|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
THE TELECOMMUNICATIONS ACT OF 1996 AND MEDICAID HEALTH
TROY CLARENCE QUAST
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
Troy Clarence Quast
To my parents, Clarence and Elaine Quast; and to my son, Andres Quast.
I thank my advisor, David Sappington, and committee members Chunrong Ai,
David Figlio, and Justin Brown. I also thank Betsy Shenkman, David Gabel, Donald
Stockdale, Richard Romano, Mark Jamison, Sanford Berg, Jonathan Hamilton, Laura
Braden, Damon Clark, Roger Clemmons, Mircea Marcu, Burcin Unel, Richard Gentry,
and Vanessa Cruz for their helpful advice and words of encouragement. I am grateful for
financial support from the University of Florida Department of Economics, the
University of Florida Public Policy Research Center, the Telecommunications Policy
Research Conference, and the University of Florida Public Utilities Research Center. I
am also grateful for technical and research support provided by the University of Florida
Department of Epidemiology and Health Policy Research. Finally, I thank my wife,
Daniela Quast, for her unwavering support and her ability to feign interest in the notion
TABLE OF CONTENTS
A C K N O W L E D G M E N T S ................................................................................................. iv
LIST OF TABLES .............. ................. ........... ................ .......... vii
LIST OF FIGURES ......... ......................... ...... ........ ............ ix
A B STR A C T ................................................. ..................................... .. x
1 IN T R O D U C T IO N ............................................................................. .............. ...
2 HOW STATE GOVERNMENTS IMPLEMENT FEDERAL POLICIES: THE
TELECOM M UNICATIONS ACT OF 1996 ........................................ ....................3
In tro d u ctio n ........................................................... ... ... .................................3
Background Information on Unbundled Network Element (UNE) Rate
Proceedings ....................................... ..... ..... 7
Steps Involved in Setting UNE Rates......................................... ............... 7
Influence of N neighboring States....... ........................................... ................. 9
Hypothesized UNE Rate Determinants ........................... ................................... 12
M odel Specification and Data U sed ..................................................................... 15
M odel Specification................. ........................ .. .... ........ .. ......15
D ata U se d ................................................................16
E stim action R results ........................ .................... .. .. .... ........ ......... 20
C oefficient E stim ates........... .............................................. ........ .............20
E conom ic E ffects............ ........................................................ .. .... ..... .. 22
C conclusion ....................................................... ...........................24
3 THE EXTENT AND MEANS OF ENTRY INTO LOCAL
TELECOM M UNICATIONS M ARKETS ........................................ .....................37
Introduction ................................................................. .........................37
Background Information on UNE-Based Entry ............................... ...............42
Loop-Based Versus Platform-Based Entry ............................... ............... 42
History of Platform-Based Entry Regulation .............................. ... ..................43
Hypothesized Determinants of Loop- and Platform-Based Entry ..............................45
M odel Specification and D ata U sed ........................................ ....................... 48
E stim ation R results ....................................................................... ........ 51
Estimates without Population Interactions.......................................................51
Coefficient estim ates .................. ...................... ...... .. ..................... ..51
E conom ic effects ........................................................ .. ....... ........... 52
Estim ates with Population Interactions .................................... ............... 54
Coefficient estim ates ............................................................... .. ........ 54
E conom ic effects ........................................................ .. ....... ........... 55
C conclusion .................................. ........................... .... ..... ........ 58
4 DOES THE FORM OF DOCTOR COMPENSATION AFFECT THE QUALITY
OF CARE IN M EDICAID HM OS? ........................................ ....................... 67
In tro d u ctio n .......................................................................................6 7
B background Inform ation................................................................. ............... 69
D ata and Em pirical Specification ................................................................... ......71
F in d in g s ..............................................................................7 5
C o n c lu sio n s........................................................................................................... 7 7
5 CON CLU SION S .................................. .. .......... .. .............82
A DATA NOTES AND ADDITIONAL RESULTS ............... .................... ..........83
D ata N o te s ......................................................................... 8 3
UNE Rates ..................................... ................................ .........83
C o st E stim ate ................................................................................... 83
PUC Characteristics.................. ..... ...... .... .. .. ...............84
State P political V ariab les ........................................................... .....................85
Section 271 Status ................................................. ... ............. ............ 86
Level of Com petitive UNE Entry ........................................... .........................86
Tests of Instrument Validity and the Exogeneity of the Average Retail Rate
V variable .............................................. ............................................... 87
Tests of Instruments' Orthogonality to the Error Process.............. ................ 87
Tests of Instrum ent Relevance .................................................. .... .........88
Test of the Exogeneity of the Lagged Average Retail Rate ...............................88
A additional R results ............................................ .. .. .... ......... ......... 89
B ADDITONAL RESULTS ..................................... ................................ 95
L IST O F R E FE R E N C E S ............................................................................. .............. 98
BIOGRAPHICAL SKETCH ............................................................. ............... 105
LIST OF TABLES
2 -1 Su m m ary Statistics ............................................. ......................... ......................... 3 1
2-2 Correlation Matrix of Non-UNE Rate Variables ........................................... 32
2-3 Coefficient Estimates from One-Year and Two-Year Lag Specifications...............34
2-4 Effects of Changes in Explanatory Variables on UNE Rates Two Years Later ......36
3-1 Sum m ary Statistics. ......................................... .. .. .... ........ ......... 62
3-2 Correlation Matrix of Explanatory Variables.............. .... .................62
3-3 Coefficient Estimates without Population Interactions..................... ..............63
3-4 Economic Effects without Population Interactions ............................................64
3-5 Coefficient Estimates with Population Interactions..........................................65
3-6 Economic Effects for the 25th and 75th Population Percentiles.............................66
4-1 H M O A ttributes.......... .................................................................... ........ .. ....... .. 79
4-2 HEDIS Success Rates by HM O Attribute..................................... ............... 79
4-3 Demographic Characteristics By Population. .................................. ...............80
4-4 HEDIS Success Rates by Population. ........................................... ............... 80
4-5 OLS Regression Estim ates. ...... ........................... ....................................... 81
A-i Acronym s U sed .................. ............................... ........ ... .......... 90
A-2 Tests of Overidentifying Restrictions ........................................... ............... 91
A-3 Tests of Instrum ent Relevance ..................................................................92
A-4 Effects of Changes in Explanatory Variables on UNE Rates Two Years Later
U sing One-Y ear Lag Specification ........................................ ....... ............... 93
A-5 Effects of Changes in Explanatory Variables on UNE Rates in the Long-Run.......94
B-l Marginal Effects from Probit Regressions. ............. ........................................96
B-2 OLS Regression Estimates with HMO Fixed Effects...................... ...............97
LIST OF FIGURES
2-1 Comparison of UNE Rates. A) New Hampshire and Vermont. B)Kentucky and
Tennessee. C) W yom ing and Utah........................................................................ 26
2-2 HCPM Estimate versus UNE Rate. A) Ameritech. B) BellSouth. C) Pacific
Telesis. D) Qwest. E) Southwestern Bell. F) Verizon. .........................................28
2-3 Cumulative effect of $1 increase in the leader's rate on the follower's rate............30
3-1 Fraction of Incumbent Lines Leased by Entrants (National Average)...................60
3-2 Fraction of Incumbent Lines Leased by Entrants by Means of Entry (National
A average .............................................................................60
3-3 Loop Share of Incumbent Lines, July 2001 and July 2004............................... 61
3-4 Platform Share of Incumbent Lines, July 2001 and July 2004. ............................61
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
THE TELECOMMUNICATIONS ACT OF 1996 AND MEDICAID HEALTH
Troy Clarence Quast
Chair: David Sappington
Major Department: Economics
We analyzed the effects of the landmark Telecommunications Act of 1996 (TA96).
We also investigated whether the method of provider compensation affects the level of
care provided by Medicaid health maintenance organizations (HMOs).
We examined the rates set by state public utility commissions (PUCs) that
competitors must pay to lease parts of the local network from the largest incumbent U.S.
telecommunications suppliers (RBOCs). The results indicate that rates in the smaller
states in each RBOC region are strongly influenced by the largest state in the region.
Rates are lower where the level of competitive entry is lower, while they are higher in
states where the governor is a Republican. The analysis suggests that different states have
employed different methodologies in implementing TA96.
We investigated entry in local telecommunications markets. Panel data are used to
analyze the number of lines that competitive local exchange carriers (CLECs) lease from
RBOCs using two alternative arrangements: leasing only the wires that connect a
customer's premises to the phone network (loop-based entry), or leasing all of the
network elements that are needed to provide phone service (platform-based entry). The
estimates suggest that while the two types of entry are generally affected by different
market factors, there appears to be cost-based substitution between them. Also, loop-
based entry is more responsive to changes in economic conditions in smaller states, while
platform-based entry is more responsive in larger states.
Using data for all of the Medicaid HMO enrollees in a large state, my coauthors
and I found that enrollees in HMOs that pay their doctors exclusively via fee-for-service
arrangements are more likely to receive services for which the HMO's doctors receive
additional compensation. Further, these enrollees are less likely to receive services for
which the HMO's doctors do not receive additional compensation. These findings
suggest that financial incentives may influence the behavior of doctors in Medicaid
HMOs, and thus the health care received by Medicaid participants enrolled in HMOs.
The federal government in the United States often attempts to harness competitive
market forces to achieve policy outcomes. Politicians and analysts frequently claim that
policies that are based largely on market incentives are best suited at resolving observed
inefficiencies in the economy. Two important areas in which this approach has been
attempted are telecommunications and health care. One of the hallmarks of the
Telecommunications Act of 1996 (TA96) was its focus on introducing competition into
the industry, which its supporters claimed would lower prices and result in greater
deployment of technology. Meanwhile, a prominent component of Medicaid reform has
been to move enrollees into health maintenance organizations (HMOs). Many believe
that when HMOs assume responsibility for the health care of Medicaid participants, costs
are lowered and quality increases. This study analyzes how these policies have been
implemented and whether they have been successful.
First, we investigate how state governments have implemented TA96. State public
utility commissions have been given somewhat vague guidelines as to how to set the rate
at which entrants can lease portions of the incumbent telephone company's network. We
find that the commissions may look to neighboring states for guidance as to how these
rates. Further, the commissions may be influenced by the level of existing entry when
determining the rates.
Second, we explore the factors that determine where entry occurs in local
telecommunications markets and the means by which entrants choose to enter. Beginning
in late 2000, entrants could either lease part or all of the incumbent firm's network to
provide service. Regression results indicate that the relative cost of the two forms of entry
may influence the means of entry chosen by firms. Further, the factors that influence the
level of entry vary by the size of the market.
Finally, we analyze how the means by which primary care physicians are paid may
influence the quality of care they provide. Using data from Medicaid HMOs, we compare
the quality of care provided by physicians who are paid a flat rate per enrollee versus the
quality of care provided by physicians who are paid per service provide. The results
suggest that doctors who are paid a flat rate per enrollee may be less likely to provide
check-up visits to children than doctors who are paid per service provided.
HOW STATE GOVERNMENTS IMPLEMENT FEDERAL POLICIES:
THE TELECOMMUNICATIONS ACT OF 1996
A primary goal of the TA96 (Table A-i) was to encourage competitive entry into
the local telecommunications market, under the presumption that such entry would lower
prices and increase social welfare. One of the means by which entrants1 were to enter this
market was by leasing unbundled network elements (UNEs). The legislation decreed that
entrants could lease certain segments of the incumbent' s2 network3, thus allowing for
competition where it would otherwise be unprofitable or infeasible. The rates at which
entrants could lease the UNEs could play an important role in the degree of entry
observed.4 The only rate-setting guidance offered by TA96 was that the rates were to be
priced at cost plus possibly a reasonable profit, determined without reference to a rate-of-
return proceeding, and set by the state public utility commissions.5
SEntrants into local telecommunication markets are referred to as Competitive Local Exchange Carriers,
2 Specifically, this paper analyzes the rates at which Regional Bell Operating Companies (RBOCs) lease
UNEs to entrants. RBOCs are the regional monopolies that were created in the split-up of AT&T in
3 The question as to which segments the incumbent should be forced to lease has been litigated
extensively (see Lichtman and Picker, 2003).
4 As discussed in Lichtman and Picker (2003, pp 22-23), excessively low rates can discourage entrant
investment in their own networks and foster excessive reliance on UNEs. Conversely, exceptionally high
rates may encourage inefficient investment by entrants. The effect of the rates on incumbent investment
5 47, USC 252(d).
Given the ambiguity in how the UNE rates were to be set, the Federal
Communications Commission (FCC) interpreted TA96 as giving it the authority to
proscribe the methodology the states should use in determining the rates. The FCC
decided that the rates should be based on total element long-run incremental costs
(TELRIC), the hypothetical costs of implementing the least-cost network given the
current locations of the incumbent's wire centers.6
The state commissions were left with the unenviable task of operationalizing the
vague "TELRIC" notion. Figure 2-1 depicts UNE loop rates for three pairs of arguably
comparable states: New Hampshire and Vermont; Kentucky and Tennessee; and
Wyoming and Utah. The charts suggest that the state commissions operationalized the
TELRIC concept differently over time. For example, when Wyoming and Utah first set
rates in 1997, the UNE rate in Utah was only $2.00 less than the rate in Wyoming. By
2003 that difference had increased five-fold. This type of variability across states and
over time is representative of the experiences in many states and suggests the states
varied a great deal as to how they implemented the TELRIC methodology.7
Understanding how the state public utility commissions set UNE rates can
provide valuable lessons, both in telecommunications regulation and in the formation of
federal policies. Any evaluation of the success of TA96 at promoting competitive entry
6 The FCC's decision was challenged numerous times in the courts, both regarding whether the
commission had the authority to dictate the rate-setting methodology and also whether the TELRIC
methodology was consistent with TA96. The Supreme Court eventually decided both issues in favor of
the FCC. (AT&T v. Iowa Utilities Board (1999), Verizon, et al v. FCC, et al (2002)) More recently, the
Washington, D.C. Court of Appeals recently ruled that incumbents should not be forced to lease
switching equipment to entrants at TELRIC rates, and the FCC subsequently revised their rules to adhere
to that decision (FCC (2005)). However, local loops, the focus of this paper, continue to be included by
the FCC as a UNE that the incumbents must lease at TELRIC rates.
7Further, the Department of Justice recommended rejecting applications by some incumbents for
permission to sell interLATA service on the grounds that the state commissions had not correctly
calculated TELRIC rates (e.g., see Department of Justice (1997, 1998, and 2000)).
must account for how states implemented the Act. Given the widespread criticism that
TA96 did not spur the level of entry that was anticipated, it is important to disentangle
effects due to the execution of TA96 versus the Act itself. Beyond telecommunications,
the UNE rate experience provides an opportunity to analyze the behavior of state officials
charged with implementing federal policy. According to TA96, the state commissions are
to base the rates only on state-specific cost factors. However, given the vague TELRIC
definition of cost and the discretion given to the states, it is possible that additional
factors influence how these rates are set. An analysis of these factors can provide insight
as to how state officials implement federal policies, and may apply to other policy areas
such as education or environmental regulation.
Previous studies have analyzed the determinants of UNE rates. De Figueiredo and
Edwards (2004) examine the UNE rate in place in the three two-year election cycles from
1998 through 2002. The authors find that states with Democratic state legislatures and
relatively large political contributions by entrants have lower UNE rates. Further, elected
commissions are found to set higher rates, while states in which the utility commission
has imposed retail price caps tend to have lower UNE rates. Lehman and Weisman
(2000) explore UNE rates set immediately following the implementation of TA96. They
also find that elected commissions set higher rates while price caps lead to lower rates.
