Data and summary statistics

In this study, we investigate the impact of CON laws on inpatient market competition in the United States. Before presenting our empirical analysis, we provide theoretical arguments to establish why such an analysis is worthwhile.

We measure market competition using the normalized Herfindahl–Hirschman Index (HHI) in each health referral region (HRR). We develop these HHI indices using inpatient volume (admissions) data from each acute care hospital in an HRR. The HHI is defined as $10,000\mathop{\sum}\nolimits_{i{\equals}1}^n {s_{i}^{2} } $ , where n is the number of firms in that market and s _i the market share of the ith firm, and market share is calculated using the proportion of patients availing inpatient services at a specific hospital out of the total inpatient volume of that HRR in a given year; i=1, 2, … , n (Tirole, Reference Vaughan-Sarrazin, Hannan, Gormley and Rosenthal1988: 221–223).Footnote ³ It can range between ${{10,000} \over n}$ (lower limit, when all firms have equal share) and 10,000 (upper limit, when one firm is a monopoly). Building on this, we adopt a normalized HHI: $10,000 \ \Big( {{{\mathop{\sum}\nolimits_{i{\equals}1}^n {s_{i}^{2} } {\minus}{1 \over n}} \over {1{\minus}{1 \over n}}}} \Big).$ The normalized HHI measures the extent of inequality of market shares in the health care industry. Unlike the traditional version that has a variable lower limit, this normalized index takes a value between 0 and 10,000 with a higher value of the index representing less competition or greater concentration (monopolization) and vice versa. This provides for a consistent comparable range for all states and has been suggested in earlier work (Hirsch and Schumacher, Reference Hirsch and Schumacher2005; Owen et al., Reference Owen, Ryan and Weatherston2007).

Our outcome measure is the normalized HHI built using inpatient volume data of acute care hospitals in each HRR for the years 2000, 2002, 2004, 2006 and 2009.Footnote ⁴ Our key variables of interest are CON law and its stringency.

We use a binary variable to describe whether a state has a CON law or not. Given that there is a distinction between states that retained CON for long-term care (states such as Iowa, Massachusetts and Wisconsin) and those who retained any sort of CON, our indicator only captures those who have CON for acute care. Namely, our indicator of CON law is zero for states which only maintain CON law for long-term care. The stringency of CON law is measured by an index based on the CON threshold for service-related expenditures in the state, beyond which hospitals need to get permission from the state government if they plan to acquire a new service. A small value of this threshold indicates a relatively stringent CON law whereas a large value of this threshold indicates a relatively lenient version of the law. It is important to include a measure on this feature of CON law not only to differentiate between states with CON laws but also to differentiate between states with and without CON laws. This is mainly because there exist some states with CON laws that have a high threshold on new service-related spending, which means very few service acquisitions are subject to review. These states are therefore not very different from states without CON laws where no spending is subject to review. The index of the threshold on equipment spending is defined as follows:

$$\left( {{\rm Stringency}\,{\rm Index}_{{{\rm State}_i}} {\equals}\left[ {{{{\rm max}_{{j \in I}} {\rm CON}\,{\rm Threshold}_{{{\rm State}_{j} }} {\minus}{\rm CON}\,{\rm Threshold}_{{{\rm State}_{i} }} } \over {{\rm max}_{{j \in I}} {\rm CON}\,{\rm Threshold}_{{{\rm State}_{j} }} }}} \right]} \right).$$

A high value of this index implies a stringent environment and a low value indicates a lenient environment.

In general, competition in the health care market is affected by three broad kinds of factors: (i) supply side factors, (ii) demand side factors and (iii) political factors. The first two factors determine the profitability of operating in a market, and consequently have a role in explaining entry as well as capacity expansion by incumbents. Apart from these two, it is also important to control for political factors. Health care is a highly regulated industry, and regulation in a state depends on the political environment in a state. Below, we explain in greater detail how we control for political factors.

