Data sources and sample
To construct an analytic sample of actively practicing dentists in the U.S., we used a number of data sources, including those collected and maintained by the American Dental Association (ADA): the ADA Distribution of Dentists (DOD) Survey and the ADA office database. These databases include information on individual dentists (e.g., dentist demographics, specialty information, employment relationship) and the organizational structure of the practices in which they work (e.g., practice size, DMSO affiliation). Since the DOD is collected from approximately one-third of U.S. dentists each year, we used data from the three most recent years available (2013, 2014 and 2015) to reflect the status of all dental practices. Based on the ADA’s masterfile of all actively practicing dentists, the 2013–2015 DOD has about a 70.6% response rate. The ADA data was merged with measures of dental insurance market concentration obtained from the FAIR Health
® Dental Module based on the year the dentist was surveyed. The FAIR Health
® Dental Module contains dental claims for approximately 75% of individuals who have commercial dental insurance in the United States (FAIR Health
® Inc.
2016). Finally, the FAIR Health
® Dental Module and ADA databases were supplemented with publicly available information from the Area Health Resource File (AHRF) and the U.S. Census Statistics of U.S. Businesses (United States Census
2018).
Using this data, the analytic sample was restricted to dentists that were actively practicing in the United States and reported information about the organizational structure of their practice. This resulted in a sample of 53,831 dentists. After we restricted the sample to urban dentists and those with non-missing data, the final analytic sample for the main analysis was 46,594 unique respondent dentists. We restricted our sample to urban dentists since rural geographic areas are typically larger in size and may not accurately represent a rational dental insurance market. While data from 2013 through 2015 was utilized, this analytic sample was cross-sectional in nature and approximated the state of the dental industry around 2015.
Dental insurance market concentration
In a national study, it is not feasible to perform a detailed analysis to define each geographic market for dental insurance; therefore, proxies are utilized. Some studies have used the state as a proxy (Moriya et al.
2010), arguing that insurers are regulated at the state level and that once entry occurs, it is easy to expand into other locations within the state. However, this may result in geographic markets that are significantly large, underestimating the level of dental insurer concentration. Therefore, we referred to previous work that defines insurance markets based on the first three digits of a zip code (Dafny et al.
2012). While market concentration has been calculated using multiple approaches (Austin and Baker
2015; Baker et al.
2014), we did not have access to the detailed claims data from FAIR Health
®. Therefore, we were restricted to calculating insurer concentration based on a fixed geographic area, which, based on the level of granularity provided by FAIR Health
®, is a 3-digit zip code.
Using the 3-digit zip code to proxy for the geographic market, FAIR Health® provided dental insurer market concentration as measured by the Herfindahl–Hirschman Index (HHI) for each year from 2011 through 2015. The calculation of HHI is based on the market share of each insurers’ number of paid dental claims. Some prior studies have used enrollment data to calculate insurer HHIs. We used paid claims to calculate market share because dentists may not have enrollment information but knew how many claims were submitted to various dental insurers. Therefore, an HHI based on paid claims may better reflect the insurance market dynamics that dentists face when determining how to organize their practice.
Empirical strategy
Our analysis evaluated the impact of dental insurance market concentration on the three measures of organizational structure of dental practices. To understand this relationship, we evaluated the following model:
$$ Y_{i} = {\text{G}}(Insurance \,Concentration_{i} *\beta + {\mathbf{X}}_{i}\varvec{\alpha}\text{)} $$
(1)
where
i indexes a dentist,
\( Y_{i} \) is a measure of the organizational structure of the practice, dental insurance concentration is measured by a HHI in the 3-digit zip code of the dentist, and
\( {\mathbf{X}}_{i} \) is a set of control variables. We used HHI in its logged and level form in Eq.
1. We reported both results and confirmed that they are quantitatively similar. Because all of the practice structure measures we examined are non-linear, we needed to use the proper functional form of
\( G\left( - \right) \) to capture the non-linear nature of each dependent variable. We used Poisson regression when analyzing the number of dentists in a practice and estimated probit models when determining the likelihood that a dentist is a non-owner or part of a DMSO. Standard errors were clustered at the 3-digit zip code level.
According to the conceptual model, we expected that more concentrated dental insurance markets will lead to consolidation among dental providers, which can be measured as larger practices, a lower percentage of dentists with ownership stakes, and a greater share of dentists in DMSOs. In all cases, this implies that we expected the marginal effect associated with \( \beta \) to be positive. To account for differences in dentists and where they practice, \( \varvec{X}_{\varvec{i}} \) is a set of control variables that are commonly adopted in the literature. These include the dentist’s experience (measured as years), experience squared, gender, race/ethnicity, and dental specialty (i.e., general practice dentist, pediatric dentist, and other specialists). We also controlled for county characteristics, including census region, population density (population per square mile), dentist density (professionally active dentists per square mile), an indicator for whether the local market is a dental health professional shortage area (HPSA), log of real median household income, log of total population, and a metro continuum categorical variable. We also included year indicators.
While a number of studies in the medical literature estimate Eq. (
1) treating the relationship between concentration in health insurance and medical provider markets as exogenous (Trish and Herring
2015; Moriya et al.
2010), it has been argued that health insurer HHI is endogenous. This could be due to measurement error (Dafny et al.
2011) in calculating insurer HHI or potential for reverse causality where insurers merge or exit the market in response to the consolidation of providers (Brunt and Bowblis
2014; McCarthy and Huang
2018; Dunn and Shapiro
2014). Therefore, we also estimated Eq. (
1) under the assumption that dental insurer HHI is endogenous. Specifically, we estimated a first-stage model as follows:
$$ Insurance\, Concentration_{i} = \varvec{Z}_{\varvec{i}}\varvec{\delta}+ {\mathbf{X}}_{\varvec{i}}\varvec{\gamma}+ v_{i} $$
(2)
where
\( {\mathbf{Z}}_{\varvec{i}} \) are excluded instruments and
\( v_{i} \) is an error term. We then applied a control function approach (Wooldridge
2015), commonly referred to as two-staged residual inclusion (2SRI) in which the estimated residual in Eq. (
2) is used as a regressor in Eq. (
1). This 2SRI approach addresses the potential for endogeneity of dental insurer HHI and will produce consistent estimates of
\( \beta \) in Eq. (
1) when there exists at least one excluded instrument that predicts insurer HHI but does not directly affect the organizational structure of dental practices.
Following the medical insurance literature, there were two sets of instrument variables that we used. Both attempted to measure the level of concentration among insurers by measuring whether a market is likely to face entry or exit by an insurer and the ability of an insurer to maintain or increase market share.
The first set of instrumental variables used in the medical literature is the number of firms and number of employees per firm (e.g. firm size), both measured at the county level (Brunt and Bowblis
2014; Bates and Santerre
2008; Town et al.
2007; Dranove et al.
1998; Baker and Brown
1999). These studies argue that because most individuals have insurance coverage through their employer, as is the case in dental markets (National Association of Dental Plans
2017), an insurance market’s attractiveness and an insurer’s market share will depend on the size and distribution of employers. That is, the number of firms approximates the size and profitability of local commercial dental insurance markets. As such, a larger number of firms is found to be negatively correlated with the insurance HHI. On the other hand, firm size is found to be positively correlated with insurance HHI because larger employers may negotiate lower premiums and hence reduce insurers’ incentives toward entry. These market dynamics also apply to the dental industry because the majority of Americans receive dental benefits from commercial entities (American Dental Association
2017c), and 92% of commercial dental plans are financed through group or employer coverage (National Association of Dental Plans
2017). Furthermore, firm size and the number of firms are not likely correlated with the organizational structure of dental practices because dentists provide care to patients from different firms. This implies that the long-term organizational decisions made by dentists are based on the potential actions of commercial dental insurers and the size of the total market, not the decision of any particular employer in the market.
The second set of excluded instruments included the unemployment rate and whether the county has a high proportion of elderly residents, defined as being among the top 5% in the nation with the highest percentage of individuals aged 65 and older (Berry and Waldfogel
2001; Davis
2006). These instruments were found to be valid in studies of physician markets because health insurance is primarily purchased through employers, and markets with stronger economies and larger working age populations are more attractive for health insurers (Dunn and Shapiro
2014). Similar to health insurance, most dental insurance is also purchased through employers, with insurance coverage rates for private dental benefits at 58.1% for working-age adults but only at 27.9% for individuals aged 65 and older (Nasseh and Vujicic
2016). This implies that dental insurers are more likely to enter markets with strong economies (e.g., lower unemployment rates). For older individuals, most have dental coverage through Medicare Advantage. If there are a large number of seniors in an area, we expected dental markets to be less concentrated as various Medicare Advantage plans may team with different dental insurers or offer their own dental insurance in order to attract seniors to their Medicare Advantage plans.
While higher unemployed and elderly populations predict dental insurer HHI, they do not necessarily affect the organizational structure of dental practices. To establish a dental practice, dentists are faced with many fixed costs, including but not limited to finding a location, investing in specialized equipment, and customizing an office to meet their needs. Most importantly, they must undergo the sunk cost of establishing a reputation and patient panel. Due to the high costs associated with establishing a dental practice, it is less likely that established dental practices will change their organizational structure in the short run in response to local economic conditions. It is possible that local economic conditions could have greater influence on the career choices of dentists who just graduated from dental school, but these individuals make up a small proportion of dentists in any market. Therefore, these instruments are likely to satisfy the criteria that they do not influence the organizational structure of dental practices. To assure our results were not sensitive to new dentists in the sample, we also ran a robustness check that is limited to dentists with at least 5 years of experience.