Beard and Ford (2004) test whether the UNE rates for certain combinations of network
elements in 2002 are correlated with the rates proposed by incumbents or entrants. They
find that the rates set by state commissions can be explained as splitting the difference
between the preferred rates of the two parties. Eisenach and Mrozek (2003) regress the
observed UNE rate against a cost estimate produced by the FCC Hybrid Cost Proxy
Model (see further description of this model in Section 4.2) and find that costs explain
only half of the variation in UNE rates.8
This paper extends these studies in several important directions. First, the present
analysis is based on a unique data set that contains the UNE loop rates for each state
since TA96 was enacted and when those rates were ordered. Further, unlike previous
studies, it is known whether the rate was the result of a voluntary reduction by the
incumbent. Second, the analysis examines the potential for information spillovers across
states.9 Given the ambiguity of the TELRIC notion and the lack of experience state utility
commissions had with the concept, it is possible if not likely that many states looked to
other states for guidance. Specifically, the influence of the rates set by the commission in
the largest state in each incumbent region (the "leader") on the other states in the region
(the "followers") is tested. Third, whereas in previous papers the form of retail rate
regulation was assumed to be exogenous, the analysis corrects for the potential
endogeneity of this variable. Fourth, the impact of the level of entry on UNE rates is
Four primary conclusions emerge from this research. First, less than 50% of a
change in the estimated cost of providing a UNE is reflected in the UNE rate two years
later. Second, a $1 increase in the leader state's rate results in a roughly $0.75 increase in
8 Another area in which state utility commission telecommunications regulation has been analyzed is how
retail rates are set. Donald and Sappington (1995) find that state commissions are more likely to choose
incentive regulation where the incumbent can gain more, rates are especially high, and elected state
leaders are churned by voters. Smart (1994) observes that retail rates are lower in states where the
commissioners are elected and that control of the governorship and state legislature only has an impact
on prices when the offices are held by different political parties.
9 There is a large literature regarding strategic interaction between governments, especially in the areas of
environmental, welfare, and tax policy (see Brueckner (2003) for a survey). However, the information
spillovers in UNE rates do not appear to be strategic, i.e., the state commissions do not appear to take
into consideration the reactions of other commissions when setting their own UNE rates.
the rates of the follower states two years later. Conversely, a state's rate does not appear
to be affected by changes in the rates of the follower states. Third, applications for
permission to sell long-distance services and lower levels of competitive entry tend to put
downward pressure on rates. Fourth, Republican leadership in a state is associated with
higher UNE rates.
These conclusions suggest that smaller state utility commissions rely on the work
of larger state commissions to help determine the regulated rates that they choose for
their state. Further, non-cost factors such as the market and political environment may
influence the rates set by the commissions. These results indicate that information
spillovers and factors beyond those specified in TA96 may influence how state
governments implemented this federal policy.
The paper is organized as follows. Section 2 presents background information on
how UNE rates are set. Section 3 describes the hypothesized UNE rate determinants.
Section 4 details the econometric methodology and the data used. Section 5 discusses the
estimation results, while Section 6 provides conclusions and areas for further research.
Background Information on Unbundled Network Element (UNE) Rate Proceedings
Steps Involved in Setting UNE Rates
Before reviewing the formal analysis, consider how UNE rates have been
determined by the states.
Immediately following the passage of TA96, state public utility commissions were
often forced to arbitrate interconnection agreements between incumbents and entrants
without having the luxury of completing a formal cost study. To prevent the delay of
interconnection agreements, the commissions often decided on a UNE rate based solely
on the proposed cost studies and testimony submitted by the two sides. When the
commission completed its own cost study, the rates from the study replaced the
temporary placeholder rates in the existing interconnection agreements.10
Following the initial cost study, the states had complete discretion as to when to
review the UNE rates, if at all. Most interconnection agreements between the incumbents
and entrants last three years, so the states often revise the UNE rates to coincide with the
expiration of the agreements. However, sometimes commissions revise the rates before
the end of this three-year cycle, while on other occasions they review the rates less
While practices differ to some extent, most states follow the same procedures
when modifying UNE rates. First, the commission announces that it will review the UNE
rates and that hearings will be held. Before these hearings take place, the incumbent and
the entrant submit their own cost models and expert testimony. Following the hearings,
the commission reviews the material and testimony and sometimes will ask for further
information from the parties. After this stage, the commission announces its decision on a
cost model and the proper inputs for it. This decision will often include what it thinks the
resulting UNE rates are, but it will ask the incumbent or entrant to run the model chosen
by the commission with the proscribed inputs and report the resulting rates. The initial
commission decision regarding the model and inputs may also be appealed. (The state
typically has discretion as to whether it will allow an appeal to be heard.) This process
can last as long as two years.
The cost studies performed by the state commissions involve decisions on many
parameters, ranging from labor costs to the costs of telephone poles. In its most general
10 Note that some states were able to complete a TELRIC cost study quickly enough so that they did not
have to implement the placeholder rates.
form, the TELRIC studies consider three types of costs: operating costs, depreciation
costs, and the cost of capital.
Influence of Neighboring States
As described above, the TELRIC methodology that the FCC ordered the
commissions to follow to set UNE rates is both complicated and vague. The states are
forced to estimate the costs that a provider of the loops would incur if they were to build
the network today using an efficient technology. Specifying the efficient technology is a
daunting task in itself. In addition, the commission must determine the hypothetical cost
of installing the efficient technology throughout the incumbent's service area. Such an
analysis requires considerable resources, and can be particularly burdensome for states
that have relatively small staffs and budgets. Given these challenges, it is not surprising
that smaller states look to larger states for guidance in setting UNE rates.
The state commissions in an incumbent's region share a working relationship that
is conducive to collaboration in setting UNE rates. For example, regional associations of
state commissions that are carved almost exactly on incumbent regional lines meet
annually or semiannually.11 The meetings typically include working sessions where
commission staff members that work in telecommunications discuss current issues and
listen to incumbent and entrant representatives present their views on upcoming
regulatory matters. The potential for information spillovers among states was further
highlighted in a recent survey by the National Association of Regulatory Commissioners
of state commissions. When the state commissions were asked if they would be interested
1 The Qwest utility commissions comprise of the Qwest Regional Oversight Committee, the BellSouth
commissions are part of the Southeastern Association of Regulatory Utility Commissioners, the Verizon
commissions are members of either the New England Conference of Public Utility Commissioners or the
Mid-Atlantic Conference of Regulatory Utilities Commissioners, while all but two of the SBC states are
members of the Mid-America Regulatory Conference.
in working with other states in a matter closely related to UNE rates, many of them
indicated that they believed such coordination would be useful and has been useful in the
past (NARUC TRIP Task Force, 2003). 12
Figure 2-2 provides additional evidence that information spillovers may be present
within incumbent regions. The charts show for each state in January 2001 the UNE rate
set by the state and an estimate of the TELRIC loop cost using a cost model developed by
the FCC known as the Hybrid Cost Proxy Model. While the ability of the this model to
accurately calculate TELRIC costs has been hotly disputed, the model provides a
consistent measure across states and over time of the relevant TELRIC cost, under
specified assumptions. For most states, the UNE rate set by the state is lower than the
model estimate. However, the difference varies considerably by incumbent. For instance,
the UNE rates in the Ameritech states are roughly half of the model cost estimate, while
in the Qwest states the UNE rates are typically roughly equal to the cost estimate. These
inter-regional differences are consistent with the notion that the state commissions within
an incumbent region base their UNE rates in part on the rates of their neighbor states.
It is unlikely that all states within an incumbent's region have an equal influence on
each other. Within each region there appears to be a leader state that other states in that
region look to for guidance in their UNE rate proceedings. The leader state not only has
the most resources available in that region to conduct a UNE rate study13, but it may also
12 The Kansas utility commission replied, "The Commission believes it would be especially beneficial and
cost effective for the five original Southwestern Bell Telephone Company states to coordinate efforts."
The Rhode Island commission stated, "RI has a very small staff and would be interested in coordinating
logistics with other states in Verizon's territory..." Wyoming's response included, "The Commission is
considering the possible benefits of a regional approach such as participating in a coordinated effort of
the Qwest ROC (Qwest Regional Oversight Committee)."
13 An appropriate measure of the resources available would be the number of commission employees or the
commission's budget. However, state utility commissions are organized differently across states and
state comparisons of these measures are problematic. For example, Arizona's is part of the state's
have been the first state in which the incumbent applied for permission to sell long-
distance services in its territory14. The FCC encouraged smaller states to use the
information provided by leader states in their UNE rate cost studies:
We recognize that many states lack the extensive resources that were dedicated to
the process by New York and Texas, as detailed in our orders in those states...
We encourage states with limited resources to take advantage of the efforts devoted
by New York and Texas in establishing TELRIC-compliant prices, by relying
where appropriate on the existing work product of those states. (FCC, 2001, p 40)
There is considerable evidence that follower states look to the leader states for
guidance. For example, in the BellSouth region, it was reported that the Kentucky
commission closely monitored the UNE rate proceeding in Florida and even discussed
Florida's findings with the Florida commission and its staff (Caldwell, 2002, paragraph
131). In the SBC region, the Kansas commission suggested that they delay their UNE
proceeding in order to wait until the ongoing Texas study was completed (Kansas
Corporation Commission, 2000, p 2). In Nevada, the commission approved a stipulation
between the incumbent and the entrants that set UNE rates based on the proceedings of
the California commission (Public Utility Commission of Nevada, 2002). According to a
recent trade press article, follower states in Verizon's region halted their UNE rate cases
to see how the New York would decide its rate (State Telephone Regulation Report,
Corporation Commission, which also handles issues regarding securities and insurance. Data on the
number of employees and budget for the Arizona commission are available only for the state corporation
commission. As a proxy for state commission resources, the number of telephone access lines is used.
14 Under Section 271 of TA96, in order for an incumbent to receive permission to sell long-distance
services in a state, it had to file an application with both the FCC and the state that demonstrated that the
local telephone service market was open to competition. One of the criteria by which the application was
judged was whether the UNE rates were based on TELRIC estimates.
The influence of a leader state's rate on rates in follower states may be influenced
by incumbent behavior. To illustrate, incumbents have encouraged the influence of leader
state rates by benchmarking follower states' rates to them when applying for permission
from the FCC to sell long-distance services. For instance, in 2002 Qwest voluntarily
lowered their UNE rates in eight states in order to benchmark them to the rates set by
Colorado (FCC, 2002). 15 The incumbents also urge follower states to base their rulings
on decisions by the leader states. For example, in an Alabama UNE rate case BellSouth
urged the Alabama commission to adopt the Florida commission's position regarding the
timing of UNE rates, while in Georgia BellSouth argued for the approach that Florida
adopted to allocate costs between regular telephone service and data service.
Hypothesized UNE Rate Determinants
If the states followed the TELRIC methodology to the letter, the TELRIC cost
would perfectly determine the UNE rate. However, as described above and demonstrated
in Figures 1-1 and 1-2, it does not appear that costs alone explain the UNE rates.
Therefore, other variables are included in the model to account for the variation in rates.
To control for information spillovers across states, the rates of the other states in
the same incumbent region are included in the model.16,17 To test for the effect of the
leader state on the follower states, the model includes the rate of the leader state
15 Specifically, Qwest voluntarily lowered the rates to the level in Colorado adjusted for the difference in
average costs according to the FCC's cost model.
16 The regions are muddled somewhat by the mergers that have taken place among the incumbents. For
instance, Pacific Telesis and Ameritech were acquired by SBC in 1998 and 1999, respectively, while
Verizon (formerly Bell Atlantic) acquired NYNEX in 1997. Given their geographic locations and the
timing of the acquisitions, Pacific Telesis and Ameritech are treated separately from SBC while the
former Bell Atlantic and NYNEX are treated as one entity.
7 The identification strategy employed to capture information spillovers is closely related on the strategic
interaction literature cited above. Specifically, the lagged rates of neighboring states are interacted with
dummy variables in order to isolate the effects of interest. See Fredriksson and Millimet (2002),
Fredriksson, List, and Millimet (I 21114, and Hayashi and Boadway (2001).
interacted with a dummy variable that equals one if the state is a follower state. This
variable will capture how the follower states respond to changes in the leader's rate. If the
coefficient on this variable is statistically significant, one can conclude that the rates in
the follower states are influenced by the leader state's rate.
However, it may be the case that the follower states also are influenced by the rates
in the other follower states. Therefore, also included in the regressions is the weighted
average18 of the follower states' rates interacted with a dummy variable that equals one if
the state is a follower state. If there are information spillovers between the follower
states, the coefficient on this variable will also be statistically significant.
It may also be the case that the leader states are influenced by the rates in the other
states in the incumbent's region. To test for this, the weighted average of the follower
states' rates is interacted with a dummy variable that equals one if the state is a leader.
Thus, any influence of the follower states on the leader states is captured in this
Characteristics of the state commissions may also influence UNE rates. Under the
theory of regulatory capture, the longer a commissioner has served, the more amenable
s/he might be to setting a rate favorable to the incumbent. Applied to UNE rates, this
theory suggests that the longer the commissioners have served the higher the UNE rates
will be. There may also be an influence due to the political affiliation of the
18 The weights are based on the number of switched access lines per state. The results do not change when
weights based on population are used.
19 The analysis in this paper focuses on the influence of neighboring states within the incumbent's region. It
is possible that some commissions might have influence that extends beyond the incumbent's region.
However, given the limited degrees of freedom available, it is not possible to simultaneously control for
nation-wide influences of all of the leader states. Initial estimates that test for the influence of one
national leader state at a time suggest that such effects may be present. However, the qualitative
conclusions reported below persist in settings that admit national leadership patterns. As stated in the
conclusion, further analysis of this issue is warranted.
commissioners. However, it is unclear as to the direction of this influence. One may
surmise that Republicans generally favor less regulation and less aggressive (i.e., higher)
UNE rates. Conversely, it may be the case that Republicans are sympathetic towards
small businesses, and therefore may favor lower UNE rates to benefit both small-business
consumers of telecommunications services and entrants. Prior studies have debated
whether states that enact incentive-based retail rate regulation enact lower UNE rates20
Further, the incumbent's retail rates may influence the level of UNE rates, as state
commissioners may regard the retail rate as an upper bound on what an entrant could
afford to pay for a UNE. Finally, UNE rates may systematically differ if they were set in
an arbitration case immediately following the enactment of TA96.
Beyond the state commissions, there may be state-specific influences that vary by
time. For instance, the political affiliation of the governor or the state legislature may
influence how the state sets UNE rates. Not only can the governor and state legislature
have a direct influence on the state utility commission through appointments and
budgetary powers, but their political affiliation may be a proxy for the political sentiment
of the citizens and reflect the general regulatory environment in the state. As described in
the previous paragraph, the direction of this effect is ambiguous.
The incumbent's federal regulatory status may play an important role in how UNE
rates are set. As noted above, the FCC and state commissions had to certify that the
incumbent's UNE rates were TELRIC-based before the incumbent was allowed to sell
long-distance phone service. Therefore, one might expect that the UNE rates in the period
immediately prior to the FCC's decision were lower than they would have otherwise
20 Lehman and Weisman (2000) argue that state commissions can unfairly shift risk to incumbent firms by
enacting a retail price cap and setting a relatively low UNE rate.
been. Some incumbents also voluntarily lowered their UNE rates during the application
process in the hope of securing permission to provide long-distance service. If the
incumbents were in fact lowering their UNE rates below the rates that would have
prevailed otherwise, one would expect these voluntary reductions to result in lower UNE
rates. However, the incumbent might have made voluntary reductions that were not as
drastic as would otherwise have been ordered by the state during the application process.
If so, the incumbent might have been able to secure a more favorable UNE rate by pre-
empting action by the state. In such a case, the marginal effect of the voluntary reduction
could be positive.
Lastly, the level of observed UNE entry may affect the UNE rates. If the
commission views the level of entry as relatively low, certes paribus, it may be inclined
to lower UNE rates to encourage additional entry. Thus, one would expect a positive
coefficient on this variable.
