pre-

redistricting levels and mean, standard deviation, minimum and maximum proportional

changes in the variables.

Table 2-1 Summary statistics: changes in demographic variables
Value before Proportion changes:
Variable Obs. redistricting Mean Std. dev. Minimum Maximum
Per capital income 322 14,214 0.0383 0.1488 -0.5715 0.9284
Fraction aged < 18 years 322 0.2458 -0.0019 0.0139 -0.0891 0.0515
Fraction aged 55 or older 322 0.3018 0.0015 0.0205 -0.0676 0.1218


Each of the proportional changes in the key variables has a standard deviation

larger than the mean. This suggests that there is sufficient variation for identification of

the effects of these changes. For example, the district's per capital income changed by

more than one standard deviation from the mean change for 64 of the 322 legislators.

The changes in the variables of interest for Florida's Representatives are shown in Table

2-2 to illustrate how some Representatives' constituencies changed.

Table 2-2 Changes in demographic variables for Florida's Representatives
Representative Per capital income Fraction aged 55 and over Fraction aged less than 18
Bacchus .0244 .0621 -.01015
Bilirakis .0782 -.0164 .0007
Gibbons .0102 -.0156 -.0012






11
Goss -.0628 -.0264 .0137
Hutto .0436 .0020 -.0016
Johnston -.0877 .0290 -.0140
Lewis .0024 .0208 -.0184
McCollum .0254 -.0095 -.0206
Peterson .1290 -.0064 -.0009
Ros-Lehtinen -.0264 -.0143 -.0061
Shaw .2886 .1218 -.0613
Steams .0515 -.0587 .0356
Young -.0909 .0068 -.0005

Voting Index

I created a database of votes on all the bills related to aging issues from 1987 to

1998; that is, 6 years before and after redistricting changes went into effect. Each

legislator's vote on each bill is recorded in the Congressional Quarterly Almanac (CQA).

I studied the descriptions given by CQA for every bill in the period of my study to

determine which votes were relevant. Votes were included if they were substantially

about aging issues, but this was a more difficult process than it might seem. Many votes,

while they include some provision for the elderly, have a variety of other provisions as

well. Most of the votes concerning Medicare, for instance, also include various other

budget items. One example is vote 84 from 1990, a substitute amendment to the Fiscal

1991 Budget Resolution, which included Medicare cuts of $3.2 billion, but also froze the

budget for defense programs and discretionary domestic programs. Most of the other

votes in this session concerning the budget resolution do not mention Medicare cuts, and

this vote does in large part concern Medicare. For these reasons, this vote was ultimately

included in the database.

Sometimes it was even difficult to determine whether a vote was pro-elderly.

For example, the 1988 Medicare Catastrophic Coverage Act was largely repealed in

1989. Vote 268 of 1989 was an "amendment to repeal...all stop-loss coverage of

hospital and doctor bills.. .while retaining broad prescription drug coverage and non-







Medicare provisions." This allowed slightly more coverage for Medicare patients than

the previous vote, so I was unable to determine whether voting "yes" on this bill would

be pro-elderly. It was ultimately excluded from the index for this reason. Similar issues

came up in other situations, such as when Medicare cuts were being argued. If a

legislator suggested a sizeable Medicare cut that was still smaller than the previous

suggestion, I was unable to record a "yes" vote as "pro" or "anti." I also excluded votes

where there was less than 20% contention; many of these votes would otherwise have

been perfect additions to the index. The final list of votes included is in the Appendix.

The Appendix shows that there were only 2 votes on elderly issues in 1991, none

in 1992, 2 in 1993 and 1 in 1994. In order to have a reasonable number of observations

for each Representative, I also excluded those legislators whose first observation was

after 1990 or whose last observation was before 1994. The remaining sample consists of

276 legislators. For each of these I created an "elderly-friendly" index for the periods

before and after redistricting. This index is a simple average of "pro-elderly" and "anti-

elderly" votes. "Pro" votes were assigned a value of I and "anti" votes a value of 0; thus,

an index value of 1 indicates a lawmaker always voting pro-elderly. The following table

contains the summary statistics for the voting index for the entire period of the analysis

and for the periods before and after the 1992 redistricting.

Table 2-3 Summary statistics: voting index
Observations Mean Std. Dev. Minimum Maximum
1987- 1998 276 0.6881 0.1646 0.3762 0.9444
1987- 1992 276 0.7226 0.1792 0.1944 1.0000
1993- 1998 276 0.6655 0.1700 0.3333 1.0000

I performed a simple F test to determine if the difference in the means of the

before and after redistricting periods is statistically significant. The F value for this test is






81.42, indicating that I can reject the null hypothesis that the means are equal at a

confidence level of 0.9999.

An alternate index was also constructed that included a larger set of votes. While

all the votes contained an elderly component, the additional votes included in the larger

set were less focused on elderly issues. The results in the regressions performed on the

alternate index were similar to those in the main regressions, but were weaker. This is to

be expected, as the effects of demographic change on elderly issues would be difficult to

separate from the other issues included in the bills.

Moreover, a regression was performed using the individual vote as the unit of

observation rather than the index. If each individual vote is a separate observation, the

much larger sample (11,797 rather than 552) means that each variable will tend to have

more statistical significance. However, there is no change in the crucial variables from

vote to vote during the period before reapportionment or the period after. The results of

this regression are similar to those in the primary regression, but the explanatory power

of the model is much lower, with an adjusted R-squared statistic of 0.4110.

Model

For my analysis of the effects of demographic change on the voting index, I

performed a panel regression with individual and time fixed effects. The individual

effects allow me to isolate the change in the voting index of legislators, rather than

capturing ideological characteristics to certain individuals that would cause them to vote

in a given way, regardless of the makeup of their districts. The dependent variable is the

voting index; I examine how it changes in response to the exogenous shock of

redistricting, which changes the districts' demographic characteristics. The explanatory







variables consist of legislator fixed effects and a vector of individual- and time-varying

characteristics, including

age (fraction 55 and older)
income (average per capital income)
an interaction term of fraction 55 and older and income
proportion of children (fraction younger than 18)
gender (fraction women)
and educational attainment (fraction with college degree).

Because the index is based on a different number of votes for each period and

because there are two observations for each Representative, the standard errors are

corrected for heteroskedasticity and clustering of errors within Congressional districts.

The sample is made up of 276 legislators for two periods, before and after redistricting.

Hypotheses

Older people should prefer spending to benefit the elderly if they are narrowly

self-interested. Thus, a district containing a large proportion of older people should be

reflected by a Representative with a higher index value.

Higher income areas will prefer less redistribution because the rich want to keep

their money, and there are fewer poor people who would prefer these programs for their

own narrow self-interest. Therefore, wealthy areas may oppose programs that

redistribute toward the elderly poor. Some programs might be viewed as being more

redistributional in nature than others. For example, legislation against age discrimination

in employment does redistribute money away from younger workers and toward older

people; however, this redistribution works through the labor market, so it is more indirect

than transfer payments. This might mean that it is viewed as less redistributional than

transfer payments.4 Programs for the elderly that are not viewed as redistributional might


4 Redistribution is likely not to the poorest if it works through the labor market. This, combined with the
Social Security tax cap, may mean that bills designed to curb age discrimination actually redistribute to the
rich.






be normal goods. In addition, as income increases, the opportunity costs of private

(i.e. family) provision of services to the elderly increase, raising the demand for public

provision of elderly services.

If income has a negative effect on support for elderly programs, this might be

mitigated by the proportion of elderly. If income has a positive effect, this should be

strengthened by a high number of elderly. Therefore, I include in my analysis an

interaction between proportion elderly and income.

Educational attainment is a measure of permanent income. This would imply

that, as educational attainment in a district increases, voting for elderly services ---

particularly those seen as strictly redistributional --- would fall. On the other hand, the

sign may be ambiguous due to the possible normality of services for the elderly, as

discussed above.

Intergenerational conflict implies that a high proportion of children in the district

should cause less support for elderly issues. A higher proportion of women should

increase support for elderly programs. Kenny and Lott (1999) found that suffrage

increased government spending at the state level and led to more liberal voting patterns

by Representatives. They suggest this is due to the lower incomes of women. Another

reason women might prefer more liberal government spending is that traditionally they

are society's caregivers, both for children and the elderly. Government programs can be

seen as substitutes for women's efforts in caregiving. Finally, women may prefer higher

government support for the elderly simply because they live longer than men.







Results

The effects of changes in the demographic makeup of the districts are shown in

the following table.

Table 2-4 Changes in elderly voting index
Effect of a 1 standard
Variable Coefficient deviation increase in X
Fraction aged 55 and up 2.4775 0.0310

(1.3463)
Per capital income 0.0289 -0.0398
(coefficient and SE 1000) (0.0162)
Interaction of income and 55 up -0.1287
(coefficient and SE 1000) (0.0531)
Fraction aged < 18 0.2901 0.0091

(2.0047)
Fraction female -0.7769 -0.0088

(2.6387)
Fraction college grads 1.1692 0.0608
(0.8016)
Adjusted R-squared 0.7752
Standard errors in parentheses. The full effects of the fraction 55 and older and income
are found using the means of the interacted variables.

The coefficients for the fraction of the population aged 55 and older and per capital

income are statistically significant at the 90% level, while the interaction of these two

variables is significant at the 95% level. The variables for the fraction of women,

children and college graduates have no statistically significant effect.

The full estimated effect of a I standard deviation increase in the proportion aged 55

and older, evaluated at the means, is to increase the voting index value by 0.03. This

positive effect is expected, if legislators respond to the perceived wishes of their

electorates; that is, if politicians respond to increases in the proportion of older people in

their constituencies by voting more often for benefits to the elderly. The effect of an






increase in the fraction aged 55 and older diminishes as income increases, becoming

negative when per capital income exceeds $19,250. This is substantially above the mean

per capital income of $14,250. If a high per capital income and high proportion of seniors

means that the seniors in that district are wealthy, as seems likely,5 these individuals may

be less likely to prefer government services for seniors.

Using the mean value of the proportion of seniors (which is 30%), a 1 standard

deviation increase in per capital income is associated with a 0.04 decrease in the index

value. This negative value is consistent with the idea that higher income areas prefer less

redistribution toward the elderly poor. The effect of an increase in per capital income

decreases as a larger fraction of the population is aged 55 and older, and becomes

negative when more than 22.5% of the population is in this age group. This is the case

in only 30 of the 552 observations.