First, we discuss the supply side factors. The factors include important hospital characteristics, such as the number of teaching hospitals per 1000 population, number of physicians working in the ED in each state per 1000 population, number of beds in for-profit hospitals per 1000 population and the percentage of population that make use of ED services. The ED-related inputs are incorporated to account for the fact that a large proportion of inpatient volume comes from the ED and therefore the ED volume and number of ED physicians that impact ED throughput can have a significant impact on inpatient competition (Paul and Lin, Reference Paul, Ni and Bagchi2012). Specifically, given their impact on the patient turnaround rate, a shortage of emergency physicians might make it harder for smaller EDs to expand, adversely affecting their ability to serve greater patient volumes. This in turn can negatively impact inpatient volume and therefore competition. That is, other factors equal, we expect an increase in the number of emergency physicians to enhance inpatient competition. Similarly, because of the greater role of resident physicians, an increase in the proportion of teaching hospitals can negatively impact the turnaround rate, adversely affecting the market share of such facilities. If the slack is picked up by one large hospital, then the Herfindahl Index would increase because it is affected more by the share of large hospitals. On the other hand, if the slack is picked up by small hospitals, then the Herfindahl Index would decrease. Therefore, the effect of teaching hospitals on competition can go either way.

Next, we discuss demand side factors. For this, we construct measures to capture the health need of the population in a given market (state). First, we control for the demographic characteristics of the state population using measures on distributions of different age, gender and ethnicity groups. An increase in population in a state increases the demand for health care and hence is expected to enhance competition. Further, on average, females have more health care needs when compared with men, and hence an increase in their proportion might increase demand for inpatient care which in turn is expected to enhance competition. An increase in the proportion of the elderly similarly might enhance competition in the inpatient market by increasing the demand for care. Health outcomes and socio-economic status have been found to significantly vary by race (NCHS, 2012; Census Bureau, 2013) and therefore any substantial changes in the racial mix of the population can influence the demand for inpatient care and related competition.

Second, we include the proportion of population covered by different types of health insurance, which could be a natural measure of individuals’ accessibility to health care. For instance, if the number of individuals with employer-provided insurance increases, it can potentially lead to a reduction in demand for inpatient care as these individuals are generally younger and healthier. The opposite might hold true for those on Medicaid. They are likely to be one of the most vulnerable groups as well. Third, we measure the health status of a state population using the prevalence of obesity, percent of population that never smoked in the state and infant mortality rate (CDC, 2014a, 2014b, 2014c). For instance, infant mortality rate is a good indicator of health status as it has been found to be directly related to health of expecting women which in turn is influenced by their lifestyle and behavioral choices (CDC, 2014d). The positive relationship noted between obesity and smoking to health costs provides another motivation for considering them in our empirical models (NCHS, 2012).

We also control for the degree of economic prosperity in a state. This is because in a richer state, inpatient care providers have a stronger incentive to obtain a bigger share of the market and hence increase the degree of competition. We use median household incomeFootnote ⁵ and poverty rate to capture the degree of economic prosperity.

Finally, we control for political factors. The political environment of a state influences the policies implemented in the health care market which can either promote or discourage competition. We construct several measures on the political environment of a state: (1) party affiliation of the state governor; (2) voting records of the two senators from a state: both the number of affirmative votes on liberal issues and deviation of their votes. We include these variables in our study for the following reasons: first, voting record of senators is an indicator of the political climate in that state since rational senators would not vote in a manner that jeopardizes their chance of winning future elections. A higher number of affirmative votes on liberal issues therefore is an indication that the political climate of the state leans liberal. Also, a low deviation in the voting record of the senators is an indication of agreement between the senators and is possibly a result of a cooperative political climate in a state, either because one party is overwhelmingly dominant or because both parties enjoy a working relationship. Legislative changes can be made relatively easily in such a state. In such an environment, powerful (influential) groups with special interests can get laws passed or modified in their favor and such changes might not be conducive to competition. The inclusion of political variables in health care-based studies is supported by prior work such as Lanning et al. (Reference Lanning, Morrisey and Ohsfeldt1991). Lanning et al. (Reference Lanning, Morrisey and Ohsfeldt1991) focus on impact of CON law on rate setting whereas our study features market competition. We include them in our study as control variables since politics can influence the share of market power by relocating scarce resources and therefore competition in a state. Additional support for impact of politics on health care can be noted in articles by Bodenheimer (Reference Bodenheimer2005b) and Kaplan and Babad (Reference Kaplan and Babad2011).

To the best of our knowledge, there is no single data set that includes all of the variables we need for our analysis. Therefore, we collect information on these variables from a variety of sources (online Appendix Table A.1). We end up with a sample of 1395 observations at the HRR level, which covers the period of years: 2000, 2002, 2004, 2006 and 2009. We lost 130 observations (1525–1395) due to missing information on variables included in our study: number of teaching hospitals per 1000 population (30), thresholds of CON laws (50), infant mortality rate (12), senators’ opinions on proposals (35) and the beds owned by for-profit hospitals per 1000 population (3). We do not notice any pattern of missing information to raise concerns about sample selection.