Model Specification and Data Used
As noted above, UNE rates changed infrequently in some states. Consequently,
UNE rates often exhibit a high degree of stationarity. However, relatively frequent data
are required to capture the exact timing of the rate decisions. To allow for frequent data
and the stationarity of the lagged dependent variable, the lagged rate is included as an
Including the lagged dependent variable as an explanatory variable in a panel data
regression complicates the econometric analysis. When an OLS fixed-effects estimator is
21 In the econometrics literature this is referred to a dynamic panel data model.
used, a negative bias of order 1/T is introduced in the coefficient on the lagged dependent
variable (Nickell, 1981). OLS estimation of the model in first differences partially
corrects the bias, but does not entirely alleviate the endogeneity of the lagged dependent
variable. Arellano and Bover (1995) and Blundell and Bond (1998) derive a generalized
methods of moments estimator (known as system GMM) that simultaneously estimates
the model in levels and first differences. Blundell and Bond (1998) perform Monte Carlo
simulations that demonstrate the system GMM estimator is superior to both the OLS
fixed effects and GMM estimations using first differences only. Further lagged values of
the levels and first difference of the dependent variables are used as instruments for the
lagged dependent variable.22 The system GMM estimator is appropriate when the
coefficient on the lagged dependent variable is 0.8 or greater. For this estimator to be
valid, the lagged dependent variable must have a constant correlation with the state
effects and be uncorrelated with present and past values of the error term. Further, it is
assumed that the error terms have a mean of zero and are not serially correlated. Robust
standard errors that are consistent in the presence of heteroskedasticity and
autocorrelation within states are used in calculating t-statistics.
Summary statistics of the data used are provided in Table 2-1. The sample is
comprised of quarterly data. The date in which a state enters the sample is determined by
the date the utility commission in that state and the corresponding leader state ordered its
initial UNE rates. That date ranges from April 1997 to October 1997. Data for all states
22 In the estimations that follow, the previous four quarters of the level of the lagged dependent variable are
used as instruments in the first differences equation, while one lag of the first difference of the lagged
dependent variable is used as an instrument in the levels equation.
included in the analysis run through the end of 2003. Correlation coefficients between the
non-UNE rate explanatory variables are provided in Table 2-2.
The dependent variable in the analysis is the statewide average recurring rate for
the local loop in a quarter.23,24,25 The local loop consists of the wires that connect a
customer's premises to the incumbent's wire center. The local loop is the network
element that is the most costly to replicate, and so is the element that entrants would most
likely lease from the incumbent.26 The rates used in the study were obtained primarily
from state commission orders and incumbent documents. Unlike data sets used in earlier
studies, the exact date on which the rate was ordered is known and will be integral in
examining the issue of interstate interdependence
The cost variable is a measure of the monthly cost of the loop. Beginning in 2000
the FCC published annual TELRIC estimates. As noted above, while parties have debated
whether the model over- or under-estimates costs, it does provide a reference over time
and across states. However, these data are not available prior to 2000. Therefore the cost
variable used here is constructed from two different data series. For the period prior to
2000 the cost variable equals the embedded, or historical, cost of the loop as reported by
the National Exchange Carrier Association regarding universal service funding. For 2000
23 Specifically, it is weighted average of the monthly rate for the 2-wire copper in the various density zones
in the state. The weights are the number of lines in each density zone.
24 Another option is to use the rate set for the loop in the densest areas of the state, known as the urban zone
rate. However, many states did not de-average rates by zone until 2000 and did so at differing times.
Thus, to be able to model rates since TA96 was implemented, statewide average rates must be used.
25 Data for Alaska and Hawaii are not included as they are not part of any of the major incumbent regions.
Arkansas is also not included because the commission in that state ruled that it did not have the authority
to conduct a cost study to set UNE rates. Rates in the District of Columbia are not included, as the focus
is on rate setting by state utility commissions. Rates for Connecticut are not included as it is
geographically surrounded by Verizon states but the incumbent in the state is owned by SBC.
26 Some entrants serving large business customers built their own networks and therefore did not have to
rent any network elements from the incumbent.
forward, the cost variable is the estimate from the FCC's model. While the variable is not
ideal, it should capture the factors that account for the discrepancies in costs, such as
population density, wire center locations, and local cost levels.27
As explained in Section 2, UNE rate proceedings are often lengthy. As such, it
takes time for commissions to incorporate new information in their cost studies. Further,
state commissions may not learn the results of proceedings in other states for some time
or may be in an earlier stage of their UNE rate study. Therefore, it is appropriate to model
with a lag the influence of the rates of the other states. However, as the number of lags
included in the model increases, the number of explanatory variables increases quickly28,
thus limiting degrees of freedom. With this constraint in mind, the model is estimated
using two different lag structures. The first includes the neighbor states' rates over each
of the last four quarters, thus spanning one year. The second structure includes the
neighbor states' rates from each of the last four half-years, thus spanning two years.29
As described above, characteristics of the state utility commissions may also
influence UNE rates. To control for the length of time commissioners have served, the
average tenure of the commissioners is included. The political ideology of the
commission is captured in the fraction of commissioners that describe themselves as
Republican. The effect on UNE rates of the form of retail rate regulation is captured in a
dummy variable that equals one if the state employs rate of return regulation on either
27 The results do not change substantially when embedded costs are used for the entire period.
28 For each lag of other states' rates that is included in the estimation, three explanatory variables are added
(the weighted average of the neighbor rates interacted with the leader and follower dummy variables and
the leader's rate interacted with the follower dummy variable).
29 Given the rates of other states enter the model with a lag, they are treated as exogenous from the
perspective of the commission setting current rates.
residential or business services.30 However, this variable may be endogenous, as
unobserved factors may be jointly determining the form of retail rate regulation and the
UNE rate. To correct for this endogeneity, variables that measure customer satisfaction as
reported by the FCC's ARMIS database are used as instruments.31 The average retail rate
for the incumbent is calculated using FCC ARMIS data and is lagged one year to account
for the information delays in UNE rate proceedings outlined above.32 Finally, a dummy
variable is also included that takes the value of one if the rate was set during an
arbitration case immediately following the passage of TA96, and zero otherwise.33
Two explanatory variables are employed to account for the political sentiment of
the state's elected officials and its citizens. A dummy variable is included that equals one
if the governor is a Republican and zero otherwise. Another variable is included that
equals the percent of the state legislators that are Republican. To determine the potential
interactive effects between the ideology of the commission and the governor, the product
of the two variables is included as an explanatory variable.
30 Retail rate regulation schemes are quite complex, as some plans can have price caps on some services
and another form of incentive regulation on other products. By using a dummy variable that reflects
whether the state employs rate-of-return regulation on either residential or business basic services, the
complications posed by the idiosyncrasies of the various incentive-based regulation plans are avoided.
As this variable takes a value of one when rate-of-return regulation is used, the expected sign of the
coefficient of this variable is positive given Lehman and Weisman's (2000) analysis.
31 These variables are used as instruments because while UNE rate proceedings typically pit incumbents
against entrants, retail rate regulation proceedings tend to be disputes between the incumbent and
consumer groups. Therefore, while these satisfaction variables affect the form of retail rate regulation,
they do not affect the level of UNE rates. The econometric evidence supports this logic, as the tests of
joint significance indicate that the instruments are correlated with the form of retail rate regulation while
the C-statistic values indicate they are not correlated with UNE rates. See Appendix A for details.
32 Given the variable enters the estimation with a lag, endogeneity is not expected to be a concern. A C-test
confirms that the variable is not endogenous. See Appendix A for details.
33 Specifically, the arbitration rates used in this study are those from arbitrations between AT&T and the
Dummy variables are also used to capture the effects of the status of the
incumbent's application to sell long-distance service in that state. A dummy variable is
included that equals one if the rate was set during the year prior to the incumbent's
application up to the date of the FCC's decision. To measure the marginal impact of the
incumbent making a voluntary reduction, a dummy variable is included that equals one if
the rate was the result of a voluntary reduction by the incumbent in conjunction with its
The level of competitive entry in the state is controlled for by the number of UNE
lines leased by entrants. To account for delays in commissions incorporating information
into their decisions, the variable is lagged one year. Endogeneity may be present, as the
level of entry is likely determined in part by the future UNE rate, which in turn (given the
stationarity of UNE rates) is likely correlated with the present rate. The state
unemployment rate lagged one year is used as an instrument for this variable.34 To allow
for meaningful comparisons across states, the number of leased lines is divided by the
standard deviation of the variable for that state.
Finally, dummy variables for each calendar year are included in the analysis to
account for shocks common to all states in a given year not captured by the other
Table 2-3 summarizes the coefficient estimates from the basic model. Column (1)
lists the estimates from the lag structure spanning one year, while the estimates
34 The logic behind the use of the variable is that the level of economic activity affects the level of UNE
entry but not UNE rates. The econometric results suggest that these instruments are valid. See Appendix
A for details.
corresponding to a lag structure spanning two years are listed in Column (2). The
estimation diagnostic tests indicate that the system GMM approach is valid with this data
set. The tests for autocorrelation demonstrate that there is first-order autocorrelation in
first differences, but no autocorrelation of higher order. This indicates that the error terms
are not serially correlated. Further, the specification passes the Hansen tests of
overidentification, which tests whether the moment conditions beyond those needed to
identify the parameters are valid.35 Finally, the coefficient on the lagged dependent
variable is over 0.8 in both specifications, which confirms that the system GMM
estimator is appropriate.
The estimates from both columns indicate that the cost variable is statistically
significant, at either a 90% or 95% statistical confidence level depending on the
specification. Of the neighbor rates, the leader's rate lagged six months has a statistically
significant impact on the followers' rates, while the followers' rates do not have a
statistically significant impact on the rates of other followers or leaders.
The dummy variables that capture the effect of the incumbent's application to sell
long-distance services are negative and statistically significant. The negative coefficient
on the variable that indicates whether the incumbent made a voluntary reduction in the
application process suggests that the incumbents were not engaging in strategic behavior.
In the two-year lag specification, the dummy variable that captures the effect of a
Republican governor is positive and statistically significant. Finally, the level of UNE
entry appears to have a statistically significant effect at a 90% confidence level.
35 See Appendix A for further details.
Somewhat surprisingly, none of the variables that capture commission characteristics are
The presence of the lagged dependent variable as an explanatory variable implies
that the coefficient estimates only measure short-run effects. Furthermore, the coefficient
estimates do not clearly convey the economic significance of the explanatory variables.
Table 2-4 details the effects on UNE rates from a change in an explanatory variable that
lasts two years.36 The effects reported in Table 2-4 are based on the coefficient estimates
in the specification that includes two years of lagged neighbor UNE rates37
According to the estimates, a one-dollar increase in the cost variable on average
leads to a $0.40 increase in the UNE rate two years later. This increase is highly
statistically significant. However, an F-test as to whether a $1 increase in the cost
variable leads to $1 increase in UNE rates is rejected at over a 99% confidence level.
Thus, UNE rates do not perfectly reflect changes in cost (as measured by the available
The estimates suggest that UNE rates are significantly affected by changes in the
leader's UNE rate. Figure 2-3 depicts the effect over two years on followers' rates after a
one dollar increase in the leaders' rate and the corresponding 95% confidence interval. As
the figure shows, the average effect after one year is to increase the followers' rate by
36 Long-run effects that are based on permanent changes in the explanatory variable are reported in
7 The corresponding effects based on the specification that includes one year of lagged neighbor rates are
very similar and are contained in Appendix A.
38 It is worth repeating that the cost variable used here (based mostly on the FCC's cost model) is not
universally considered a valid cost proxy. Thus, while a $1 increase in the cost variable is not realized in
the UNE rate, one could argue that the cost variable overstates TELRIC costs to such a degree that
changes in costs are in fact fully realized in changes in UNE rates.
roughly $0.60, while after two years, the effect reaches almost $0.80. Table 2-4 reports
that a one standard deviation ($3.37) in the leader's rate leads to a $2.47 increase in the
followers' rates. Conversely, the estimates suggest that the effect of followers' rates on
other followers is not statistically significant. Thus, the results suggest that the leader
states do in fact have a significant impact on the other states in the incumbent's region.39
Of the variables that capture characteristics of the state utility commission, the
effect of average tenure has the expected sign, but is not statistically significant. In
regards to political ideology, neither the effects of the ideology of the commissioners nor
the ideology of the commissioners interacted with the ideology of the governor are
statistically significant. This may be due to the potentially conflicting effects described
above. The effect of the rate-of-return retail rate regulation variable is positive as
expected but statistically insignificant and only a third of the effect found by Lehman and
Weisman (2000). The lack of statistical significance and smaller estimated effect may be
due in part to the evolving nature of incentive regulation plans and the difficulty in
classifying them (as noted above). Furthermore, the average retail rate also is statistically
insignificant. Finally, while the AT&T arbitration rate has the expected sign, it too is not
However, the election of a Republican governor is associated with a $1.15 increase
in UNE rates two years later that is statistically significant. One may attribute this effect
39 The estimate of the effect of the followers' rate on the leader's rate is somewhat surprising. While it is
only roughly two-fifths of the size of the leader's effect on the followers, it is statistically significant at a
90% confidence level. From the coefficient estimates, it appears this effect exists four quarters after the
followers change their rates. A possible explanation is that there is a feedback effect. For example, if
rates are generally falling, the following pattern may be present: The leader lowers its rate, which is
followed six months later by the followers. One year after the followers' rate change, the leader again
lowers its rate again, which causes the leader's rate to appear to be influenced by the followers' rate
change. Also, given that the coefficient estimates of the effect of the followers' rates on the leaders are
not statistically significant, the statistical significance of the overall effect is being driven largely by
covariance between the coefficients on the followers' rates and the lagged dependent variable.
to two, perhaps not mutually exclusive effects: a preference for less-aggressive regulation
and favorable treatment for large corporations.40 The effects of both the percent of
Republican state legislators and the interaction of the Republican governor and
commission political affiliation are statistically insignificant.
The variables that capture the long-distance application status of the incumbent
indicate that those applications had a strong influence on the UNE rates. On average,
UNE rates set during the period prior to applications to enter the long-distance market are
roughly $2.00 lower two years later. When the incumbent voluntarily lowered the UNE
rate this effect more than doubled.
Finally, the observed level of entrant use of incumbent UNE lines is both
statistically and economically significant. A one standard deviation decrease in UNE
lines is associated with UNE rates falling roughly $4.00. As UNE rates are generally
falling during this period, the positive coefficient may also reflect that UNE rate
reductions are more modest when UNE entry is relatively strong.
This paper analyzes the factors that determine the rates that state utility
commissions set for access to the incumbent's telecommunications network. The results
suggest that factors other than cost influence the rates. The rate set by the largest state in
each incumbent region appears to influence the rates set by other states in that region.
Further, rates tend to be higher in states where the governor is a Republican and lower in
the period prior to an incumbent applying for permission to sell long-distance services
and when observed competitive entry is lower.
40 One concern with this result could be that Republican governors tend to be elected in less urban, and thus
higher UNE-cost, states. However, the average UNE rate in states with Republican governors is virtually
identical to the average in states with Democratic governors.
Beyond providing insights into telecommunications regulation, these results may
have implications for other areas in which the federal government delegates
implementation of policies to state agencies. While state agencies may possess valuable
local knowledge, it is possible that the intent of the federal policy will not be perfectly
realized through the actions of the states. Information spillovers across states and factors
outside the scope of the federal mandate may affect how the policy is implemented. Other
policy areas in which this type of analysis may provide some insight include state
implementation of the No Child Left Behind Act and environmental policies as dictated
by the EPA.
The findings in this paper could be extended in several directions. For example,
the rates used in this paper are the monthly rates for the local loop. The commissions also
set one-time connection fees for these loops that are also important to entrants
considering entering a state. Also, while the local loop is arguably the most important
network element, the states set rates for many other network elements. Analyses of these
additional rates could shed further light on the factors that influence state commissions.
Finally, as noted in footnote 20, initial tests for nation-wide information spillovers
suggested that states may have influence outside the incumbent's region. Further analysis
into this issue could provide additional insight into how state agencies implement federal
an97 Jan98 Jan99 JanOO Jan01 Jan02 Jan03 Jan04
----- New Hampshire -Vermont
----- Kentucky Tennessee
Figure 2-1 Comparison of UNE Rates. A) New Hampshire and Vermont. B)Kentucky
and Tennessee. C) Wyoming and Utah.
