Input and Technology Choices in Regulated Industries: Evidence from the Health Care Sector
Author(s): Daron Acemoglu and Amy Finkelstein
Source: Journal of Political Economy, Vol. 116, No. 5 (October 2008), pp. 837-880
Published by: The University of Chicago Press
Stable URL: http://www.jstor.org/stable/10.1086/595014 .
Accessed: 06/07/2015 05:51
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
.
The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to Journal
of Political Economy.
http://www.jstor.org
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
837
[ Journal of Political Economy, 2008, vol. 116, no. 5]
᭧ 2008 by The University of Chicago. All rights reserved. 0022-3808/2008/11605-0006$10.00
Input and Technology Choices in Regulated
Industries: Evidence from the Health Care Sector
Daron Acemoglu and Amy Finkelstein
Massachusetts Institute of Technology
This paper examines the implications of regulatory change for input
mix and technology choices of regulated industries. We study the
increase in the relative price of labor faced by U.S. hospitals that
resulted from the move from full cost to partial cost reimbursement
under the Medicare Prospective Payment System (PPS) reform. Using
the interaction of hospitals’ pre-PPS Medicare share of patient days
with the introduction of PPS, we document substantial increases in
capital-labor ratios and declines in labor inputs following PPS. Most
interestingly, we find that PPS seems to have encouraged the adoption
of a range of new medical technologies.
I. Introduction
There is broad agreement that differences in technology are essential
for understanding productivity differences across nations, industries,
and firms. Despite this agreement, we know relatively little about the
empirical determinants of technology choices and of adoption of capital
goods embodying new technologies. The lack of empirical knowledge
is even more pronounced when we turn to regulated industries, such
We are grateful to David Autor, David Cutler, Jon Gruber, Jerry Hausman, Rob Huckman,
Paul Joskow, Steve Levitt, Nolan Miller, Joe Newhouse, Joshua Rauh, Nancy Rose, Catherine
Wolfram, two anonymous referees, and seminar participants at the MIT public finance
and industrial organization lunches, the National Bureau of Economic Research’s Productivity
conference and Industrial Organization conference, the International Industrial
Organization conference, the Harvard–MIT–Boston University health seminar, Chicago
Graduate School of Business, the NBER Labor Studies conference, Brown, the Bureau of
Labor Statistics, and the NBER Health Care conference for helpful comments. We also
thank Adam Block, Amanda Kowalski, Matthew Notowidigdo, Erin Strumpf, and Heidi
Williams for excellent research assistance.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
838 journal of political economy
as health care, electricity, and telecommunications, which not only are
important for their sizable contributions to total GDP but have been at
the forefront of technological advances over the past several decades.
In this paper, we investigate how input and technology choices respond
to changes in the regulatory regime.
Starting in the mid-1980s, a number of different industries in a variety
of countries experienced a change in the regulatory regime away from
full cost reimbursement toward some type of “price cap.”1
These new
regulatory regimes often entailed a mixture of “partial cost reimbursement”
and “partial price cap.” Under this mixed regime—which we refer
to hereafter as “partial cost reimbursement”—only expenditures on capital
inputs are reimbursed, whereas labor expenses are supposed to be
covered by the fixed price paid per unit of output. Consequently, a
change from full cost to partial cost reimbursement increases the relative
price of labor inputs, among other things.
Despite many examples of this type of partial cost reimbursement,
including the Medicare Prospective Payment System (PPS) reform in
the United States, which we study in this paper, partial cost reimbursement
has received little theoretical or empirical attention. For example,
in his recent survey, Joskow (2005, 36) notes that “Although it is not
discussed too much in the empirical literature, the development of the
parameters of price cap mechanisms . . . [has] typically focused primarily
on operating costs only, with capital cost allowances established
through more traditional utility planning and cost-of-service regulatory
accounting methods.”
In this paper, we empirically investigate the impact of Medicare PPS
in the United States. Introduced in October 1983, PPS changed reimbursement
for hospital inpatient expenses of Medicare patients from
full cost to partial cost reimbursement, resulting in significant changes
in the relative factor prices faced by hospitals. The motivation behind
the reform was to reduce the level and growth of hospital spending,
which had been rising rapidly (as a share of GDP) for several decades.
We first present a simple neoclassical framework highlighting how
the change in relative factor prices faced by hospitals should affect their
demand for capital and labor and their technology adoption decisions.2
1
Examples include the telecommunications sector in the United States and United
Kingdom; gas, electric, and water utilities in the United Kingdom, New Zealand, Australia,
and parts of Latin America (see, e.g., Laffont and Tirole 1993; Armstrong, Cowen, and
Vickers 1994; Joskow 2005); and the Medicare Prospective Payment System for U.S. hospitals,
which is the focus of this paper.
2
Our framework is related to that of Averch and Johnson (1962) and Baumol and
Klevorick (1970) but focuses on cost reimbursement rather than rate of return regulation
and on the relative price changes implied by the switch from full cost reimbursement to
partial cost reimbursement. More important, it derives the implications of the change in
regulation regime on technology choices. These implications are not discussed either in
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 839
In terms of our framework, PPS increases the price of labor faced by
hospitals but leaves the price of capital unchanged. Before PPS, hospitals
were reimbursed retrospectively for capital and labor costs associated
with the care of Medicare patients. The PPS reform left the reimbursement
for capital unchanged while introducing prospective payments for
labor. After PPS, hospitals bore the full marginal costs of labor inputs
but received a fixed payment for each admitted Medicare patient, regardless
of the actual labor costs incurred for that patient. We refer to
this post-PPS fixed payment as a “price subsidy.”
The increase in the price of labor faced by hospitals relative to the
price of capital implies that PPS should increase the capital-labor ratio
of hospitals. However, the impact of the change in regulation regime
on labor inputs, capital inputs, and technology adoption decisions depends
on the level of the price subsidy, the extent of decreasing returns
to capital (or technology) and labor, and the elasticity of substitution
between these factors. In practice, the price subsidy seems to have been
limited, so that our theoretical framework suggests that the demand for
labor should decline. Nevertheless, when there is sufficient substitution
between capital (or technology) and labor, the demand for capital may
increase. Intuitively, PPS, by making labor more costly, induces hospitals
to downsize. In the standard case, with constant returns to scale to capital
and labor, this will necessarily be associated with a decline in the demand
for both labor and capital. However, with decreasing returns to scale
and sufficient substitution between capital and labor, affected hospitals
can increase rather than decrease their demand for capital. The same
also applies for their technology decisions, so that PPS may induce more
rapid adoption of new technologies, even as hospitals downsize.3
The PPS reform provides an attractive setting for studying the impact
of regulatory change on firm input and technology choice for several
reasons. First, the health care industry is one of the most technologically
intensive and dynamic sectors in the United States. Indeed, rapid technological
change is believed to be the major cause of both the dramatic
increase in health spending as a share of GDP and the substantial health
improvements experienced over the last half century (Newhouse 1992;
Fuchs 1996; Cutler 2003). Understanding the determinants of technological
progress in the health care sector, and the role played by
Averch and Johnson or Baumol and Klevorick or in the optimal regulation models (e.g.,
Laffont and Tirole 1993).
3
This result also highlights a parallel between our simple framework and the labor push
theory of innovation suggested by Hicks (1932) and Habakkuk (1962), where higher wages
may encourage technology adoption. Note, however, that in our empirical setting, the
introduction of PPS is associated with both an increase in the price of labor and some
increase in the price of output (increased reimbursement provided per Medicare patient
admitted). Our empirical results on the effect of PPS on technology adoption therefore
do not provide direct evidence on the labor push theory.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
840 journal of political economy
government policy, is therefore of substantial interest in its own right.
Second, government regulation is ubiquitous in this industry. Finally,
because of substantial differences in the importance of Medicare patients
for different hospitals, there is an attractive source of variation
to determine the effects of such a regulatory reform on input and technology
choices: hospitals with a higher Medicare share experience a
larger increase in the relative price of labor from the PPS reform to
reimbursement of Medicare patients.
Our empirical strategy is to exploit the interaction between the introduction
of PPS and the pre-PPS share of Medicare patient days (Medicare
share for short) in hospitals. We document that before the introduction
of PPS, hospitals with different Medicare shares do not display
systematically different trends in their input or technology choices. In
contrast, following PPS, hospitals with different Medicare shares show
significantly different trends.
Consistent with the predictions of our motivating theory, there is a
significant and sizable increase in the capital-labor ratio of higher–
Medicare share hospitals associated with the change from full cost to
partial cost reimbursement. This change in the capital-labor ratio is
made up of a decline in the labor inputs of higher–Medicare share
hospitals, with approximately constant capital inputs. Perhaps most interestingly,
we find that the introduction of PPS is associated with a
significant increase in the adoption of a range of new health care technologies.
We document this pattern both by looking at the total number
of different technologies used by hospitals and also by estimating hazard
models for a number of specific high-tech technologies that are in our
sample throughout. The increase in technology adoption and the decline
in labor inputs associated with the increase in the relative price
of labor also suggests that there is a relatively high degree of substitution
between technology and labor. We present suggestive evidence of one
possible mechanism for this substitution: the introduction of PPS is
associated with a decline in length of stay, which may represent substitution
of high-tech capital equipment for relatively labor-intensive hospital
stays.
Finally, we present evidence that the introduction of PPS is associated
with an increase in the skill composition of hospital nurses. This pattern
buttresses our results on increased capital-labor ratios and technology
adoption, since the consensus view in the literature is that skilled labor
is complementary to capital and/or technology (e.g., Griliches 1956;
Berman, Bound, and Griliches 1994; Autor, Katz, and Krueger 1998;
Krusell et al. 2000; Acemoglu 2002).
We consider a number of alternative interpretations for our findings
and conclude that the evidence for them is not compelling. We therefore
interpret our finding of PPS-induced changes in input mix and techThis
content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 841
nology adoption in the health care sector to be a response to the changes
in relative factor prices induced by the change in regulation regime.
Consequently, to our knowledge, this makes ours the first paper to
document that technology adoption in the health care sector is affected
by relative factor prices.4
It is also noteworthy that our findings run counter to the general
expectation that PPS would slow the growth of expensive technology
diffusion (see, e.g., Sloan, Morrissey, and Valvona [1988], Weisbrod
[1991], and the discussion of initial expectations in Coulam and Gaumer
[1991]). However, most prior analyses of PPS have conceived of it as a
move from full cost reimbursement to full price cap reimbursement
and have overlooked the fact that it was only a partial price cap on
noncapital expenditures; both our theoretical and empirical results
show the importance of the increase in the relative price of labor resulting
from the partial price cap structure.5
The rest of the paper proceeds as follows. Section II outlines a simple
neoclassical model of regulation and the implications of the change in
regulation regime on input and technology choices. Section III reviews
the relevant institutional background on Medicare reimbursement. Section
IV describes the data and presents some descriptive statistics. The
econometric framework is presented in Section V. Our main empirical
results are presented in Section VI, and Section VII shows that they are
robust to a number of alternative specifications. Section VIII briefly
summarizes our findings and discusses some directions for further work.
II. Motivating Theory
In this section, we discuss a simple neoclassical model of regulation and
derive its implications for a switch from full to partial cost reimbursement.
To conserve space, the claims made in this section are formalized
and proved in online Appendix B.
As noted in the introduction, PPS corresponds to a switch from full
retrospective cost reimbursement to partial cost reimbursement. Under
4
In this respect, our paper is related to the paper by Newell, Jaffe, and Stavins (1999),
who study the effect of energy price increases on the energy efficiency of a variety of
appliances. See also Greenstone (2002) on the effect of environmental regulations on
plant-level investment. In the hospital sector, past work has suggested that hospital technology
adoption appears to increase in response to traditional fee-for-service health insurance
(Finkelstein 2007) and to slow in response to managed care organizations (Cutler
and Sheiner 1998; Baker 2001; Baker and Phibbs 2002). In the context of the health
sector more generally, the rate of pharmaceutical innovation appears to increase in response
to increased (expected) market size (Acemoglu and Linn 2004; Finkelstein 2004)
or to tax subsidies for R&D (Yin 2005). Also related to our paper is the paper by Cawley,
Grabowski, and Hirth (2006), who study capital-labor substitution in nursing homes in
response to relative factor prices.
5
The literature on PPS is reviewed in Sec. III.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
842 journal of political economy
the partial cost reimbursement regime, capital costs associated with the
care of Medicare patients are reimbursed retrospectively whereas labor
expenses are reimbursed prospectively. Moreover, hospitals receive a
fixed payment for each Medicare patient admitted in lieu of the direct
reimbursement of actual labor costs incurred for the care of Medicare
patients they previously received. We refer to this fixed payment as a
“price subsidy” since it increased (from zero) the price Medicare paid
hospitals for each patient; it could also be referred to as a “price cap”
because the switch to PPS corresponds to partial price cap regulation.