The proportion of children in the district has no statistically significant effect. This

may suggest that intergenerational conflict is not an issue in this context. The proportion

of women in the district also has no statistically significant effect, offering no support for

the hypothesis that women prefer higher spending on social programs for seniors. The

fraction of college graduates in the district also has no statistical significance, suggesting

that per capital income captures most of the effects of permanent income.

Pairs of Bills

It is possible that the dependent variable, the elderly voting index, does not clearly

measure preferences toward generation-specific redistribution. This could be because

politicians vote for other reasons, such as political favor-trading or logrolling, or because

s The correlation coefficient for percent 55 and older and per capital income is -0.1308. If seniors tend to
have lower incomes on average than people more apt to be in the working population, a high proportion of
seniors and a high mean per capital income in a district may imply that the seniors are wealthier than
average in this district.







some of the votes in the index bundle elderly services with unrelated issues. In order

to address this concern, I undertook a second analysis. I searched for pairs of bills that

are very close in content and specific to aging issues, one before and one after

redistricting.6 I attempted to keep the votes as chronologically close together as possible

to control for unobserved time-specific variables. However, there was difficulty in

finding pairs of bills relating to a specific topic, so in some cases the bills are close to the

chronological edges of the dataset.

I then applied probit regression to estimate the relationship between voting on a

bill after reapportionment and demographic changes, controlling for voting on an earlier

bill. The dependent variable is the vote after redistricting. As in the index, these votes

were assigned a value of 1 if pro-elderly and 0 if anti-elderly. The explanatory variables

other than the previous vote are similar to those used in the index regressions, except that

changes in these variables are used rather than levels. These variables include the

changes in per capital income, fractions aged 55 and over and less than 18, and the age-

income interaction, with changes in the fraction female and fraction with a college degree

included as controls.

The first pair of bills examined is Vote 171 from 1987, "Housing and Community

Development/Elderly Rents", and Vote 156 from 1996, "Housing Overhaul/Family Rent

Caps." Both of these bills concern limits on rent or rent increases to elderly tenants of

subsidized housing, which are antipoverty measures.

I expect the probability that a Representative will vote for this antipoverty bill for

the elderly to increase if a larger fraction of his constituency is aged 55 and older.

However, this effect should diminish as the district becomes wealthier. If a high per
6 In their 1998 article on direct foreign investment, Bruce Blonigen and David Figlio examine pairs of
related bills to analyze legislators' voting behavior on issues of direct foreign investment.






capital income and high proportion of seniors means that the seniors in that district are

wealthy, these individuals will prefer less redistribution to the elderly poor. This implies

the interaction coefficient should be negative.

Antipoverty measures should be seen as strongly redistributional in nature. This

should allow me to unravel the ambiguous prior on the sign of the income coefficient: I

expect the income coefficient to be negative, as the wealthy prefer less redistribution.

The hypotheses for the coefficients on the other variables are the same as in the first

regression. That is, I expect the effect of an increase in the proportion of children or the

proportion of college graduates to be negative, while an increase in the proportion of

women should have a positive effect.

While an attempt was made to use pairs of bills that are chronologically close

together, there were no other votes in the time frame of this analysis on the specific issue

of redistribution toward the elderly. This necessitated using a pair of bills nine years

apart. Table 2-5 shows the summary statistics for the constituent demographics of the

164 Representatives who voted on this pair of bills.

Table 2-5 Summary statistics: fraction changes in constituent demographic
characteristics for representatives voting on antipoverty measures for elderly
Standard
Variable Mean Deviation Minimum Maximum
Aged 55 and older 0.0022 0.0798 -0.2640 0.3540
Per capital income 0.0313 0.1711 -0.5715 0.9284
Aged less than 18 -0.0039 0.0659 -0.3765 0.2869
College degree 0.0337 0.2071 -0.6843 0.9186
Female -0.0012 0.0108 -0.0668 0.0338

The second pair of votes concerns age discrimination. Vote 371 from 1986, "Age

Discrimination in Employment," addressed whether employers should be exempted from

age discrimination laws in hiring and terminating firefighters and police officers. Vote

291 from 1996, "Fiscal 1997 Transportation Appropriations/Commercial Pilot






Retirement," was to determine the appropriate age for mandatory pilot retirement.

Note that the first vote is from the year before the beginning of the index. There were no

votes related to age discrimination from 1987 to 1992, which made it necessary to use

this earlier vote.

In this pair, I separated the effects of aging by using age variables of 65 and older

and 55 to 64. This is because most people in the latter age group are still in the labor

force, and are thus more likely to be impacted by age discrimination in employment. By

the age of 65, however, most people are retired, and should care less about discrimination

in employment. Therefore, I expect the effect of an increase in those aged 55 to 64 to be

positive, but do not expect a significant and positive coefficient for those aged 65 and

older.

I have no strong prior expectations for the other coefficients, as their effects on

labor discrimination are unclear. The coefficient for the fraction of women may be

negative due to the lower labor force participation rates of women. Table 2-6 shows the

summary statistics for the constituent demographics of the 133 Representatives who

voted on this pair of bills.

Table 2-6 Summary statistics: fraction changes in constituent demographic
characteristics for representatives voting on age discrimination issues
Variable Observations Mean Std. Dev. Minimum Maximum
Aged 55 to 64 133 0.0160 0.1715 -0.3245 0.9737
Aged 65 and older 133 0.0189 0.1967 -0.5903 0.9960
Aged less than 18 133 -0.0103 0.1774 -0.4381 0.9694
College degree 133 0.0509 0.2586 -0.7001 1.0188
Female 133 -0.0150 0.1638 -0.3658 0.9258
Per capital income 133 0.0394 0.0760 -0.5715 0.9284

Table 2-7 shows the results of the probit regressions for the pairs of bills relating

to aging issues.






Table 7 Factors influencing congressional votes on elderly issues (probit regressions)
Antipoverty Antidiscrimination
CQ Almanac bill number and year 156, 1996 291,1996
Pre-redistricting vote 1.8571 -0.1669
(Poverty: Bill #171, 1987) (0.2715) (0.3017)
(Discrimination: Bill #371, 1986) 0.6380 -0.0582
Change in 55 and over 3.3730
(2.6352)
0.1130
Change in 55 to 64 3.1566
(3.5927)
0.1315
Change in 65 and older -2.4558
(2.3521)
-0.1132
Change in per capital income -3.9144 2.4361
(1.1913) (1.1244)
-0.1130 0.1499
Interaction of income and 55 and older 6.0760
(5.7405)
Interaction of income and 55 to 64 -25.4219
(13.3691)
Interaction of income and 65 and older 21.2754
(9.5832)
Change in younger than 18 0.9701 1.8442
(2.9549) (3.7198)
0.0254 0.1165
Change in women 9.9898 -4.9843
(12.8983) (6.5793)
0.0431 -0.2906
Change in those with college degree 1.2499 0.2385
(1.0093) (1.0264)
0.1030 0.0220
Pseudo R-squared 0.3734 0.1442
Standard errors in parentheses. The bottom number is the effect of a 1 standard deviation
increase in the explanatory variable. These numbers are computed using the mean values
for changes in income and age variables. For the pre-redistricting vote, the bottom
number is the effect of moving from a vote of 0 to a vote of 1.

When the antipoverty bills are isolated, the estimated effect of increasing per

capital income is negative and significant, as expected. The full effect, including the

interaction with age, is quite large. Calculated at the mean value for the proportion of 55






and up, a I standard deviation increase in per capital income in the district is associated

with a 0.27 decrease in the probability that the Representative will make a pro-elderly

vote on the bill.

The estimated effect of an increase in the fraction 55 and older is positive, as

expected, but only significant at the 80% confidence level. The coefficient for the

interaction term is also positive, which is not expected, but it is not statistically

significant. The full effect of the fraction aged 55 and higher, including the interaction

with income, is not as large as the income effect, but is still substantial. Calculated at the

mean value for per capital income, a 1 standard deviation increase in the proportion of

seniors is associated with a 0.11 increase in the probability that the Representative will

vote pro-elderly on the bill..

The Representative's action on the pre-redistricting vote is positive and highly

significant, as expected. Increases in the fractions of children, women and those with a

college degree have little effect.

In the age discrimination pair, the coefficient for the pre-redistricting bill is not

statistically significant and has an unexpected negative sign. This, combined with the

low pseudo R-squared statistic for this regression, suggests that this may not be a well

matched pair of bills. This could be due to the fact that the votes took place 10 years

apart. In addition, age considerations for police officers, firefighters and airline pilots

may be considered safety issues by some, which makes the age discrimination effects

difficult to separate.

The only statistically significant coefficients in this regression are those for the

change in per capital income and the interaction term. While the primary coefficients for

seniors are not statistically significant, the positive sign of the coefficient for the younger






cohort is as expected. The full effect of a I standard deviation increase in the fraction

of the district's population aged 55 to 64, evaluated at the means, is to increase the

probability the Representative will vote to limit age discrimination in this bill by 0.13.

This sign is expected because people in this age group are likely to still be in the work

force, and thus be affected by issues of age discrimination. The full effect of a 1 standard

deviation increase in the fraction aged 65 and older is to reduce the probability by 0.11.

This suggests that older people not impacted by employment issues --- namely, those

above the typical working age --- are less concerned with these issues.

The full effect of a I standard deviation increase in per capital income, evaluated

at the means, is to increase in the probability the Representative will vote to limit age

discrimination in this bill by 0.15. As noted above, none of the other variables have a

significant effect. It is interesting to note that there is no significant effect due to an

increase in the proportion of children in either of the regressions above. This suggests

that intergenerational conflict does not occur in this direction, i.e. a high proportion of

children does not negatively impact votes on elderly issues.

Children's Issues

In order to test whether intergenerational conflict does occur in the opposite

direction, that is, whether a high proportion of aging persons negatively impacts votes on

children's issues, I considered 2 pairs of bills on children's issues. The first of these pairs

is Vote 372 from 1992, "Child Welfare and Nutrition Programs/Passage" and Vote 543

from 1995, "Fiscal 1996 Agriculture Appropriations/WIC Food Program." These votes

deal with antipoverty measures for children, specifically with nutrition programs for

children.






If intergenerational conflict is a factor in the support of antipoverty programs

for children, the coefficient for 55 and older should be negative. As in the regression on

antipoverty programs for the elderly, I expect the sign on the income coefficient to be

negative. The coefficient for fraction children should be positive, as districts with a high

proportion of children should prefer more children's services. The coefficient on fraction

women should also be positive, since women tend to be the primary caregivers for

children, and government provision of services is a substitute for private provision. If

education is related to permanent income, the education coefficient should be negative.