Figure 2-1 Continued
IN MI OH WI
U HCPM estimate U UNE rate
AL FL GA KY LA MS NC SC TN
* HCPM estimate U UNE rate
* HCPM estimate U UNE rate
Figure 2-2 HCPM Estimate versus UNE Rate. A) Ameritech. B) BellSouth. C) Pacific
Telesis. D) Qwest. E) Southwestern Bell. F) Verizon.
AZ CO IA ID MN MT ND NE NM OR SD UT WA WY
U HCPM estimate U UNE rate
* HCPM estimate U UNE rate
Figure 2-2 Continued.
DE MA MD ME NH NJ NY PA RI VA VT WV
HCPM estimate U UNE rate
1 2 3 4 5 6 7 8
Number of Quarters Since Increase
Figure 2-3 Cumulative effect of $1 increase in the leader's rate on the follower's rate.
Table 2-1 Summary Statistics
Variable Mean Minimum Maximum Deviaion
Laging Neighbor States' Rates One Year (n = 1103)
UNE Rate (All States) $16.62 $7.01 $28.82 4.59
UNE Rate (Leader States) $15.30 $9.81 $20.65 3.41
UNE Rate (Weighted Average of $15.80 $8.13 $27.22 3.14
Cost Variable $21.27 $13.97 $41.78 4.38
Percent of Commissioners Republican 46.3% 0.0% 100.0% 30.7
Average Tenure of Commissioners (yeai 5.2 0 22.5 3.1
Rate of Return Retail Rate Regulation 0.18 0 1 0.39
Governor Republican 0.6 0 1 0.49
Percent of Legislature Republican 0.5 0.13 0.89 0.14
AT&T Arbitration 0.16 0 1 0.37
Period Prior to Section 271 Decision 0.25 0 1 0.43
Voluntary Reduction 0.07 0 1 0.26
UNE Entry (Lagged One Year, Divided 0.54 0 2.77 0.69
by Standard Deviation)
Average Retail Rate $29.16 $14.45 $44.44 5.16
Laging Neighbor States' Rates Two Years (n = 919)
UNE Rate (All States) $16.44 $7.01 $28.29 4.49
UNE Rate (Leader States) $15.11 $9.81 $20.65 3.37
UNE Rate (Weighted Average of $15.65 $8.13 $20.51 3.13
Cost Variable $21.24 $13.97 $35.86 4.22
Percent of Commissioners Republican 45.8% 0.0% 100.0% 30.8
Average Tenure of Commissioners (yeai 5.2 0.1 22.5 3.0
Rate of Return Retail Rate Regulation 0.17 0 1 0.37
Governor Republican 0.6 0 1 0.49
Percent of Legislature Republican 0.5 0.13 0.89 0.14
AT&T Arbitration 0.11 0 1 0.31
Period Prior to Section 271 Decision 0.29 0 1 0.45
Voluntary Reduction 0.09 0 1 0.28
UNE Entry (Lagged One Year, Divided 0.64 0 2.77 0.71
by Standard Deviation)
Average Retail Rate $29.53 $14.45 $44.44 5.22
Table 2-2 Correlation Matrix of Non-UNE Rate Variables
% Comms. Tenure of ROR Retail % Legis.
Cost Var. Rep. Comms. Rate Reg. Gov. Rep. Rep.
Cost Var. 1.00
% Comms. -0.23 1.00
Avg 0.24 0.00 1.00
ROR Retail 0.23 -0.08 -0.02 1.00
Gov. Rep. -0.07 0.44 0.01 -0.04 1.00
% Legis. 0.04 0.20 0.11 0.27 0.12 1.00
AT&T Arb. 0.17 -0.14 0.00 0.10 0.14 0.10
Long-Dist. 0.00 0.03 -0.01 -0.26 -0.06 0.04
Vol. Reduct. 0.01 0.10 0.06 -0.12 0.08 -0.01
Gov. Rep. -0.18 0.80 -0.01 0.00 -0.08 0.78
UNE Entry -0.11 0.07 -0.07 -0.29 -0.12 -0.05
Avg. Retail 0.43 0.01 0.15 -0.09 -0.12 -0.06
Table 2-2 Continued
Gov. Rep. *
Long-Dist. Vol. Reduct. % Comms. Avg. Retail
AT&T Arb. App. By Incumb. Rep. UNE Entry Rate
AT&T Arb. 1.00
Long-Dist. -0.06 1.00
Vol. Reduct. 0.13 0.49 1.00
Gov. Rep. 0.19 0.00 0.05 1.00
UNE Entry 0.04 0.50 0.37 -0.01 1.00
Avg. Retail -0.19 0.21 0.08 -0.04 0.12 1.00
Table 2-3 Coefficient Estimates from One-Year and Two-Year Lag Specifications
Explanatory Variable Specification Specification
0.876 *** 0.815 ***
UNE Rate, Lagged One Quarter
Effect of Leader's Rate on Followers
Lagged One Quarter
0.164 0.281 ***
Lagged Two Quarters
Lagged Three Quarters
Lagged Four Quarters -
Lagged Five Quarters
Lagged Six Quarters-0.
Lagged Seven Quarters
Lagged Eight Quarters (1.16
Effect of Followers' Rates on Other Followers
Lagged One Quarter
Lagged Two Quarters 0
Lagged Three Quarters
Lagged Four Quarters-0
Lagged Five Quarters
Lagged Six Quarters0.0
Lagged Seven Quarters
Lagged Eight Quarters 0.1
Effect of Followers' Rates on Leaders
Lagged One Quarter
Lagged Two Quarters-0
Lagged Three Quarters
Lagged Four Quarters 0
Lagged Five Quarters
Lagged Six Quarters0.0
Lagged Seven Quarters
Lagged Eight Quarters 0.018
Table 2-3 Continued
Percent of Commissioners
Average Tenure of
Rate of Return Retail Rate
Percent of State Legislature
Percent of PUC Republican
Interacted with Governor
Period Prior to Decision on -0.481 ** -0.479 ***
Long-Distance Application (-2.65) (-2.48)
Voluntary Reduction by -0.532 ** -0.636 **
Incumbent (-2.45) (-2.16)
UNE Entry (lagged 1 year, 0.770 0.999 *
divided by standard deviation (1.72) (1.92)
Average Retail Rate (lagged 1 0.008 0.011
year) (0.77) (0.79)
Overall F-Statistic 1894 1412
Number of Observations 1103 919
For AR(1) in First -2.85 *** -2.57 ***
Differences (0.00) (0.01)
For AR(2) in First -0.22 -0.8
Differences (0.82) (0.42)
Hansen Test of Overidentified 1.76 2.59
Restrictions (p-value) (0.94) (0.86)
Notes Year dummy variables are included and are generally statistically significant.
Unless otherwise noted, t-statistics are reported in parentheses.
*** statistically significant at 99% confidence level
** statistically significant at 95% confidence level
* statistically significant at 90% confidence level
Table 2-4 Effects of Changes in Explanatory Variables on UNE Rates Two Years Later
Effect of a One
Explanatory Variable Unit Increase Economic Effect1
Effect of Neighbors' Rates
0.73 ** 2.47 **
Leader's Rate on Followers (C) .72.7
Followers' Rates on Followers (C)-.031.
(-1.03) (-1.03)97 *
Followers' Rates on Leaders (C) 1..91
Effect of Other Explanatory Variables
0.40 *** 1.68 ***
Cost (C) 0.40 1.68
Percent of Commissioners -0.01 -0.31
Republican (C) (-1.10) (-1.10)
Average Tenure of Commissioners 0.07 0.21
(C) (0.96) (0.96)
Rate of Return Retail Rate 0.67 0.67
Regulation (B) (0.20) (0.20)
1.15 1.15 *
Governor Republican (B)1.
Percent of State Legislature -1.63 -0.23
Republican (C) (-0.74) (-0.74)
Percent of PUC Republican -0.01 -0.37
Interacted with Governor Republican (-0.81) (-0.81)
AT&T Arbitration (B)3
Period Prior to Decision on Long- -2.08 ** -2.08 **
Distance Application (B) (-2.27) (-2.27)
Voluntary Reduction by Incumbent -2.77 ** -2.77 **
(B) (-2.19) (-2.19)
UNE Entry (lagged 1 year, divided by 4.35 ** 4.35 **
state standard deviation) (C) 2 (2.42) (2.42)
Average Retail Rate (lagged 1 year) 0.05 0.24
(C) (0.81) (0.81)
Economic effects for continuous explanatory variables are based on a one standard
deviation increase in that variable, while economic effects for binary explanatory variables
are based on a change in the variable from zero to one.
2 This variable is already scaled by dividing by the standard deviation, so the economic
effect is based on a one-unit increase.
Notes (B) binary explanatory variable, (C) continuous explanatory variable,
t-statistics are reported in parentheses.
*** statistically significant at 99% confidence level
** statistically significant at 95% confidence level
statistically significant at 90% confidence level
THE EXTENT AND MEANS OF ENTRY INTO LOCAL TELECOMMUNICATIONS
A primary goal of the Telecommunications Act of 1996 (TA96) is to facilitate
competition in the local telecommunications market. TA96 includes provisions that allow
entrants1 to lease parts of the incumbent' s2 network, known as unbundled network
elements (UNEs), at relatively low rates determined by state public utility commissions.
There are two ways in which entrants can provide phone service by leasing UNEs. Under
a loop arrangement3, the entrant rents from the incumbent the phone line that connects a
customer's residence or premises to the local wire center. However, the entrant provides
the equipment that connects the customer's line to the broader telephone network. The
other means by which entrants can offer service is by leasing UNEs through a platform
arrangement, which became widely available in 2000.4,5 The key difference between
loop- and platform-based entry is that in platform arrangements, the entrant leases all of
the UNEs needed to provide telephone service. In other words, the entrant simply
SEntrants into local telecommunications markets are known as Competitive Local Exchange Carriers, or
2 Incumbents in local telecommunications markets are known as Incumbent Local Exchange Carriers, or
3 This is also referred to as UNE-L entry.
4 This is also referred to as UNE-P entry.
5 As described below, recent court rulings and revisions to FCC regulations have altered the regulatory
treatment of platform-based entry.
rebundles the UNEs that are required to provide service and does not have to own any of
the necessary equipment.6
A sufficiently long time series of data is now available to investigate the factors
that determine the extent of entry via these two alternative arrangements. Figure 3-1
indicates that from July 2001 to July 2004 the national average of the fraction of
incumbent7 lines leased to entrants increased steadily. The average increased from just
over 5% in July 2001 to almost 20% three years later. However, Figure 3-2 indicates that
the increases in entry differ greatly by means of entry. During the 2001 2004 period,
while the fraction of lines leased through a loop arrangement ranged between 2% and
4%, the fraction leased through a platform arrangement tripled from less than 4% to over
Figures 2-3 and 2-4 suggest that the evolution of entry has also differed
significantly across states. Each dot in Figure 3-3 represents for a given state the fraction
of lines leased through a loop arrangement in July 2001 and in July 2004. Dots near the
45-degree line represent states where the share of loop-based entry changed little between
the two dates, while dots above the 45-degree line indicate that the share increased. As
the figure indicates, while there are some states in which the share changed little, there
6 Incumbents have been quick to point out that there is no actual difference between platform-based entry
and reselling the incumbents' services, another option provided for under TA96. Given the cost of
platform arrangements and the resale discount rates, reselling the incumbents' services is typically more
costly for entrants than providing service under a platform arrangement. However, platform-based entry
is more risky for entrants as they are not guaranteed a positive gross profit margin as they are under a
resale arrangement. This risk would be realized if the retail price fell below the platform cost, in which
case the entrant would face a negative gross profit under a platform arrangement whereas under a resale
arrangement the entrant's cost would fall by the same percentage as the retail price.
The largest incumbents in each state are the regional monopolies that were created in the court-ordered
split-up of AT&T in 1984 and are commonly referred to as RBOCs (Regional Bell Operating
Companies). While entrants can rent UNEs from other incumbents in areas not served by an RBOC, the
vast majority of UNE entry has occurred in RBOC regions. For the remainder of this paper, the term
incumbent will refer exclusively to RBOCs.
are also a number of states where the share increased substantially. Figure 3-4 displays
the corresponding data for platform-based entry. While most states are above the 45-
degree line, the extent to which they line above the line varies substantially.
Figures 2-1 through 2-4 raise important questions regarding telecommunications
market dynamics, including the following two: What market factors influence the level of
competitive entry via the two alternative means? What effects did state regulatory
policies or the political environment have on the observed level of entry?
The answers to these questions are of interest to both policymakers and
researchers. On one level, the answers shed light on the effectiveness of TA96 and its
implementation. Not only can these answers inform the ongoing debate regarding
regulation of U.S. telecommunications markets, but they may also provide guidance to
regulators in other countries attempting to fashion policies. Further, the results
supplement the existing body of literature regarding competitive entry. There has been
scant prior research regarding how entrants choose whether to lease all of the inputs from
their competitors (as in a platform arrangement) or buy some of the inputs themselves (as
in a loop arrangement). The choices of entrants could have important implications for
Several authors have analyzed the determinants of competitive entry into local
telecommunications markets.8 These papers often use cross-sectional data sets to explain
the level of competitive entry, as measured by the number of entrants or the number of
telephone lines entrants have acquired. Typically these papers find that variables
associated with higher demand correspond to higher entry, while variables associated
8 Examples include Abel (2002), Abel and Clemments (2001), Alexander and Feinberg 21i '14), Brown and
Zimmerman (2"'114), Jamison (2"'114), Lehman (2002), Lehman (2003), Roycroft (2005), and Zolnierek,
Eisner, and Burton (2001).
with higher costs are correlated with a lower degree of entry. In regards to the effects of
regulation, areas where retail price caps are used are generally associated with lower
levels of entry, while the effects of the political composition of state public utility
commissions differ across papers.
This paper is closely related to Beard and Ford (2002) and Beard, Ford, and
Koutsky (2005). Beard and Ford (2002) use a pooled data set to analyze the determinants
of loop- and platform-based entry. Among their findings is that, for both types of entry,
as the cost of leasing the UNEs used in that type of entry increases, the level of entry
falls. Their estimates of the cross-price elasticity of demand also suggest that entry via
loop and platform arrangements are not substitutes. Beard, Ford, and Koutsky (2005)
examine a cross-sectional data set to examine the deployment of equipment by entrants.
Using proprietary data, they come to the conclusion that higher UNE leasing costs lead to
decreased entrant equipment investment. Sappington (2005) constructs a theoretical
model that suggests the price at which an entrant can lease an input may have little effect
on its decision to lease the input from the incumbent supplier or make the input itself.
This conclusion arises because the lease price influences the intensity of downstream
competition such that the incumbent tends to price less aggressively when the lease price
paid by the competitor is higher.
This paper improves upon the existing literature in at least four important ways.
First, panel data are used rather than cross-sectional data. Thus, the data will allow for a
more precise estimate of the effects of dynamic changes in the market environment on the
level of entry observed. Second, the influence of the political composition of state public
utility commissions is estimated. Third, the effect of the connection charge that loop
entrants must pay whenever they gain a new customer is measured. The potentially
dampening effect of the connection charge on loop-based entry was cited by the Federal
Communication Commission as an important basis for its regulatory treatment of
platform-based entry.9 Fourth, the empirical specification allows for the estimation of
differential effects of the explanatory variables for varying levels of market size.
Four significant conclusions are offered. First, while the two types of entry are
generally affected by different market factors, there appears to be cost-based substitution
between them. Second, changes in the own monthly costs of the two forms of entry have
limited effects, but loop-based entry decreases in response to increases in the connection
charges entrants must pay incumbents when a customer is acquired. Third, loop-based
entry tends to be more pronounced relative to platform-based entry as the degree of
Republican representation on state public utility commissions increases. Fourth, loop-
based entry is more responsive to changes in market conditions in smaller states, while
platform-based entry is more responsive to market conditions in larger states.