Table 1 provides the summary statistics of our sample with a comparison between states with and without CON laws. A detailed description of the variables can be found in the online Appendix Table A.2.

Table 1

Summary statistics for variables

CON=Certificate of Need; ED=emergency department.

From Table 1, it seems the inpatient market is quite competitive since the HHI averages at 1200. Compared with HRRs that are in states without CON laws, those with CON laws on average have smaller populations, higher black population, lower proportion of people with no health insurance rate. A higher proportion of population use inpatient service in HRRs in states with CON, where we observe a lower proportion of beds are owned by for-profit hospitals. The governors of HRRs in states with CON are more likely to be Democratic, and their senators seem to be more cooperative as well.Footnote ⁶ HRRs in states with CON also have (1) higher Gini indices, which indicates a higher dispersion in income; (2) lower tech index, which indicates a slower speed to implement newer technology and (3) higher tax on beer, which implies the local government has a stronger incentive to promote healthy life style.Footnote ⁷

Methodology

Following the literature, we begin our analysis using a binary variable to indicate whether a state has CON law or not. In this model, we also control for other important variables as described previously that measure both the demand and supply side of inpatient care market, and the political environment of a state.Footnote ⁸ The estimated model is specified belowFootnote ⁹ :

(1) $${\rm HHI}\, {\equals}\, \alpha _{0} {\plus}\alpha _{1} {\rm CON}{\plus}{\mib{\alpha }} _{2} {\mib{X}} {\plus}{\epsilon},$$

where HHI measures inpatient market competition for each HRR, CON is a dummy variable that indicates if a state has the CON law or not, X includes all the other covariates and ε represents the error term.

Since the decision to retain the CON law in a state depends on state-specific characteristics, some of which are unobserved (state’s attitude towards inpatient care market competition for instance), we cannot rule out the possibility that the CON indicator is correlated with the error term in equation (1) without testing for it. That is, there is a possibility that the CON indicator may be an endogenous variable, which could result in biased estimates if we fail to tackle this issue properly. Therefore, we use a two-stage least squares (2SLS) modelFootnote ¹⁰ to estimate the effects of the CON laws treating CON law indicator as endogenous. We then use the Durbin–Wu test to investigate whether we have strong empirical evidence to believe that the CON indicator should be treated as an endogenous variable. The 2SLS estimation is done using the following specifications:

(2) $${\rm Stage}\,{\rm one\,\colon\,}\,{\rm CON}\, {\equals}\, \theta _{0} {\plus}{\mib{\theta }} _{1} {\mib{IV}} {\plus}{\mib{\theta }} _{2} {\mib{X}} {\plus}\zeta ,$$

(3) $${\rm Stage}\,{\rm two\,\colon\,}\;{\rm HHI}\, {\equals}\, \gamma _{0} {\plus}\gamma _{1} \,\widehat{{{\rm CON}}}{\plus} \gamma _{2} \mib X{\plus}\mu .$$

In the first stage, we estimate the likelihood of a state having CON law as a function of instrumental variables (IVs) and other covariates. And in the second stage, we estimate the market competition as a function of the estimated likelihood of having CON law from the first stage and other covariates. Theoretically, we need to include at least one IV for each endogenous variable in order to identify the model. In this specification, we use the index of science and technology in a state, the excise tax rate on beer and GINI index as our IVs.

These three variables are likely to influence whether a state has CON law but not likely to influence the inpatient care market competition for the following reasons: (1) the index of science and technology in a state is likely to be associated with the attitude of the state administration regarding how quickly new technologies can be adopted. A technologically advanced state will usually have a large technology sector that provides both jobs and taxes. Everything else remaining constant, such a state is less likely to have CON law since the law can hurt business interests in the state. Moreover, states worried that benefits yielding from such technology and innovation do not match the associated costs are highly likely to have the law. Extant literature supports these concerns that the efforts to innovate often overtake the effort to economize (Bodenheimer, Reference Bodenheimer2005a). As a result, these states are more likely to be behind those without the law from a technology and innovation standpoint. (2) The excise taxes on beer are a way to discourage drinking and hence prevent related chronic diseases in a state. As taxes paid per gallon of beer increase, the population will be motivated to pursue a healthier lifestyle in the form of reduced alcohol consumption. This would in turn reduce the state spending on population availing inpatient services on health issues as a consequence of drinking (NCHS, 2014). It could serve as a proxy for the attitude of a state towards promoting health of the state population, as well as towards reducing health care cost. A state with higher taxes on beer is more likely to have CON in effect because such a state is likely to be more concerned about health care expenditure. (3) Gini index is a good measure of the income inequality (World Bank, Reference Zeta2013).