Table 2-8 shows the summary statistics for the constituent demographics of the 214

Representatives voting on this pair of bills.

Table 2-8 Summary statistics: fraction changes in constituent demographic
characteristics for representatives voting on antipoverty measures for children

Variable Observations Mean Std. dev. Minimum Maximum
Aged 55 and older 214 0.0009 0.0724 -0.2640 0.3540
Aged less than 18 214 -0.0028 0.0611 -0.3765 0.2869
College degree 214 0.0278 0.1875 -0.6843 0.9186
Female 214 -0.0010 0.0102 -0.0668 0.0338
Per capital income 214 0.0298 0.1554 -0.5715 0.9284

The second pair of votes on children's issues is Vote 385 from 1992,

"Neighborhood Schools Improvement Act," and Vote 496 from 1993, "School

Improvement/Passage." Again, if intergenerational conflict is an issue, the 55 and

older coefficient should be negative. The fraction children coefficient should be positive.

The income variable should also be positive here, as education is generally thought to be

a normal good. This idea is discussed more fully in Chapter 3. This positive effect

should particularly be the case since the votes concern national education policy;

therefore, Tiebout endogeneity is not likely to be an issue. The coefficient for fraction

women may be positive for the same reason as in the previous regression; that is, women






25
traditionally are the primary caregivers for children. They should be expected to prefer

more education as a substitute for their care, and their close relationship with children

suggests that altruism may be a factor. Fraction with college degree should be positive,

as higher educational attainment is direct evidence of preference for education. Table 2-9

shows the summary statistics for the constituent demographics of the 263 Representatives

voting on this pair of bills, while Table 2-10 shows the results of the probit regressions

for the pairs of bills relating to children's issues.

Table 2-9 Summary statistics: fraction changes in constituent demographic
characteristics for representatives voting on education improvements
Variable Observations Mean Std. dev. Minimum Maximum
Aged 55 and older 263 0.0052 0.0706 -0.2640 0.3540
Aged less than 18 263 -0.0060 0.0577 -0.3765 0.2379
College degree 263 0.0277 0.1914 -0.6843 1.0023
Female 263 -0.0002 0.0101 -0.0668 0.0427
Per capital income 263 0.0312 0.1435 -0.5715 0.9284

In the first regression, the only significant coefficient is the one for the pre-

redistricting vote. As expected, the earlier vote's coefficient is positive, and very large.

The estimated effect of moving from an anti-child to a pro-child position on the earlier

bill is associated with a 0.65 increase in the probability of voting for antipoverty

measures for children on the later bill. The coefficient for the fraction change in income

is significant at only the 85% confidence level. The negative sign is consistent with the

result found earlier: there is a negative relation between income and votes on antipoverty

measures. However, note how much smaller is the effect of a one standard deviation

increase in per capital income (-0.07) here than in the regressions concerning elderly

antipoverty issues (-0.2654).

In the education regression, the previous vote is again the only variable to have a

statistically significant effect. The estimated effect of the previous vote is again positive






Table 2-10 Factors influencing congressional votes on children's issues (probit
regressions)
Antipoverty Education
CQ Almanac vote number and year 543, 1995 496, 1993
Pre-redistricting vote 2.1539 3.3201
(Antipoverty: Bill #372, 1992) (0.2550) (0.4015)
(Education: Bill #385, 1992) 0.6106 0.7585
% change in 55 and older -0.7968 -10.3924
(10.4750) (13.8709)
0.0000 -0.0183
% change in per capital income -1.4852 0.0101
(1.1316) (1.4030)
-0.0629 0.0012
Interaction of income and 55 and older 24.2130 47.6047
(27.9514) (69.2474)
% change in younger than 18 -2.9415 11.3532
(15.5450) (21.9858)
-0.0115 0.0189
% change in women -9.7333 42.5152
(31.5709) (32.2058)
-0.0139 0.0253
% change in those with college degree -2.5580 13.2747
(7.4047) (11.5623)
-0.0187 0.0434
Pseudo R-squared 0.4389 0.6592
Standard errors in parentheses. The bottom number is the effect of a I standard deviation
increase in the explanatory variable. These numbers are computed using the mean values
for changes in income and age variables. For the pre-redistricting vote, the bottom
number is the effect of moving from a vote of 0 to a vote of 1.


and significant, with a very large estimated effect (0.78) of moving from an anti-child to

a pro-child stance on the earlier vote.

In both of these pairs, the pre-redistricting vote is responsible for nearly all the

explanatory power of the model. The pseudo R-squared statistic for the antipoverty pair

is 0.4764, but is 0.4538 when only the pre-redistricting vote is used as an explanatory

variable. Similarly, the pseudo R-squared for the education pair is 0.6373, but is 0.6266

when only the pre-redistricting vote is used as an explanatory variable. This suggests that






demographic change has little effect on how Representatives vote on issues pertaining

to the well-being of children.


In the education regression, the effect of the previous vote is again positive and

significant, with a very large estimated effect (0.7585) of moving from an anti-elderly to

a pro-elderly stance on the earlier vote. The income coefficient is not significant, and is

very small in any case. The coefficient for the fraction of women is positive and

significant, as expected. However, a 1 standard deviation increase in the proportion of

women is associated with only a 2.5% increase in the dependent variable. None of the

other variables are significant.

Conclusion

An important question in the political economy literature is whether politicians

vote in a way that reflects the presumed preferences of their electorates, or simply vote

according to their own preferences. Using demographic data from the 1990 Census and

vote records of individual members of the House of Representatives from 1986 to 1998,

and redistricting as a natural experiment, this chapter offers evidence that politicians are

indeed Representatives when voting on issues related to the elderly.

It should be noted that this analysis does not take several factors into account.

More work should be undertaken to determine the effects of differential voting

participation rates of competing cohorts, as well as the effects of interest group power

(e.g. fraction membership in AARP). Also, if the newness of a Representative to his

constituency matters, the chronological distance from the point of redistricting should

play a role in voting behavior. Another key question is the extent to which politicians

whose voting records do not reflect the demographic makeup of their electorates are

punished by defeat in election. Finally, it would be interesting to see if politicians vote






28
strategically to ensure reelection, voting more in line with the characteristics of their

constituencies close to elections.

This chapter offers a unique and effective identification strategy which can be

used in addressing the additional questions raised here. This area of inquiry will

contribute to a better understanding of voting behavior in a representative democracy as

well as issues of intergenerational conflict.













CHAPTER 3
IT TAKES A VILLAGE? INTERGENERATIONAL CONFLICT AND
COOPERATION IN EDUCATION EXPENDITURES

Introduction

A major theme in the public economics literature involves the determinants of

spending on public education. Researchers including Poterba (1997, 1998), Ladd and

Murray (2001), and Harris, Evans and Schwab (2001) have examined the effects of

intergenerational conflict, and have found that the elderly have either a negative effect or

no effect on education spending. A consistent finding in these analyses is that differences

in the races of the elderly and the younger cohort are associated with significantly lower

education spending. This paper provides evidence as to why support of education by the

elderly varies. My results offer a new way of thinking about altruism, and help to explain

the disturbing results others have found that might suggest racism among the elderly. If

the elderly develop connections to their communities, the migration patterns of the

elderly in an area should affect education spending in the area. My model includes

unique instruments to explicitly explain the migration behavior of the elderly.

In a state-level analysis, Poterba (1997) finds that education spending falls as the

state's fraction of elderly rises. This negative result is not robust to the addition of an

urbanicity measure, however. Poterba also finds a strong negative effect of differences in

the racial composition of the school-aged and elderly cohorts, which henceforth will be

called cross-cohort racial heterogeneity.








The relevant Tiebout market for education is generally accepted to be the county

or metropolitan statistical area (MSA). Ladd and Murray (1999) point out that a county-

or district-level analysis can capture many important features of education finance that

would be missed at the state level. They conduct an analysis similar to Poterba's, but at

the county level with county and time fixed effects. They find that the proportion of

elderly has no significant effect on per child education spending in any specification.

However, they too find cross-cohort racial heterogeneity to have a negative effect. Ladd

and Murray also perform a state-level analysis that includes measures for how retirees

and children are distributed within states and how segregated the cohorts are. Their

results indicate that segregation from the school-aged cohort makes the elderly less

willing to support education.

The consistent negative effect of the "racial mismatch" leads to an interesting

question. Are the elderly simply racist? I propose a different explanation. The negative

effects of cross-cohort racial heterogeneity, combined with Ladd and Murray's

segregation result, suggest that some sense of connectedness might matter to the elderly

when they make decisions about funding for education.

The elderly do not enjoy the direct benefits of public education, unlike families

with children. Other researchers including Poterba (1998) and Kemnitz (2000) have

discussed the external benefits of education that might make the elderly more willing to

improve it. These include decreased crime, a higher level of services such as health care,

and increased Social Security contributions from the better educated younger cohort. In

the real world, though, the elderly will not live to enjoy these external benefits of paying

to improve education now. For example, improving a child's education now may make








him less likely to commit crimes now, but the effect presumably (and hopefully) extends

to the child's full lifetime. If this is the case, the elderly will not live long enough to

benefit from the cumulative reduction in crime. The immediate gains to the welfare of

the elderly will be small; thus, the elderly should be unwilling to fund education.

Altruism would mitigate this tendency. Following Becker's (1981) work on the

economics of the family, grandparents are very likely to support those children who mean

the most to them: their grandchildren. If grandparents living in the same locale as their

grandparents shift the position of the median voter, this will lead to higher spending on

education. Once we are outside the confines of the family, however, the literature does

not explain why we might be more altruistic toward some people than others.

I propose that the extent to which we are altruistic depends on how connected we

are to the beneficiaries of our good will. Charity begins at home, and then extends to

others. That is, if the elderly subscribe to the idea that "it takes a village to raise a child,"

they will also act to help other children who are part of their village --- close to them in a

geographic sense or with whom they identify in some other way.

It may also take time for people to identify with their communities; if this is the

case, they will become more altruistic with longer residence. If recent residents of a

community maintain some sense of connection to their previous residence, support for

education among the elderly should be highest where more of the elderly are long-time

residents (i.e. moved a distance of zero); and lowest where more of the elderly have

moved in from far away.