The paper is organized as follows. Section 2 presents background information on
the functional and regulatory differences between loop- and platform-based entry.
Section 3 presents the determinants that are hypothesized to affect the level of entry.
Section 4 details the econometric methodology and the data used. Section 5 presents the
estimation results, while Section 6 provides conclusions and areas for further research.
9 For example, see FCC (2003), 295-298.
Background Information on UNE-Based Entry
Loop-Based Versus Platform-Based Entry
To properly interpret the results below, it is important to understand the
differences between loop-based and platform-based entry.
As stated above, when an entrant acquires a customer under a loop arrangement,
the entrant leases only the UNEs associated with the wire that connects the customer's
premises to the incumbent's wire center. This wire is referred to as the local loop. In
order to provide service using a loop strategy, an entrant must provide its own switching
equipment10 and pay the incumbent for the space it rents and the power it uses in the
incumbent's buildings where the switching and related equipment are housed.11,12
Conversely, under a platform arrangement, the entrant does not have to own any of
the network equipment needed to provide phone service. The entrant in effect "re-
bundles" all of the UNEs it needs to provide service.
An important difference between loop- and platform-based entry is how the
incumbent's customers are switched to the entrant's service. Lines that are used to serve
customers under a loop arrangement must be physically disconnected from the
incumbent's switching equipment and reconnected to the entrant's equipment. This
transfer, known as a hot-cut, requires that both an incumbent and entrant technician be
10 While switches can have many functions, the key role that they serve for entrants to provide telephone
service is that they connect the incumbent's loop to the entrant's network.
1 Entrant switches require economies of scale to be cost-effective. Thus, entrants may install equipment in
remote locations that allows them to aggregate traffic before reaching a switch that it owns (this process
is known as back-hauling). However, the entrant then not only has to pay for the aggregation equipment,
but also must pay the incumbent to move its traffic to its switch.
12 This practice of renting space in an incumbent telephone facility is known as collocation.
present in order to perform a "seamless migration"13 of the customer to the entrant's
network. For each hot-cut that is performed, the entrant must pay the incumbent a
connection charge to compensate the incumbent for the labor involved. Conversely, to
transfer an incumbent's customer to an entrant who employs a platform strategy,
computer software is used that allows for the process to be fully automated. The entrant
must only pay a nominal administrative fee to the incumbent to transfer the customer.
History of Platform-Based Entry Regulation
While there has been widespread agreement that allowing entrants to lease the local
loop is beneficial to long-run competition, there has been heated disagreement as to the
effects of allowing entrants to follow a platform strategy. Advocates of allowing
platform-based entry typically point to the technical and financial difficulties entrants
face when using loop arrangements to serve residential customers.14 Those critical of
forcing incumbents to provide platform arrangements argue that the option discourages
entrant investment in telecommunications equipment, which is one of the goals of
TA96.15 Among the FCC commissioners, this disagreement typically splits along party
lines, with Democratic commissioners in support of the platform-based entry option and
Republicans opposed. 16
13 A seamless migration is an industry term that describes a hot-cut where the customer does not lose
phone service for any noticeable length of time.
14 The connection charges and labor costs involved in a hot cut are claimed to be prohibitively expensive
for entrants, especially as the chur rate among their customers is relatively high (WorldCom (2002)). In
addition, when the hot cut is performed, the customer may lose service for a brief period of time.
Entrants complain that customers associate the delay with the entrant to whom they are transferring
service, and thus the customer immediately perceives a lower quality of service with the entrant (FCC
15 See, for example, Crandall, Ingraham, and Singer (2i 1'4).
16 A notable exception to this generalization is the current Republican FCC Chairman Kevin Martin, who
sided with the two Democratic commissioners in the 2003 Triennial Review Order (FCC (2003)) to
continue the availability of platform-based entry.
Given the enormous financial stakes involved in the local telecommunications
market, perhaps it is not surprising that there has been a great deal of regulation and
litigation concerning platform-based entry. There are two main regulatory requirements
for platform-based entry to be feasible: entrants must be able to "re-bundle" UNEs and all
of the necessary network elements must be unbundled by the FCC.
Entrants eventually realized that for certain customers they could realize greater
profits if they simply leased all of the UNEs necessary for them to provide phone service,
rather than providing any of the necessary equipment themselves. 17 While incumbents
began allowing entrants to lease the UNEs necessary in a platform arrangement in some
states in 1999, the incumbents often charged additional fees for "re-bundling" the
network elements. The FCC then ruled that the incumbents could not charge these re-
bundling fees. Litigation soon followed, culminating with the Supreme Court ruling in
2002 in Verizon vs FCC that entrants can legally re-bundle UNEs at no additional charge
from the incumbents.
In terms of the availability of UNEs, the network element that has been at the heart
of the platform-based entry debate is the incumbent switches, specifically those that are
used to service residential and small business customers.18 While the debate among the
FCC commissioners has often been very contentious, until recently most incumbent
switches19 have been available to entrants as UNEs at relatively low rates20. However, in
17 To be precise, the entrant must still provide equipment necessary for billing, marketing, and customer
18 The FCC defines this classification as any customer with four access lines or fewer.
19 In its 1999 UNE Remand Order (FCC (1999)), the FCC ordered that incumbents did not have to lease
certain switches located in the 50 largest MSAs.
20 These rates are set by state public utility commissions and based on a TELRIC (total element long-run
incremental cost) methodology. In essence, the UNE cost is to be based on the costs a hypothetical
March 2004, the U.S. Court of Appeals in Washington, D.C. ruled that the FCC rules did
not comply with TA96. In June 2004, the Bush administration announced that it would
not appeal the decision by the U.S. Court of Appeals, thus ringing the death bell of
platform-based entry as known by market participants.21 Partly in response to these
developments, AT&T and MCI subsequently announced that they were exiting the
residential market (Young (2004)). Finally, in February 2005, the FCC (2005) ruled that
as of February 2006 incumbent switches would not be available to entrants at the low
Hypothesized Determinants of Loop- and Platform-Based Entry
This section outlines the likely determinants of loop- and platform-based entry.
The determinants can be classified as describing the revenue potential, regulated costs,
and political effects.
Measures of revenue potential reflect the profits an entrant would expect to earn.
An obvious candidate would be the current retail price in that market, in so far as, ceteris
paribus, the higher the retail price the more attractive a market is to a potential entrant.23
The incumbent's average net revenue per line in that state is used as a proxy for these
prices. Another variable of potential interest to entrants is the growth prospects of the
market. One may expect that entrants will focus their efforts in areas where they expect
incumbent would incur using current technology and is not to be based on the actual historical
investment by the incumbent. See Quast (2005) for further details.
21 Rather than abruptly make incumbent switches unavailable as UNEs, the FCC ordered a 12-month
transition phase during which switches would be available at slightly higher costs. However, after the
12-month transition the parties are to negotiate the rates at which switches can be leased.
22 Many entrants have negotiated agreements with incumbents to continue leasing switches after they are
no longer available at TELRIC rates. For instance, Qwest has negotiated such agreements with more
than 60 entrants, while Verizon and BellSouth have negotiated such agreements with over 50 and 45
CLECs, respectively. (See BellSouth (t' "4); State Telephone Regulation Report (2005); and Telecom
23 Unfortunately, within a state and across customer types, several retail prices may be charged.
increases in the number of potential customers. To control for the effects of changes in
the growth level in a state, the change in the unemployment rate is included as an
The level of entry is also likely to be affected by the costs an entrant expects to
incur in providing service. A main source of costs for entrants is the payments they must
pay to incumbents to lease UNEs. As noted above, during the sample period those rates
are determined by state public utility commissions and do not vary by entrant within a
given state. For loop-based entry, there are two UNE rates that are especially important to
entrants: (1) the monthly price they must pay to incumbents to rent the local loop and a
connection device known as a line port; and (2) the fee charged by the incumbent to
transfer one of its customers line to the entrant's equipment (i.e., to perform a hot cut).
One would expect both rates to have a negative effect on the level of loop-based entry. In
terms of platform-based entry, the process used to transfer a customer to an entrant's
network is fully automated and incurs only a nominal charge. However, under a platform
arrangement the entrant must pay the incumbent to lease its switching equipment and to
transport calls over the long distance network, as well as to lease the local loop. This total
charge is referred to as the platform rate and, as it increases, platform-based entry is
expected to decline.
Political factors also may influence entry in ways that are not entirely clear. A
potential influence is the political composition of the state public utility commission.
Among the ways state public utility commissions affect entry (beyond setting UNE rates)
is by forcing the incumbent to adjust their hot-cut processes and computer systems
through which entrants order and are billed for services for newly acquired customers.24
Also, state commissions set right-of-way access regulations that can either help or hinder
entrants who install their own equipment. One could imagine that a commission that is
majority Republican would oppose governmental involvement in the market and thus
limit their efforts at facilitating a relatively more involved regulatory design such as
platform. However, it may also be the case that Republican allegiance to small
businesses may predispose them to treat entrants favorably. Likewise, the effect of a
Republican governor on competitive entry is not clear.25,26
While the factors described above may be expected to have certain effects on
entry, these effects may differ substantially by the size of the market. For instance, a $1
increase in revenue per customer may lead to entrants attaining a greater additional
market share in larger markets than in smaller markets, as entrants may already have a
presence in larger markets and the incremental cost of acquiring an additional customer is
relatively low. Conversely, changes in market conditions may have a greater impact in
smaller markets, as entry in these markets is less certain and changes in potential profits
can determine whether entrants choose to enter the market. To capture these potential
differences across states, the model includes interaction terms between the factors listed
above and the population of the state.
24 For example, states have implemented various approaches in forcing incumbents to implement
procedures to migrate customers to entrants (FCC (2003), 309-310). In the former Ameritech region,
entrants accused SBC of not having adequate computer systems to allow for platform-based entry (see
25 Note that while the discussion that follows refers to measurements of Republican influence, given the
virtual binary political party environment, the results can be cast in terms of Democratic influence
merely by replacing the reported result with its opposite.
26 Another regulatory variable that could have a significant effect on entry is whether the incumbent is
under price or rate of return regulation for their retail prices. During the sample period the type of retail
rate regulation did not change for any of the incumbents. Therefore, this effect is captured in the fixed
One of the advantages of a panel data analysis is the ability to control for effects
specific to a state that are not captured elsewhere. State fixed effects control for time-
invariant effects that are not captured in the other explanatory variables and that are
specific to a given state (e.g., operating cost differences).27 In addition, time fixed effects
are included to control for national effects specific to a given period.28
Model Specification and Data Used
Given that the levels of loop- and platform-based entry are highly related,29 it is
appropriate to estimate them simultaneously. Thus, a seeming unrelated regression (SUR)
approach is employed.30
Summary statistics of the data used are provided in Table 3-1. The sample is based
on semi-annual data by state for the period January 2001 to July 2004.31'32'33 Correlation
coefficients between the explanatory variables are provided in Table 3-2.
27 Given that in the sample no state is served by more than one incumbent, state fixed effects also control
for incumbent fixed effects.
28 The model was also estimated without state and time fixed effects. Compared with the results reported
below that include these fixed effects, the estimates without fixed effects indicate a stronger negative
relationship between the level of entry and the monthly cost of that form of entry. These results are
similar to those in Beard and Ford (2002), who do not include state fixed effects in their regressions.
However, in the regressions without state and time fixed effects, the monthly cost of entry may reflect
the attractiveness of entry into that state. For example, a high loop rate may be associated with lower
entry, but high loop rates exist in areas where loop entry is relatively unattractive (e.g., Wyoming).
29 Insofar as when an entrant decides to enter a market by leasing unbundled elements, the entrant must
choose either one of these entry methods.
30 As specified, the SUR model results in the same coefficient estimates as when each dependent variable
is regressed separately. However, the SUR estimator is more efficient (assuming the model is correctly
specified) and thus results in lower standard errors of the coefficient estimates.
31 Alaska, Hawaii, and Washington, DC are not included in the analysis because of their particular
geographic circumstances. Arkansas is not included due to difficulties in obtaining connection cost data,
while for New Mexico those data could only be obtained for January 2003 forward. The fraction of lines
leased in a platform arrangement by Verizon are not available for Maine and Vermont for January and
July 2002, and for Delaware, New Hampshire, and West Virginia for July 2002 only. The resulting data
set contains 271 observations.
32 The data regarding the level of entry, UNE rates, and the incumbent's average revenue are based on the
incumbent's service territory within a state, rather than on the entire state. Typically the two differ in that
the incumbent service area may not include some of the relatively rural areas within a state. However, as
The dependent variables in the estimations are the fraction of incumbent lines
leased by entrants using a loop arrangement and the fraction of lines leased using a
platform arrangement. These variables are based on data reported by incumbents to the
FCC in their Form 477 data submissions (http://www.fcc.gov/wcb/iatd/comp.html).34
In regards to revenue potential, the incumbent's revenue per line for a state is
obtained from ARMIS data submitted by the incumbents to the FCC
(http://www.fcc.gov/wcb/armis/).35 The change in the unemployment rate is based on
seasonally adjusted data reported by U.S. Bureau of Labor Statistics
As the level of entry may affect the average revenue, the average revenue may be
endogenous to the level of competitive entry. To attempt to control for this potential
endogeneity, three alternative strategies were also employed: using the lagged value of
average revenue as the explanatory variable, using the lagged value of average revenue as
an instrumental variable, and using as an instrumental variable the percent of the
incumbent's lines that are used by participants in a low-income assistance program
known as Lifeline.36 The estimates from each of these specifications are very similar to
those reported below.37
most of the entry occurred in non-rural areas, this discrepancy between the data sets should not
materially affect the results.
33 The incumbent's revenue per line is reported annually. The missing data points are calculated by linearly
interpolating the data.
34 As the UNE cost data are not available, the service areas of Verizon that were formerly GTE service
areas are not included in the analysis.
35 Specifically, the average revenue is calculated by dividing net revenue (ARMIS Report 43-01, Line
1090) by the number of lines (ARMIS Report 43-08). While a variable that more precisely measures the
retail rate would be preferred, this variable should approximate the potential revenue an entrant could
obtain and has been used by other authors, e.g., Ai and Sappington (2005) and Abel (2002).
36 The Lifeline program is an effort coordinated by the FCC and state public utility commissions that helps
pay for basic phone service for low-income consumers. Incumbents are obligated to publicize the
The monthly loop and platform UNE costs are obtained from surveys of state
public utility commissions by Billy Jack Gregg of West Virginia's Office of the
Consumer Advocate (http://www.cad.state.wv.us/Une Page.htm).38 During the sample
period examined in this paper, most state public utility commissions set rates that varied
by different density zones within the state. For the loop monthly cost, the monthly loop
and port cost in the urban zone used39, as these are the areas most likely to be entered by
entrants who utilize a UNE-L entry strategy. For the platform monthly cost, the variable
used is the incremental cost of providing service to customers via a platform-based entry
strategy in the suburban zone, as entrants typically serve residential customers via a
availability of this service and are encouraged to sign up as many participants as possible, which
therefore influences their customer mix. Further, one can imagine that low-income customers are more
likely to participate in the Lifeline program the higher the retail price charged by the incumbent. As
such, it is likely that the extent to which the incumbent's customers participate in this program is
correlated with the incumbent's average revenue. The first-stage F-statistic confirms that the variable is
sufficiently correlated with the incumbent's average revenue (or, in the language of the instrumental
variable literature, it is sufficiently relevant). Conversely, entrants typically do not participate in this
program. Also, differences in how state utility commissions administer the Lifeline program result in the
fraction of participants being uncorrelated with state income. Thus, the effect on entry of the fraction of
the incumbent's customers that are Lifeline subscribers should be minimal and the variable should be
exogenous to the level of entry. While these arguments suggest that the variable is exogenous, there is
potential correlation between how state utility commissions administer the Lifeline program and how
they treat competitive entry. Further, there does not exist a test to determine if an instrument is
exogenous in an exactly identified model such as this. Thus, it is not clear that this candidate instrument
is sufficiently exogenous to be valid.