In this case, it is possible that a large fraction of the population may be unable to pay for health care, particularly for inpatient services. Hospitals in states that have a higher proportion of low-income patients get subsidies towards the cost of care. For instance, as per the Medicare Inpatient Prospective Payment System (IPPS), the base payment rate (determined by the diagnosis-related group of a patient) to hospitals is adjusted by an add-on payment known as the disproportionate share hospital (DSH) adjustment (CMS, 2014). DSH basically provides for a percentage increase in Medicare payment for hospitals that serve a disproportionate share of low-income patients. In addition, it includes a provision to increase this IPPS payment for expensive patient cases. Therefore, the governments of such states would be concerned if hospitals could easily pass on the cost of unnecessary treatment to them and are more likely to have CON laws. This would help the state curb excessive expansion of hospital. This indicates that GINI index is less likely to influence the competition but rather is more likely to impact whether a state has a CON law or not. We use statistical tests to investigate the validity of these three IVs as well as apply a Durbin–Wu (Hausman type) test to determine if there is empirical evidence to suggest that the CON indicator is an endogenous variable.

Next, we extend our analysis by controlling for the stringency of the CON laws on service spending as well. By using the CON indicator, we can differentiate between CON states and non-CON states. However, within CON states, there is a large variation in the features of the law and one source of variation we are interested in is the stringency of the law. Moreover, states with CON laws can act similarly to those without it by setting a very high threshold (in some cases, threshold is listed as ‘None’) beyond which hospital expenditures are subject to review. As a result, it is vital to take this measure of the stringency of CON laws into account.Footnote ¹¹ We construct an index to measure the stringency of CON law. This index takes value of 0 for states with CON law where a high threshold (where the law states ‘None’) is suggested, and it takes value of one for states with CON law where a low threshold (where the law states ‘any amount’) is suggested. Similar to the previous section, we start our analysis by treating the stringency of CON laws as exogenous as follows:

(4) $${\rm HHI}\, {\equals}\, \beta _{0} {\plus}\beta _{1} {\rm CON}{\plus}\beta _{2} {\mathop{\rm CON}\nolimits} \,stringency{\plus}{\mib{\beta }} _{3} {\mib{X}} {\plus}\eta .$$

For the reasons highlighted earlier, it is possible that the CON stringency may be an endogenous variable as well. In order to tackle this empirical issue and test for the endogeneity of the law’s stringency, we estimate the effect of the CON laws on market competition by using a 2SLS model as followsFootnote ¹² :

(5) $${\rm Stage}\,{\rm one}\,\colon\,\,{\rm CON}\,{\rm stringency}\, {\equals}\, \varphi _{0} {\plus}{\mib{\varphi }} _{1} IV {\plus}\varphi _{2} {\rm CON}{\plus}{\mib{\varphi }} _{3} {\mib{X}} {\plus}\omega ,$$

(6) $${\rm Stage}\,{\rm two\,\colon\,}\;{\rm HHI}\, {\equals}\, \delta _{0} {\plus}\delta _{1} \,{\rm CON}\widehat{{{\rm String}}}{\rm ency}{\plus}\delta _{2} {\rm CON}{\plus}\delta _{3} \mib X{\plus}\vartheta .$$

In this specification, we use three IVs for stringency: tech index, GINI index and the excise tax rate on beer. The relationship between tech index and service stringency could go either way. On the one hand, in a technologically advanced state, hospitals would have a greater tendency to upgrade technology frequently which can lead to an increase in the price of health care. To counter this, the government of the state would want a stringent CON law. On the other hand, such technologically advanced states will also have a large technology sector that provides both jobs and taxes. Since a strict CON law hurts the technology industry, therefore this reduces the incentive of a technologically advanced state to have a strict CON law.

As noted above, the governments of states that serve a high proportion of low-income patients and have concerns about hospitals easily passing on the cost of unnecessary treatment to patients due to support from the IPPS setup, would have an incentive to curb excessive expansion of hospital by lowering the threshold on service expenditures by health care providers. This indicates that GINI index is less likely to influence the competition but rather is more likely to impact the CON law in a state.

Similarly, as noted above, a state with higher taxes on beer is more likely to have a stringent CON law because such a state is likely to be more concerned about health care expenditures that result from poor behavioral (increased alcohol consumption) choices of citizens. The validity of these IVs and the endogeneity of CON law stringency are tested as well.