The next section of the paper discusses the data and the empirical strategy, which is

to estimate of the effects of elderly migration on school spending with ordinary least








squares (OLS). The following section presents the results of the ordinary least squares

regressions. However, this approach does not take into account the possible endogeneity

of migration with respect to education spending. This is followed, then, by a section

including a discussion of this Tiebout endogeneity and how this issue is addressed. I use

several creative instruments, including the number of golf courses per square mile in each

county, to directly explore the migration patterns of the elderly. The following section

presents the results of the first-stage regressions, which predict elderly migration from

out of state, as well as the results of two-stage and three-stage least squares regressions to

explain school spending. Finally, some concluding remarks are presented.

Empirical Strategy and Data Collection

In this paper I explore the effect of elderly migration on education expenditures.

Migration statistics from the United States Census of Population are available only at the

county and state levels, precluding a district-level analysis employing these data. This

information comes from respondents' answers to questions about where they lived five

years earlier. Migration data that describe previous residence are available at the county

level for 1990 only 7, so a panel analysis is also impossible. I therefore perform a county-

level cross section analysis for 1990, using demographic and migration data from the

1990 Decennial Census of Population and Housing and school finance data from the

1992 Census of Governments.








7County-to-county migration data are also available from the 1980 Census, but do not include information
on migrants' previous residences.








My first approach is to estimate an ordinary least-squares model with state dummy

variables. The dependent variable in my county-level model is the natural logarithm of

per pupil state plus local revenues to education in county i in state j.8 The basic model is:

ln(per pupil revenuesij) = Xgl + Zgy + Sj8 + ej

All variables are in natural logarithms to yield elasticities. The explanatory

variables of primary interest are those in Zy, which is a vector of migration data

consisting of the fraction of each county's population that is both aged 65 and higher and

that lived in the same county in 1990 as in 1985; the fraction that migrated to the county

from within the same state; and the fraction that migrated from a different state.

If the elderly are self-interested and the external benefits they receive from

education are relatively small, a higher proportion of elderly in a county should be

associated with lower education spending. However, the migration patterns of retirees

may mitigate this tendency. Community-based altruism implies that elderly nonmovers

should not have a large negative effect on education spending. A larger proportion of

retirees moving within the state may have a positive or negative impact on education; if

positive, the effect should be smaller than that of the elderly nonmovers. If negative, it

should be smaller in magnitude than the negative effect of a higher proportion of elderly

inmigrants from out of state.

X is a vector of demographic control variables. These include the number of

school districts per hundred thousand population, fraction of the population that is



' Poterba (1997, 1998) and Ladd and Murray (2001) use per child measures to help prevent any
endogeneity issues with the number of public school pupils. However, Harris, Evans and Schwab (2001)
use per pupil measures. The education revenues or expenditures and number of school-aged children are
from two different data sources. I follow HES and construct the education finance variables from a single
data source to avoid measurement error.








nonwhite, a term for cross-cohort racial heterogeneity, median household income,

fraction of those aged 25 and older with at least some college, a tax price variable,

fraction of homes owner occupied and fraction of the population in an urban area. The

number of school districts is from the 1992 Census of Governments; the remaining

variables in X are from the 1990 Census of Population and Housing.

Community-based altruism should be higher where the communities are smaller.

This would suggest that a larger number of districts per hundred thousand persons should

be reflected by higher education spending, ceteris paribus. However, there should also be

an offsetting competition effect, which will drive costs, and therefore expenditures, down

as the number of districts increases.

A large literature suggests that the quality of education received by nonwhites is

less than that of whites; this is presumably due to the historically inferior educational

opportunities afforded to nonwhites. Lower educational quality among adults is likely to

be associated with lower demand for education for children. This would imply lower per

pupil expenditures where a higher proportion of the population is nonwhite. Racial

differences between the elderly and children may also cause lower education spending

because the elderly may feel more altruism toward children who are more like them.

This cross-cohort racial heterogeneity is measured by the absolute value of the difference

between the nonwhite share of the population aged 5 to 17 and the nonwhite share of the

population aged 65 and higher.9'10



S0.0000000 was added to the fractional values in each county to prevent undefined values in the natural
logarithm, and the resultant decrease in sample size.

"0 The linear distance between the nonwhite fractions of each cohort measures cross-cohort racial
heterogeneity. Previous researchers use a simple difference, but the linear distance is the absolute value.
97 of the 2520 counties in my sample have a cross-cohort racial heterogeneity value less than zero. In








Education is generally accepted to be a normal good; higher median household

income should be associated with higher spending on education. Educational attainment

is a commonly used proxy for permanent income, but higher educational attainment is

also direct evidence of a taste for education. In either case, a larger fraction of the

population aged 25 and older with at least some college should be associated with higher

education spending.

Increasing the price of education, however, should lead to less demand for it. The

tax price variable is equal to median household income divided by mean household

income, multiplied by the mean number of children per household. If taxes for education

are proportional to income, this is the median voter's cost of increasing per child

education spending by one dollar." As median income increases relative to mean income

--- that is, as the income distribution becomes less skewed --- the median voter's

education subsidy from wealthier families shrinks.

Several authors, including Black, Figlio, and Harris, Evans and Schwab have

shown or suggested that increases in education spending are capitalized into housing

prices. A higher fraction of homeowners in a county should thus increase per pupil

spending. However, some researchers also have found homeownership to have a

negative effect on school spending. This has been explained as a renters' illusion effect,

where renters do not believe property tax increases are borne by them. Hence, they will

prefer higher school spending than otherwise identical homeowners.12


these counties, there is a higher fraction of nonwhite elderly than nonwhite children. None of the results
shown later change appreciably when the alternate measure is used.

" See Lovell (1978) for a thoughtful discussion of this tax price variable and empirical evidence of its
validity.

2 Denzau and Grier (1984).








An increase in the fraction of a county's population living in an urban area is

expected to have a negative effect on per pupil spending because of economies of scale

and other cost differences between urban and rural areas, such as the higher cost of

transporting pupils for longer distances in rural areas. This will be offset to some extent

by the higher wages paid in urban areas.

The education finance data from the 1992 Census of Governments include

revenue to education from local, state and federal sources and the number of pupils

enrolled in the fall semester, to yield the dependent variable and federal aid per pupil.

These data are aggregated from the school district level to the county level. Of the

15,868 districts listed in the Census of Governments, 98 were dropped because they

could not be matched to a county. These include special districts for regional technical

schools and other schools that appear to draw students from more than 1 county. Eight

hundred and seventeen districts were dropped because their total revenue or enrollment

was recorded as zero; 19 additional observations were dropped because of missing or

obviously incorrect data. Many of these districts truly have no students; after district

consolidation, it often takes as long as 10 years for a district to be dissolved. Districts in

the independent cities of Virginia were not included in my county-level analysis, and

some of the demographic data are not available for Hawaii or Alaska counties or the

District of Columbia.

Some states mandate equalization of educational resources across districts within

the state. In these states, there is little or no choice in education spending at the local

level, so that expenditures on public education will not be affected by differences in

demand due to differences in the characteristics of the residents. The twelve states with









the lowest variation in per pupil spending, as measured by the coefficient of variation of

per pupil revenues, are thus omitted from the sample.13 The final aggregated sample

consists of 2,520 counties from the remaining 36 continental states.

Sj is the vector of state dummies. These serve as controls for characteristics, such

as state law and the structure of public education, that affect the state's education

spending. State contributions make up a large fraction of a school district's revenues for

education. Finally, the error term ij is assumed to have zero mean and constant variance.

Differences in costs between rural and urban areas (e.g. teacher wages,

transportation costs and economies of scale) also affect expenditures; therefore, I

separately consider the urban portion of the sample. This subsample consists of the 1,850

counties from the full sample for which less than 100% of the population lived in a rural

area. Table 3-1 shows the key summary statistics for the 2 county-level samples.

Table 3-1 Sample summary statistics
Full sample
Variable Mean Std. dev. Min. Max.
Per pupil state plus local revenue
4,592 1,526 1,032 23,250
Fraction of the county's population that is aged 65
and older and that did not move between
1985 and 1990' 0.1185 0.0364 0.0108 0.2810
Fraction of the county's population that is aged 65
and older and that moved from within the
state but not the county 0.0254 0.0107 0.0000 0.1363
Fraction of the county's population that is aged 65
and older and that moved from out of state 0.0055 0.0063 0.0000 0.0931
School districts per hundred thousand population 29.20 63.35 0.05 1,298.70
Fraction nonwhite 0.1356 0.1563 0.0000 0.9490


Twelve states (Arizona, Arkansas, Connecticut, Delaware, Indiana, Iowa, Kentucky, Maine, New
Hampshire, New Jersey, Rhode Island, and Wisconsin) have coefficients of variation less than 0.1. These
twelve include four (Arkansas, Connecticut, Kentucky and New Jersey) of the ten states that had state
Supreme Court decisions overturning the education finance systems and mandating equity and/or adequacy
in educational resources by 1990. See Figlio, Husted, and Kenny (2001) for more discussion of these
decisions.









Nonwhite share aged 5 to 17 less nonwhite share
aged 65 and older
Median household income

Tax price

Fraction of the county's population aged 25 and
older with at least some college
Per pupil federal revenue

Fraction homeowner

Fraction urban


Urban subsample
Per pupil state plus local revenue


Fraction of the county's population that is aged 65
and older and that did not move between
1985 and 1990 '
Fraction of the county's population that is aged 65
and older and that moved from within the
state but not the county

Fraction of the county's population that is aged 65
and older and that moved from out of state

School districts per hundred thousand population
Fraction nonwhite
Nonwhite share aged 5 to 17 less nonwhite share
aged 65 and older
Median household income
Tax price
Fraction of the county's population aged 25 and
older with at least some college
Per pupil federal revenue
Fraction homeowner


0.0762 0.0801 -0.1096 0.4596

23,636 6,185 9,791 54,801

2.1413 0.2546 0.8005 5.6337

0.3572 0.1097 0.1170 0.8120


387.49 332.57 0.00 6,775.44


0.6146 0.0915 0.1010

0.3528 0.2950 0.0000


0.8270

1.0000


4,413 1,247 1,032 20,340


0.1106 0.0319 0.0108


0.0247 0.0096 0.0016



0.0056 0.0069 0.0000


15.62 22.16 0.05
0.1425 0.1466 0.0006
0.0828 0.0777 -0.1096

24,712 6,390 9,791
2.1610 0.2444 1.2427
0.3692 0.1113 0.1410


0.2402


0.0826



0.0931


475.66
0.9490
0.4374

54,801
5.2211
0.8120


359.62 284.40 60.26 6,775.44
0.6225 0.0818 0.1910 0.8270


Fraction urban 0.4806 0.2390 0.0006 1.0000
Demographic and migration data are from 1990 and school finance data are from 1992. The full
sample is made up of the 2520 counties from the 36 continental states that have a coefficient of
variation of per pupil state plus local revenues greater than 0.1. The urban subsample is made up
of the 1850 observations in the full sample for which the fraction rural is less than 1.