7 While none of these three strategies is perhaps ideal, only one existing telecommunications entry paper
reviewed by the author has attempted to address this issue (Abel (2'' i' I?', and it employs the first strategy
listed above. Further, concerns about endogeneity are mitigated by the fact that often many of the
incumbent's retail rates are either regulated or published in tariffs that must be approved by state utility
commissions. Thus, the response in average revenue to changes in entry may be delayed or diminished.
38 Some participants in legal proceedings have disputed the way in which platform arrangement costs are
calculated in these data (see Willig, et al, 2002). Switching rates are typically based on minutes of use,
and the rates reported by Gregg are based on 1000 minutes of use. Criticisms of the Gregg rates center
on whether 1000 minutes of use is an appropriate benchmark. However, there is no other source of
consistently reported UNE rates over this period. Perhaps more importantly, as the fixed effects
estimator is based on deviations from the state mean for each explanatory variable, mismeasurement bias
is unlikely. Specifically, if the minutes of use used to calculate the platform arrangement cost is
consistently either less or more than the actual minutes of use, the coefficient estimates will not be
39 Data on collocation costs are not available.
platform-based entry strategy.40 The loop connection charges are obtained by reviewing
state public utility commission and FCC documents and trade press, and by contacting
the state public utility commissions directly.41
The percent of Republicans serving on state public utility commissions is derived
by reviewing membership directories of the National Association of Regulatory
Commissioners,42 while the political affiliation of the Governor in a state is obtained
from the U.S. Statistical Abstract (http://www.census.gov/prod/www/statistical-abstract-
04.html) and the National Governors Association website (http://www.nga.org).
Estimates without Population Interactions
Table 3-3 summarizes the coefficient estimates and t-statistics from estimating the
fraction of lines entrants acquired via loop and platform arrangements. The estimations
are first performed including each group of explanatory variables individually, and then
including all of the explanatory variables. The table indicates that the estimates do not
vary substantially according to whether the other groups of explanatory variables are
included in the regression.
40 Due to difficulty in comparing these costs across states, costs for transporting calls between switches are
not included in the Gregg UNE-P cost estimate.
41 A concern regarding the endogeneity of the UNE rates could arise due to the potential for reverse
causation. In particular, one could surmise that low levels of entry may persuade state public utility
commissions to lower UNE rates. However, any such response by state commissions could only occur
with a substantial lag, as the commissions would only learn of the level of entry with a lag and can only
revise UNE rates after a lengthy set of proceedings. Quast (2005) finds that it takes approximately one
year for the level of UNE entry to affect UNE rates set by state commissions. Thus, UNE rates can be
treated as exogenous.
42 The dates of the membership directories used to construct this variable are February 1999, February
2002, February 2003, July 2003, and March 2004
Of the variables that measure revenue potential, the average revenue has a
positive and statistically significant effect on the platform share, while its effect on loop
share is statistically insignificant. The lack of significance in the loop share equation may
reflect that this variable is a better indicator of the average residential price than the
average business price. The change in the unemployment rate is statistically insignificant
for both types of entry.
In regards to the variables that measure costs, those that measure the own monthly
cost have negative coefficients. Specifically, the coefficient on the loop monthly cost in
the loop share equation and the coefficient on the incremental platform cost in the
platform share equation are negative. However, neither is statistically significant. In
contrast, the loop connection charge is negative and statistically significant. In the
platform share equation, the coefficient on the loop monthly cost is positive, suggesting
that there may be substitution to platform-based entry as the monthly loop cost increases.
The coefficients on the regulatory variables indicate that Republican state public
utility commissions are associated with higher shares of loop-based entry and lower
shares of platform-based entry.43 Conversely, Republican governors are associated with
less loop-based entry.
To obtain a sense of the relative importance of the explanatory variables, Table 3-4
details the effect of a one-standard deviation increase in each explanatory variable on the
share of loop and platform-based entry.44
43 Conceivably, the Republication state commission variable could simply reflect that Republican
commissions set lower loop rates and higher platform rates. However, the very low correlation between
the UNE rates and commission political affiliation (see Table 3-2) mitigates this concern.
44 For binary explanatory variables, the economic effect is calculated as the effect from a change in the
The largest effect in either equation is associated with average revenue in the
platform share equation. A one-standard deviation increase in the average revenue
($9.60) is associated with an increase of over 2 percentage points in the share of the
incumbent's lines leased in a platform arrangement, which translates to a roughly 33%
increase in the platform share when measured at the mean of 6.1%.
Of the cost variables, a one-standard deviation increase in the loop connection
charge ($25.80) decreases the level of entry by almost 0.2 percentage points. Measured at
the mean of 2.6%, this effect translates to an approximately 6% decrease in the loop
share, which validates to some extent the FCC's concern regarding the dampening effect
of this charge on loop-based entry.45 The economic effect of the monthly loop cost on
platform-based entry is non-trivial, which suggests that for some customers entrants
utilize a platform strategy when the monthly loop cost increases. Surprisingly, the own-
cost economic effects are not only statistically insignificant but also relatively small.
The political affiliation of the state utility commissions has divergent effects on the
two means of entry. When adjusted for the difference in means, the absolute value of the
economic effects is roughly equal for loop and platform-based entry. As mentioned
above, this may reflect a preference on the part of Republican commissioners to promote
the entry option that entails less regulatory intervention. The negative effect of
Republican governors on loop-based entry is interesting, given the estimated preference
of Republican commissioners and the absence of an effect on platform-based entry. This
effect may reflect a concern on the part of entrants that Republican governors discourage
explanatory variable from zero to one.
45 Note that this effect does not capture the effects of logistic difficulties entrants may face when
attempting to connect a customer to its network via a loop arrangement.
regulation and are thus more likely to advocate policies that are favorable to incumbents.
The lack of effect on platform-based entry may reflect the low fixed costs that platform-
based entry requires, and the resulting ability of an entrant to exit a market relatively
Estimates with Population Interactions
As described above, the estimates in Table 3-3 and Table 3-4 and estimates in prior
studies assume that the effects of the explanatory variables do not vary across states.
However, the effects of market and political factors may well differ by the size of the
state. To account for such potential variation, the model is also estimated such that each
explanatory variable is interacted with the state population.
Table 3-5 details the coefficient estimates and t-statistics that result from including
population interactions. For each explanatory variable, two coefficients are reported: the
coefficient on that variable and the coefficient of the variable multiplied by the state
population. As such, one can interpret the second coefficient as the additional effect of
the explanatory variable as the population is increased.
For the variables that measure revenue potential, the strong effect of the average
revenue on the platform share persists when population interactions are included.
However, whereas previously average revenue did not have a statistically significant
effect in the loop share equation, the average revenue interacted with the population does.
Further, the coefficient on the uninteracted average revenue variable is larger than the
estimate in Table 3-3 and falls just short of a statistical significance level of 90%.
Market size effects also appear to be important in measuring the effects of costs on
entry share. For the loop connection charge in the loop share equation, the negative
coefficient on the uninteracted variable and the positive coefficient on the interacted
variable suggest that the connection charge has a negative effect on entry in smaller
markets, but the effect diminishes in larger markets. Conversely, the negative effect of
the monthly incremental platform cost on platform-based entry tends to increase as the
population increases. The effects of the own monthly costs also appear to differ by
Of the political variables, in contrast to the results in Table 3-3 and Table 3-4, it
appears that there may be an effect on platform-based entry associated with Republican
governors when population effects are taken into account. Specifically, in small states
Republican governors are associated with greater loop-based entry, but the effect
diminishes as the state's population increases.
To estimate the economic effects implied by the estimates in Table 3-5, Table 3-6
calculates the effect on entry share of a one-standard deviation in each explanatory
variable for two representative population sizes: the 25th percentile and the 75th percentile
(2.2 and 7.4 million, respectively).46 The first two columns detail the economic effects on
the loop share while the last two columns detail the economic effects on the platform
The economic effects on the loop share differ significantly by population. None of
the statistically significant explanatory variables for the 25th percentile of population has
a statistically significant effect for the 75th percentile of population. Specifically, the
effects of the loop connection charge and the political affiliation of the utility commission
46 The economic effect is calculated as the standard deviation of the explanatory variable times the sum of
the coefficient on the uninteracted variable plus the population times the coefficient on the interacted
and governor become statistically insignificant as the size of the state increases. Further,
the effect of the monthly loop cost is nearly statistically significant at the 25th percentile
but not at the 75th percentile.
The effects on loop share may be diminished in larger states because loop-based
entry is less responsive to changes in revenue and costs in larger markets. Loop-based
entrants may already be present in larger markets because they contain the most potential
customers, and thus even a small profit margin per customer translates to large profits in
that market. Further, incumbent prices to business customers have historically been well
above cost in order to cross-subsidize lower prices to residential customers. Loop-based
entrants may have previously entered these larger markets to undercut the inflated
incumbent business prices and may therefore be less affected by changes in revenue and
The differential effect of Republican commissioners on loop share in smaller states
may also be due in part to differences in entry conditions. In large markets loop-based
entry may be sufficiently profitable that substantial entry will occur regardless of state
regulatory policy. In contrast, given the relative difficulty of attracting loop-based entry
in small states, Republican state commissions may adopt policies that are especially
favorable to loop-based entry and thus the commissions have a larger effect on the level
However, the effects of the explanatory variables on platform share offer more of a
mixed picture. The positive effect of the loop monthly cost is quite similar in the two
market sizes, but the effects of the other explanatory variable differ. In contrast to the
effects on loop share, the economic effects on platform share tend to become more
pronounced as the population increases. For instance, the effect of the average revenue
per line is over 20% larger at the 75th population percentile than at the 25th percentile.
Also, the monthly incremental platform cost has a statistically significant effect in larger
states whereas in smaller states it does not.
These findings may reflect in part how platform-based entrants attract customers.
Platform-based entrants tend to target residential customers in broad geographic markets
(such as an MSA or state) and use mass marketing to acquire customers. In smaller states
like Wyoming, an incremental increase in per customer profit may not justify the fixed
marketing costs of entering the state due to the limited number of potential customers.
Conversely, when per customer profit levels increase in a large state such as California,
the same increase in marketing efforts can reach a much larger population and thus may
justify the additional expense.
The negative effect of Republican commissioners on platform-based entry may also
be due to a difference in the ability of small and large states to attract entry. Republican
commissions may have a preference for loop-based entry over platform-based entry, but
in small states it may be difficult to attract loop-based entry. Thus, the desire to have any
form of entry may offset whatever preferences the commissioners possess. In larger
states, commissioners can perhaps be more selective as to which type of entry to
encourage, thus Republican commissioners may discourage platform entry.
The economic effect of the loop connection charge on platform-based entry also
differs by market size. The positive effect in small states can be explained by substitution
from loop-based to platform-based entry as the cost of loop-based entry increases.
However, the negative effect in large states implies that increases in the loop connection
charge deter both types of entry. A possible explanation is that some platform-based
entrants in larger states enter those markets with the intent to in the long-term to convert
to a loop-based arrangement. When the cost of loop-based entry increases, some of these
entrants may decide that it is less profitable to pursue such a strategy and they may elect
to curtail their platform-based operations.
This paper analyzes the factors that determine the level of entry in local
telecommunications markets given two alternative entry strategies. The estimates suggest
that while generally the two types of entry are affected by different market factors, there
appears to be cost-based substitution between them. Changes in the own monthly costs of
the two forms of entry have limited effects, but loop-based entry decreases in response to
increases in the connection charges entrants must pay incumbents when a customer is
acquired. Loop-based entry tends to be more pronounced relative to platform-based entry
as the degree of Republican representation on state public utility commissions increases.
Finally, loop-based entry is more responsive to changes in economic conditions in
smaller states, while platform-based entry is more responsive to market conditions in
The results in this paper may suggest some potential lessons regarding the effects
of TA96 and potential revisions to it. Proponents of platform-based entry can point to the
negative effect of loop connection charges as evidence of the need for an alternative to
loop-based entry. On the other hand, critics of platform-based entry can argue that, given
the positive effect of the monthly loop cost on the platform share, loop-based entry is
hindered by the existence of the platform-based entry option. Also, the results suggest
that the local interests of state regulators need to be taken into account when they are
charged with implementing federal policies. Finally, given the differing effects across
states of different sizes, policymakers need to consider how to fashion policies that
achieve national goals but recognize local market conditions.
The findings in this paper could be extended in several directions. First, if the data
become available, a more disaggregated analysis of the entry decision could provide
more precise results. Market conditions can vary greatly within a state, but the available
data do not allow for an analysis of that granularity. Also, as noted above, the monthly
loop cost variable used in the regressions does not include collocation costs, while the
estimate of the incremental cost of platform-based service does not include costs related
to transporting calls between switches. Including these additional costs would allow for a
more complete analysis. Additionally, a more precise measure of the retail price may
uncover a more important role for it in determining the level of entry. Finally, a complete
analysis would simultaneously estimate the effects of other types of telecommunications
entry, such as cellular phones and the emerging presence of new technologies such as
Jan02 Jul02 Jan03 Jul03 Jan04
Figure 3-1 Fraction of Incumbent Lines Leased by Entrants (National Average).
Jan02 Jul02 Jan03 Jul03 Jan04
Figure 3-2 Fraction of Incumbent Lines Leased by Entrants by Means of Entry (National
0 5 10 15
Share in July 2001 (Percent)
Figure 3-3 Loop Share of Incumbent Lines, July 2001 and July 2004.
0 5 10 15 20 25 30
Share in July 2001 (Percent)
Figure 3-4 Platform Share of Incumbent Lines, July 2001 and July 2004.
Table 3-1 Summary Statistics.
Variable Mean Minimum Maximum Deviation
Share of Entrant Loop Lines 2.6 0.04 11.9 2.0
Share of Entrant Platform Lines 6.1 0.00 24.4 5.2
Incumbent Average Revenue per Line $61.12 $45.16 $101.19 $9.60
Change in Unemployment Rate 0.13 -1.3 2.2 0.61
Loop Monthly Cost $13.90 $4.77 $28.14 $3.80
Loop Connection Charge $45.69 $3.33 $159.76 $25.80
Platform Incremental Monthly Cost $7.25 -$0.17 $24.81 $4.70
Percent of Commission Republican 53.2 0.0 100.0 33.3
Governor Republican 0.56 0.00 1.00 0.5
Population (millions) 6.3 0.5 35.9 6.5
Table 3-2 Correlation Matrix of Explanatory Variables.
Unemploy Loop Mthly Loop Con Plat Incr % PUC
Avg Rev Rate Cost Chrg Mthly Cost Rep Gov Rep
Avg Rev 1.00
Unemploy Rate -0.14 1.00
0.25 0.10 1.00
Loop Mthly Cost 0.25 0.10 1.00
Loop Con Chrg 0.16 0.14 0.37 1.00
Plat Incr Mthly 0.31 0.08 0.25 0.15 1.00
% PUC Rep 0.23 0.07 0.05 0.02 -0.01 1.00
Gov Rep 0.13 0.05 0.07 -0.04 0.04 0.41 1.00
Table 3-3 Coefficient Estimates without Population Interactions.
Specification #1 Specification #2 Specification #3 Specification #4
Explanatory Variable Loop Platform Loop Platform Loop Platform Loop Platform
Average Revenue per 0.010 0.185 *** 0.001 0.217 ***
Line (0.48) (3.05) (0.04) (3.58)
Change in 0.115 -0.051 0.164 -0.157
Unemployment Rate (1.10) (-0.17) (1.56) (-0.54)
p M C -0.032 0.194 *** -0.033 0.186 ***
Loop Monthly Cost
(-1.32) (2.87) (-1.37) (2.79)
Loop Connection -0.006 ** 0.002 -0.007 *** 0.008
Charge (-2.10) (0.31) (-2.72) (1.04)
Platform Incremental 0.010 -0.023 0.014 -0.028
Monthly Cost (0.51) (-0.44) (0.76) (-0.53)
Percent State Utility 0.009 ** -0.026 ** 0.009 ** -0.026 **
Commission (2.24) (-2.36) (2.34) (-2.40)
Governor Republican -0.225 0.001 -0.306 ** 0.004
(Binary) (-1.72) (0.00) (-2.33) (0.01)
Within R-Squared 0.34 0.71 0.35 0.71 0.35 0.70 0.37 0.72
Number of Observations 314 314 314 314 314 314 314 314
Notes t-statistics are reported in parentheses, coefficient estimates for state and time fixed effects and constant term omitted for
brevity, *** statistically significant at 99% confidence level, ** statistically significant at 95% confidence level, statistically
significant at 90% confidence level
Table 3-4 Economic Effects without Population Interactions.