This new payment structure implies that under PPS, hospitals bear their
own labor costs. In contrast, the reimbursement for capital expenses
remains unchanged, so that at the margin, a hospital can still pass on
additional capital costs associated with Medicare. Compared to the full
retrospective reimbursement regime, this corresponds to an increase in
the price of labor faced by hospitals relative to their capital costs.
To derive the implications of this change in regulation regime, we
consider a stylized world in which hospitals maximize profits (and we
ignore the interconnection between the demand for health care and
various private and social insurance programs). Although nonprofit or
public hospitals have other objectives as well, profit maximization both
is a useful benchmark and also is consistent with a large empirical literature
that finds essentially no evidence of differential behavior across
for-profit and nonprofit hospitals (see Sloan [2000] for a recent review
of this literature). Suppose that the revenue function of hospital i can
be written as
˜F(A , L , K , z ). (1)i i i i
This function combines the production of health services by the hospital
and the prices it faces for these services (which may be potentially affected
by the supply by this hospital).6
In (1), is a one-dimensionalAi
index of technology determined by the technology adoption decisions
of the hospital, is total labor (which can further be divided into skilledLi
and unskilled labor), and is capital (including structures capital).Ki
Finally, represents other inputs, in particular, managerial inputs andzi
doctor services, which are typically not reported as part of hospital em-
6
The composition of the patients of the hospital, in particular, between Medicare and
non-Medicare patients, is also incorporated into this function (see online App. B). In
practice, hospitals may not be able to choose the total number of Medicare patients. Either
a hospital is the only one in the area, thus facing an essentially constant demand for
Medicare services, or it may be competing with other hospitals in the area, in which case,
the number of Medicare patients will depend on the quality of service. This would require
a more involved analysis in which the firm chooses both quantity and quality, and there
is quality competition. Although we believe that this is an important area for theoretical
analysis, it falls outside the scope of our paper.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 843
ployment. Throughout, we assume that is homothetic and exhibits˜F
“decreasing returns” to labor and capital (and also to labor and technology).
This implies, for example, that when labor inputs and capital
inputs are doubled, hospital revenues will increase less than twofold.
We also assume that hospitals are price takers in input markets, facing
a wage rate of w per unit of labor and a cost of capital equal to R per
unit of capital.7
To simplify the discussion, let us also make two further
assumptions in the text. First, at the margin there is fungibility between
labor and capital inputs used for Medicare and non-Medicare purposes.
Second, the Medicare share of the patients of a hospital, , is exoge-mi
nous. Online Appendix B shows how the results discussed here generalize
when these assumptions are relaxed (so that the Medicare share
of patients changes endogenously and there is only limited fungibility
between Medicare and non-Medicare expenses).
The implications of the switch to partial cost reimbursement for input
mix and technology can be derived simply by considering the profit
maximization (cost minimization) of a hospital with a given Medicare
share and investigating how its decisions change in response to anmi
increase in the relative price of labor w and how this response depends
on .mi
Result 1. The switch to partial cost reimbursement increases the
capital-labor ratio, and this effect should be stronger for hospitals with
a greater Medicare share.
This result is straightforward and follows simply from the cost minimization
of a firm faced with a change in relative factor prices; the
magnitude of the change in relative factor prices is increasing in the
hospital’s Medicare share since it is only reimbursement of Medicare
patients that is affected.
Less obvious are the implications for the level of input demands.
These depend on the generosity of the price subsidy introduced by
partial cost reimbursement in lieu of the direct reimbursement of labor
costs. If the price subsidy is sufficiently generous, high–Medicare share
firms may expand their scale because of the greater profitability of their
activities or they may try to attract more Medicare patients. In contrast,
if the price subsidy is sufficiently low, then the switch to partial cost
reimbursement approximates a pure increase in the price of labor, with
no compensating increase in the price of “output.” The preexisting
evidence suggests that this latter case is likely to be more relevant in
7
In practice, certain hospitals might have monopsony power for some component of
their labor demand. For example, Staiger, Spetz, and Phibbs (1999) find evidence of
hospital monopsony power in the market for registered nurses. Incorporating any such
monopsony power would have no effect on our main results.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
844 journal of political economy
practice.8
Also consistent with a relatively low price subsidy, our empirical
work below shows that the price subsidy appears to have been less than
sufficient to overturn the effects of decreased cost subsidies.
The next result summarizes the implications of the switch to partial
cost reimbursement for the levels of input demand when the price
subsidy is sufficiently low.
Result 2a. When the price subsidy in the partial reimbursement
scheme is sufficiently low, the switch to partial cost reimbursement reduces
labor demand. This decline is greater for hospitals with a greater
Medicare share.
Result 2b. Moreover, if the (local) elasticity of substitution between
capital (respectively, technology) and labor is sufficiently large, then the
switch to partial cost reimbursement increases the demand for capital
(respectively, technology). This increase is greater for hospitals with a
greater Medicare share.
The most notable feature here is that the demand for capital (or
technology) can increase even though labor demand decreases. The
decrease in labor demand is intuitive in view of the fact that the price
subsidy is limited, so that the switch to partial cost reimbursement approximates
an increase in the price of labor, with the price of capital
and the price of “output” remaining constant. Consequently, the greater
price of labor (with limited price subsidies) induces hospitals to downsize
and reduce employment. If, in addition, there were constant returns
to scale to capital and labor, the decline in labor demand would be
associated with a decline in the demand for capital. This standard result
does not necessarily apply, however, when there are decreasing returns
to capital and labor, as we have assumed above, and when there is
sufficient substitutability between capital and labor. In this case, the
greater labor costs resulting from the change in regulation regime may
induce hospitals to substitute capital for labor with only limited change
in their scale and revenues, thus increasing their overall demand for
capital.
The same analysis establishes an analogous relationship between Medicare
share and technology adoption. For concreteness, suppose that
technology is embodied in equipment capital and recall that there are
decreasing returns to technology and labor, in the sense that a doubling
of equipment capital and labor leads to a less than twofold increase in
output. In addition, define the elasticity of substitution between technology
and labor analogously to the elasticity of substitution between
8
Qualitative descriptions of the PPS suggest a relatively low level of the price subsidy,
particularly after the first year of the program (Coulam and Gaumer 1991). The empirical
evidence reviewed by Cutler and Zeckhauser (2000) and Coulam and Gaumer (1991)
indicates that the introduction of PPS was associated with a decline in hospital profit
margins, which is also consistent with a relatively low level of the price subsidy.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 845
capital and labor (see online App. B). Then, result 2b also establishes
that when there are decreasing returns to technology and labor and
when the elasticity of substitution is sufficiently large, PPS can induce
technology adoption. Intuitively, hospitals now substitute technology for
labor, and this substitution can be sufficiently pronounced that hospitals
may adopt new technologies even as they downsize.
We note that the effects on capital and technology need not be the
same since the elasticities of substitution of each of these inputs with
labor are not the same (e.g., structures capital may be more complementary
to labor than labor-saving technology). Below, we will see that
PPS appears to be associated with increased technology adoption combined
with more or less constant overall capital expenditures, suggesting
that there was likely a decline in some other type of capital expenditures,
such as structures.
Finally, we also investigate the effect of the switch to partial cost reimbursement
on the ratio of skilled to unskilled nurses. The following
result summarizes the implications of this switch on the skill composition
of employment.
Result 3. If there is technology-skill or capital-skill complementarity,
then the switch to partial cost reimbursement will increase the
ratio of skilled to unskilled workers. This effect will be stronger for
hospitals with a greater Medicare share.
Since technology-skill and/or capital-skill complementarities appear
to be a good approximation to the technology of many sectors, result
3 suggests that we may see a significant change in the skill composition
of employment of affected hospitals and thus acts as a check on our
capital-labor ratio and technology adoption results.
The above results take the Medicare share of patients in a hospital
as given. In response to a switch to partial cost reimbursement, hospitals
will also change the composition of their patients. Online Appendix B
shows that when the price subsidy is low, there may also be a decline
in the Medicare share of all hospitals, but this does not affect the results
we focus on in our empirical analysis.
III. Overview of Medicare Reimbursement Policies
Medicare PPS was introduced in October 1983 (fiscal year 1984) in an
attempt to slow the rapid growth of health care costs and Medicare
spending. Under the original (pre-PPS) system of cost reimbursement,
Medicare reimbursed hospitals for a share of their capital and labor
inpatient expenses, which was proportionate to Medicare’s share of patient
days in the hospital (OTA 1984; Newhouse 2002, 22). By contrast,
under PPS, hospitals are reimbursed a fixed amount for each patient
based on his or her diagnosis, but not on the actual expenditures inThis
content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
846 journal of political economy
curred on the patient. At a broad level, this reform can be thought of
as a change from cost reimbursement to fixed price cap reimbursement,
and indeed, it is often described in these terms (e.g., Cutler 1995).
However, an important but largely overlooked feature of the original
PPS system—and a central part of our analysis—is that initially only the
treatment of inpatient operating costs was changed to a prospective
reimbursement basis. For the first 8 years of PPS, capital costs continued
to be fully passed back to Medicare under the old cost-based reimbursement
system, and capital reimbursement became fully prospective
only in 2001. Thus for almost its first 20 years, PPS continued to reimburse
capital costs at least partly on the margin.9
The reason for the
differential treatment of operating and capital costs, both in this case
and more generally in other regulated industries, appears to be the
greater difficulty in designing a prospective payment system for capital
(CBO 1988; Cotterill 1991; Joskow 2005).
The PPS reform is thus an example of a switch from full cost reimbursement
to partial cost reimbursement, as described in Section II. To
our knowledge, this feature of PPS has received no theoretical or empirical
attention, even though almost all empirical examinations of the
impact of PPS focus on the initial PPS period when partial cost reimbursement
was in effect.
Coulam and Gaumer (1991) and Cutler and Zeckhauser (2000) review
the extensive empirical literature on the effects of PPS. Broadly speaking,
this literature concludes that PPS was associated with declines in
hospital spending and in Medicare utilization (both admissions and
length of stay), but not with substantial adverse health outcomes. However,
most of this literature is based on simple pre-post (time-series)
comparisons. Important exceptions include Feder, Hadley, and Zuckerman’s
(1987) study of the impact of PPS on spending and Staiger
and Gaumer’s (1990) and Cutler’s (1995) studies of the impact of PPS
on health outcomes. Staiger and Gaumer pursue an empirical approach
similar to our strategy below, which exploits the interaction between the
introduction of PPS and hospital-level variation in the importance of
Medicare patients. Our empirical findings below are consistent with the
time-series evidence from this literature suggesting that there has been
a decrease in hospital expenditures and in utilization associated with
PPS. To our knowledge, our work is the first to investigate the impact
9
The original legislation specified that the treatment of capital costs would be unchanged
for the first 3 years of PPS (i.e., through October 1, 1986) and instructed the
Department of Health and Human Services to study potential methods by which capital
costs might be incorporated into a prospective payment system. In practice, a series of
eleventh-hour delays postponed any change in Medicare’s reimbursement for capital costs
until October 1, 1991, at which point a 10-year transition to a fully prospective payment
system for Medicare’s share of inpatient capital costs began (GAO 1986; CBO 1988; Cotterill
1991).
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 847
of PPS on labor and capital inputs and the skill composition of the
workforce.
Finally, a small empirical literature also uses pre-post comparisons to
study the impact of PPS on technology adoption. This literature finds
little conclusive evidence of any effect of PPS (PPAC 1988, 1990; Sloan
et al. 1988). To our knowledge, ours is also the first theoretical or
empirical study to show that PPS might have been associated with an
overall increase in technology adoption.
IV. Data and Descriptive Statistics
A. The AHA Data
Our analysis of the impact of PPS uses seven years of panel data from
the American Hospital Association’s (AHA) annual census of U.S. hospitals.
PPS took effect at the start of each hospital’s fiscal year on or
after October 1, 1983. Our data consist of 4 years prior to PPS (fiscal
years 1980–83) and 3 years after PPS (fiscal years 1984–86). We interpret
the year of the data as corresponding to the hospital’s fiscal year.
We restrict our analysis to the first 3 years of PPS, during which the
treatment of capital was specified in advance.10
We also exclude from
the analysis the four states (Maine, New York, Maryland, and New Jersey)
that received waivers exempting them from the federal PPS legislation.
Because these four states also experienced their own idiosyncratic
changes in hospital reimbursement policy during our period of analysis
(often right around the time of the enactment of federal PPS), the
states are not useful for us as controls (HCFA 1986, 1987; Antos 1993).
These four states contain about 10 percent of the nation’s hospitals,
leaving us with a sample of about 6,200 hospitals per year.