The characteristics of the full and urban samples are strikingly similar. The

dependent variable is per pupil state plus local revenues to education. Mean revenues are

$4,592 in the full sample and $4,413 in the subsample, with standard deviations of $1,526








and $1,247. The range of revenues is large, from approximately $1,000 to over $20,000,

but per pupil spending increases smoothly over the entire range. Twenty-three counties

in the full sample and 9 in the urban sample have per pupil spending greater than twice

the mean value.

The explanatory variables of main interest are those for elderly migration. Those

aged 65 and higher make up 14 to 15% of the county's population, on average. This

elderly group is broken down by migration characteristics. On average, the elderly who

lived in the same county in 1985 and 1990 make up 11 to 12% of the county's

population. Approximately 2.5% of the population consists of the elderly who migrated

within a state, and an additional .06% are elderly from out of state.

Results

The results of the ordinary least squares regressions with robust standard errors

are presented in Table 3-2. State dummy variables are included in all specifications, and

all variables in these regressions are in natural logarithms to yield elasticities. The

second column for each specification shows the result (in dollars of per pupil revenue) of

a 1 standard deviation change in each explanatory variable. The relative magnitudes of

the elasticities and the dollar effects may differ because of the variation in the standard

deviations of the explanatory variables. The first page contains the results of the

regressions performed on the full sample; the second page contains the results for the

urban sample.

The variables of primary interest are those related to age and migration. The

large, positive estimates for the elderly nonmovers in both samples are surprising. The

benchmark group is all households not aged 65 and older, who may or may not have

children. The positive estimated elasticity suggests that elderly nonmovers are more








Table 3-2 Determinants of per pupil state and local revenue for education county-level
cross-section (1990), ordinary least squares regressions
Effect of a 1 SD
Full sample Elasticity change in X
Fraction of the county's population that is aged 65
and older and that did not move between 1985
and 1990' 0.1360 205.27
(0.0280)
Fraction of the county's population that is aged 65
and older and that moved from within the state
but not the county -0.0170 -49.22
(0.0082)
Fraction of the county's population that is aged 65
and older and that moved from out of state -0.0131 -94.89
(0.0034)
School districts per hundred thousand population 0.0685 426.50
(0.0087)
Fraction nonwhite 0.0074
(0.0057)
Absolute value of the nonwhite share aged 5 to 17
less the nonwhite share aged 65 and older -0.0050
(0.0060)
Median household income 0.3416 384.14
(0.0424)
Tax price -0.1292 -64.30
(0.0692)
Fraction of the county's population aged 25 and older
with at least some college 0.0802 110.13
(0.0241)
Per pupil federal revenue -0.0284
(0.0182)
Fraction homeowner -0.2655 -205.22
(0.0348)
Fraction urban -0.0043 -126.56
(0.0006)
R squared 0.6947
Urban subsample
Fraction of the county's population that is aged 65 and
older and that did not move between 1985 and
1990' 0.1121 155.21
Fraction of the county's population that is aged 65 and
older and that moved from within the state but not
the county -0.0007
(0.0143)








Fraction of the county's population that is aged 65 and -0.0094 -45.50
older and that moved from out of state
(0.0041)
School districts per hundred thousand population 0.0592 318.33
(0.0102)
Fraction nonwhite -0.0029
(0.0085)
Absolute value of the nonwhite share aged 5 to 17 less
the nonwhite share aged 65 and older 0.0017
(0.0058)
Median household income 0.4272 465.50
(0.0604)
Tax price -0.2061 -94.67
(0.0749)
Fraction of the county's population aged 25 and older
with at least some college 0.0595 76.72
(0.0283)
Per pupil federal revenue 0.0627
(0.0362)
Fraction homeowner -0.1540 -95.39
(0.0418)
Fraction urban -0.0022
(0.0061)
R squared

The second column is the effect, in dollars of per pupil revenue, of a I standard
deviation change in the explanatory variable.
All variables are in natural logarithms. All specifications include state dummies.
Individuals are considered nonmovers if they moved within the county. Demographic
and migration data are from 1990 and school finance data are from 1992. The full
sample is made up of the 2520 counties from the 36 continental states that have a
coefficient of variation of per pupil state plus local revenues greater than 0.1. The urban
subsample is made up of the 1850 observations in the full sample for which the fraction
rural is less than 1.

likely to support education, compared younger households. These younger households

include those without children that moved, and these households might be expected to

have very low demands for education. A 1 standard deviation increase in the log of the

fraction of nonmovers increases revenues by $205 in the full sample and $155 in the








urban subsample. Recall that the mean value of per pupil revenues is approximately

$4,500.

The estimated elasticities of the migration coefficients only partially follow the

predicted ordering. In the full sample, the estimated elasticity for instate movers is -0.02

and statistically significant, with a 1 standard deviation rise leading to an estimated

decrease of $49 in per pupil revenues. The elasticity for movers from another state is

predicted to be more negative than this estimate, but the elasticity is -0.01. However, the

dollar effect of -$95 is indeed more negative than the estimate for instate movers.

The migration coefficients in the urban subsample do follow the predicted

ordering: the estimated elasticity for instate movers is not significantly different from

zero, while the elasticity for movers from another state is estimated to be -0.009. The

dollar effects of a 1 standard deviation change in these variables also follow this ordering;

the instate movers' effect is essentially zero while the effect of the movers from another

state is estimated to be -$46. It is likely that these estimates are biased, however, because

this model does not control for the endogeneity of location choice with respect to

education spending.

It is interesting to note that when the migration patterns of retirees are taken into

account, cross-cohort racial heterogeneity has no effect. The estimated elasticity of this

racial mismatch term is not significantly different from zero in either sample. One

potential explanation for this is that the mismatch term is simply picking up the effects of

migration.

Recall that my definition of racial mismatch differs somewhat from that of

previous researchers. Poterba and Harris, Evans and Schwab use the simple difference








between the nonwhite fraction of school-aged children and the nonwhite fraction of those

aged 65 and older, while I use the absolute value of this difference. The absolute value

measures cross-cohort racial heterogeneity, with the hypothesis that the elderly will be

less likely to support education if they are of a different race than the local

schoolchildren. The simple difference, however, has a different interpretation: an

increase in this measure means there are proportionally either more nonwhite children or

more white elderly, or both. The negative estimated effect that others have found for this

variable implies that the white elderly may be racist. Ninety-seven counties in my

sample have a negative mismatch value, where there is a larger fraction of nonwhite

elderly than nonwhite children; for these counties, the difference in definition matters.

The results of my analysis do not change when the simple difference is used rather than

the absolute value. This may be simply because such a small fraction of the sample is

affected by the change in definition. However, using the absolute value rather than the

simple difference does lead to a slightly better fit, as measured by the adjusted R squared.

The estimated elasticity for the number of school districts per hundred thousand

persons in the population is positive and significant in both samples, as predicted. This

result is consistent with the idea that support for education is likely to be higher where the

community is smaller, because of the higher level of community-based altruism. The

estimated elasticities are not large at 0.06 to 0.07, but the dollar effects associated with a

1 standard deviation increase in this variable range from a $318 to $426 increase in per

pupil revenues.

The other variables in the models are intended as controls, but bear some

discussion here. The estimated effect of increased median household income is positive








and significant, as predicted, with estimated elasticities of 0.34 and 0.43. These

elasticities are larger than those obtained by Ladd and Murray, but are consistent with the

income elasticities more commonly found in the literature.14 The estimated dollar effects

are $384 in the full sample and $466 in the urban subsample.

An increase in the nonwhite proportion of the population has no significant effect

in either sample. This suggests that median household income captures all of the effects

of educational quality. The estimated elasticity of the tax price variable is negative and

significant in both samples, as predicted; as the price of education rises, less education is

demanded. The estimated price elasticity is -0.13 for the full sample and -0.21 for the

urban sample; the dollar effects of a 1 standard deviation increase in the tax price variable

are -$64 and -$95, respectively.

As predicted, an increase in the fraction of the adult population with at least some

college education has a positive effect on per pupil revenues. The fact that the estimated

elasticities are significant is consistent with the idea that educational attainment is a

signal of taste for education, and is not just a proxy for income. It could also be that the

education variable is picking up the part of permanent income not captured by current

income. The effects are not large, though, with estimated elasticities of 0.06 to 0.08 and

estimated dollar effects of $76 to $110.

The estimated effect of an increase in the fraction of homeowners is negative and

significant in both samples. This supports the renters' illusion hypothesis, where renters

do not believe property tax increases are borne by them. Hence, they will prefer higher




" See Denzau and Grier (1984) for consistent estimates of key parameters affecting education spending.








school spending than otherwise identical homeowners.15 These negative elasticities are

fairly large, at -0.15 to -0.27, but the estimated dollar effects are only -$85 to -$95.

Finally, an increase in the fraction of the population living in an urban area has a

negative effect in the full sample and no significant effect in the urban subsample. The

estimated elasticity is small, at only -0.004, with estimated 1 standard deviation effect of

-$127 on per pupil revenues. The small size of the estimates is likely due to the offsetting

cost differences between urban and rural areas, although the economy of scale effects and

higher rural transportation costs dominate the higher costs of urban salaries in the full

sample.

Controlling for the Effects of Tiebout Sorting

Researchers including Hoyt and Rosenthal (1997) offer evidence consistent with

Tiebout's idea that households sort efficiently across locations to receive their preferred

levels of public goods. The elderly can control their tax liabilities through location

choice as well as through voting. Therefore, the proportion of elderly in a county is

endogenous with respect to education expenditures. This makes bias likely in the

coefficients from the ordinary least squares regressions. Harris, Evans and Schwab and

Ladd and Murray attempt to control for this endogeneity using an instrumental variables

approach. They instrument for the fraction elderly in a given year with the fraction aged

55 to 64 in the district (or county) 10 years before. Instead of using the fraction elderly in

the county as the key explanatory variable, my model employs the migration variables,

which sum to the fraction elderly. I use several instruments to directly explore the

migration patterns of the elderly.


5 See, again, Denzau and Grier.








These instruments are employed in two-stage least squares and three-stage least

squares frameworks. The advantage of the three-stage procedure is that the first stage

includes only the variables that are dictated by theory. The two-stage process includes all

the second-stage variables in the first stage regressions, which is less intuitively

appealing. It avoids a serious problem of three-stage least squares, however, where

misspecification of either equation in the system will contaminate the coefficients for the

entire system.