Explanatory Variable Loop Platform
Average Revenue per Line
Change in Unemployment
Loop Monthly Cost
Loop Connection Charge
Percent State Utility 0.30 ** -0.87 **
Commission Republican (2.34) (-2.40)
-0.31 ** 0.00
Governor Republican (Binary .
Notes Economic effects for continuous explanatory variables are based
on a one standard deviation increase in that variable, economic effects for
binary explanatory variables are based on a change in the variable from
zero to one. t-statistics are reported in parentheses, *** statistically
significant at 99% confidence level, ** statistically significant at 95%
confidence level, statistically significant at 90% confidence level
Table 3-5 Coefficient Estimates with Population Interactions.
Explanatory Vanable Loop Platform
Average Revenue per Line
Change in Unemployment
Loop Monthly Cost
Loop Connection Charge
Percent State Utility 0.011 ** -0.010
Commission Republican (2.03) (-0.63)
Interacted with -0.001 -0.010
Population (-1.25) (-0.75)
-0.342 0.895 *
Governor Republican (Binary .7
Interacted with 0.019 -0.176 ***
Population (0.99) (-3.32)
Within R-Squared 0.44 0.76
Number of Observations 314 314
Notes t-statistics are reported in parentheses, coefficient estimates for
state fixed effects and constant term omitted for brevity,
*** statistically significant at 99% confidence level, ** statistically
significant at 95% confidence level, statistically significant at 90%
Table 3-6 Economic Effects for the 25th and 75th Population Percentiles.
25th 75th 25th 75th
Explanatory Variable Percentile Percentile Percentile Percentile
0.13 -0.36 2.47 *** 3.20 ***
Average Revenue per Line
(0.65) (-1.65) (4.24) (5.25)
Change in Unemployment 0.03 0.07 0.10 -0.05
Rate (0.52) (1.02) (0.57) (-0.25)
p M C -0.17 -0.05 0.72 ** 0.75 **
Loop Monthly Cost
(-1.62) (-0.42) (2.44) (2.55)
-0.28 *** 0.08 0.52 ** -0.59 **
Loop Connection Charge
(-3.47) (0.83) (2.31) (-2.29)
Platform Incremental 0.13 0.04 0.14 -0.67 **
Monthly Cost (1.28) (0.33) (0.49) (-2.04)
Percent State Utility 0.33 ** 0.20 -0.40 -0.63
Commission Republican (2.06) (1.54) (-0.95) (-1.79)
Governor Republican -0.30 -0.20 0.51 -0.41
(Binary) (-1.92) (-1.56) (1.19) (-1.15)
Notes t-statistics are reported in parentheses, coefficient estimates for state fixed effects and
constant term omitted for brevity, statistically significant at 99% confidence level,
** statistically significant at 95% confidence level, statistically significant at 90%
DOES THE FORM OF DOCTOR COMPENSATION AFFECT THE QUALITY OF
CARE IN MEDICAID HMOS?1
The Medicaid program, already one of the largest social programs in the United
States, is growing rapidly. Between 1996 and 2004, the number of Medicaid enrollees
increased by roughly one-third, from 33 million to 44 million (U.S. HHS (2004), 3). In an
attempt to control the cost of the Medicaid program and to improve the quality of care
provided to enrollees, many states have moved enrollees into HMOs (also known as
managed care organizations). During the same 1996 to 2004 time period, the fraction of
Medicaid enrollees in managed care organizations increased by approximately one-half,
from 40% to over 60% (U.S. HHS (2004), 3). The trend towards managed care does not
appear to be slowing. For instance, in December 2005 the governor of Florida signed a
bill that requires all of the state's Medicaid enrollees eventually be enrolled in HMOs
State Medicaid programs often contract with HMOs to care for the health care
needs of Medicaid participants. The HMOs, in turn, contract with doctors and other
health care providers to deliver necessary health care services. HMOs can differ in many
ways, including the manner in which they pay their doctors. FFS and capitated payment
arrangements are common. Under a FFS arrangement, the doctor is paid according to the
1 This paper is co-authored with Betsy Shenkman (Department of Epidemiology, Health Policy Research
and Pediatrics, University of Florida) and David Sappington (Department of Economics, University of
services she provides to an enrollee. Under a capitated payment arrangement, the doctor
is paid a fixed amount per enrollee regardless of the health care services actually
provided to the enrollee. Consequently, a doctor paid via FFS can increase her revenue by
providing additional services. In contrast, the revenue a capitated doctor receives is not
affected by the health care services she provides.
This paper investigates whether the means by which HMOs compensate their
doctors influence the quality of care the HMO's enrollees receive. Using data for all of
the Medicaid HMO enrollees in a large state,2 we find that enrollees in HMOs that pay
their doctors exclusively via FFS arrangements are more likely to receive services for
which the HMO's doctors receive additional compensation. Further, these enrollees are
less likely to receive services for which the HMO's doctors do not receive additional
compensation. These findings suggest that financial incentives may influence the
behavior of doctors in Medicaid HMOs, and thus the health care received by Medicaid
participants enrolled in HMOs.
Numerous studies have analyzed whether the form of physician pay influences the
level of care provided to HMO enrollees. The papers suggest such a link often is present.
For example, Hillman, Pauly, and Kerstein (1989) report that hospitalization rates are
higher for enrollees whose physicians are paid via FFS rather than capitation. Steams,
Wolfe, and Kindig (1992) find that specialist referrals, hospital admissions, and hospital
length of stays fell when an HMO switched from FFS physician payment to capitation
payment. Ransom et al. (1996) report that gynecologists tend to provide fewer elective
surgical procedures when their payment method is changed from FFS to capitation.
2 The state is not identified, to preserve confidentiality of key data.
Shrank et al. (2005) find that fewer cataract procedures are performed when physicians
are moved to capitation. In contrast, Conrad et al. (1998) do not find any significant
effects of physician payment method on health care utilization.
Our analysis enhances the literature in three important respects. First, we examine
differences in the behavior of FFS and capitated doctors for services that have different
effects on the doctors' expected profits. As noted, we find that FFS doctors deliver more
services that increase their profit than capitated doctors. However, corresponding
differences are not detected on services that do not increase physician profit. Second, we
consider services for which there are clearly stated and widely accepted norms for the
proper level of care. Consequently, we are able to assess whether financial incentives
affect the extent to which actual care departs from the most appropriate level of care.
Third, we examine the health care services delivered to Medicaid enrollees, who often are
especially at risk of not obtaining the proper level of care.
The paper is organized as follows. Section 2 describes the services we analyze and
explains how the profits of FFS and capitated doctors are affected by the delivery of these
services. Section 3 describes our data and empirical specification. Section 4 presents our
findings. Section 5 provides conclusions and directions for future research.
Preventive care is a major focus of the Medicaid program, especially for children.
Preventative care can both reduce treatment costs and avoid painful and debilitating
illness. Routine well-child visits and the provision of asthma medications are two
important forms of preventive care. Annual well-child visits allow doctors to monitor
enrollees' health and to deliver essential, routine health care services such as
immunization shots. Asthma is considered to be an epidemic and affects over 4 million
children in the U.S. (US HHS (2000), 2). Proper treatment of asthma conditions can
reduce asthma attacks, emergency room visits, and morbidity.
While both well-child visits and asthma medications can be beneficial to enrollees,
these two forms of preventive care can affect the profits of doctors differently. In the state
studied here, the cost of a well-child visit is borne by the doctor that performs the well-
child. In contrast, the Medicaid program pays for prescribed asthma medications.
Consequently, the amount of asthma medication prescribed has no direct financial impact
on either the HMO or its doctors. This difference in the incidence of the costs of well-
child visits and asthma medications is an important element of the ensuing analysis.
Because they are reimbursed for each service they provide, FFS doctors increase
their revenue with every well-child visit they perform. In contrast, FFS doctors do not
receive greater revenue when they prescribe additional asthma medications. In fact, to the
extent that the prescribed asthma medications control the symptoms of asthma and
thereby reduce future office visits, additional prescriptions can reduce the doctor's
Capitated doctors receive no additional revenue when they perform a well-child
visit, but do incur the cost of providing the visit. The well-child visit may reduce future
costs by allowing the doctor to detect an ailment and treat the enrollee before
complications arise and the requisite care becomes more costly. However, Medicaid
enrollees often have limited spells in the program, which reduces the likelihood that the
capitated doctor would bear the costs of later treatment. As noted above, capitated
doctors do not bear the costs of prescribed asthma medications. Furthermore, the
medications can reduce the costs of the capitated doctor by limiting the need for office
This paper compares the extent to which FFS and capitated doctors in Medicaid
HMOs provide well-child visits and asthma medications. Given the differences in the
compensation structures the FFS and capitated doctors face, our findings may provide
some useful evidence about how financial incentives affect the quality of care received
by HMO enrollees.
Data and Empirical Specification
Our data is of two types: enrollment and encounter data and data from interviews.
The enrollment and encounter data is for every healthy3 Medicaid enrollee in an HMO in
the state in question in 2004. This enrollment data contains demographic information for
each enrollee; the encounter data records the enrollee's usage of medical services,
including office visits, medical treatments, and pharmaceutical prescriptions. The
encounter data also documents diagnoses made by the physician when treating the
The interview data is derived from interviews with personnel from the HMOs. The
HMOs in our sample are required to answer questions posed by the state regarding
various characteristics of their organization. Among the questions asked is how the HMO
compensates its doctors.4
The dependent variables in the analysis are based on measures established in the
Health Plan Employer Data and Information Set (HEDIS) developed by the National
3 Specifically, the sample is limited to enrollees with a clinical risk grouping (CRG) score of
4 Unfortunately, data for each HMO are aggregate data, not data on how the HMO compensates
each individual provider.
Committee for Quality Assurance (NCQA) (NCQA, 2002). HEDIS measures are used to
evaluate the care received by HMO enrollees and are widely used by industry
participants. These measures are based on diagnosis codes and treatment codes found in
enrollee encounter data and the ages of the enrollees.5 The measures are binary: one
indicates that the proper care was provided; zero indicates otherwise. The measures are
based on treatment received over a twelve month period and specify the age ranges of the
enrollees to be included. The treatment period examined in the analysis below is January
2004 December 2004.
The first HEDIS measure analyzed is whether the enrollee received at least one
well-child visit during the treatment period. Two age cohorts are analyzed: children
between 3 and 6 years of age; and adolescents between the ages of 12 and 21.6 The
success rates for the younger and older age cohorts are 50% and 32%, respectively, in our
The second HEDIS measure we analyze is whether children with persistent asthma
were prescribed appropriate medications.7 This measure is based on two years of data.
Data from 2003 are examined for evidence of persistent asthma. Data for 2004 are
examined for evidence of appropriate medication. The age cohorts employed for this
analysis are 5 through 9 and 10 through 18. The success rates for these cohorts on this
measure in our sample are 51% and 54%, respectively.
5 The measures used in this paper are based on the administrative specification of the measures.
6 The two measures are named, "Well-Child Visits in the Third, Fourth, Fifth, and Sixth Years
of Life" (page 177) and "Adolescent Well-Care Visits" (page 180).
7 The measure is named "Use of Appropriate Medications for People with Asthma" (page 104).
The explanatory variables reflect HMO operating characteristics and enrollee
demographics. The HMO operating characteristic of primary interest is whether the HMO
compensates all of its doctors via FFS arrangements. During the HMO interviews, each
HMO is asked to report the percent of its doctors that are paid via FFS arrangements.
The distribution of answers was bimodal, with five of the eight HMOs in our sample
stating that they pay roughly 100% of their doctors via FFS8 and three HMOs stating that
they pay 85%9 of their doctors via FFS. This bi-modal distribution underlies our
treatment of this variable as binary.
Two other variables are included to control for the practices of the HMO: whether
HMO case managers work directly with the primary care physicians10 and whether the
HMO makes reminder calls to enrollees immediately prior to their well-child visits. (This
latter variable is included only in the analyses of the HEDIS well-child visit measure.)
The other variables included to control for characteristics of the HMO are the for-
profit/non-profit status of the HMO, the number of enrollees (Medicaid and otherwise),
the percent of the HMO's enrollees that are in Medicaid, and the number of years the
HMO has been operating in the state in any capacity.
Table 4-1 lists the HMO attributes included in the estimation. As the table
indicates, there is significant heterogeneity among the HMOs for each of the variables
employed in the analysis. The smallest HMO has roughly 35,000 enrollees while the
largest has approximately nine times that number. The HMOs also vary significantly in
8 Specifically, the reported percentages were 100%, 100%, 100%, 99%, and 99%.
9 Specifically, the reported percentages were 84%, 85%, and 85%.
10 HMO case managers are responsible for ensuring that children with chronic conditions receive
their operating practices and in the extent to which their enrollment is limited to Medicaid
Table 4-2 presents the success rate for the two HEDIS measures for each value of
the binary HMO attributes. Children in HMOs that paid all doctors via FFS arrangements
had higher well-child visit success rates than HMOs that paid only some of their doctors
via FFS. In addition, children in HMOs where case managers worked directly with
primary care physicians had higher asthma medication success rates.
The demographic variables we employed are gender, race, age, and whether the
enrollee resides in a rural area. Table 4-3 presents the values of the demographic
variables in our sample. Hispanics outnumber blacks and whites, while the vast majority
of enrollees reside in non-rural areas. The data in Table 4-4 indicate that, relative to
whites and blacks, Hispanic children had a higher success rate for well-child visits and
lower success rate for asthma medications. Also, non-rural children had higher success
rates for well-child visits and lower success rates for asthma medications.
The two equations estimated are:
(1) WCHILD,, = a + ,X,+ /2Z, + E,
(2) ASTHMA,, = 3 + yX, + y,2Z + ,J
WCHILD, is the HEDIS well-child measure for enrollee i in HMOj
ASTHMA, is the HEDIS asthma medication measure for enrollee i in HMOj
X, are enrollee demographic variables
Z are HMO attribute variables
Although the dependent variables are binary, the large number of observations in
our sample ensures that regression via ordinary least squares is consistent.11 To account
for unobserved HMO-level effects (see Moulton (1990)), the observations are clustered
by HMO and location.12
Table 4-5 presents the regression estimates. The first two columns in Table 4-5
contain the estimates for the well-child visit measure for the two age cohorts. The last
two columns contain the estimates for the asthma medication measure for the two age
The first row of data in Table 4-5 presents the coefficient estimates for the variable
that indicates whether the HMO pays all of its doctors via a FFS arrangement. For both
age cohorts, the well-child visit success rate for enrollees in HMOs that pay all of their
doctors via FFS is six percentage points higher than for those enrolled in HMOs that pay
some of their doctors via capitation. Given the mean success rates, this difference implies
that the average probability that an enrollee receives a well-child visit is 10-20% higher
in an HMO that pays all of its doctors via FFS.
The opposite conclusion arises with regard to the asthma medication measure. For
both age cohorts, the success rate is lower for HMOs that pay all of their doctors via FFS.
The effect is statistically significant for the 5-9 year old cohort. The estimates imply that
1 As a specification test, the model was also estimated via probit. The results are largely
unchanged and are reported in Appendix Table B-1.