The data contain information on total input expenditures and its
components, employment and its components, and a series of binary
indicator variables for whether the hospital has a variety of different
technologies. All these input and employment data refer to the total
amounts for the hospital and therefore are unaffected by any potential
reallocation of factor usage within the hospital, for example, to nursing
home or outpatient units that may be affiliated with the hospital. In
10
As explained in n. 9, the years following the initial 3-year period after the introduction
of PPS were characterized by considerable uncertainty concerning the treatment of capital.
Given this uncertainty, it is not ex ante clear how hospitals should be expected to behave.
In light of this, we limited our main sample to include only the initial 3 years. Nevertheless,
we also investigated how the capital-labor ratio and technology adoption changed in subsequent
years. Our results (not reported) show that the increase in the capital-labor ratio
and in technology adoption that we find below for the first 3 years persisted during the
subsequent period of uncertainty (i.e., through 1990). These and other results mentioned
in the paper but not shown in the tables are available on request from the authors.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
848 journal of political economy
addition, we also observe inpatient hospital utilization, specifically admissions
and patient days. The expenditure and utilization data for year
t are in principle measured for the 12-month reporting period from
October 1, , through September 30, t; the employment and tech-t Ϫ 1
nology variables are in principle measured as of September 30, t. Note
that hospital employment and payrolls consist of nurses, technicians,
therapists, administrators, and other support staff; most doctors are not
included since they are not directly employed or paid by the hospital.
With the exception of patient days, none of the variables are reported
separately for Medicare. We use Medicare’s share of patient days in the
hospital as the key source of our cross-sectional variation in the impact
of PPS across hospitals (see below).
Medicare explicitly defines a hospital’s reimbursable capital costs to
include interest and depreciation expenses (OTA 1984; GAO 1986; Cotterill
1991), each of which we can identify in the AHA data.11
Since
changes in interest expenses may reflect financing changes rather than
real input changes, we focus on depreciation expenses (which are about
two-thirds of combined interest and depreciation expenses). Medicare
uses straight-line depreciation to reimburse hospitals for the depreciation
costs of structures and equipment (CBO 1988). The estimated life
of an asset is determined by the AHA; during the time period we study,
it ranged from 4 to 40 years depending on the asset; lives of 5 and 10
years tend to be the most common (AHA 1983). Depreciation expenses
therefore measure past and current capital expenditures rather than
the capital stock (which would have been the ideal measure). Since the
cost of capital and equipment prices should not vary systematically across
hospitals with different Medicare shares, depreciation expenses should
be a good proxy for the capital stock.
Our baseline measure of the capital-labor ratio, in terms of theK /Li i
model, is therefore the “depreciation share,” defined as depreciation
expenses divided by operating expenses. We define operating expenses
as total input expenses net of interest and depreciation expenses. Just
under two-thirds of operating expenses are payroll expenses (including
employee benefits), with the remainder consisting of supplies and purchased
services. Although payroll expenses are a more direct measure
of labor costs, they are not our preferred measure since they do not
11
Capital-related insurance costs, property taxes, leases, rents, and return on equity (for
investor-owned hospitals) are also included in capital costs. In practice, however, capital
costs are primarily interest and depreciation expenses, which are also the items reported
separately in the AHA data and used by the overseers of Medicare to study Medicare
capital costs (e.g., CBO 1988; PPAC 1992, MedPAC 1999).
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 849
TABLE 1
Summary Statistics
Variable Average
Standard
Deviation
Medicare share of inpatient days
in 1983 ( )mi 38% 21%
Real operating expenditures
(000’s) $30,976 $44,539
Real depreciation expenditures
(000’s) $1,379 $2,224
Depreciation share (depreciation/operating)
4.50% 2.50%
Skill ratio (RNs/RNs ϩ LPNs) 70% 16%
Note.—The table reports averages for the various hospital characteristics. All dollar estimates
are in thousands of 2004 dollars. All share estimates are multiplied by 100. The number of
observations is 43,188, except for skill ratio, which is 43,162. Data consist of a total of 6,280
hospitals, of which 5,881 (94 percent) are in the data for all 7 years, and all are in the data for
at least 2 years. All hospitals in the sample have information on Medicare share in 1983; this
variable is defined as zero for hospitals that were exempted from the PPS reform (see the text
for more details).
include the full set of costs that experienced the relative price change
under PPS.12
Depreciation expenses are on average about 4.5 percent of operating
expenses (see table 1), indicating that the hospital sector is much less
capital intensive than many other regulated industries.13
B. Descriptive Statistics and Time-Series Evidence
Table 1 gives the basic descriptive statistics for our key variables over
the entire sample. Changes in some of the outcome variables over time
are depicted in figures 1–3.
Figure 1 shows the simple time-series average of hospitals’ capitallabor
ratio (depreciation share). Consistent with result 1, the time series
displays a striking increase in the average capital-labor ratio at the time
of PPS’s introduction (fiscal year 1984) both in absolute terms and
relative to the preexisting time-series pattern. Result 2 suggests that if
the level of the price subsidy is sufficiently low, labor inputs should fall;
but even in this case, capital inputs may rise or remain unchanged.
Consistent with this, the time series shows a pronounced decrease in
12
In practice, our results are not sensitive to alternative measures of capital or labor
inputs, such as measuring labor inputs on the basis of payroll expenses or employment
or measuring capital expenses on the basis of interest expenses or interest plus depreciation
expenses (see Acemoglu and Finkelstein 2006).
13
The National Income and Product Accounts indicate that the share of capital in value
added in health services is 12.8 percent during the period of our data. In contrast, the
share of capital in electric, gas, and sanitary services is 64.1 percent and in telephone and
telegraph is 49 percent. We are grateful to Veronica Guerrieri for help with the National
Income and Product Accounts.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
Fig. 1.—Capital-labor ratio (shown in units per 100)
Fig. 2.—Log labor inputs (dollar amounts are measured in 2004 dollars)
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 851
Fig. 3.—Log capital inputs (dollar amounts are measured in 2004 dollars)
labor inputs (real operating expenditures) relative to the preexisting
trends (fig. 2). It also shows no evidence of a deviation in capital inputs
(real depreciation expenditures) from the preexisting time-series trend
(fig. 3).14
The time-series evidence is only suggestive, however, since it may be
driven by other secular changes in the hospital sector or the macroeconomy
more generally. Our empirical work below exploits the variation
across hospitals in the impact of PPS focusing in particular on the interaction
between the introduction of PPS and the hospital’s pre-PPS
Medicare share (the empirical counterpart of in the model). It ismi
nonetheless interesting that this very different empirical strategy will
show patterns quite similar to those visible in figures 1–3.
V. Econometric Framework
As discussed in Section II, the impact of PPS on hospital input and
technology choices should be larger for hospitals with a higher (prePPS)
Medicare share, . On the basis of this reasoning, our basic es-mi
timating equation is
y p a ϩ g ϩ X 7 h ϩ b 7 (POST 7 m ) ϩ ␧ , (2)it i t it t i it
14
To match the empirical work below, the time series in figs. 2 and 3 are presented on
a log scale. In practice, the pattern is similar if we look at absolute levels.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
852 journal of political economy
where is the outcome variable of interest in hospital i at time t. Inyit
our first empirical models, will represent the capital-labor ratio (mea-yit
sured as the depreciation share) to investigate the predictions discussed
in Section II. We will later use the same framework to investigate the
responses of a number of other outcomes.
In our estimating equation (2), represents a full set of hospitalai
fixed effects, stands for a full set of year dummies, and is a vectorg Xt it
of other covariates, which are not included in the baseline regressions
but will be added in several of the robustness checks below. Finally,
is a random disturbance term capturing all omitted influences.␧it
The main variable of interest is the interaction term ( ) withPOST 7 mt i
coefficient b. Here is a dummy variable that takes a value equalPOSTt
to one for the three post-PPS years (1984–86). A useful variant of this
equation is
y p a ϩ g ϩ X 7 h ϩ b 7 (POST 7 m ) ϩ f 7 (d 7 m ) ϩ ␧ , (3)it i t it t i 1983 i it
where is a dummy for the year 1983. The interaction termd1983
acts as a prespecification test; it will be informative on whetherd 7 m1983 i
there are any differential trends in the variables of interest by Medicare
share before the introduction of PPS.
We also estimate a more flexible version of these equations of the
form
y p a ϩ g ϩ X 7 h ϩ b 7 (d 7 m ) ϩ ␧ , (4)͸it i t it t t i it
t≥1981
where the term with the summation stands for a separate coefficient
for 1981 and each subsequent year, and is an indicator variable fordt
year t. Relative to (2) or (3), the model in (4) allows both time-varying
post-PPS effects and also a more flexible investigation of whether there
are any differential trends in the variables of interest by Medicare share
in any of the pre-PPS years.
In all models, to account for potential serial correlation of the observations
from the same hospital, we adjust the standard errors by
allowing for an arbitrary variance-covariance matrix within each hospital
over time (see Wooldridge 2002, 275). In practice, this does not have
much effect on the standard errors.
A key question is how to measure empirically. Because reimburse-mi
ment in the pre-PPS regime was based on Medicare’s share of patient
days in the hospital (Newhouse 2002, 22), we define as the share ofmi
Medicare inpatient days. Since, as discussed in the motivating theory,
the Medicare share is likely to respond endogenously to the regu-mi
latory change, we measure in 1983, the year prior to the implemen-mi
tation of PPS.
Figure 4 shows the considerable variation across hospitals in their
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 853
Fig. 4.—Distribution of hospitals by Medicare share in 1983
Medicare share in 1983. The average hospital’s Medicare share is almost
two-fifths (38 percent), with a standard deviation of over one-fifth (21
percent). The mass point of almost 15 percent of hospitals that we have
coded as having a zero Medicare share reflects the fact that certain types
of hospitals—specifically federal, long-term, psychiatric, children’s, and
rehabilitation hospitals—were exempt from Medicare PPS (OTA 1985;
Newhouse 2002, 27). The exemption stems from the extremely low
Medicare share of these hospitals. For our purposes, we code their mi
as zero since they would not be affected by the reform. In the robustness
analysis below, we show that the main results can be obtained when we
identify the effect of PPS using only the variation between zero share
and nonzero share hospitals or using only the variation in amongmi
hospitals coded with a nonzero .mi
The identifying assumption in estimating equations (2), (3), and (4)
is that, without the introduction of PPS, hospitals with different ’smi
would not have experienced differential changes in their outcomes in
the post-PPS period. However, is not randomly assigned. For example,mi
when we examine the variables summarized in table 1, we find that in
the cross section prior to PPS (i.e., in 1983), a higher Medicare share
is correlated with lower operating expenses, a higher depreciationmi
share, and a lower skill ratio. Any fixed differences across hospitals will
be absorbed by the hospital fixed effects, the ’s. However, such sys-ai
tematic differences raise concerns about whether, without the introduction
of PPS in fiscal year 1984, hospitals with a different wouldmi
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
854 journal of political economy
have experienced similar changes in the outcomes of interest. Equations
(3) and (4) allow us to use the pre-PPS data to investigate the validity
of this identifying assumption by looking for differential trends prior
to PPS. The results below will show little evidence of such preexisting
trends, supporting our identifying assumption.
Motivated by the theoretical predictions, we estimate equations (2),
(3), and (4) for various dependent variables: capital-labor ratio (depreciation
share), log labor inputs (log operating expenditures), log capital
inputs (log depreciation expenditures), Medicare share of patient days,
log average length of hospital stay, and the share of nurse employment
that is high-skill. When the dependent variable is not already a share,
we estimate the equation in logs.15
VI. Main Results
A. Results on Capital-Labor Ratio
Result 1 in Section II suggests that the move from full cost to partial
cost reimbursement should increase the capital-labor ratio. We investigate
this in table 2, which shows that the introduction of Medicare
PPS is associated with a statistically and economically significant increase
in the capital-labor ratio (depreciation share).
Column 1 shows the estimation of our most parsimonious equation,
(2). The variable is simply a dummy for the 3 years in which PPSPOSTt
is in effect in our sample (1984–86). In this specification, the coefficient
b on the key interaction term ( ) is estimated as 1.129 (stan-m 7 POSTi t
dard error p 0.108). This is both a highly statistically significant and
economically large effect. Given that the average hospital has a 38 percent
Medicare share prior to PPS, this estimate suggests that in its first
3 years, the introduction of PPS was associated with an increase in the
depreciation share of about 0.42 ( ) for the average hos-Ӎ 1.129 # 0.38
pital. Since the average depreciation share is about 4.5, this corresponds
to a sizable 10 percent increase in the capital-labor ratio of the average
Medicare share hospital.
Column 2 investigates whether the differential growth in the capitallabor
ratio between high– and low–Medicare share hospitals was present
before the introduction of PPS by estimating equation (3). The estimate
of the key parameter, b, is essentially unchanged, and the coefficient f
on the interaction between the 1983 dummy and the Medicare share
( ) is very small (practically zero) and highly insignificant. Thisd 7 m1983 i
indicates that relative to the years 1980–82, hospitals with a larger mi
15
A level specification would constrain the outcomes of each hospital to grow by the
same absolute amount within a year, which is inappropriate given the considerable variation
in size across hospitals.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 855
TABLE 2
The Impact of PPS on the Capital-Labor Ratio
(1) (2) (3) (4) (5)
POST#mi 1.129
(.108)
1.122
(.121)
POSTTREND#mi .538
(.050)
.532
(.053)
d # m1981 i .153
(.114)
d # m1982 i Ϫ.388
(.131)
d # m1983 i Ϫ.028
(.098)
Ϫ.109
(.136)
Ϫ.060
(.088)
d # m1984 i .601
(.163)
d # m1985 i 1.068
(.172)
d # m1986 i 1.474
(.189)
Note.—The dependent variable is depreciation share. The table reports results from estimating eqqs. (2)–(5) by
ordinary least squares (OLS). All regressions include hospital and year fixed effects. The mean dependent variable is
4.5. POST is an indicator variable for the years 1984–86. The variable POSTTREND p 0 through 1983 and then takes
the values 1, 2, and 3 in 1984, 1985, and 1986, respectively. The variable is an indicator variable for year t. Thedt
variable measures the Medicare share of the hospital’s inpatient days in 1983. Standard errors are in parentheses.mi
Standard errors are adjusted to allow for an arbitrary covariance matrix within each hospital over time. The number
of observations is 43,188. In col. 3, the omitted category is . To interpret the magnitudes, recall that the averaged # m1980 i
Medicare share in 1983 is about two-fifths.
did not experience a statistically or economically significant change in
their capital-labor ratio in 1983 (the year before PPS) relative to hospitals
with a smaller . This is supportive of the identifying assumption thatmi
without the introduction of PPS, hospitals with different Medicare shares
would have experienced similar changes in their capital-labor ratios.
Column 3 shows the results from estimating the more flexible equation
(4) in which each year dummy is interacted with the hospital’s 1983
Medicare share; the omitted year is 1980. This allows a further investigation
of the identifying assumption as well as an examination of the
timing of the response to PPS. The results indicate that relative to their
1980 spending, hospitals with a larger Medicare share did not experience
a significant change in their capital-labor ratio relative to hospitals
with a smaller Medicare share in the pre-PPS year 1981 or 1983, but
there is a one-time downward blip in 1982.16
Thus, the pattern over all
four pre-PPS years suggests that, if anything, the capital-labor ratio may
have been declining in hospitals with a larger Medicare share relative
to hospitals with a smaller Medicare share. There is a pronounced shift
16
This 1-year blip in 1982 is not a major concern for our results. We show in Sec. VII
that our results are robust to numerous alternative specifications that deal flexibly with
potential mean reversion and to including a linear trend interacted with Medicare share.
Moreover, our results are also robust to excluding the 1982 data from the analysis, with
or without including a linear trend interacted with Medicare share (not shown).
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
856 journal of political economy
in this pattern starting in 1984, the first year that PPS is in place. In
this year, hospitals with a larger Medicare share experience a statistically
significant increase in the capital-labor ratio relative to hospitals with a
smaller Medicare share, confirming the results in the previous two
columns.
In a pattern that will repeat itself for many of the other dependent
variables that we analyze, the results in column 3 also indicate that the
magnitude of the increase in the capital-labor ratio associated with PPS
grows from 1984 to 1985 and again from 1985 to 1986. This likely
reflects, at least in part, lags in the implementation of PPS both in
actuality and as measured in our data. PPS was effective at the beginning
of the hospital’s fiscal year starting on or after October 1, 1983. Hospitals
were therefore added to the new regime throughout its first year in
operation, with some not entering the new system until midway or late
in the 1984 calendar year (OTA 1985). Moreover, not all hospitals follow
the AHA instructions to report data for year t for the 12-month period
from October 1, , to September 30, t; in any given year, about halft Ϫ 1
appear to instead report data for the 12-month period corresponding
to their fiscal year. This also contributes to a staggered implementation
of PPS in the data. However, the fact that the increase in the size of
the effect from 1984 to 1985 (i.e., from a year in which only some
hospitals were fully under the system to a year in which all were) is quite
similar to the increase in the size of the effect from 1985 to 1986 (two
years in which all affected hospitals were under the system) suggests
that lags in implementation alone cannot fully account for the time
pattern we observe. Lags in the hospital response to the new reimbursement
regime (perhaps due to adjustment costs) may have also
played a role.
Whatever its underlying cause, the empirical evidence in column 3
that the impact of PPS appears to grow over time suggests that a more
appropriate parameterization of the post-PPS period may be a trend
rather than a single post-PPS dummy. This motivates yet another slight
variation on our estimating equation,
˜y p a ϩ g ϩ X 7 h ϩ b 7 (POSTTREND 7 m )it i t it t i
ϩ f 7 (d 7 m ) ϩ ␧ , (5)1983 i it
which imposes that the post-PPS effects increase linearly.17
This equation
has the advantage of summarizing the post-PPS patterns more parsimoniously
than equation (4).
Columns 4 and 5 estimate equation (5) with and without the prespec-
17
More specifically, , so that following PPS there is aPOSTTREND p ͸ (t Ϫ 1983)t t≥1984
linear trend difference among hospitals with different Medicare shares.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 857
TABLE 3
The Impact of PPS on Other Variables
Log Labor
Inputs
(1)
Log Capital
Inputs
(2)
Medicare
Share
(3)
Log Length
of Stay
(4)
Skill Share
of Nurse
Employment
(5)
POSTTREND#mi Ϫ.068
(.008)
Ϫ.023
(.016)
Ϫ.032
(.003)
Ϫ.030
(.011)
1.67
(.272)
d # m1983 i .022
(.013)
.049
(.039)
Ϫ.002
(.006)
.019
(.015)
.876
(.567)
Observations 43,188 40,888 36,611 36,609 43,162
Mean dependent
variable 15.83 12.61 .38 2.16 70
Note.—The table reports results from estimating eq. (5) by OLS for the dependent variable indicated in the column
heading. Specifically, the dependent variable is log operating expenses (col. 1); log depreciation expenses (col. 2);
Medicare share of inpatient days (col. 3); log length of stay, with length of stay defined as patient days/admissions (col.
4); and the ratio of the number of full-time-equivalent RNs to the sum of full-time-equivalent RNs plus full-time-equivalent
LPNs (col. 5). All regressions include hospital and year fixed effects. The variable POSTTREND p 0 through 1983
and then takes the values 1, 2, and 3 in 1984, 1985, and 1986, respectively. The variable is an indicator variable fordt
year t. In all columns but cols. 3 and 4, measures the Medicare share of the hospital’s inpatient days in 1983. Inmi
cols. 3 and 4, measures the Medicare share of the hospital’s inpatient days in 1980, and data from 1980 are excludedmi
from the analysis. Note also that while we code the regressor to be zero for the approximately 15 percent of hospitalsmi
that are exempt from PPS (see fig. 4), we allow the dependent variable Medicare share in col. 3 to take its actual value
for all hospitals. Standard errors are in parentheses. Standard errors are adjusted to allow for an arbitrary covariance
matrix within each hospital over time. To interpret the magnitudes, recall that the average Medicare share in 1983 is
about two-fifths.
ification test term, . In both cases, there is a very preciselyd 7 m1983 i
estimated coefficient of of about 0.53 (standard error approximately˜b
0.05). In column 5 as in column 2, there is no evidence of a pre-PPS
differential effect. With a calculation similar to that above, the estimate
of 0.53 suggests that, in its first three years, PPS was associated, on
average, with an approximately 4 percent per year increase in hospitals’
capital-labor ratio.
Columns 1 and 2 of table 3 investigate changes in the demand for
capital and labor separately. In the interest of brevity, table 3 reports
only results from estimating equation (5).18
Column 1 investigates the
differential change in (log) labor inputs (log operating expenses) across
hospitals with different pre-PPS Medicare shares. Consistent with result
2a, the results suggest that the move from full cost to partial cost reimbursement
was associated with a decline in labor inputs; the estimate
of b, the coefficient on the interaction term ( ) inPOSTTREND 7 mt i
column 1, is Ϫ0.068 (standard error p 0.008). The coefficient on
shows some evidence of a small and marginally statisticallyd 7 m1983 i
insignificant increase in labor inputs in more affected hospitals prior
to PPS; although this may raise concerns about the potential for mean
reversion that may contaminate our estimate of the impact of PPS,
18
For all the outcomes examined in table 3, results from the other specifications shown
in table 2 are similar and can be found in the working paper version of the paper (Acemoglu
and Finkelstein 2006).
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
858 journal of political economy
Section VII shows that the results are highly robust to a number of
specifications that deal flexibly with mean reversion issues.
The results for log capital inputs (log depreciation expenses) in column
2 indicate essentially no statistical or substantive effect on capital
inputs. This is again consistent with the implications summarized in
result 2b that even when the price subsidy is low enough that labor
inputs decline, capital inputs need not decrease and may in fact increase
when there is sufficient substitutability between capital and labor.
Consistent with a low price subsidy, the time-series evidence also shows
a decline in the average Medicare share of patient days across hospitals
after the introduction of PPS (not shown). In addition, column 3 of
table 3 shows that there is a more pronounced decline in Medicare’s
share of patient days in hospitals that initially had a higher Medicare
share.19
We did, however, confirm that hospitals that initially had a
higher Medicare share also had a higher Medicare share after the introduction
of PPS, so that the initially higher–Medicare share hospitals
are the ones that would be the most affected by the policy.20
(The
remaining columns in table 3 are discussed below.)
B. Technology Adoption
The AHA data contain a series of binary indicators for whether the
hospital has various “facilities,” such as a blood bank, open-heart surgery
facilities, computed tomography (CT) scanner, occupational therapy,
genetic counseling, and neonatal intensive care. These data have been
widely used to study technology adoption decisions in hospitals (e.g.
Cutler and Sheiner 1998; Baker and Phibbs 2002; Finkelstein 2007).
Since they contain only indicator variables for the presence or absence
of various facilities, we cannot study upgrading of existing technology
or the intensity of technology use, but we can study the technology
adoption decision on the extensive margin.
Overall, during our time period, the AHA collects information on the
presence of 113 different facilities. These are listed, together with their
sample means and the years in which they are available, in Appendix
table A1. On average, a given facility is reported in the data set for 4.6
out of the possible seven years; only one-quarter of the technologies are
19
To prevent a mechanical correlation between the cross-sectional variation, , and themi
dependent variable, in col. 3 we define on the basis of the hospital’s Medicare sharemi
in 1980 and exclude 1980 from the analysis. All our previous results are robust to this
alternative specification. Also, while we code the regressor to be zero for the approx-mi
imately 15 percent of hospitals that are exempt from PPS, we allow the dependent variable
in this case to take its actual value for all hospitals. On average over 1981–86, the dependent
variable is 0.38; it is 0.09 (0.43) for exempt (nonexempt) hospitals.
20
In particular, the Spearman rank correlation between Medicare share in 1983 and
the average Medicare share between 1984 and 1986 is 0.86.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 859
in the data for all seven years. Moreover, as is readily apparent from
Appendix table A1, the list encompasses a range of very different types
of facilities. Given these two features of the data, we pursue two complementary
approaches to analyzing the impact of the change from full
to partial cost reimbursement on technology adoption.
Our first approach treats all facilities equally and estimates equations
(2)–(5) using the (unweighted) number of facilities that hospital i has
in year t as the dependent variable (in this specification, year fixed effects
take care of the unbalanced panel nature of the technology data). This
has the advantage of looking broadly across a wide range of different
technologies but the disadvantage that it treats all the very different
facilities symmetrically. This would be appropriate if all the technologies
were perfect substitutes; this is the case in the model in online Appendix
B, but may be far from reality. Our second approach therefore estimates
separate hazard models of the time to adoption for specific “high-tech”
technologies that are in the data for all the years of our sample.
In our first approach, the dependent variable is the raw count of the
number of facilities of each hospital. This dependent variable ranges
from zero to 77 with an average of 25. Approximately 10 percent of the
hospital-years in the sample have zero facilities. The results are shown
in table 4.21
The estimates suggest that the change from full to partial
cost reimbursement is associated with a statistically and economically
significant increase in the number of hospital facilities. For example,
the point estimate in column 1 is 2.622, suggesting that, on average,
the regulatory change is associated with an increase of about one new
facility ( ) in a hospital over its first 3 years; this corre-Ӎ 2.622 # 0.38
sponds to about a 4 percent increase over the average number of facilities
in a hospital (which is about 25).
The results are also broadly supportive of our identifying assumption
of no differential trends across hospitals in the number of facilities prior
to PPS. Column 3 shows some evidence of a differential decline in the
number of facilities in higher–Medicare share hospitals in 1981 relative
to 1980, but reassuringly, there is no similar pattern among any of the
other pre-PPS years 1981, 1982, or 1983. Although the differential decline
in 1981 may raise concerns about mean reversion, Section VII will
show that the results are robust to several different checks against mean
reversion.
One difference with the previous set of findings is the time pattern
of the impact of PPS in the flexibly estimated specification (col. 3);
21
Since there are a large number of zeros, we cannot estimate this equation in logs;
nor is there a natural scaling factor to use in the denominator to turn this into a share
estimate. Nevertheless, both the magnitude and the statistical significance of the estimates
are robust to estimating a conditional fixed-effects Poisson model (Hausman, Hall, and
Griliches 1984) instead. These results are shown in Acemoglu and Finkelstein (2006).
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
860 journal of political economy
TABLE 4
The Impact of PPS on Technology Adoption I: Number of Facilities
(1) (2) (3) (4) (5)
POST#mi 2.622
(.357)
2.501
(.401)
POSTTREND#mi 1.156
(.164)
1.094
(.177)
d # m1981 i Ϫ2.423
(.526)
d # m1982 i Ϫ2.965
(.541)
d # m1983 i Ϫ.467
(.354)
Ϫ2.281
(.517)
Ϫ.631
(.326)
d # m1984 i Ϫ.496
(.567)
d # m1985 i 1.894
(.634)
d # m1986 i .696
(.619)
Note.—The dependent variable is number of facilities. All regressions include hospital and year fixed effects. The
mean dependent variable is 25. The table shows results from estimating eqq. (2)–(5) by OLS. The variable POST is an
indicator variable for the years 1984–86. The variable POSTTREND p 0 through 1983 and then takes the values 1, 2,
and 3 in 1984, 1985, and 1986, respectively. The variable is an indicator variable for year t. The variable measuresd mt i
the Medicare share of the hospital’s inpatient days in 1983. Standard errors are in parentheses. Standard errors are
adjusted to allow for an arbitrary covariance matrix within each hospital over time. The number of observations is
43,188. In col. 3, the omitted category is . To interpret the magnitudes, recall that the average Medicare shared # m1980 i
in 1983 is about two-fifths.
rather than the approximately linear growth for the other variables
studied so far, the number of facilities in the affected hospitals shows
a statistically significant increase from 1983 to 1984 and again from 1984
to 1985, but the effect then appears to decline somewhat from 1985 to
1986.
An important drawback to the preceding analysis is that it treats all
technologies as perfect substitutes. As an alternative, we estimate separate
hazard models of the time to adoption for specific technologies
that are in the data for all the years of our sample. We focus on 10
technologies that were identified as high-tech by previous researchers
(Cutler and Sheiner 1998; Baker 2001; Baker and Phibbs 2002) and
that are present in our data in all years. Two of these are cardiac technologies
(cardiac catheterization and open-heart surgery), two are diagnostic
technologies (CT scanner and diagnostic radioisotope facility),
four are radiation therapies used in cancer treatment (megavoltage
radiation therapy, radioactive implants, therapeutic radioisotope facility,
and x-ray radiation), and the remaining two are the neonatal intensive
care unit and organ transplant. Figure 5 plots the diffusion pattern over
our sample period of each of these 10 technologies; they differ both in
their initial diffusion level and in whether and how rapidly they are
diffusing over our sample period.
In the hazard model analysis, we exclude hospitals that have a given
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 861
Fig. 5.—Technology diffusion
technology in 1980 (since they are not “at risk” of failure, i.e., of adoption)
and treat hospitals that have still not adopted the technology by
1986 (the end of our sample period) as censored. Our first model is
an exponential (i.e., constant baseline) proportional hazard model of
the form
l p a exp [g ϩ f 7 d 7 m ϩ b 7 (POST 7 m ) ϩ X 7 h], (6)t t 1983 i t i i
where denotes the conditional probability that the hospital adoptslt
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
862 journal of political economy
the technology in question at time t, given that it has not yet adopted
the technology, and a denotes the constant baseline hazard parameter
(which we estimate). The assumption of the proportional hazard model
is that the covariates shift the baseline hazard proportionally. Our second
estimation strategy uses the Cox semiparametric proportional hazard
model, which allows for a fully flexible, nonparametric baseline
hazard (see Kiefer 1988). In the Cox model, we do not include yearl0
fixed effects since the fully flexible baseline hazard is also specified with
respect to calendar time.
Since we have at most a single transition (adoption) for each hospital,
we cannot include hospital fixed effects in the hazard model analysis.
Instead, we control for a range of time-invariant hospital characteristics
(denoted by ). These are (i.e., the hospital’s 1983 Medicare share),X mi i
the square of , the number of beds in 1983, and dummy variables formi
whether the hospital is a general (nonspecialty) hospital, whether it is
a short-term (as opposed to a long-term) hospital, whether it is a federal
hospital, whether it is located in an urban area, and a complete set of
state fixed effects. The results reported so far are very similar if we
control for these covariates instead of hospital fixed effects (not shown).
Table 5 reports the results from the exponential and Cox proportional
hazard models. To conserve space, we report results only from a specification
similar to equation (3), which includes a single interaction
between the Medicare share, , and the post-PPS period dummy,mi
, as well as the prespecification test with the interaction betweenPOSTt
and the dummy for the year 1983. Panel A reports results from themi
exponential proportional hazard model, and panel B reports results
from the Cox proportional hazard model. For each technology in each
panel, we report the coefficient and the standard error on POST 7 mt i
and . To illustrate the magnitude of our estimates, we alsod 7 m1983 i
translate the hazard model coefficient on into the impliedPOST 7 mt i
change in the proportion of hospitals that adopt the technology between
1981 and 1986 associated with changing from zero to its mean value.mi
Since we look at 10 different technologies, the per-technology p-values
will be lower than when each technology is viewed as part of a “family
of hypotheses” that PPS had no effect on any of the 10 technologies.
We therefore also report the family-wise error rate adjusted p-value (in
brackets). This p-value corresponds to the probability of rejecting the
null hypothesis of no effect on a given technology under the null family
of hypotheses of no effect on any of the technologies.22
The family-wise
adjusted p-values are about five times larger than the standard p-values.
22
We calculate these family-wise error rate adjusted p-values on the basis of 10,000
iterations of the free step-down resampling method of Westfall and Young (1993). This
is more powerful than the standard Bonferroni correction because it does not assume
independence across the 10 outcomes and sequentially removes hypotheses from the family
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 863
Both panels of table 5 show similar results and suggest that the shift
from full cost to partial cost reimbursement was associated with increased
technology adoption. At a 5 percent cutoff, the results from the
exponential (respectively, Cox) model using the standard p-values suggest
that PPS is associated with increased adoption of seven (respectively,
six) of the 10 specific technologies. The results using the family-wise
adjusted p-values suggest that PPS is associated with an increased adoption
of three of the 10 technologies. Two of these three technologies,
open-heart surgery and CT scan, are likely to be used disproportionately
by Medicare patients. Our interpretation of the increase in adoption
following PPS is thus along the lines of result 2b and relies on technologylabor
substitution.23
While we cannot definitively pinpoint the mechanism for this technology-labor
substitution, we can provide some evidence of one natural
mechanism, the use of technology to reduce the length of stay. The
typical hospital day is relatively nurse or custodial care intensive. By
increasing the intensity of treatment up-front, hospitals may be able to
reduce length of stay on the margin. Consistent with this, column 4 of
table 3 presents evidence that Medicare PPS is associated with declines
in log average length of stay, defined as log(patient days/admissions).24
It is also noteworthy that the other technology for which the familywise
adjusted p-values show a statistically significant increase in adoption
is the neonatal intensive care unit (NICU), which is likely to be used
almost exclusively by non-Medicare patients. Although an effect of PPS
on NICU adoption may be viewed as problematic for our identification
strategy, it is consistent with the growing body of evidence of “spillovers”
in the health care sector.25
Such spillovers could be incorporated into
the framework in Section II by relaxing the assumption that in thezi
after they are rejected. See Kling and Liebman (2004) for an application and a more
detailed discussion of this method.
23
As noted before, hospital labor consists of nurses, orderlies, administrators, and custodial
staff but does not include doctors (who are neither employed by nor paid by the
hospitals). Thus the technologies may well be complementary with physicians (or particular
physician specialties) but still substitutes for hospital labor.
24
Because (as in col. 3 of table 3) the dependent variable in col. 4 of table 3 is mechanically
related to the cross-sectional variation of the Medicare share of patient days in
1983, we again drop 1980 from the sample and redefine the cross-sectional variation as
the Medicare share of patient days in 1980.
25
For example, Baicker and Staiger (2005) find that increases in the hospital reimbursement
rate of Medicaid—which primarily reimburses for childbirth and pediatrics—
are associated with declines not only in infant mortality but also in heart attack mortality
among the elderly Medicare population. Similarly, Baker (1997) finds that higher managed
care penetration in private insurance is associated with decreased hospital spending on
fee-for-service Medicare patients. Most closely related to our findings, Dafny (2005) finds
that in response to increases in average reimbursement rates for Medicare patients with
specific diagnoses, hospitals spread the increased revenue uniformly across the treatment
of all patients.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
864
TABLE5
TheImpactofPPSonTechnologyAdoptionII:HazardModelsofTechnologyAdoption
CardiacTechnologies
Diagnostic
Radiology
Other
TechnologiesRadiationTherapy(CancerTreatment)
Cardiac
Catheterization
Open-Heart
SurgeryCTScan
Diagnostic
Radioisotope
Neonatal
Intensive
Care
Organ
Transplant
Megavoltage
Radiation
Therapy
Radioactive
Implants
Therapeutic
Radioisotope
Facility
X-Ray
Radiation
A.ExponentialProportionalHazardModel
POST#mi1.23
(.481)
[.071]
2.61
(.683)
[.001]
.928
(.259)
[.003]
.666
(.265)
[.072]
3.83
(.663)
[!.0001]
1.74
(.785)
[.116]
1.76
(.782)
[.116]
Ϫ.74
(.508)
[.383]
Ϫ.096
(.490)
[.975]
Ϫ.081
(.601)
[.975]
d#m1983i.24
(.692)
[.978]
1.48
(1.054)
[.745]
.769
(.343)
[.218]
Ϫ.024
(.312)
[.996]
1.14
(.843)
[.745]
1.07
(1.30)
[.885]
Ϫ1.397
(1.026)
[.745]
Ϫ.72
(.769)
[.883]
.008
(.738)
[.996]
1.70
(1.01)
[.584]
Changein%
adoptifchange
POST#frommi
0tomean.025.004.086.054.008.003.004Ϫ.021Ϫ.003Ϫ.001
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
865
B.CoxProportionalHazardModel
POST#mi1.13
(.480)
[.129]
2.48
(.680)
[.002]
.783
(.259)
[.024]
.577
(.266)
[.165]
3.69
(.659)
[!.0001]
1.61
(.783)
[.165]
1.69
(.786)
[.165]
Ϫ.826
(.511)
[.281]
Ϫ.183
(.490)
[.921]
Ϫ.188
(.601)
[.921]
d#m1983iϪ.795
(.510)
[.511]
Ϫ.086
(.657)
[.984]
.204
(.300)
[.932]
Ϫ.659
(.253)
[.070]
.205
(.641)
[.982]
.023
(.894)
[.984]
Ϫ2.54
(.725)
[.003]
Ϫ1.598
(.568)
[.040]
Ϫ1.160
(.513)
[.145]
Ϫ.562
(.556)
[.838]
Changein%
adoptifchange
POST#frommi
0tomean.053.053.090.088.12.016.061Ϫ.029Ϫ.002Ϫ.004
Meanadoption
rate.066.033.41.36.066.040.036.082.092.073
Observations4,8615,1304,7392,7585,3015,4374,9504,5424,4854,741
Note.—Thetableshowscoefficientsfromproportionalhazardmodels.CensoringoccursifPPShasnotbeenadoptedby1986.Allestimatesincludecovariatesforthefollowinghospitalcharacteristics
in1983:Medicareshareofpatientdays,Medicareshareofpatientdayssquared,numberofbeds,andindicatorvariablesforstate,whetherinametropolitanstatisticalarea,whetherageneralhospital,
whetherashort-termhospital,andwhetherafederalhospital.Estimatesusingtheexponentialmodelalsoincludeyearfixedeffects.ThevariablePOSTisanindicatorvariablefortheyears1984–86.
Thevariableisanindicatorvariableforyear1983.ThevariablemeasurestheMedicareshareofthehospital’sinpatientdaysin1983.“Changein%adoptifchangePOST#from0tomean”dmm1983ii
denotesthedifferenceintheaveragehospitaladoption(by1986)rateifallvariablesaresettotheirmeanexceptforPOST#,whichissettozero,relativetoifallvariables(includingPOST#)mmii
aresettotheirmean.Heteroskedasticity-robuststandarderrorsareinparentheses.Family-wiseadjustedp-valuesbasedon10,000iterationsareinbrackets(seethetextformoredetails);thesetend
tobeaboutfivetimeslargerthantheunadjustedper-technologyp-values(notshown).Numberofobservationsdenotesthesizeoftheat-risksample(i.e.,thenumberofhospitalsthathavenotadopted
by1980).Meanadoptionratedenotesthepercentageoftheat-risksamplethatadoptedby1986.Thismaydifferslightlyfromtheimpliedchangeintheproportionofhospitalsthathavethetechnology
between1980and1986infig.5becausethatfigureisdoneinacrosssectionwhereastheestimatesinthetablearedoneinapanel,andhospitalsoccasionallychangetheirreportfromhavingtonot
havingatechnology;wehaveverifiedthattheresultsinthistablearerobusttoalternativewaysoftreatingthismeasurementerror.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
866 journal of political economy
production function (1) is constant. In this case, PPS-induced technology
adoption may increase managerial effort (a component of ) thatzi
is complementary to technology, thus reducing the cost of adopting
other, non-Medicare technologies. Perhaps more plausibly, PPS may also
induce a switch in managerial effort from Medicare-related activities
that have now become less profitable to non-Medicare activities, potentially
inducing the adoption of non-Medicare technologies. Of course,
a more prosaic explanation for the apparent “spillover” may be that, in
practice, Medicare’s cost-based reimbursement rules permitted hospitals
considerable latitude in determining which costs to assign to Medicare
(OTA 1984; CBO 1988), allowing some degree of fungibility in the
reimbursement of capital expenses.
C. Changes in Skill Composition
Finally, result 3 suggests that when technology (or capital) is more complementary
to skilled than to unskilled labor, the induced increase in
technology (or in the capital-labor ratio) should cause a change in the
composition of the workforce toward more skilled employees. We can
identify full-time-equivalent employment of two types of nurses in the
data, registered nurses (RNs) and licensed practical nurses (LPNs); together
these constitute about one-quarter of total hospital employ-
ment.26
RNs are considerably more skilled than LPNs.27
Column 5 of table 3 shows that the introduction of PPS appears to
be associated with an increase in the proportion of nurses that are
relatively more skilled (the RNs). These results are somewhat weaker
than our previous findings because there is evidence of preexisting
trends prior to PPS in the same direction as PPS and of magnitude
about half of that estimated for PPS. Overall, we interpret the finding
as broadly suggestive of a potential increase in the skill content of employment
associated with the induced increase in technology adoption.
Since, as discussed in the introduction, the existing view in the literature
is that capital and technology are more complementary to skilled labor
than to unskilled labor, evidence that PPS is associated with increases
in the skill composition of hospitals’ workforces provides indirect support
for our results concerning the effect of PPS on capital-labor ratios
and technology adoption.
26
The total amount of hospital employment accounted by nurses is about one-third,
but the other nursing categories do not have consistent names across years, making it
impossible for us to use them in this exercise.
27
RN certification requires about twice as many years of training as LPN certification,
which is reflected in the approximately 50 percent higher hourly wage of RNs relative to
LPNs. We are grateful to Doug Staiger for providing us with the estimates of hourly wages
by occupation from the 2000 Merged Outgoing Rotation Groups of the Current Population
Survey.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 867
D. Alternative Interpretations
We have so far offered our preferred interpretation that the observed
changes in factor demands and technology are a response to the change
in relative factor prices induced by PPS. There are a number of alternative
interpretations for the results reported so far. But the evidence
suggests that these are less plausible than our preferred interpretation.
One potential alternative is that the increase in the depreciation
share, documented in table 2, may be a mechanical effect. Depreciation
is a backward-looking measure, and thus the ratio of depreciation to
operating expenses may mechanically increase in response to a proportional
scaling back of capital and labor inputs. But in practice, we
see no scaling back of capital inputs, and moreover, this alternative
explanation would suggest that the effect should attenuate over time,
whereas the results in column 3 of table 2 indicate that the effect appears
to grow over time. Another related concern would be that the PPSinduced
reduction in hospitals’ Medicare share (see col. 3 of table 3)
could mechanically cause an increase in the capital-labor ratio if Medicare
patients are treated in a less capital-intensive manner than nonMedicare
patients. Empirically, however, Medicare patients appear to be
more capital intensive than non-Medicare patients: in the 1983 cross
section, hospitals with a higher Medicare share have a statistically significantly
higher depreciation share, with or without controlling for a
rich set of covariates. In addition, neither of these two “mechanical”
explanations is consistent with the evidence of PPS-induced changes in
technology adoption and skill composition.
Another possible interpretation is that the increase in the capitallabor
ratio may partly reflect a strategic response by hospitals to the
possibility that capital reimbursement may be made prospective in the
future; if so, hospitals may wish to build up their historical capital costs
to increase their future prospective capital reimbursement rates. The
incentive for such a strategic response is not obvious, however, since it
was not a priori clear if and when capital reimbursement would be made
prospective, nor how or whether a hospital’s own historical costs would
affect any prospective reimbursement rates (see, e.g., GAO 1986; CBO
1988). Moreover, to the extent that the response reflects the results
from such “gaming,” we might expect it to occur predominantly—or at
least disproportionately—on the more easily manipulatable financing
dimension (e.g., interest expenditures or leveraging) rather than on
the depreciation share per se. However, we found no evidence that PPS
is associated with an increase in debt financing of capital expenditures
(“leveraging up”).28
Finally, this type of gaming response should also
28
These results are reported in Acemoglu and Finkelstein (2006).
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
868 journal of political economy
not translate into effects on other margins, such as technology adoption
or the skill composition of the workforce.29
One potential concern with the technology adoption results is the
presence of secular increases in medical technology during this time
period. Since the elderly are among the most intensive users of medical
technology, there may be a spurious association between Medicare share
and technology adoption trends. However, several of the technologies
for which we find an impact of PPS are in fact not diffusing over our
sample period (see fig. 5). Most important, the results from our prespecification
test ( ) in tables 4 and 5 show no systematic evidenced 7 m1983 i
of differential trends in technology adoption across hospitals with different
Medicare shares before the introduction of PPS.
An alternative interpretation for our technology findings is that PPS
reimbursement for Medicare patients (i.e., the price subsidy) is not fully
prospective (McClellan 1996, 1997); the reimbursement a hospital receives
for a Medicare patient varies on the basis not only of the patient’s
diagnosis, but also, in some cases, of the type of treatment he or she
receives, particularly the type of surgery, if any. These features may have
increased hospitals’ incentives to perform these surgeries and consequently
induced them to adopt the technologies needed to perform
them. However, the evidence suggests that this type of incentive effect
is unlikely to be the driving factor behind our technology adoption
results since we find equally strong results for procedures that are not
reimbursed more generously after PPS. As noted by McClellan (1996),
for ad hoc reasons, there are separate reimbursement rates for patients
who have a heart attack if they undergo percutaneous transluminal
coronary angioplasty or coronary artery bypass graft, but not if they
spend time in the cardiac care unit (CCU). Hazard models estimated
for adoption of the CCU show that the introduction of PPS is associated
with an increased rate of adoption of the CCU even though this was
not a technology whose use was associated with any increased reimbursement
rate.30
Moreover, there is no evidence that hospitals vary
resources per patient in response to the subsequent changes in Medicare’s
relative reimbursement rates of various health services (though
29
Even if the effect is not merely a strategic one, the magnitude of the input response
may be affected by hospitals’ expectations that continued reimbursement of capital costs
might be temporary. A priori, however, it is not clear how such expectations (even if they
were important) would affect magnitudes. On the one hand, the response might be larger
because the relative subsidy to capital is expected to be temporary and hospitals may
attempt to incur and pass through their capital costs while they still can. On the other
hand, if there are adjustment costs, the response may be smaller than the case in which
the change in the regulatory regime is expected to be permanent.
30
Information on whether a hospital has a CCU is available from 1980–85 (see App.
table A1). The other technology adoption results in table 5 are robust to excluding 1986
from the data.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 869
there is evidence of nominal responses, so-called up-coding; see, e.g.,
Dafny 2005).
Finally, since PPS applied only to hospital inpatient expenditures, it
may have encouraged a real or nominal reallocation of some inpatient
hospital activity to outpatient or nursing home units of the hospital.
Previous empirical evidence suggests that PPS was, in fact, associated
with reallocation of some inpatient hospital activity to hospital outpatient
units, although there is no evidence of reallocation to nursing
homes (Coulam and Gaumer 1991). However, any such reallocations
within the hospital cannot explain our findings regarding the impact
of PPS on input and technology choices since our input and technology
measures are inclusive of hospital-based outpatient units and hospitalbased
nursing home facilities.31
Of course, if some activities were spun
off out of the hospital completely, these might potentially contribute to
our estimated decline in labor inputs. Nevertheless, they cannot explain
the estimated increase in technology adoption. Moreover, the empirical
evidence is not suggestive of such spinoff behavior. We find no evidence
that the introduction of PPS is associated with an increased probability
of having a hospital-based nursing home unit (results not reported),
and although we cannot use our empirical strategy to investigate the
impact of PPS on the creation of freestanding nursing home facilities,
there is no evidence in the time series of an increase in total nursing
home use over our time period (HCFA 1999, 178).
VII. Robustness Checks
Table 6 reports the results for some of the robustness checks we performed
on the depreciation share (capital-labor ratio) and the number
of facilities (one of the measures of technology adoption).32
Column 1
reproduces the baseline results from estimating equation (3). To investigate
the concern that our results may be spuriously picking up
underlying differential trends by hospitals with different pre-PPS Medicare
shares, our first robustness exercise adds an interaction between
the Medicare share (in 1983), , and a linear trend (i.e., in terms ofmi
our estimating equations above, the vector of covariates now includesXit
). The estimates in column 2 show that our main results are robustm 7 ti
to the inclusion of this linear trend.
A related but different concern is that of mean reversion. In particular,
if high–Medicare share hospitals are adjusting back to some hospital-
31
The estimated impact of PPS on the capital-labor ratio looks quite similar if we instead
use measures of capital and labor that exclude any inputs used in a hospital-based nursing
home (results not reported).
32
Additional robustness analyses are presented and discussed in Acemoglu and Finkelstein
(2006).
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
TABLE6
RobustnessAnalysis:AlternativeSpecifications
Base
Case
(1)
Add
Linear
Trend
#mi
(2)
AddYear
Dummies#
Dependent
Variable
in1982
(3)
First
Differences
(4)
Instrument
formi
withPast
Values
(5)
UsingOnly
Within
Variation
(6)
Using
Only
Between
Variation
(7)
A.DependentVariable:CapitalLaborRatio(DepreciationShare)
POSTTREND#mi.532
(.053)
.633
(.114)
.657
(.097)
.532
(.059)
.511
(.052)
.176
(.092)
.787
(.057)
d#m1983iϪ.060
(.088)
.040
(.110)
Ϫ.068
(.182)
.064
(.079)
Ϫ.131
(.087)
.041
(.146)
Ϫ.129
(.101)
Observations43,18843,18843,04136,90042,42843,18843,188
B.DependentVariable:NumberofFacilities
POST#mi2.501
(.401)
4.254
(.731)
2.267
(.377)
2.478
(.427)
3.010
(.383)
2.418
(.632)
2.561
(.494)
d#m1983iϪ.467
(.354)
.410
(.459)
Ϫ.566
(.347)
.688
(.333)
Ϫ.204
(.353)
.135
(.553)
Ϫ.896
(.473)
Observations43,18843,18843,04136,90042,42843,18843,188
Note.—Thetablereportsresultsfromestimatingeqq.(3)and(5)byOLS.Allregressionsincludehospitalandyearfixedeffects.Thedependentvariableisgiveninthepanelheading.The
variablePOSTisanindicatorvariablefortheyears1984–86.ThevariablePOSTTRENDp0through1983andthentakesthevalues1,2,and3in1984,1985,and1986,respectively.Thevariable
isanindicatorvariablefortheyear1983.ThevariablemeasurestheMedicareshareofthehospital’sinpatientdaysin1983.Thebasecaseincludesyearandhospitalfixedeffects.Columndm1983i
2addsalineartimetrendinteractedwithMedicaresharein1983tothebasecase.Column3addsyeardummiesinteractedwithlogtotalexpendituresin1982tothebasecase.Column4redoes
thebasecaseinfirstdifferencesinsteadoffixedeffects.Column5instrumentsforMedicaresharein1983withpastvaluesofthehospital’sMedicareshare(specifically,thevaluesin1980,1981,
and1982).Column6addsafullsetofyeardummiesinteractedwitheachofthethreecategoricalvariablesthatcanmakeahospitalexemptfromPPS(federalownership,long-termstays,or
specialtyhospital)sothattheidentificationofthevariablesofinterestcomesonlyfromwithin–hospitaltypevariationinMedicareshare.Column7instrumentsforPOSTTREND#andmi
withafullsetofyeardummiesinteractedwiththethreecategoricalvariablesthatcanmakeahospitalexemptfromPPS,sothattheidentificationofthevariablesofinterestcomesd#m1983i
onlyfrombetween–hospitaltypevariationinMedicareshare.Standarderrorsareinparenthesesandareadjustedtoallowforanarbitrarycovariancematrixwithineachhospitalovertime.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 871
specific equilibrium level, this may be picked up by our post-PPS#
Medicare share interaction. To investigate this issue, column 3 interacts
the value of the dependent variable for each hospital in 1982 with a
full set of year dummies. This specification thus flexibly controls for any
mean-reverting dynamics as well as any potential differential trends that
depend on hospital baseline characteristics (e.g., based on technology
levels). The estimates are remarkably similar to the baseline and show
no evidence that mean reversion or differential trends based on pretreatment
characteristics had any significant effect on our results.
As another check on the serial correlation properties of the error
term and patterns of mean reversion, column 4 estimates the model in
first differences rather than in levels. This specification is also useful as
a check on the strict exogeneity assumption necessary for consistency
of the fixed-effects estimator (Wooldridge 2002, 284) and on the potential
importance of measurement error in the data (Griliches and
Hausman 1986). For the depreciation share, the first-differenced results
in column 4 are very similar to the baseline results. However, for the
number of facilities, the results now show a pre-PPS effect of the same
sign as the estimated PPS effect that is significant at 5 percent. However,
since this is the only specification among many in which we find a samesigned
significant pre-PPS effect for the number of facilities, we interpret
this as partly driven by sampling variability.
Column 5 deals with concerns about measurement error in our key
variable, the Medicare share, by instrumenting for the 1983 Medicare
share with past values. The results are again similar to the baseline
estimates.
We explored the heterogeneity in the estimated effect of PPS based
on the type of variation in used to identify its effects. Recall thatmi
federally owned hospitals, long-term hospitals, and certain specialty
hospitals—together totaling 15 percent of all hospitals—were exempted
from PPS and were coded as having a zero Medicare share
(see fig. 4). We explored how the estimated effect of PPS varies depending
on whether we use the variation in Medicare share provided
by these exempt hospitals to identify the effect of PPS. In column 6,
we add a full set of year dummies interacted with each of the three
categories that provide an exemption from PPS to equations (3) and
(5). As a result, identification of the effect of PPS comes only from
within–hospital type variation in , and the three types of hospitalsmi
that are exempt from Medicare PPS are not used to estimate its impact.
Column 7 presents the complementary approach in which identification
of PPS comes only from between–hospital type variation in
. Here, we instrument the interaction terms andm d 7 mi 1983 i
or with the full set of year dum-POSTTREND 7 m POSTTREND 7 mt i t i
mies interacted with each of the three exemption categories. The reThis
content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
872 journal of political economy
sults indicate that the basic findings are robust to using either the
within or between variation, although the estimated impact of PPS on
the capital-labor ratio is substantially larger using the between variation
than the within variation.
We also briefly explored potential heterogeneity in the impact of PPS
across hospitals of different ownership types and across hospitals facing
different degrees of competition (again results not shown here). The
estimated impact of PPS appears to be quite similar across publicly
owned, for-profit, and nonprofit hospitals. It is also quite similar across
hospitals that are monopoly providers in their county compared to hospitals
that face one or more competitor hospitals in their county. The
one exception is that the technology adoption effects of PPS appear to
be more pronounced in publicly owned hospitals (results not shown).
Finally, we considered the sensitivity of our results to allowing for
differential trends in different parts of the country. In particular, since
the price subsidy of Medicare PPS was phased in over a 4-year period
as a combination of hospital-specific historical rates, regional average
rates, and national rates (CBO 1988; Staiger and Gaumer 1990), regional
differences in the level of the price subsidy might contribute to differential
regional effects of PPS. We therefore verified that the results are
not affected by including a full set of interactions between the (nine)
census region dummies and year effects (not shown). Another potential
source of confounding time-varying geographic factors is that seven
states—constituting about one-fifth of the hospitals in our sample—
experienced some change in their Certificate of Need (CON) laws during
our time period. We verified that our results are robust in both
statistical significance and magnitude to excluding these seven states
(not shown).33
VIII. Conclusions
This paper has investigated the impact of regulatory change on firm
input mix and technology choices, focusing on the introduction of the
Medicare Prospective Payment System in the United States. This reform
changed the reimbursement for Medicare-related inpatient hospital expenses
from a full cost reimbursement system for both labor and capital
inputs to a partial cost reimbursement system and thereby raised the
relative price of labor.
We argued that the introduction of PPS increased the price of labor
faced by hospitals, in particular, for hospitals with a high share of Medicare
patients, while leaving the price of capital unchanged. A simple
33
More details on the CON reforms can be found at http://www.ncsl.org/programs/
health/cert-need.htm.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 873
neoclassical framework indicates that this change in relative prices
should lead to an increase in the capital-labor ratio. More interestingly,
our framework shows that PPS could also increase the overall demand
for capital and induce technology adoption.
Consistent with these implications, our empirical results suggest that
the PPS reform is associated with an increase in the capital-labor ratio.
This decline stems mainly from a decline in labor inputs. We also found
that the introduction of PPS is associated with a significant increase in
the adoption of a range of new health care technologies. This increase
in technology adoption would be predicted when there is a relatively
high degree of substitutability between technology and hospital labor.
We presented suggestive evidence of technology-labor substitution working
through declines in the length of stay. We also found an increase
in the skill composition of these hospitals, which is consistent with technology-skill
(or capital-skill) complementarities.
Our empirical findings suggest that relative factor prices are an important
determinant of technology diffusion in the hospital sector and
perhaps in other sectors as well. They raise an interesting question for
further research of whether other factors that increase the relative
price of labor for hospitals, such as labor unions or the tax treatment
of capital expenditures, also encourage capital deepening and technology
adoption.
Our findings do not address the efficiency and welfare consequences
of the increase in relative labor prices associated with a move to PPS.
These depend in part on the relative generosity of labor and capital
reimbursement under the previous full cost reimbursement system and
also on other preexisting distortions in the health care sector. However,
the evidence that the move to PPS was not associated with substantial
adverse health outcomes (Staiger and Gaumer 1990; Cutler 1995) suggests
that at least on the health dimension there were no substantial
adverse welfare effects.
It is also worth emphasizing that our findings regarding technology
adoption run counter to the general expectation that PPS would likely
reduce the pace of technology adoption (Sloan et al. 1988; Coulam and
Gaumer 1991; Weisbrod 1991). Such expectations were formed by considering
PPS as a full price cap system and hence overlooked the associated
relative factor price changes. This highlights the potential importance
of the details of regulation policy in determining its ultimate
impact. The particular detail we have focused on—the exemption of
capital costs from prospective reimbursement systems—appears to be
quite common in practice. For example, the subsequent Medicare prospective
payment reimbursement system for home health care also excluded
capital-related costs (in particular, “durable medical equipment”
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
874 journal of political economy
such as hospital beds or oxygen equipment) from the price cap (Federal
Register, July 3, 2000).
Naturally, our empirical results directly speak only to the impact of
regulatory change in the hospital sector. It is possible that the health
care sector is not representative of regulatory effects in other sectors,
for example, because most hospitals are nonprofit or public entities or
because the health care sector is significantly less capital intensive than
many other regulated industries (recall n. 13). Nevertheless, the ideas
presented here should also apply to other regulated industries, many
of which operate under some form of partial cost reimbursement (see
Joskow 2005). In light of this, we may expect that a switch to partial
cost reimbursement in other regulated industries may have also increased
capital-labor ratios and perhaps even encouraged technology
adoption, though we are not aware of any direct evidence on this. An
investigation of the response of input and technology choices to similar
regulatory changes in other industries is another interesting area for
future research and would be particularly useful for understanding the
extent to which the results presented here generalize to other industries.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
Appendix A
TABLE A1
Description of 113 Binary Facilities in the Data for 1980–86
Facility Description Years in Data
Sample
Mean
Abortion services (inpatient or outpatient) 1980–85 .22
Adult day care 1986 .05
AIDS services 1986 .28
Alcoholism/chemical dependency acute and
subacute inpatient care 1980–85 .26
Alcoholism/chemical dependency services
(outpatient) 1981–86 .17
Ambulance services 1980–81 .17
Ambulatory surgical services 1981–86 .80
Anesthesia service 1980–81 .72
Autopsy services 1980–81 .47
Birthing room 1985–86 .44
Blood bank 1980–86 .64
Burn care 1980–85 .09
Cancer tumor registry 1980–81 .30
Cardiac catheterization 1980–86 .16
Cardiac intensive care 1980–85 .67
Chaplaincy services 1980–85 .55
Clinical psychology services 1980–86 .33
Community health promotion 1986 .54
Continuing care case management 1986 .15
Contraceptive care 1986 .09
CT scanner (head or body unit) 1980–86 .34
Day hospital 1981–86 .17
Dental services 1980–85 .48
Diagnostic radioisotope facility 1980–86 .56
Diagnostic x-ray 1985–86 .89
Electrocardiography 1980–85 .91
Electroencephalography 1980–81 .50
Electromyography 1980–81 .27
Emergency department 1981–86 .85
Extracorporeal shock wave lithotripter 1985–86 .02
Family planning 1980–85 .10
Fertility counseling 1986 .09
Fitness center 1986 .09
General laboratory services 1980–81, 1984–85 .88
General surgical services 1980–81, 1983–85 .87
Genetic counseling 1980–86 .06
Genetic screening 1986 .06
Geriatric acute-care unit 1986 .12
Geriatric assessment services, comprehensive 1982–86 .13
Geriatric clinics, satellite 1986 .02
Health promotion 1981–85 .40
Hemodialysis (home care/mobile unit) 1980–81 .04
Hemodialysis services (inpatient or outpatient) 1980–86 .21
Histopathology services 1980–86 .56
Home care program 1980–86 .18
Hospice 1980–86 .08
Hospital auxiliary 1980–86 .75
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
TABLE A1
(Continued)
Facility Description Years in Data
Sample
Mean
Intermediate care for mentally retarded 1980–85 .03
Intermediate care, other 1980–85 .13
Intravenous admixture services 1980–85 .71
Intravenous therapy team 1980 .25
Medical library 1980–81 .84
Medical/surgical acute care 1980–85 .91
Medical/surgical intensive care 1980–85 .74
Megavoltage radiation therapy 1980–86 .14
Neonatal intensive care 1980–85 .09
Neurosurgery 1980–81 .29
Newborn nursery 1980–85 .70
Nuclear magnetic resonance facility 1983–86 .04
Obstetrical care 1980–85 .70
Occupational health services 1986 .23
Occupational therapy 1980–86 .40
Open-heart surgery 1980–86 .11
Optometric services 1981–85 .16
Organ bank 1980–81 .03
Organ transplant (including kidney) 1980–86 .05
Organized outpatient department 1981–86 .49
Patient education 1986 .67
Patient representative services 1980–86 .49
Pediatric acute care 1980–85 .75
Pediatric intensive care 1980–85 .18
Percutaneous lithotripsy 1985 .11
Pharmacy service (full- or part-time) 1980–85 .91
Pharmacy unit dose system 1980–85 .71
Physical therapy 1980–86 .79
Podiatric services (inpatient or outpatient) 1980–85 .31
Postoperative recovery room 1980–82 .83
Premature nursery 1980–85 .26
Psychiatric acute care 1980–85 .36
Psychiatric consultation and education 1980–86 .29
Psychiatric emergency services 1981–86 .32
Psychiatric foster and/or home care program 1980–86 .03
Psychiatric intensive care 1980–82 .13
Psychiatric liasion services 1983–86 .16
Psychiatric long-term care 1980–85 .06
Psychiatric outpatient services 1981–86 .18
Psychiatric partial hospitalization program 1980–86 .13
Psychiatric services, pediatric 1981–86 .14
Pulmonary function laboratory 1980–81 .58
Radioactive implants 1980–86 .20
Recreational therapy 1980–86 .30
Rehabilitation 1980–85 .30
Rehabilitation services (outpatient) 1981–86 .32
Residential care 1980 .05
Respiratory therapy 1980–86 .81
Respite care 1986 .09
Self care 1980–85 .06
Sheltered care 1981–85 .02
Skilled nursing, long-term 1980–85 .16
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 877
TABLE A1
(Continued)
Facility Description Years in Data
Sample
Mean
Social work services 1980–85 .77
Speech therapy 1980–86 .36
Special care, other 1981–85 .22
Sports medicine clinic/service 1986 .11
Sterilization 1986 .23
Therapeutic radioisotope facility 1980–86 .21
Toxicology/antidote information 1980–81 .38
Trauma center 1984–86 .17
Tuberculosis and other respiratory diseases 1980–86 .34
Ultrasound 1981–86 .69
Volunteer services 1980–86 .65
Women’s center 1986 .09
Work site health promotion 1986 .35
X-ray radiation therapy 1980–86 .16
Note.—All facilities are coded directly from a single variable in the data except for neonatal intensive care unit,
where we followed the coding procedure of Baker and Phibbs (2002), and the following seven variables, which we
generated as a consistent series using combinations of different variables in different years: (1) Abortion services
(inpatient or outpatient): coded 1 in 1980 and 1981 if the hospital reports having either inpatient abortion services or
outpatient abortion services or both; coded 1 in 1982–85 if the hospital reports having abortion services. (2) Alcoholism/
chemical dependency acute and subacute inpatient care: coded 1 in 1984 if the hospital reports having alcohol/chemical
dependency acute inpatient care or alcohol/chemical dependency subacute inpatient care or both; coded 1 in 1980–
83, 1985 if hospital reports having alcohol/chemical dependency inpatient care. (3) CT scanner (head or body unit):
coded 1 in 1980 and 1981 if the hospital reports having either a CT scanner head unit or a CT scanner body unit or
both; coded 1 in 1982–86 if the hospital reports having a CT scanner. (4) Pharmacy service (full- or part-time): coded
1 in 1980 or 1981 if the hospital reports having either a full-time or a part-time pharmacist or both; coded 1 in 1982–
85 if the hospital reports having pharmacy services. (5) Hemodialysis services (inpatient or outpatient): coded 1 in
1980 and 1981 if the hospital reports having either hemodialysis inpatient services or hemodialysis outpatient services
or both; coded 1 in 1982–86 if hospital reports having hemodialysis services. (6) Organ transplant (including kidney):
coded 1 in 1980–85 if the hospital reports having either organ transplant capability (other than kidney) or kidney
transplant capability or both; coded 1 in 1986 if hospital reports having organ transplant capability (including kidney).
(7) Podiatric services (inpatient or outpatient): coded 1 in 1981 if the hospital reports having inpatient or outpatient
podiatric services or both; coded 1 in 1980, 1982–85 if hospital reports having podiatric services.
References
Acemoglu, Daron. 2002. “Technical Change, Inequality and the Labor Market.”
J. Econ. Literature 40 (1): 7–72.
Acemoglu, Daron, and Amy Finkelstein. 2006. “Input and Technology Choices
in Regulated Industries: Evidence from the Health Care Sector.” Working
Paper no. 12254 (May), NBER, Cambridge, MA.
Acemoglu, Daron, and Joshua Linn. 2004. “Market Size in Innovation: Theory
and Evidence from the Pharmaceutical Industry.” Q.J.E. 119 (3): 1049–90.
AHA (American Hospital Association). 1983. Estimated Useful Lives of Depreciable
Hospital Assets. Chicago: American Hospital Assoc.
Antos, Joseph. 1993. “Waivers, Research, and Health System Reform.” Health
Affairs 12 (1): 179–83.
Armstrong, Mark, Simon Cowen, and John Vickers. 1994. Regulatory Reform: Economic
Analysis and British Experience. Cambridge, MA: MIT Press.
Autor, David, Lawrence F. Katz, and Alan B. Krueger. 1998. “Computing Inequality:
Have Computers Changed the Labor Market?” Q.J.E. 113 (November):
1169–1214.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
878 journal of political economy
Averch, Harvey, and Leland Johnson. 1962. “Behavior of the Firm under Regulatory
Constraint.” A.E.R. 52 (5): 1053–69.
Baicker, Katherine, and Douglas Staiger. 2005. “Fiscal Shenanigans, Targeted
Federal Health Care Funds, and Patient Mortality.” Q.J.E. 120 (1): 345–86.
Baker, Laurence. 1997. “The Effect of HMOs and Fee-for-Service Health Care
Expenditures: Evidence from Medicare.” J. Health Econ. 16 (4): 453–81.
———. 2001. “Managed Care and Technology Adoption in Health Care: Evidence
from Magnetic Resonance Imaging.” J. Health Econ. 20 (3): 395–421.
Baker, Laurence, and Ciaran Phibbs. 2002. “Managed Care, Technology Adoption,
and Health Care: The Adoption of Neonatal Intensive Care.” Rand J.
Econ. 33 (3): 524–48.
Baumol, William, and Alvin Klevorick. 1970. “Input Choices and Rate-of-Return
Regulation: An Overview of the Discussion.” Bell J. Econ. and Management Sci.
1 (1): 162–90.
Berman, Eli, John Bound, and Zvi Griliches. 1994. “Changes in the Demand for
Skilled Labor within U.S. Manufacturing: Evidence from the Annual Survey
of Manufacturers.” Q.J.E. 109 (May): 367–97.
Cawley, John, David Grabowski, and Richard Hirth. 2006. “Factor Substitution
in Nursing Homes.” J. Health Econ. 25 (2): 234–47.
CBO (Congressional Budget Office). 1988. “Including Capital Expenses in the
Prospective Payment System.” Report, Congressional Budget Off., Washington,
DC.
Cotterill, Philip. 1991. “Prospective Payment for Medicare Hospital Capital: Implications
of the Research.” Health Care Financing Rev. (suppl.): 79–86.
Coulam, Robert, and Gary Gaumer. 1991. “Medicare’s Prospective Payment System:
A Critical Appraisal.” Health Care Financing Rev. (suppl.): 45–77.
Cutler, David. 1995. “The Incidence of Adverse Medical Outcomes under Prospective
Payment.” Econometrica 63 (1): 29–50.
———. 2003. Your Money or Your Life: Strong Medicine for America’s Health Care
System. New York: Oxford Univ. Press.
Cutler, David, and Louise Sheiner. 1998. “Managed Care and the Growth of
Medical Expenditures.” In Frontiers in Health Policy Research, edited by Alan
Garber. Cambridge, MA: MIT Press.
Cutler, David, and Richard Zeckhauser. 2000. “The Anatomy of Health Insurance.”
In Handbook of Health Economics, vol. 1A, edited by Anthony Culyer and
Joseph Newhouse. Amsterdam: Elsevier.
Dafny, Leemore. 2005. “How Do Hospitals Respond to Price Changes?” A.E.R.
95 (5): 1525–47.
Feder, Judith, Jack Hadley, and Stephen Zuckerman. 1987. “How Did Medicare’s
Prospective Payment System Affect Hospitals?” New England J. Medicine 317
(14): 867–73.
Finkelstein, Amy. 2004. “Static and Dynamic Effects of Health Policy: Evidence
from the Vaccine Industry.” Q.J.E. 119 (2): 527–64.
———. 2007. “The Aggregate Effects of Health Insurance: Evidence from the
Introduction of Medicare.” Q.J.E. 122 (1): 1–37.
Fuchs, Victor. 1996. “Economics, Values, and Health Care Reform” (presidential
address of the American Economic Association). A.E.R. 86 (1): 1–24.
GAO (General Accounting Office). 1986. “Medicare: Alternative for Paying Hospital
Capital Costs.” Report, Gen. Accounting Off., Washington, DC.
Greenstone, Michael. 2002. “The Impacts of Environmental Regulation on Industrial
Activity: Evidence from the 1970 and 1977 Clean Air Act Amendments
and the Census of Manufactures.” J.P.E. 110 (6): 1175–1219.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
input and technology choices 879
Griliches, Zvi. 1956. “Capital-Skill Complementarity.” Rev. Econ. and Statis. 51
(3): 465–68.
Griliches, Zvi, and Jerry Hausman. 1986. “Errors in Variables in Panel Data.” J.
Econometrics 31 (1): 93–118.
Habakkuk, H. J. 1962. American and British Technology in the Nineteenth Century:
Search for Labor Saving Inventions. New York: Cambridge Univ. Press.
Hausman, Jerry, Bronwyn Hall, and Zvi Griliches. 1984. “Econometric Models
for Count Data with an Application to the Patents-R&D Relationship.” Econometrica
52 (4): 909–38.
HCFA (Health Care Financing Administration). 1986. Status Report: Research and
Demonstrations in Health Care Financing. Baltimore: U.S. Dept. Health and Human
Services.
———. 1987. Status Report: Research and Demonstrations in Health Care Financing.
Baltimore: U.S. Dept. Health and Human Services.
———. 1999. Statistical Supplement. Baltimore: U.S. Dept. Health and Human
Services.
Hicks, John. 1932. The Theory of Wages. London: Macmillan.
Joskow, Paul. 2005. “Incentive Regulation in Theory and Practice: Electricity
Distribution and Transmission Networks.” Manuscript, Dept. Econ., Massachusetts
Inst. Tech.
Kiefer, Nicholas. 1988. “Economic Duration Data and Hazard Functions.” J. Econ.
Literature 26 (2): 646–79.
Kling, Jeffrey, and Jeffrey Liebman. 2004. “Experimental Analysis of Neighborhood
Effects on Youth.” Manuscript, Dept. Econ., Princeton Univ.
Krusell, Per, Lee Ohanian, Victor Rios-Rull, and Giovanni Violante. 2000. “Capital-Skill
Complementarity and Inequality: A Macroeconomic Analysis.” Econometrica
68 (5): 1029–53.
Laffont, Jean-Jacques, and Jean Tirole. 1993. A Theory of Incentives in Procurement
and Regulation. Cambridge, MA: MIT Press.
McClellan, Mark. 1996. “Medicare Reimbursement and Hospital Cost Growth.”
In Advances in the Economics of Aging, edited by David Wise. Chicago: Univ.
Chicago Press.
———. 1997. “Hospital Reimbursement Incentives: An Empirical Analysis.” J.
Econ. and Management Strategy 6 (1): 91–128.
MedPAC (Medicare Payment Advisory Commission). 1999. “Report to Congress:
Medicare Payment Policy.” http://www.medpac.gov.
Newell, Richard, Adam Jaffe, and Robert Stavins. 1999. “The Induced Innovation
Hypothesis and Energy-Saving Technological Change.” Q.J.E. 114 (3): 941–
75.
Newhouse, Joseph. 1992. “Medicare Care Costs: How Much Welfare Loss?” J.
Econ. Perspectives 6 (3): 3–21.
———. 2002. Pricing the Priceless: A Health Care Conundrum. Cambridge, MA: MIT
Press.
OTA (Office of Technology Assessment). 1984. “Medical Technology and the
Costs of the Medicare Program.” Report, Government Printing Off., Washington,
DC.
———. 1985. “Medicare’s Prospective Payment System: Strategies for Evaluating
Cost, Quality, and Medical Technology.” Report, Government Printing Off.,
Washington, DC.
PPAC (Prospective Payment Assessment Commission). 1988. “Medicare Prospective
Payment and the American Health Care System: Report to the Congress.”
Report, Prospective Payment Assessment Comm., Washington, DC.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions
880 journal of political economy
———. 1990. “Medicare Prospective Payment and the American Health Care
System: Report to the Congress.” Report, Prospective Payment Assessment
Comm., Washington, DC.
———. 1992. “Medicare Prospective Payment and the American Health Care
System: Report to the Congress.” Report, Prospective Payment Assessment
Comm., Washington, DC.
Sloan, Frank. 2000. “Not-for-Profit Ownership and Hospital Behavior.” In Handbook
of Health Economics, vol. 1B, edited by Anthony Culyer and Joseph Newhouse.
Amsterdam: Elsevier.
Sloan, Frank, Michael Morrissey, and Joseph Valvona. 1988. “Medicare Prospective
Payment and the Use of Medical Technologies in Hospitals.” Medical Care
26 (9): 837–53.
Staiger, Douglas, and Gary Gaumer. 1990. “The Impact of Financial Pressure on
Quality of Care in Hospitals: Post-admission Mortality under Medicare’s Prospective
Payment System.” Report, Abt Assoc., Cambridge, MA.
Staiger, Douglas, Joanne Spetz, and Ciaran Phibbs. 1999. “Is There Monopsony
in the Labor Market? Evidence from a Natural Experiment.” Working Paper
no. 7258 (July), NBER, Cambridge, MA.
Weisbrod. Burton. 1991. “The Health Care Quadrilemma: An Essay on Technological
Change, Insurance, Quality of Care, and Cost Containment.” J. Econ.
Literature 29 (2): 523–52.
Westfall, Peter H., and Stanley S. Young. 1993. Resampling-Based Multiple Testing:
Examples and Methods for P-Value Adjustment. New York: Wiley.
Wooldridge, Jeffrey. 2002. Econometric Analysis of Cross Section and Panel Data.
Cambridge, MA: MIT Press.
Yin, Wesley. 2005. “Do Market Incentives Generate Innovation or Balkanization?
Evidence from the Market for Rare Disease Drugs.” Manuscript, Dept. Econ.,
Univ. Chicago.
This content downloaded from 147.251.189.14 on Mon, 6 Jul 2015 05:51:06 AM
All use subject to JSTOR Terms and Conditions