Graves (1979) showed that areas with high variations in temperature or other

manifestations of undesirable climates have lower inmigration for groups of all ages,

ceteris paribus. This result should be particularly strong for elderly movers, as they do

not face labor market constraints. Mean January temperature is included as an instrument

to capture the effects of climate. It is a proxy for temperature variation, as there is less

variation in summer temperatures than in winter temperatures across regions. In addition,

the elderly are particularly susceptible to the dangers of cold winters, such as slipping on

ice and shoveling snow. Finally, researchers including Dewey, Husted and Kenny have

shown that compensating wage differentials exist that make wages higher in areas with

low January temperatures. This implies that cold winters are undesirable. Once people

retire and wages no longer affect their choice of residence, they would be expected to

move to warmer climates. For all of these reasons, a higher mean January temperature

should have a positive effect on elderly inmigration.

Graves considered weather as an example of amenities that affect migration

decisions. Other amenities that appeal to the elderly include favorite leisure activities

such as golf and going to the beach. Thus, coastal counties and those with more golf








courses should attract more elderly movers. Each county's number of golf courses per

square mile and a coastal dummy are included to capture these amenities. An interaction

of temperature with the coastal variable is also included.16 I expect the estimates for

these variables to be positive, as beaches are best enjoyed in warm weather.

The standard deviation of land slope is also included, but the expectation for its

effect is less clear. While anecdotal evidence suggests that retirees enjoy the scenic

beauty of the mountains, the increased difficulty of transportation in rugged terrain may

be undesirable to the elderly.

The cost of living has also been shown to be an important predictor of elderly

migration 7; the urbanicity measure is intended to capture the cost of living. Urban areas

have higher costs of living due to higher land costs, so more urban areas should attract

fewer elderly movers.

Data on the number of golf courses in each county are supplied by the National

Golf Foundation, and land slope data are from David Figlio and Joe Stone.18 The coastal

dummy was obtained by inspection of state maps. The fraction of the county's

population living in an urban area is from the 1990 Census of population.

The mean January temperature of each county (in degrees Fahrenheit) is derived

from the National Oceanic and Atmospheric Administration's National Climactic Data

Center files. These files include mean January temperature by weather station. The

mean January temperatures for all stations in the county were averaged to derive the

'6 The interaction between golf courses and temperature could not be included because of collinearity; the
simple correlation between the natural logarithms of the number of golf courses and the interaction term is
0.90.

17 See Fournier, Rasmussen and Serow (1988).

" Special thanks to David Figlio and Joe Stone for the land slope data.








county mean January temperature. No weather stations are listed for 467 counties; for

these counties, the mean January temperatures of the adjacent counties (or closest

counties where there are no weather stations in the adjacent counties) were averaged.

State maps showing county boundaries were used to determine the surrounding counties

for this purpose.

Table 3-3 Summary statistics for the determinants of elderly migration
Full sample Mean Std. Dev. Minimum Maximum
Fraction of the population that is aged 65 and
older and that lived in the same county in
1985 and 1990 0.0055 0.0063 0.0000 0.0931
Mean January temperature 38.21 12.31 1.20 73.23
Number of golfcourses per square mile 0.0079 0.0143 0.00 0.1742
Standard deviation of land slope 0.8185 1.0902 0.0097 6.9052
Fraction of the population living in an urban
area 0.3528 0.2950 0.0000 1.0000

Urban Subsample
Fraction of the population that is aged 65 and
older and that lived in the same county in
1985 and 1990 0.0056 0.0069 0.0000 0.0931
Mean January temperature 39.54 11.91 1.20 73.23
Number of golf courses per square mile 0.0100 0.0161 0.00 0.1742
Standard deviation of land slope 0.8095 1.0918 0.0097 6.9052
Fraction of the population living in an urban
area 0.4806 0.2390 0.0006 1.0000
The full sample is made up of the 2520 counties from the 36 continental states that have a
coefficient of variation of per pupil state plus local revenues greater than 0.1. The urban
subsample is made up of the 1850 observations in the full sample for which the fraction rural is
less than 1.

As mentioned in the discussion of Table 3-1, the mean fraction of elderly from out

of state is 0.006 in both samples. There is a great deal of variation in this fraction: the

standard deviation is greater than the mean at approximately 0.007, and the elderly from

out of state make up more than 9% of the population at the maximum. Mean January

temperature is 38 to 400 Fahrenheit, and ranges from 1 to 730 Fahrenheit.

The supply of golf courses is quite varied across counties. The mean number of

golf courses per square mile is 0.008 for the full sample and 0.01 for the urban sample,








with a standard deviation of 0.014 to 0.016. The number of golf courses per square mile

ranges from zero to 0.1742. There is also a great deal of variation in the land slope

variable. The mean of the standard deviation of land slope is approximately 0.81 with a

standard deviation of over one, and a range of 0.01 to 6.9.

Results

The hypotheses for the determinants of elderly migration are, for the most part,

supported by the results shown in Table 3-4. The dependent variable is the natural

logarithm of the fraction of the population that is elderly and that migrated from out of

state, and has a mean of -5.68 in the full sample (-5.49 in the subsample) and a standard

deviation of 1.34 (0.89 in the subsample). It is in natural logarithm form to be consistent

with the variables in the main regressions. The explanatory variables are also in

logarithms.

The estimated effect of warmer January weather is positive and significant in all

specifications except the three-stage regression performed on the subsample. This effect

is even stronger in the coastal counties in the three-stage specifications, but the

interaction term has no significant effect in the two-stage regressions. These results

suggest that warmer winter weather is an important consideration for retirees deciding to

move out of state, and becomes even more important in coastal areas.

A 1 standard deviation increase in the natural logarithm of January temperature is

associated with a 0.0006 increase in the fraction of elderly inmigrants from out of state in

both two-stage specifications, where the interaction of January temperature with the

coastal dummy has no significant effect. The coastal interaction does have a significant

effect in the three-stage regressions. If the county does not have a coast, a 1 standard








Table 3-4 Determinants of elderly migration
From county-level cross-section (1990) 2


Mean January temperature

Number of golf courses per square mile

Coastal dummy

Interaction of January temperature and the
coastal dummy

Standard deviation of land slope

Fraction of the population living in an urban


*n from out-of-state
2SLS 3SLS
Full Urban Full sample Urban
sample subsample pie subsample
0.4043 0.3772 0.2421 0.0993
(0.1868) (0.1583) (0.0866) (0.0770)
0.0664 0.0617 0.1039 0.1014
(0.0121) (0.0149) (0.0109) (0.0128)
0.1230 -0.1908 -0.7030 -1.2909
(0.4181) (0.3027) (0.3349) (0.2622)

0.0009 0.0062 0.0247 0.0379
(0.0091) (0.0065) (0.0072) (0.0056)
0.1120 0.0884 0.2236 0.2090
(0.0444) (0.0362) (0.0282) (0.0226)


area 0.0030 -0.2165 0.0031 -0.1296
(0.0061) (0.0491) (0.0055) (0.0412)
All variables are in natural logarithms. The two-stage specifications include state dummies, as
well as the natural logarithms of the number of school districts per hundred thousand
population, nonwhite fraction of the population, absolute value of the nonwhite share aged 5 to
17 less the nonwhite share aged 65 and older, median household income, mean income divided
by mean income times the mean number of school-aged children per household, fraction of the
population aged 25 and older with at least some college, per pupil federal revenue, fraction
homeowner and fraction urban. Demographic and migration data are from 1990 and school
finance data are from 1992. The full sample is made up of the 2520 counties from the 36
continental states that have a coefficient of variation of per pupil state plus local revenues
greater than 0.1. The urban subsample is made up of the 1850 observations in the full sample
for which the fraction rural is less than 1.

deviation increase in the natural logarithm of January temperature is associated with a

0.0003 increase in the fraction of elderly immigrants from out-of-state in the full sample,

and has no significant effect in the subsample. If the county does have a coast, the full

effect is larger: an increase of 0.0010 in the full sample and 0.0006 in the urban sample.

More golf courses per square mile are also associated with higher elderly

migration from out of state in all specifications. A 1 standard deviation increase in the

natural logarithm of the county's number of golf courses is associated with an increase in

elderly migration from out of state ranging from 0.0005 to 0.0008 in the two-stage


regressions, and from 0.0009 to 0.0014 in the three-stage regressions.








The land slope variable has a positive effect on the migration of retirees from

out-of-state in all specifications. This effect is largest in the three-stage specifications,

where it is estimated to increase migration by 0.0010 in both the full and urban samples.

The estimated effect in both two-stage regressions is 0.0005. This suggests that retirees

enjoy the beauty of mountainous regions, notwithstanding the difficulties this may cause

for transportation.

The full effect of coastal counties on migration is slightly more complex. The

coastal dummy has no significant effect in the two-stage specifications. In the three-stage

regressions, the coast has a positive effect on elderly migrants from out of state if the

mean January temperature is above 28.50 Fahrenheit for the full sample and 130 for the

urban sample. Only 520 counties of 2,520 have mean January temperatures below 28.5.

No counties have both a seacoast and a mean January temperature below 28.50, while 31

counties with a coast on a Great Lake have such cold winters. Of the 1,850 counties in

the urban subsample, 66 have mean January temperatures below 130, but none of these

has a coast. These effects suggest that the elderly can better enjoy the coast in warmer

weather.

The effect of urbanicity on elderly migration is mixed. The fraction of the

county's population living in an urban area has a significant effect on elderly migration

from out of state only in the regressions performed on the urban subsample. A higher

level of urbanicity has the predicted negative effect in these specifications, with estimated

elasticities of -0.13 to -0.22. A 1 standard deviation increase in the fraction of the

population in an urban area is estimated to decrease elderly migration from out of state by

0.0003 to 0.0005.








Bound, Jaeger and Baker (1995) present evidence that using instruments that are

only weakly correlated with the endogenous explanatory variable may lead to

inconsistent instrumental variables estimates, even if the correlation between the

instruments and the error term in the main equation is very small. In addition, the finite-

sample bias of the estimates approaches the OLS bias as the first-stage R squared

approaches zero. They suggest that the F values and R squared statistics from the first

stage equations be examined to help determine the validity of the instruments. Stock and

Watson (2003) suggest that a first-stage F value greater than ten indicates an acceptable

instrument. The F statistics from the first stages of the two-stage least squares

regressions are 12.47 for the full sample and 13.28 for the urban subsample. The R

squared values from the first stage regressions are 0.1885 and 0.2530, respectively, which

are reasonable.