12 By clustering the observations, the estimates of the coefficient standard errors are adjusted to allow for
the possibility that the observations within each group are not independent. The enrollees are grouped
here by the HMO in which they are enrolled and the metropolitan area in which they reside.
the probability that the recommended asthma medications are prescribed is approximately
5% lower in HMOs that pay all of their doctors via FFS than in other HMOs.13
These findings suggest that financial incentives may affect the services doctors
provide to Medicaid enrollees. In particular, the services that increase the revenue of FFS
doctors (well-child visits) are provided more frequently in HMOs where all doctors are
compensated via FFS. Furthermore, the services that could reduce the future revenues of
FFS doctors (asthma medication prescriptions) are provided less frequently in HMOs
where all of the doctors are paid via FFS.
The other variables that measure HMO attributes generally are not statistically
significant in our regressions. A larger number of enrollees is associated with a higher
success rate on the well-child visit measure. However, the effect is of limited statistical
and economic significance. The probability that an enrollee in the 5-9 year cohort
receives the recommended asthma medications is higher in HMOs where the case
manager works directly with the primary care physician. This effect may be due to case
managers working with children with severe asthma to ensure that they receive the
appropriate medications. The finding that the probability that an enrollee in the 10-18
year cohort receives the recommended asthma medication declines as the percentage of
HMO enrollees in Medicaid increases may reflect practice style effects in HMOs that
serve both commercial and Medicaid populations. HMOs may tend to provide relatively
high service quality to commercial populations in an effort to retain these profitable
13 To control for the possibility that the effects are being driven by spurious interaction between
the variables measuring HMO attributes, the model was also estimated where the other HMO
attributes were replaced by HMO dummy variables. The results, reported in Appendix Table
B-2, are largely unaffected. (The HMO attributes and HMO dummy variables cannot be
included simultaneously due to perfect multicollinearity among the two sets of variables.)
clients. To the extent that Medicaid and commercial enrollees receive the same basic
health care services within an HMO, HMOs with a larger concentration of commercial
enrollees may provide higher quality care to their Medicaid enrollees.
Hispanic children in our sample have a higher likelihood of receiving an annual
well-child visit and a lower likelihood of receiving asthma medications than black
children and white children. This result is interesting in light of the "Hispanic paradox"
that suggests Hispanics tend to have better health outcomes than non-Hispanics of similar
socio-economic status (Franzini, Ribble, and Keddie (2001)). Within age cohorts,
younger children generally are more likely than their older counterparts to receive well-
child visits and asthma medications. (The exception is for asthma medications for the 5-9
year cohort.) The finding that 4-year-olds receive well-child visits with relatively high
frequency likely reflects the fact that parents often take their children to the doctor to
obtain the immunizations required to enter kindergarten. Rural residence is associated
with reduced (but statistically insignificant) performance on the well-child measure and
increased (and statistically significant) performance on the asthma medication measure
for the younger cohort. Families that live in rural regions likely have to travel farther for
well-child visits, which may reduce the likelihood of such visits. However, because they
tend to live farther from the doctor's office or the emergency room, parents of rural
families may take particular precautions to be sure their asthmatic children do not
develop serious conditions that would require long trips to receive immediate care.
We have examined whether the form of doctor compensation affects the quality of
care received by Medicaid HMO enrollees. Our findings suggest that financial incentives
may influence the services that doctors deliver to enrollees.
Further research is required to determine whether our findings persist in other
settings. The HEDIS measures we employed require that an enrollee be a member of the
HMO for almost the entire period in question. 14 Therefore, our findings pertain only to
enrollees with relatively stable enrollment. These enrollees may not be entirely
representative, as many Medicaid enrollees move in and out of Medicaid frequently.
It would be ideal to be able to identify exactly which doctors are paid via FFS and
which are paid via capitation. This information would permit more precise measurement
of the effects of doctor compensation arrangements on the quality of care they provide.
Finally, time series data for each enrollee would allow for the inclusion of enrollee
fixed effects. Such effects would control for time-invariant, unobservable characteristics
of each enrollee, and would thereby improve the precision of the analysis.
14 For the well-child visit measure, the enrollee must be enrolled in the same HMO for 11 of the
12 months in question. For the asthma medication measure, the minimum enrollment in the
same HMO is 22 of the previous 24 months.
Table 4-1 HMO Attributes.
Attribute HMO#1 HMO#2 HMO#3 HMO#4 HMO#5 HMO#6 HMO#7 HMO#8
Providers Paid Only via FFS no no yes yes yes yes no yes
Markets Served 4 1 4 1 1 1 3 1
Total Enrollees 292,091 97,606 36,311 34,016 115,230 70,503 116,853 100,710
Years in State 6 8 5 4 17 5 6 7
Fraction of Enrollees in Medicaid 76% 37% 100% 46% 10% 100% 80% 24%
For-Profit yes no no no yes no yes no
Check-Up Visit Reminder Calls yes no no no no yes no yes
Case Manager Works with PCP no yes no yes no yes no no
Table 4-2 HEDIS Success Rates by HMO Attribute.
Check-Up Visits Asthma Medications
Ages 3-6 Ages 12-21 Ages 5-9 Ages 10-18
Total 50% 32% 51% 54%
Are All Providers Paid on a Fee-For-Service Basis?
Yes 51% 35% 50% 54%
No 48% 30% 52% 53%
Does HMO Place Reminder Calls?
Yes 51% 33% n/a n/a
No 47% 31% n/a n/a
Does Case Manager Work with Primary Care Physician?
Yes 50% 33% 55% 57%
No 48% 32% 50% 53%
Is HMO a For-Profit HMO?
Yes 51% 32% 50% 52%
No 48% 32% 52% 57%
Table 4-3 Demographic Characteristics By Population.
Check-Up Visits Asthma Medications
Ages 3-6 Ages 12-21 Ages 5-9 Ages 10-18
Total 82,227 62,475 1,971 1,756
Table 4-4 HEDIS Success Rates by Population.
Ages 3-6 Ages 12-21
Total 50% 32%
Ages 5-9 Ages 10-18
Table 4-5 OLS Regression Estimates.
HMO Operating Characteristics
Dep. Var.: Check-Up Visit
Aged 3-6 Aaed 12-21
Dep. Var.: Asthma Medication
Aged 5-9 Aaed 10-18
Providers Paid Only via FFS
Total Enrollees (Medicaid &
Commercial, hundred thousands)
Percent of Enrollees in Medicaid
Check-Up Visit Reminder Calls
Case Manager works with PCP
Years Operating in State
Age Dummy 1
Age Dummy 2
Age Dummy 3
Age Dummy 4
Age Dummy 5
The t-statistics reported in parentheses are based on Huber-White robust standard errors clustered at the
The age dummy variables differ across specifications:
In the check-up visit regressions, age dummy 1 corresponds 3 years, age dummy 2 corresponds to 4
years, age dummy 3 corresponds to 5 years, age dummy 4 corresponds to 12-15 years, and age dummy 5
corresponds to 16-18 years.
In the asthma medications regressions, age dummy 1 corresponds to 5-7 years and age dummy 2
corresponds to 10-14 years.
*** 99% confidence level, ** 95% confidence level, 90% confidence level
This dissertation analyzed government attempts at using market forces to achieve
policy outcomes in two important areas, telecommunications and health care. The
chapters analyzed how state governments implemented TA96 and the effects of market-
oriented polices in the telecommunications and health care sector.
A number of significant results emerge. First, the manner in which the policies are
executed may be inconsistent with the goals of the policy. Our analysis of TA96 suggests
that state public utility commissions may have been influenced by decisions in larger
states and by factors outside of the policy's guidelines. Second, firms may respond to
incentives in ways that policymakers may not foresee. In the investigation of local
telecommunications markets, it was shown that responses by entrants to market factors
may vary by the size of the market. Thus, national policies may not be well-suited to
heterogeneous markets. Third, the details of how firms operate may have substantial
effects on the level of quality they provide. Our results demonstrate that the way in which
physicians in Medicaid HMOs are paid may influence the level of care that they provide
DATA NOTES AND ADDITIONAL RESULTS
This appendix describes some of the details of the data and specification tests used
in the analysis in Chapter 2.
For 18 of the 919 observations de-averaged rates were reported and a statewide
average was neither reported nor could be calculated based on the available data. In those
instances a simple average of the de-averaged rates was used as the statewide average
In December 1997 the Texas PUC set both a statewide rate that was effective
immediately and de-averaged rates that were to take effect a month later. In this analysis
the initial statewide rate is ignored.
The former Pacific Telesis region includes only two states, California (the leader)
and Nevada. Thus, Nevada is the only follower state and the rates of other follower states
in the region do not exist. To keep Nevada in the sample and given that SBC now
controls the former Pacific Telesis and Ameritech regions, the rates of the follower states
in the former Ameritech region are used as a proxy for the other follower rates for
The HCPM data were received via email from the FCC.
Both the embedded and HCPM cost estimates are reported annually. Quarterly
values are obtained via linear interpolation of the data.
For each year except 2000, the HCPM cost estimate is provided only for the cost of
the entire line, which includes, in addition to the loop cost, the cost for line port, EO
usage, signaling, transport, billing/bill inquiries, and directory listing. To obtain an
estimate of the loop cost for those years, the fraction of the line cost that is attributable to
the loop in 2000 is applied to the total line costs in the other years.
Data regarding PUC commissioners were derived from Profiles ofRegulatory
Agencies of the United States & Canada: Yearbook 1995-1996 (NARUC) and NARUC
membership directories (specifically, directories dated January 1998, February 1999,
February 2002, February 2003, July 2003, and March 2004).
Besides being reported as either a Democrat or Republican, a commissioner could
also be listed as independent or have no reported political affiliation. For the purposes of
this analysis, those commissioners who were reported as independent of for whom a
political affiliation was not reported are equally Democrat and Republican. (For example,
if a state's PUC is composed entirely of independents and/or commissioners for whom
their political affiliation is not reported, the value of the variable percent of
commissioners that are Republican for that PUC would be 0.5.)
The type of retail rate regulation employed in each state is derived from reports in
the State Telephone Regulation Report (1/25/96, 2/8/96, 3/20/97, 4/3/97, 4/3/98, 4/17/98,
8/20/99, 9/3/99, 9/29/00, 10/13/00, 10/27/00, 2/15/02, 3/1/02, 3/15/02, 5/9/03, 5/23/03,
6/6/03, 7/30/04, 8/13/04, and 8/27/04). For some of the descriptions of the regulatory
plans, only a year was given for the beginning or the end of the plan's duration. In those
instances, the exact dates were inferred from the prior or succeeding plan.
The three residential customer satisfaction variables used as instruments for the
form of retail rate regulation come from the annual FCC ARMIS Report 43-06.
Specifically, they are the percent of customers surveyed that are dissatisfied with the
RBOC's installation, repair, and billing services. Quarterly values are obtained via linear
interpolation of the data.
Estimates of retail rates are somewhat problematic as retail rates can vary across
customers and regions within a state. The proxy used in this analysis is the RBOC's local
network services revenue (Row 520 from ARMIS report 43-03) divided by the number of
switched access lines.
State Political Variables
The gubernatorial data are obtained from the Book of the States (The Council of
State Governments, 1996-1997, 1998-1999, 2001-2002, 2002, 2003) and the CNN.com
web page "2004 Election Results"
(http://www.cnn.com/ELECTION/2004/pages/results/governor/ full.list/). The state
legislature data are obtained from Statistical Abstracts of the United States (U.S. Census
Bureau, 2002, 2003, 2004-2005) and the National Conference of State Legislatures
website (2005 Partisan Composition of State Legislatures,
As Nebraska's legislature is non-partisan, for this analysis the percent of state
legislatures that are Republican is assumed to be 50%.
Section 271 Status
Data regarding RBOC applications to provide long-distance service are obtained
from the FCC web page "RBOC Applications to Provide In-region, InterLATA Services
Under 271" (http://www.fcc.gov/Bureaus/Common_Carrier/in-region_applications/).
Level of Competitive UNE Entry
The data reported are from two series of (non-overlapping) reports of RBOC
survey responses. The 1997 and 1998 data are from voluntary surveys completed by the
RBOCs and are reported in the December 1998 and August 1999 FCC Local Competition
reports produced by the Industry Analysis Division in the Common Carrier Bureau. The
data for 1999 forward are based on RBOC responses to the mandatory Form 477 survey
and are obtained from reports entitled SelectedRBOC Local Telephone Data. Both data
series can be found at http://www.fcc. gov/wcb/iatd/comp.html.
The UNE line count used in this analysis includes both UNE-L and UNE-P lines.
UNE-L lines are where the CLEC leases only the local loop, whereas UNE-P lines
involve the CLEC leasing switching unbundled elements in addition to the local loop.
For some states the first report of the number of UNE lines is made some time (at
most one year) after the CLECs are able to begin leasing the lines. The data for the period
prior to this first report is linearly interpolated by assuming that zero UNE lines were
being leased prior to a rate being set.
Finally, the data are reported on a quarterly, annual, or semi-annual basis
depending on the time period. Quarterly values are obtained when necessary via linear
interpolation of the data.
Tests of Instrument Validity and the Exogeneity of the Average Retail Rate Variable
This section outlines how the validity of the instruments used in the UNE rates
regressions is tested. The first sub-section describes the tests performed to determine if
the instruments are not correlated with the error term, while the second sub-section
details the test used to determine if the instruments are sufficiently correlated with the
endogenous explanatory variables.
The third sub-section explains the econometric test used to confirm that the lagged
average retail rate is exogenous to the determination of UNE rates.
(This section borrows heavily from Baum, Schaffer, and Stillman (2004).)
Tests of Instruments' Orthogonality to the Error Process
When there are more instruments than endogenous variables (i.e., the model is
overidentified), one can test whether all of the instruments are orthogonal to the error
term. In GMM estimation the overidentifying restrictions can be tested with Hansen's J
statistic, which if found to be greater than a threshold value indicates that the instruments
are not exogenous or that they should be included as explanatory (rather than
instrumental) variables in the regression. The J statistics are reported in Table A-2. The J
statistics for the base model suggest that the instrument set as a whole is orthogonal to the
One can also test whether subsets of instruments are orthogonal to the error
process. The test statistic (referred to as the C statistic) is the difference in J statistics
between the specification that includes all the instruments and the specification that
excludes the instruments to be tested. If the C statistics exceeds a threshold value, there is
cause for concern that the tested instruments are not valid. Table A-2 also details the C
statistics for the instrument sets used for each of the three endogenous explanatory
variables. The C-statistics indicate that the instrument sets used for the three endogenous
variables are neither correlated to UNE rates nor endogenous.
Tests of Instrument Relevance
In addition to being orthogonal to the error term, instruments also must be
sufficiently correlated with the endogenous explanatory variables. To test for this, each
endogenous explanatory variable is regressed on all of the exogenous and instrumental
variables in the model. The coefficients on the instruments are then tested for whether
they are jointly equal to zero. To be valid, the coefficients should not jointly equal zero.
This test is complicated by the use of the system GMM estimator. By definition,
the system GMM estimator estimates two equations simultaneously, one in levels and the
other in first differences. The equation in levels is estimated along with the equation in
first differences because estimations in first differences with highly persistent dependent
variables result in weak instruments (Bond, 2002, p 154).
Table A-3 reports the results of the tests of instrument relevance for each of the
three endogenous explanatory variables for the equations in both levels and first
differences. The chi-square values suggest that the instruments easily pass the threshold
test for relevance except for the form of retail rate regulation variable in the first
differences equation. As noted above, this result is not worrisome as it is addressed by the
system GMM estimation.
Test of the Exogeneity of the Lagged Average Retail Rate
One may be concerned that the average retail rate variable is endogenous, even
though it is lagged one year. To test for this the C statistic can be used where the J
statistic from the model assuming the variable is endogenous is subtracted from the J
statistic assuming the variable is exogenous. The results of this test are shown at the
bottom of Table A-2 and confirm that the variable is exogenous.
This section describes additional results from the model.
Table A-4 contains the economic effects two years after a change in an explanatory
variable based on the coefficient estimates the one-year lag specification. Table A-5
details the long-run economic effects based on the coefficient estimates from both the
two-year and one-year lag specifications.