In Table 3-5, the variables of primary interest are those related to age, racial

mismatch and migration. The elderly nonmover variable retains the positive estimates of

the OLS specifications in all specifications, with a range in estimated elasticities of 0.11

to 0.15, which is quite close to the estimates in the OLS regressions. Again, the positive

estimated elasticity suggests that elderly nonmovers are more likely to support education,

compared to all residents under the age of 65. A 1 standard deviation increase in the log

of the fraction of nonmovers increases revenues (which have a mean of approximately

$4,500) by $158 to $232.

The estimated elasticities of the migration variables follow the predicted ordering

in three of the four regressions. In the three-stage urban sample specification, the only

migration variable that has an effect significantly different from zero is the nonmover








category. The estimated elasticity for the nonmovers in this specification is positive and

significant, so the ordering is partially correct.

Table 3-5 Determinants of per pupil state and local revenue for education
County-level cross-section (1990), Two-stage and three-stage least squares regressions
Two-stage least squares Three-stage least squares
1 SD 1 SD
Full sample Elasticity change in X Elasticity change in X


Fraction of the county's population that
is aged 65 and older and that did
not move between 1985 and 1990 0.1534 232.28 0.1362
(0.0293) (0.0180)
Fraction of instate movers 0.0061 -0.0216
(0.0143) (0.0090)
Fraction of out of state movers -0.0748 -241.43 -0.0462
(0.0251) (0.0187)
School districts per hundred thousand
population 0.0519 319.58 0.0662
(0.0115) (0.0069)
Fraction nonwhite 0.0030 0.0065
(0.0065) (0.0043)


Absolute value of the nonwhite share
aged 5 to 17 less the nonwhite
share aged 65 and older

Median household income

Tax price

Fraction of the county's population
aged 25 and older with at least
some college

Per pupil federal revenue

Fraction homeowner

Fraction urban


-0.0062 -0.0044
(0.0063) (0.0035)
0.3482 391.80 0.3484
(0.0514) (0.0303)
-0.2161 -106.97 -0.1129
(0.0754) (0.0511)


0.1079 148.85 0.0813
(0.0303) (0.0241)
-0.0081 -0.0351
(0.0223) (0.0085)
-0.2272 -176.23 -0.2817
(0.0370) (0.0272)
-0.0036 -106.97 -0.0037
(0.0008) (0.0007)


205.66

-62.57

-88.93


411.73







392.03

-56.21



111.64

-104.60

-217.44

-109.63


Urban subsample
Fraction of the county's population that
is aged 65 and older and that did
not move between 1985 and 1990 0.1233 171.08 0.1144 158.43
(0.0397) (0.0244)
Fraction of instate movers 0.0352 -0.0128
(0.0307) (0.0231)
Fraction of out of state movers -0.0739 -434.95 -0.0401
(0.0414) (0.0300)








School districts per hundred thousand
population 0.0492 262.66 0.0565 303.03
(0.0131) (0.0073)
Fraction nonwhite -0.0079 -0.0047
(0.0095) (0.0075)
Absolute value of the nonwhite share
aged 5 to 17 less the nonwhite
share aged 65 and older 0.0018 0.0014
(0.0063) (0.0061)
Median household income 0.4580 500.99 0.4330 472.12
(0.0669) (0.0392)
Tax price -0.2561 -117.34 -0.2047 -94.03
(0.0900) (0.0597)
Fraction of the county's population
aged 25 and older with at least
some college 0.0996 129.24 0.0522
(0.0406) (0.0333)
Per pupil federal revenue 0.0700 0.0625 133.44
(0.0370) (0.0131)
Fraction homeowner -0.1686 -104.31 -0.1568 -97.05
(0.0454) (0.0362)
Fraction urban -0.0146 -0.0032
(0.0103) (0.0097)
All variables are in natural logarithms. All specifications include state dummies.
Demographic and migration data are from 1990 and school finance data are from 1992. The
full sample is made up of the 2520 counties from the 36 continental states that have a
coefficient of variation of per pupil state plus local revenues greater than 0.1. The urban
subsample is made up of the 1850 observations in the full sample for which the fraction rural
is less than 1.

The elderly movers from out of state have negative and significant estimated

elasticities in 3 of the 4 specifications; these estimates, at -0.04 to -0.07, are much larger

than those from the ordinary least squares specifications (-0.01 for both samples). This

suggests that the endogeneity of elderly location choice with respect to education

spending is important in determining education spending. The dollar effects ofa I

standard deviation increase in the log of the fraction of immigrants from out-of-state'9

range from -$89 to -$435 where they are significantly different from zero. These

estimates are also much larger in absolute value than those obtained in the OLS


9 Using the predicted standard deviations from the first stage regressions.








regressions (-$46 to -$75). Cross-cohort racial heterogeneity again has no effect in any

specification.

The estimated elasticity for the number of school districts per hundred thousand

persons in the population is positive and significant in all specifications, with a range of

0.05 to 0.07. This result is consistent, as in the OLS regressions, with the idea that

support for education is likely to be higher where the community is smaller. The dollar

effects associated with a 1 standard deviation increase in this variable range from a $263

to $412 increase in per pupil revenues.

The estimated effect of increased median household income is positive and

significant in all specifications, with estimated elasticities ranging from 0.35 to 0.46.

These elasticities are close to the ones found in the OLS specifications, and are larger for

the urban sample in both specifications.

An increase in the nonwhite proportion of the population has no significant effect

in any regression. This offers more evidence that median household income captures all

of the effects of educational quality. The estimated elasticity of the tax price variable is

negative and significant in all regressions, as we would expect of an own price elasticity.

The price elasticities, which range from -0.11 to -0.26, are close to the OLS estimates.

The dollar effects of a I standard deviation increase in the tax price variable range from -

$56 to -$117.

An increase in the fraction of the adult population with at least some college

education has a positive effect on per pupil revenues in 3 of the 4 specifications, with no

significant effect in the three-stage urban regression. This positive effect is consistent

with the idea that educational attainment is a signal of taste for education. The effects are








close to those found in the OLS specifications, with estimated elasticities ranging from

0.05 to 0.11 and estimated dollar effects of $111 to $149.

The estimated effect of an increase in the fraction of homeowners is negative and

significant in all specifications, as it was in the OLS regressions. Again, this may be due

to a renters' illusion effect, where renters do not believe property tax increases are borne

by them, causing them to prefer higher school spending than otherwise identical

homeowners. These estimated elasticities are somewhat smaller than those found in the

OLS regressions; they range from -0.16 to -0.28 and have estimated dollar effects of -$97

to -$217.

Finally, an increase in the fraction of the population living in an urban area has a

negative effect in the full-sample specifications. The estimated elasticity remains small

at -0.004, and has an estimated I standard deviation effect of -$107 to -$110 on per pupil

revenues where it is significant. The small size of the estimates is again likely due to the

offsetting cost differences between urban and rural areas. The negative sign indicates

that economy of scale effects and higher rural transportation costs dominate.

Conclusion

Previous researchers have found the proportion of elderly in an area to have a

negative or insignificant effect on education spending. However, long-time older

residents should be more willing to support education than new elderly residents if

duration of residence fosters a sense of community. The results from my county-level

cross-section model are consistent with this new idea of community-based altruism. An

increase in the fraction of elderly nonmovers is shown to increase per-pupil spending,

while the elderly who migrate from another state or county are less willing to support








education. This effect is especially large when instruments to control for Tiebout

endogeneity are included, suggesting that controlling for this endogeneity is important

when estimating school spending.

The effects of elderly migration on school spending have large estimated dollar

effects on education spending compared to the effects of other variables commonly found

in the literature. In the regressions controlling for Tiebout endogeneity, these effects are

only slightly smaller in absolute value than the effects of household income and the

number of districts per hundred thousand persons. They are close in magnitude to the

effects of homeownership, and larger than the effects of adult educational attainment, the

tax price of education and the fraction of the population living in an urban area.

I also find cross-cohort racial heterogeneity to have no significant effect on

education spending. Racial differences may be a reason the elderly do not relate to the

youth of a community, but my results indicate that long-term residence of the elderly may

offset this potential source of discord.















Table A-1 Votes with 20% or greater contention, 1987 to 1998
Year Vote Bill Pro/ Pro Anti Total


number number Bill name
152 HR 1451 Older Americans Act/Funding Levels
To reduce total authorizations under the
bill for fiscal 1988 by $108 million
154 HR 1451 Older Americans Act/Home Care
To authorize $4 million to test ways to
assist seniors who receive home care
171 HR 4 Housing and Community
Development/Elderly Rents
To limit rent increases to elderly residents
of subsidized housing
279 HR 2470 Catastrophic Health Insurance
Bill/Republican Substitute
This version would cost an estimated
$18.2 billion, vs. $33.9 billion for original
bill
281 HR 2470 Catastrophic Health Insurance
Bill/Passage
Passage of catastrophic coverage and
other expansions (estimated cost of $33.9
billion)
407 HR 1212 Polygraph Tests/Nursing Home
Employees
Amendment to permit nursing home
employers to use lie detector tests


Number of votes
1988 176 HR 3436


Number of votes
1990 436 H Con Res Fiscal 1991 Budget
310 Resolution/Conference Report
Same spending targets as in budget
summit agreement, with smaller cuts in
Medicare
474 HR 5835 Fiscal 1991 Omnibus Reconciliation
Act/Democratic Alternative
To provide smaller increases in Medicare
premium and deductible, with other tax


1987
1987


provisions


Number of votes


anti votes votes votes
A 297 95


P 273 116


P 284 137



A 242 190




P 302 127




P 187 237



4 2 6
P 169 243


1 0 1
P 238 192




P 250 164


Long-Term Health Care/Rule
To increase Medicare coverage of long-
term home care services


2 0 2









Table A-I continued
1991 70 H Con Fiscal 1992 Budget Resolution/Spending A 336 89
Res 121 Caps
Decreased domestic outlays, including $25.2
billion in Medicare cuts over 5 years
Number of votes 0 1 1
1995 50 H J Balanced Budget Amendment/Recommit P 184 247
Res 1
To place Soc. Sec. trust funds off budget and
exempt them from balanced budget
calculations
69 HR 5 Unfunded Mandates/Older Americans and P 126 296
Juvenile Justice
To exempt mandates related to Older
Americans and Juvenile Justice Acts
301 ItR 483 Medicare Demonstration Program P 175 246
Expansion/Waxman Substitute
To bar age-based premium increases, allow
individuals to switch back to fee-for-service
plan
355 HR 483 Medicare Select Policies/Motion to Instruct A 224 197
Conferees
Reminder offiscal limitations in extending
Medicare Select policies
458 H Con Fiscal 1996 Concurrent Budget A 194 239
Res 67 Resolution/Adoption
Budget balance by 2002 via $894 billion
spending cuts, including $270 billion from
Medicare
729 HR 2425 Medicare Revisions/Democratic Substitute P 149 283
Medicare reductions of $90 billion over 7
years (Republican bill called for $270 billion
cuts)
730 HR 2425 Medicare Revisions/Motion to Recommit P 183 249
To remove Medicare Part B premium
increases from legislation
731 HR 2425 Medicare Revisions/Passage A 201 231
Medicare reductions of $270 billion over 7
years
742 HR 2491 1995 Budget -Reconciliation/Recommit P 180 250
Instructions to protect the health and income
security of children and the elderly
743 HR 2491 1995 Budget- Reconciliation/Passage A 203 227
Cut spending by $900 billion over 7 years,
including $270 billion in Medicare cuts
Number 6 4 10
of votes









Table A- continued
1996 156 HR 2406 Housing Overhaul/Family Rent Cap P 196 221
Rent cap for subsidized housing for
elderly and disabled residents
291 HR 3675 Fiscal 1997 Transportation A 159 247
Appropriations/Commercial Pilot
Retirement
Prohibit money for a study of whether
the mandatory retirement age for pilots
should be raised
Number of votes 1 1 2
1997 102 HR2 Public Housing System P 181 216
Overhaul/Community Service
Exemption
Exempt primary caregivers ofyoung
children and the elderly from service
requirement in subsidized housing
300 HR 2003 Budget Enforcement/Recommit P 148 279
To take Soc. Sec off budget and limit
Medicare Part B premium increases
Number of votes 2 0 2
1998 115 HR 3546 National Dialogue on Social P 174 236
Security/Recommit
Reserve budget surplus until Soc. Sec
is solvent for the future
463 HR 4578 Surplus to Social Security/Social P 210 216
Security Trust Fund
To transfer all Soc. Sec. Trust Fund
surpluses to be held in trust for Soc.
Sec.
464 HR 4578 Surplus to Social Security/Passage A 188 240
Set aside 90% of any budget surplus in
a special account until Soc. Sec.
solvent for the future
468 HR 4579 Tax Cuts/Democratic Substitute P 197 227
Prohibit tax cuts from taking effect
until Soc. Sec. solvent for the future
Number of votes 3 1 4

Total Votes: 19 9 28















REFERENCES


Andreoni, J, Cooperation in Public-Goods Experiments: Kindness or Confusion?
American Economic Review v85, n4 (September 1995): 891 904

Baker, D and Weisbrot, M, Social Security: The Phony Crisis, Chicago and London:
University of Chicago Press (1999)

Becker, Gary S., Altruism in the Family and Selfishness in the Market Place, Economica,
v48, n189 (1981): 1 15

Black, SE, Do Better Schools Matter? Parental Valuation of Elementary Education,
Quarterly Journal of Economics, vl 14, n2 (May 1999): 577 599

Blonigen, BA and Figlio, DN, Voting for Protection: Does Direct Foreign Investment
Influence Legislator Behavior?, American Economic Review, v88 n4 (September 1998):
1002- 1014

Bound, J, Jaeger, DA, and Baker, RM, Problems with Instrumental Variables Estimation
When the Correlation between the Instruments and the Endogenous Explanatory Variable
is Weak, Journal of the American Statistical Association, v90, issue 430 (June 1995):
443 450

Center for International Earth Science Information Network (CIESIN), Socioeconomic
Data and Applications Center (SEDAC). 1996. Archive of Census Related Products
[online]. Palisades, NY: CIESIN/SEDAC.

Congressional Quarterly Almanac v42 v54, Washington, D.C.: Congressional
Quarterly Inc. (1986 1998)

Denzau, A and Grier, K, Determinants of Local School Spending: Some Consistent
Estimates, Public Choice, v44 (1984): 375 383

Dewey, J, Husted, TA, and Kenny, LW, The Ineffectiveness of School Inputs: A Product
of Misspecification?, Economics of Education Review, v19, nl (February 1999): 27 45

Downs, A, An Economic Theory ofDemocracy, New York: Harper Collins (1957)

Figlio, D, Husted, TA, and Kenny, LW, Constitutions, Court Decisions, and Inequality in
School Spending, University of Florida Working Paper (June 2001)









Figlio, D, and Lucas, M, What's in a Grade? School Report Cards and House Prices,
NBER working paper 8019 (November 2000)

Figlio, D and Stone, JA, Competition and Stratification, University of Florida Working
Paper (January 2001)

Foumier, GM, Rasmussen, DW, and Serow, W, Elderly Migration as a Response to
Economic Incentives, Social Science Quarterly, v69, n2 (June 1988): 245 260

Graves, PE, Income and Migration Revisited, The Journal of Human Resources, v14, nI
(Winter 1979): 112 121

Harris, AR, Evans, WN, and Schwab, RM, Education Spending in an Aging America,
Journal of Public Economics, v81, n2 (September 2001): 449 472

Hoyt, WH, and Rosenthal, SS, Household Location and Tiebout: Do Families Sort
According to Preferences for Locational Amenities? Journal of Urban Economics, v42
(1997): 159- 178

Inman, RP, Markets, Governments and the New Political Economy, in Alan J. Auerbach
and Martin Feldstein (eds.), Handbook of Public Economics, Volume 2, Amsterdam:
North Holland (1987)

Kemnitz, A, Social Security, Public Education and Growth in a Representative
Democracy, Journal of Population Economics, v13 n3 (August 2000): 443 462

Kenny, LW, and Lott, JR, Jr.., Did Women's Suffrage Change the Size and Scope of
Government?, Journal of Political Economy, v107, n6 (December 1999)

Ladd, HF, and Murray, SE, Intergenerational Conflict Reconsidered: County
Demographic Structure and the Demand for Public Education, Economics of Education
Review v20 n4 (August 2001): 343 357

Logan, J, and Spitze, G, Self-Interest and Altruism in Intergenerational Relations,
Demography v32 (August 1995): 353 364

Lovell, MC, Spending for Education: The Exercise of Public Choice, The Review of
Economics and Statistics, v60, n4 (November 1978): 487 495

Maddala, GS, Limited-Dependent and Qualitative Variables in Econometrics,
Cambridge, New York, Melbourne and Madrid: Cambridge University Press (1983)

National Oceanic and Atmospheric Administration, NNDC Climate Data Online,
generated by Deborah Fletcher;








Palfrey, TR, and Prisbey, JE, Anomalous Behavior in Public Goods Experiments: How
Much and Why? American Economic Review v87, n5 (December 1997): 829 846

Poterba, JM, Demographic Change, Intergenerational Linkages and Public Education,
American Economic Review v88, n2 (May 1998): 315 320

Poterba, JM, Demographic Structure and the Political Economy of Public Education,
Journal of Policy Analysis and Management vl 6, n1 (Winter 1997): 48 66

Preston, S, Children and the Elderly in the United States, Demography v 21 (1984): 435-
457

Rothenberg, LS, and Sanders, MS, Severing the Electoral Connection: Shirking in the
Contemporary Congress, American Journal of Political Science v44, n2 (April 2000)

Schultz, JH, The Economics ofAging, 6th Edition, Westport, CT and London: Auburn
House (1995)

Stock, JH, and Watson, MW, Introduction to Econometrics, Boston, San Francisco, New
York: Addison Wesley (2003)

Stratmann, T, Congressional Voting over Legislative Careers: Shifting Positions and
Changing Constraints, American Political Science Review v94, n3 (September 2000):
665 676

U.S. Census Bureau, 1990 Census of Population and Housing, Summary Tape Files I and
3; generated by Deborah Fletcher, using American Fact Finder;


U.S. Census Bureau, "Federal, State and Local Governments: 1992 Public Elementary-
Secondary Education Finance Data," released November 29, 1995,


U.S. Census Bureau, County and City Data Book: 1996, generated by Deborah Fletcher
using WWW data access from the University of Virginia's GEOSTAT;


U.S. Department of Education, National Center for Education Statistics, Digest of
Education Statistics, 1997, NCES 98-015, by Thomas D. Snyder. Production Manager,
Charlene M. Hoffman. Program Analyst, Claire M. Geddes. Washington, DC: 1997





64








BIOGRAPHICAL SKETCH

Deborah Fletcher received a Bachelor of Arts in Economics from the University

of South Florida in Tampa, Florida in 1998. She continued her studies at the University

of Florida in Gainesville, Florida, where she received a Master of Arts degree in

Economics in 2001. Deborah earned a Doctor of Philosophy degree in Economics from

the University of Florida in August of 2003.








I certify that I have read this study and that in my opinion it conforms to acceptable
standards of scholarly presentation and is fully adequate, in scope and quality, as a
dissertation for the degree of Doctor of Philosophy.



David N. Figlio, Chair
Associate Professor of Economics

I certify that I have read this study and that in my opinion it conforms to acceptable
standards of scholarly presentation and is fully adequate, in scope and quality, as a
dissertation for the degree of Doctor of Philosophy.



Lawrence W. Kenny V
Professor of Economics

I certify that I have read this study and that in my opinion it conforms to acceptable
standards of scholarly presentation and is fully adequate, in scope and quality, as a
dissertation for the degree of Doctor of Philoso y.



Steven M. Slutsky
Professor of Economics

I certify that I have read this study and that in my opinion it conforms to acceptable
standards of scholarly presentation and is fully adequate, in scope and quality, as a
dissertation for the degree of Doctor of Philoso hy.



Renee J. Joh n
Assistant Professor of Political Science



This dissertation was submitted to the Graduate Faculty of the Department of
Economics in the College of Business and to the Graduate School and was accepted as
partial fulfillment of the requirements for the degree of Doctor of Philosophy.


August 2003
Dean, Graduate School
















LD
17)0














UNIVERSITY OF FLORIDA
3 2llll ll2 08554 7411
3 1262 08554 7411




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 ER64Z8O34_8VDVCG INGEST_TIME 2013-03-12T14:16:29Z PACKAGE AA00013583_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES


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 EK0QT8HBU_QZIXYO INGEST_TIME 2013-04-08T23:01:21Z PACKAGE AA00013583_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES