- Top of page
- THEORETICAL AND EMPIRICAL APPROACH
- CONFLICT OF INTEREST
Chronic tooth decay is the most common chronic condition in the United States among children ages 5–17 and also affects a large percentage of adults. Oral health conditions are preventable, but less than half of the US population uses dental services annually. We seek to examine the extent to which limited dental coverage and high out-of-pocket costs reduce dental service use by the nonelderly privately insured and uninsured. Using data from the 2001–2006 Medical Expenditure Panel Survey and an American Dental Association survey of dental procedure prices, we jointly estimate the probability of using preventive and both basic and major restorative services through a correlated random effects specification that controls for endogeneity. We found that dental coverage increased the probability of preventive care use by 19% and the use of restorative services 11% to 16%. Both conditional and unconditional on dental coverage, the use of dental services was not sensitive to out-of-pocket costs. We conclude that dental coverage is an important determinant of preventive dental service use, but other nonprice factors related to consumer preferences, especially education, are equal if not stronger determinants. Copyright © 2013 John Wiley & Sons, Ltd.
- Top of page
- THEORETICAL AND EMPIRICAL APPROACH
- CONFLICT OF INTEREST
According to the Centers for Disease Control (USDHHS 2000), chronic tooth decay is the most common chronic condition among children ages 5–17, having five times the prevalence rate of asthma, for example. A large percentage of adults also experience chronic tooth decay and other oral health problems. These conditions frequently cause chronic pain and nutritional difficulties that can adversely impact physical and mental health. Most oral health conditions are preventable with regular treatment, but less than half of the US population uses dental services of any kind annually (Manski and Brown 2007). Among the factors determining access, limited dental insurance coverage and high out-of-pocket costs are commonly cited reasons for why the use of dental care is not greater among Americans of all ages (Mueller and Monheit 1988; Manning et al. 1986). For example, Medicare covers dental procedures only under exceptional circumstances. Dental coverage is an optional benefit under Medicaid, but even when states choose to cover it, few dental providers participate because of low payment rates (Decker 2011). Although two-thirds of people with private health insurance have dental coverage, this coverage is often limited. As a consequence, Americans pay almost half of the total cost of dental services out-of-pocket (Brown and Manski 2004).
We seek to determine the extent to which limited dental coverage and high out-of-pocket costs reduce consumers' use of preventive and restorative dental services. Previous estimates of the price elasticity of demand for dental care are now more than two decades old (Mueller and Monheit 1988; Manning et al. 1986; Hay et al. 1982). The RAND Health Insurance Experiment (HIE), conducted between 1977 and 1982, (Manning et al. 1986) found 66% of the population used dental services in the free care plan, whereas approximately 50% of the population used dental services in the 25%, 50%, and 95% cost-sharing plans, as well as in the deductible plan. Although dental expenditures were 46% higher in the free care plan than the 95% cost-sharing plan, most of the increase in use and expenditures occurred between free care and the 25% coinsurance rate and resulted primarily from the higher likelihood of visiting a dentist under the free care plan.
Mueller and Monheit (1988), using nationally representative survey data from 1977 and a nonexperimental design, found that between 55% and 57% of the population of White Americans aged 16–64 with some form of dental coverage used dental services annually compared with 47% without coverage. Consistent with the RAND researchers, Monheit and Mueller found that among those with coverage, use did not vary significantly with level of coverage. This finding is confirmed by the analysis of claims data from 1978 of Hay et al. (1982), which suggests that consumer demand for dental visits was not very responsive to differences in out-of-pockets costs. Overall, both the RAND HIE and contemporary observational studies indicate that while variation in cost sharing did not significantly impact the demand for dental services, having dental insurance increased utilization and expenditures.
More recent evidence on the effect of dental coverage per se (but not level of generosity) is available (Manski and Cooper 2007; Munkin and Trivedi 2009). In particular, Munkin and Trivedi (2009), using data from the nationally representative 1996–2000 Medical Expenditure Panel Survey (MEPS) and controlling for selection, found that dental insurance coverage increased the total number of general dental visits by 0.37 on average. Manski and Cooper (2007) found that the increase in dental visits due to insurance extends to all types of medical insurance, not just dental insurance. Although those with dental or dental and medical insurance were most likely to visit providers, having medical insurance but no dental coverage was positively associated with visiting a dentist after controlling for socioeconomic and demographic factors.
The majority of recent studies on dental care utilization in the United States have focused on access to care by enrollees in public coverage, and specifically the Medicaid or State Children's Health Insurance Programs (SCHIP). Although Wang et al. (2007) demonstrated that the introduction of SCHIP reduced the probability of unmet dental care needs for low income children by 40%, and Choi (2011) estimated that Medicaid dental benefits increased the probability that adults visited a dentist by 16–22%, several other studies indicate that low reimbursement rates by these programs have limited access to care for many. In particular, both Decker (2011) and Nietert et al. (2005) found that increasing Medicaid and SCHIP payment levels to dentists increased dental service provision to enrolled children. Similarly, Chi et al. (2011) found that newly Medicaid enrolled children living in areas of Iowa with dental health professional shortages made their first visits to providers much later than those in nonshortage areas.
Recent changes in US health care policy in the form of the 2010 Affordable Care Act (ACA) have increased the need for reliable estimates of the impact of public and private insurance coverage as well as out-of-pocket costs on the demand for dental services. For example, the studies described earlier provide important information that can be used to predict the impact on dental care demand of the optional expansions of Medicaid benefits to 133% of the federal poverty line by 2014. However, the ACA also has numerous provisions that will expand private market coverage, including incentives to small businesses to provide coverage, penalties on large employers that drop coverage, and the creation of state-run Health Insurance Exchanges where individuals can purchase credible coverage in the individual market. Furthermore, a recent Institute of Medicine report commissioned by the US Department of Health and Human Services has recommended that dental care coverage for children be designated as an essential benefit that must be incorporated into all plans sold to individuals and small businesses, including those offered through the exchanges (Institute of Medicine (IOM) 2011). In some cases, dental benefits will be exempt from the mandates and regulations that apply to general medical coverage, but overall, the expansion of insurance markets will likely increase access to private dental coverage.1
We extend the previous literature in several directions to estimate the dental demand elasticities needed to evaluate recent policy reforms. First, we use more recent data from the 2001–2006 MEPS supplemented with county-level price data of various dental procedures from an American Dental Association (ADA) survey of dental practices. Second, we estimate both the effects of dental coverage and the level of out-of-pocket price and provide separate estimates for children and adults. Third, we distinguish among the following different types of dental care that are likely to have different cost-sharing and demand characteristics (National Association of Dental Plans 2008): preventive services (for example, exams, cleanings, and x-rays), basic restorative services (for example, fillings and extractions), and major restorative services (crowns and root canals).
We first present an economic model of consumer demand and then jointly estimate the probabilities of using each of the three types of preventive and restorative service (basic and major) for a sample of individuals in MEPS with private or no health insurance coverage (n = 53,133 persons, T = 2 years). To control for the potential endogeneity of out-of-pocket price, dental coverage, and certain control variables, we exploit the panel data nature of the MEPS and estimate a correlated random effects (CRE) specification (Chamberlain 1980). The CRE model relaxes the untenable assumption of standard random effect models that price and health insurance are uncorrelated with unobserved individual attributes.
THEORETICAL AND EMPIRICAL APPROACH
- Top of page
- THEORETICAL AND EMPIRICAL APPROACH
- CONFLICT OF INTEREST
Following the standard neoclassical approach to consumer demand, suppose that an individual i = 1, …, N in time t = 1, …, T has preferences over their oral health, Hit, and a composite commodity of all other goods, Cit, defined by the following utility function:
Further, assume that oral health is a stock variable defined by an initial level of health carried over from the previous period, investments in oral health made through the consumption of dental services, mkit, k = 1, …, K, and random shocks, εit, such that
To determine the optimal investment in oral health and consumption of other goods, the individual maximizes (1) and (2) subject to a budget constraint:
where Yit represents total disposable income, pkit is the price of dental service k faced by person i in time t, and the price of the composite commodity has been normalized to one. The resulting demand equations for dental services, which can be easily modified to include socio-demographic determinants, Zit, take the form:
Although the aforementioned derivation of a standard health care demand model is straightforward, complete specification of the demand equations is complicated by a number of issues specific to the market for dental services. First, dental services are nonhomogenous. To limit heterogeneity, we classify all dental services into one of three categories using common cost-sharing tiers found in a survey conducted by the National Association of Dental Plans (2008): preventive services (for example, exams, cleanings, sealants, x-rays), basic restorative services (fillings, extractions, periodontics, endodontics, oral surgery), and major restorative services (crowns, bridges, root canals, dentures). We omit orthodontia treatment from the analyses because very few individuals report visits for this category of care.
Second, units of consumption vary by dental service. Preventive dental visits are discrete and fairly homogenous units, such as cleanings or exams. In our sample, 91% of individuals with preventive care use report one or two such visits per year. However, episodes of basic and restorative treatment can span two or more visits. For example, an impression may be taken for a crown on one visit and the crown cemented on the tooth on a later visit, with one charge covering both visits. The majority of people with any basic or major restorative treatment report only one (70% and 64%, respectively) or two (19% and 23%, respectively) visits during the year. For the remaining few with three or more visits during the year, most are clearly part of the same episode of treatment. Thus, the vast majority of individuals appear to have only one episode of treatment during the year. Because our main interest is in the probability of any treatment episode, we group together multiple visits for the same type of care. In particular, for respondents who were charged a flat fee covering multiple visits, we considered all these visits part of the same episode, and we also grouped all visits for the same type of care occurring within 12 weeks of one another into a single episode for basic and major treatment services, respectively. We used the date of the initial visit for an episode to allocate episodes to years. Thus, we are estimating the probability of treatment episode initiation for basic and major services.
Third, the presence of negotiated prices, deductibles, and coverage limits in dental insurance plans can complicate the specification of prices. A major component of the value of dental insurance to consumers is access to price discounts negotiated between select (in-network) providers and plans. Thus, the relevant out-of-pocket price to consumers is net of any discounts they obtain by using in-network providers. Of course, out-of-pocket price is also a function of coverage. The National Association of Dental Plans (2008) describes a proto-typical dental plan as having the following cost-sharing tiers: 0% for preventive, 20% for basic, and 50% for major services with a $50 deductible (not applicable to preventive services). We note that while this proto-typical plan is common, there are many variants in terms of cost-sharing, amount of the deductible, and whether the deductible applies to the family or to each individual. In addition, many plans impose an annual maximum on total covered expenditures, often between $1000 and $2500. The effective price of an episode of basic or major dental treatment differs depending on whether the consumer exceeds their deductible and expects to be below or above the annual maximum. Keeler et al. (1977) derived the true shadow price of health care in the presence of deductibles and coverage limits throughout the year using a dynamic programming framework and showed that the price changes nonlinearly as consumers approach their coverage limit.
Newhouse et al. 1980) showed that defining pkit in ((4) as the average or marginal price of medical care can lead to biased estimates. However, because we almost always observe just one episode of basic or major treatment, we assume that the true shadow price is equal to the observed out-of-pocket price; that is, the true shadow, average and marginal prices are all the same. For preventive service visits, we assume that the average out-of-pocket price approximates the true shadow price, which is reasonably given than preventive care is often exempt from deductibles and is inexpensive relative to plan maximums. We use the average rather than the marginal price of the last visit, largely because dentists often include an x-ray with one of the preventive visits occurring during the year. In practice, the average and marginal out-of-pocket prices are highly correlated, and our results are not sensitive to this choice.
Fourth, there are potential dynamic aspects to the demand for basic and major services. A preventive visit initiated by the patient may lead to the dentist recommending further basic or major services in a sequential decision process. However, basic or major treatment may also be initiated independently from a preventive visit by a patient when, for example, encountering tooth pain or cracking a crown. Analyses of the temporal sequence of visits in our data suggest that patient initiated basic or major treatment that is not clearly linked to a prior preventive visit predominates. Because of this and because we wish to exploit the time series dimension of our dataset to control for unobserved heterogeneity, we opt to model the joint demand for each type of treatment as part of a nonsequential process but recognize the potential for dynamic, sequential effects.2
The basic empirical model for a demand system composed of K types of dental services can thus be specified to take the form
where mkit is now explicitly defined as a binary variable indicating whether the individual had any preventive dental visits, or episodes of basic and major treatment during the calendar year, ci is a stochastic time-invariant individual specific effect measuring unobserved heterogeneity, and the index l = 1, …, L runs over dental services such that L = K. We interpret ci as unobserved oral and physical health status and propensity to consume treatment; in cross-sectional formulations of the model, ci is typically either assumed to be zero or uncorrelated with all other regressors. Given that we include self-reported physical health in our models to capture changes in health status over time, we assume the remaining unobserved health status and preferences for treatment are time invariant. If the vector of disturbances εit = (ε1it, …, εKit)′ is assumed to be jointly distributed , then the system of K equations defined by (5) is correlated through both ci and εit.
We model prices in Equation (5) using the level out-of-pocket prices pkit for simplicity. Our results were not sensitive to alternative specifications using either log prices or the inclusion of variables to separately account for the mass point of the price distribution at zero. In contrast to out-of-pocket costs, the health insurance premiums are an anticipated expense not linked to any particular health care transaction. We, therefore, subtract the reported premiums associated with all the health and dental insurance policies covering family members from gross family income in order to create a measure of total disposable appropriate for inclusion in the demand model.
Given our interpretation of ci as the individual's unobserved physical and oral status and propensity to consumer care, we cannot assume that this factor is uncorrelated with prices, dental coverage, self-reported physical health status, or net income. Therefore, ci is modeled as a CRE and assumed to be potentially correlated with this set of regressors in each time:
where υi is assumed to be independent of the exogenous regressors and εit and is distributed .3 This specification was originally derived by Chamberlain 1980) and has been applied to demand systems by Meyerhoefer et al. (2005) and Meyerhoefer and Zuvekas (2008, 2010). Defining xit as a row vector containing all the model regressors for time t (plus a constant for the intercept), ((6) is substituted into (5) to derive the demand system with reduced form parameter vector and normally distributed disturbance ukit = εkit + υi:
Econometric estimation of the CRE model generally proceeds by first obtaining consistent estimates of the reduced form parameters in (7) followed by identification of the structural parameters of interest in (5).
Because all of the left-hand side variables, mkit, are binary indicators for preventive, basic, or major restorative treatment, the dental services demand system defined by Equations (5)–(7) is a system of three correlated Probit equations. However, single equation Probit estimation yields consistent estimates of the reduced form parameters in Equation (7). Meyerhoefer et al. 2005) demonstrate how to identify the structural parameters in Equation ((5) from the reduced form estimates using a minimum distance estimator of the form:
Here, ψ denotes the vector of structural parameters, is the estimated covariance matrix of the reduced form parameter estimates, and H is a design matrix mapping the structural parameters to the reduced form estimates.
- Top of page
- THEORETICAL AND EMPIRICAL APPROACH
- CONFLICT OF INTEREST
Our principal source of data is the 2001–2006 MEPS. The MEPS is a comprehensive, nationally representative household survey of the US civilian noninstitutionalized population, conducted annually since 1996. The MEPS uses an overlapping panel design, combining two panels to produce annual estimates. One respondent generally reports for each household over the course of two years through five interview rounds. The MEPS asks that this be the person most knowledgeable about the health and health care use of the family. Respondents are interviewed in detail about individual and household characteristics and the health care use and expenses of each family member, including dental visits, during each round. We pooled panels 6 through 10 of the MEPS and restricted our sample to persons aged 1–64 who appear in both years of their eligibility.4 We excluded children less than 1 year old because clinical guidelines recommend that children receive their first dental exam between the ages of 6 and 12 months (American Academy of Pediatric Dentistry 2009). We further excluded people covered by Medicaid because the combination of poor dental coverage for adults, low co-payment amounts where coverage exists, and limited provider participation complicates our ability to model consumer demand in comparison with the privately insured and uninsured populations. The final sample size was 53,133 persons with two observations each (106,266 person–year observations in cross-sectional analyses). Splitting the sample by age yields 12,801 children ages 1–18 and 39, 490 adults ages 19–64.5
We merged onto the MEPS mean county-level dental procedure list price data provided by the ADA from a proprietary 2004 survey of dental practices. Specific procedures we included with sufficient sample sizes were D0120-Oral Exam, D0274-Bitewing Exam, D1110-Prophylaxis (cleaning), D2150-Amalgam 2 surface, D2386-Resin-based composite, and D2750-Crown. Figure 1 presents price distributions weighted by MEPS sample weights for three representative procedures corresponding to preventive, basic, and major services, respectively: oral exam (mean = $28), amalgam 2 surface (mean = $89), and crown (mean = $692). We note that these prices represent the dentists' list prices and do not reflect any discounts that may occur for some patients with dental plans.
Table 1 lists descriptive statistics for all the variables used in the dental services demand model. The first column presents means for the full sample of privately insured and uninsured individuals ages 1–64. The next three columns contain means for the sample with any preventive, basic, or major services each year. We describe the construction of each in turn, noting which variables we treat as potentially endogenous in the CRE specification.
Table 1. Means, privately insured, and uninsured population aged 1–64, 2001–2006
| ||Full sample||Any preventive||Any basic||Any major|
|Any use|| || || || |
|Number of visits|| || || || |
|Observed out-of-pocket price|| || || || |
|Observed total price|| || || || |
|Expected out-of-pocket price|| || || || |
|Control Variables|| || || || |
|High school diploma||0.29||0.24||0.27||0.28|
|BA or higher||0.32||0.44||0.37||0.39|
|No. of family members 0–5||0.32||0.28||0.24||0.18|
|No. of family members 6–17||0.79||0.85||0.74||0.50|
|No. of family members 18–64||2.09||2.06||2.04||2.00|
|No. of family members 65+||0.05||0.04||0.04||0.06|
|No. of chronic conditions||0.43||0.44||0.53||0.71|
|Log family income||10.41||10.65||10.56||10.67|
|N∙T (person–year observation)||106,266||38,357||15,060||6,673|
Preventive, basic, and major restorative services
The household respondent reports in each round of the MEPS all dental visits made by household members and is asked to provide details about the types of dental services rendered, total charge for the visit or groups of visits, out-of-pocket payments, and third party payments. On the basis of the services described by the household, we classify each visit as for either preventive, basic, or major dental care according to the National Association of Dental Plans (2008) description of common cost-sharing tiers as described in Section 2. Preventive services include exams, cleanings, sealants, and x-rays. Basic services include fillings, periodontics, endodontics, and oral surgery. Major services include crowns, bridges, root canals, and dentures. We then constructed binary measures of whether preventive, basic, and major, treatment services, respectively, occurred during the year. In some cases where individuals reported both preventive and basic, preventive and major, or basic and major services during the same visit, we counted the individual as having both types of treatment. As indicated in Table 1, 41% of the privately insured and uninsured population had at least one preventive care visit during the year, whereas only 16% and 7% had one or more episodes of basic and major treatment, respectively.
Potentially endogenous: price measures
We constructed the following three measures of price: average out-of-pocket price for preventive care visits during the year, total out-of-pocket spending incurred during the first (if more than one) episode of basic treatment, and total out-of-pocket spending incurred during the first (if more than one) episode of major treatment. We note again that these prices are net of both any payments made by a dental, health maintenance organization (HMO), or private plan and any discounts obtained from using an in-network provider under a dental plan or HMO. Prices were observed directly for individuals using each of the three types of dental services. In cases where individuals reported using multiple services, we used a hedonic regression-based approach to proportionately allocate out-of-pocket spending for these visits to their respective components based on the specific procedures reported, individual and family characteristics, and the county-level dental procedure price data from the ADA.
Mean observed out-of-pocket price among users in our sample was $41 per preventive visit, $210 for the first episode of basic treatment, and $605 for the first episode of major treatment (Table 1). This suggests that the National Association of Dental Plan's (2008) classification scheme does a good job of distinguishing services by level of out-of-pocket spending.
We then used a regression-based predictive-mean matching approach to impute prices for nonusers of each of the three types of dental services. We tested a number of potential specifications for the out-of-pocket price prediction regression models, including standard one and two-part models with both logged and unlogged prices, one and two-part generalized linear models (gamma with log and square root links, poisson with log and square root links), and multiple-imputation procedures (using Stata's MI command). In general, the one-part models out-performed their two-part counterparts, whereas most of the one-part generalized linear models (GLM) provided similar levels of performance. On the basis of standard criteria for comparing performance, including mean squared error, mean absolute prediction error, and ratios of predicted price to actual price by deciles of predicted price (Manning and Mullahy 2001; Buntin and Zaslavsky 2004), we selected the one-part GLM model with gamma distribution and log link for the price imputation. Our results were not sensitive to this choice.
We used separate imputations for each type of service and reported our complete regression results for the imputation models for the privately insured population in Appendix Table AI. (Results of separate imputation models for the uninsured population were similar and are available from the authors). All price imputation regressions included controls for age, race/ethnicity, sex, education level, marital status, family income, Metropolitan Statistical Area (MSA) status, census region, self-reported health status, generosity of employer benefits, unionization, industry and occupation classification, employer size, HMO or managed-care enrollee, dummy variables for the year of interview, and dental insurance coverage. To these, we added ADA county-level mean list price data to improve the predictions of out-of-pocket price taking into account local market variation in prices. For preventive services, we included mean prices for exams, bitewings, and cleanings; for basic services, these prices plus amalgam and composite prices; and for major, these prices plus the price for a standard crown.
Potentially endogenous: dental coverage, private health insurance, and HMO coverage
Our main measure of dental coverage is whether individuals reported a private plan that covered dental services in each MEPS survey round during the year. We also constructed an equivalent measure of dental coverage for only part of the year if, for example, it was reported for only one round. Because the estimated effects for this part-year dental coverage variable were always small and not statistically significant, we omitted it for parsimony. No other information was available concerning the generosity of coverage in the 2001–2006 MEPS.
We also include dichotomous indicators for whether the individual reported being covered by private health insurance and whether their private insurance was through an HMO or managed care plan. Many HMOs contain some type of coverage for dental services. Some HMOs provide comprehensive dental coverage through their own dental networks, often using a fixed out-of-pocket payment benefit design, whereas others provide only minimal coverage for preventive visits. Although non-HMO private health insurers rarely include dental coverage as part of their benefit packages, enrollment in a private insurance plan may independently impact the demand for dental services. This is because insurance increases the probability of visits to primary care providers who, in accordance with guidelines developed by physician associations and the federal government, may counsel patients to visit dentists for preventive screening (Chu et al. 2007; Douglas et al. 2004; American Academy of Pediatrics 2003; U.S. Preventive Services Task Force 1996).
Potentially endogenous: income and self-reported health status
Our measure of family income is the sum of earned and unearned income for all family members minus the family's annual out-of-pocket premiums, which we divide by the square root of the household size in order to adjust for household economies of scale and then log. In the case of both income and prices, we inflate all measures to 2006 USD using the all-item urban consumer price index. To capture differences in respondent health status, we included a measure of whether the responded reported poor or fair physical health at any point in time for each year. The MEPS does not contain a health status measure that is specific to oral health.
Our models control for other sociodemographic characteristics as well, including race/ethnicity (Hispanic, Black, and other with White being the omitted category), sex, age (dichotomous indicators for age 0–18, 19–30, 31–54 with age 55–64 being the omitted category), education (high school diploma, some college, bachelors or higher with less than a high school degree omitted), marital status, number of family members by age group (0–5, 6–17, 18–64, 65 plus), MSA status, and census region. We use the parent's education information for respondents under the age of 18. We also include the variable ‘proxy’ to indicate whether the sample individual was the household respondent. We use a deterministic time trend and its square to control for secular changes during period 2001 through 2006 and include two variables to capture the time costs associated with seeking dental care. The first of these is an indicator variable of whether the individual has sick leave from his or her employer, and the second is the number of dentists per 1000 capita in the individual's county. The latter was obtained from the Census Area Resource File. Finally, we include two additional measures of health status: a count of the number of chronic conditions the individual suffers from and an indicator for whether the individual has an activity limitation.
We used the MEPS longitudinal weights in all of our analyses to ensure that our estimates are national representative. In addition, we adjusted the standard errors of our estimates to account for the stratified and clustered sample design of the MEPS and the variance due to the first stage imputations of the price variables using the method of balanced repeated replications. The balanced repeated replication survey adjustments also have the property of adjusting for within-family correlations and within-person correlation across observation year (Williams 2000).
- Top of page
- THEORETICAL AND EMPIRICAL APPROACH
- CONFLICT OF INTEREST
Table 2 contains the average marginal effects, survey adjusted standard errors, and p-values for all of the variables used in our system of Probit equations corresponding to any preventive service use, any basic treatment, and any major treatment. In Appendix Table AII, we present analogous results for the cross-sectional version of the model, which pools yearly observations for each person in the sample and includes no correction for potential endogenity of dental price, dental insurance, HMO coverage, health status, or income. The two sets of results are qualitatively similar, although the cross-sectional marginal effects are larger in some cases.
Table 2. Marginal effects from system of correlated random effect probit equations, privately insured, and uninsured population, aged 1–64, 2001–2006
| ||Any preventive||Any basic||Any major|
|High school diploma||0.037||0.008||<0.001||0.001||0.005||0.798||0.008||0.004||0.043|
|BA or higher||0.171||0.009||<0.001||0.015||0.006||0.008||0.021||0.004||<0.001|
|# 65 plus||−0.056||0.011||<0.001||−0.025||0.006||<0.001||−0.009||0.004||0.021|
|No. of chronic conditions||0.007||0.003||0.008||0.012||0.002||<0.001||0.006||0.001||<0.001|
|Log family income||0.008||0.003||0.021||−0.006||0.002||0.004||−0.003||0.001||0.085|
|Correlation parameters||Coefficient||SE||p-value|| || || || |
|Income period 1|| ||0.036||0.008||<0.001|| || || || |
|Income period 2|| ||0.036||0.008||<0.001|| || || || |
|Health period 1|| ||−0.070||0.021||0.001|| || || || |
|Health period 2|| ||−0.054||0.027||0.045|| || || || |
|HMO period 1|| ||−0.028||0.018||0.117|| || || || |
|HMO period 2|| ||−0.049||0.023||0.037|| || || || |
|Dental insurance period 1|| ||0.021||0.018||0.226|| || || || |
|Dental insurance period 2|| ||0.025||0.023||0.286|| || || || |
|Private insurance period 1|| ||0.108||0.028||<0.001|| || || || |
|Private insurance period 2|| ||0.138||0.036||<0.001|| || || || |
|Preventive price period 1|| ||0.010||0.080||0.903|| || || || |
|Preventive price period 2|| ||−0.184||0.213||0.389|| || || || |
|Basic price period 1|| ||0.010||0.023||0.654|| || || || |
|Basic price period 2|| ||0.006||0.027||0.837|| || || || |
|Major price period 1|| ||0.004||0.012||0.756|| || || || |
|Major price period 2|| ||−0.011||0.017||0.494|| || || || |
Among the covariates presumed exogenous, we find that race, gender, childhood, and education are strong predictors of all types of dental care utilization. For example, Hispanics, Blacks, and those of another minority race were significantly less likely to have dental visits than Whites. The probability that Blacks, in particular, had any preventive care visits, basic episode, or major episode was 15.0 percentage points (36%), 4.2 percentage points (27%), and 2.7 percentage points (38%) lower than for Whites, respectively. Women were more likely to visit dental professionals for all types of care than men. The probability that children 1–18 had preventive visits was 11 percentage points (27%) higher than those for ages 55–64, who were the group of adults most likely to receive preventive care. However, children were less likely to have basic or major episodes of care than older adults. Consistent with prior expectations, education was strongly related to the use of all types of services and preventive services in particular. Those with a bachelor's degree or higher were 17 percentage points (41%) more likely to have preventive visits than those with less than a high school education. Individuals living in MSAs were more likely to use preventive and major services than those living in rural areas, whereas those living in the Northeast and Midwest census regions were more likely to use preventive services than those in the West. Individuals from the South census region were the least likely to use all three types of dental services.
The demand system contains two variables to measure the time costs of treatment: an indicator for whether the individual received sick pay and the number of dentists per 1000 capita. Although sick pay was not significantly related to the probability of seeking treatment for either preventive or restorative care, those living in areas with a high concentration of dentists were more likely to receive dental services. Increasing the number of dentists in a given area by one per 1000 people (a 151% increase) was associated with a 2.6 percentage points (6%) increase in the probability of preventive visits. Overall, these results suggest that time costs do not play a major role in the demand for dental services among the privately insured.
In the bottom panel of Table 2, we report the correlation parameters corresponding to Equation (6) of our CRE specification. The individual parameter estimates suggest that income, overall self-reported health status, HMO enrollment, and private health insurance are correlated with the model's error term. In addition, chi-square specification tests based on the minimum distance function reject the null hypothesis that the error term is orthogonal to regressors specified as endogenous in the CRE model. Although the correlation parameters corresponding to price parameters are not individually significant, we maintain the specification with potentially endogenous prices for conceptual reasons and because chi-square specification tests reject the null hypothesis that the prices are jointly exogenous. Tables 3 and 4 summarize the results for our main variables of interest. For ease of interpretation, Table 3 presents income elasticities of demand, own-price elasticities of demand, and semi-elasticities corresponding to the dichotomous indicators for HMO and dental insurance coverage, whereas Table 4 presents cross-price elasticites of demand for all three services. Further, in Table 4, we present results for the full sample of privately insured individuals ages 1–64 as well as subgroup estimates for children ages 1–18 and adults ages 19–64.
Table 3. Probit demand system elasticities
| ||Cross-sectional (full sample)||CRE (full sample)||CRE (child sample)||CRE (adult sample)|
|Income elasticity|| || || || |
|Out-of-pocket own-price elasticity|| || || || |
|HMO semi-elasticity|| || || || |
|Dental insurance semi-elasticity|| || || || |
|Private insurance semi-elasticity|| || || || |
|No. of individuals||53,133||53,133||12,801||39,490|
Table 4. Probit demand system cross-price elasticities
|Cross-sectional model (full sample, N = 53,133)|
|CRE model (full sample, N = 53,133)|
|CRE model (child sample, N = 12,801)|
|CRE model (adult sample, N = 39,490)|
Income elasticities of demand from the cross sectional model reported in Table 3 were relatively small, ranging from 0.04 for basic service to 0.09 for preventive care. After accounting for endogeneity, the elasticity estimates were uniformly lower, and negative for basic and major services.
In both specifications and all estimation samples, own-price elasticities were approximately zero for all three services (Table 3). However, there was a small but statistically significant positive cross-price effect of basic price on use of major services in both the full and adult samples (Table 4), indicating that basic and major services are substitutes. A 10% increase in the price of basic services was associated with a 0.2% increase in the demand for major services for adults and overall.
Having dental insurance, either through a supplemental plan or an HMO, was a more important determinant of dental service utilization than either income or out-of-pocket costs. For preventive services having private health insurance was also important. Conclusions about the impact of insurance were sensitive to model specification, suggesting there is significant selection into insurance by those with varying service needs and preferences for oral health. On the basis of the cross-sectional model, individuals enrolled in HMOs and managed care plans were less likely to use both preventive and restorative services than those in more traditional plans. However, after controlling for selection into HMOs, we find that HMO enrollment increased the probability of using dental services by 2% (not significant), 8%, and 12% for preventive, basic, and major services, respectively (Table 3). This is consistent with positive selection into HMOs by generally healthier individuals in less need of services. HMO enrollment was not significantly correlated with dental service use among children and had the largest impact on the use of basic and major services by adults.
Given that not all HMOs cover dental care, and the coverage they do provide is typically more limited than supplemental dental insurance associated with traditional private health insurance plans, one might expect that traditional dental insurance to have a larger impact on utilization. Our results support this conjecture and are consistent with adverse selection into supplemental traditional dental insurance by those with greater needs or preferences for dental services. For example, having traditional dental insurance increased the probability of preventive, basic, and major dental care by 24%, 20%, and 27%, respectively, in the cross-sectional models, but by 19%, 11%, and 16%, respectively, in the CRE models. Our point estimates suggest that preventive care service use is more responsive to dental insurance for adults (23% more likely) than for children (12% more likely). We also find that enrollment in private health insurance independently increases the probability that both children (14%) and adults (32%) receive preventive but not basic or major dental services.
We tested a number of alternative specifications of our price measures to assess the robustness of our findings of zero own-price elasticity of demand for preventive, basic, and major services. These included logging prices, adding a dichotomous indicator to account for the mass point at zero price, and expressing out-of-pocket price as a percentage of total price. All yielded substantially similar results. Furthermore, we confirmed that the price effects were not absorbed by the HMO, private, dental insurance variables in our fully specified model. Removing the insurance variables from the analysis had almost no impact on the estimated price elasticities of demand.
We also estimated instrumental variable (IV) models as an alternative to our CRE specification, with the ADA price data as instruments. Although results concerning the effect of dental insurance on preventive care use were similar to our CRE specification, other results were not and all the IV estimates were fragile and sensitive to the exact specification of the IV models.
- Top of page
- THEORETICAL AND EMPIRICAL APPROACH
- CONFLICT OF INTEREST
We find that, even after accounting for selection, traditional dental coverage increases the use of all types of dental services. This is consistent with the findings of Munkin and Trivedi (2009) who estimate that dental coverage increased the number of general dental visits by 0.37, after accounting for selection effects using methods that are similar to, although more general than ours. Thus, increased dental coverage in the population is likely to improve access to dental services. HMO enrollment also increased the use of dental services but had a larger impact on restorative dental care than preventive care. In contrast, enrollment in private health insurance increased the use of preventive but not restorative dental services. Although traditional health insurance plans do not typically include dental services among their covered benefits, this is consistent with counseling by primary care providers to comply with recommended dental check-ups (Chu et al. 2007).
Unexpectedly, we found that both conditional and unconditional on dental coverage, use of preventive and restorative dental services was insensitive to out-of-pocket price. In fact, our estimated price elasticities of demand were near zero for all models and samples. We did find that basic and major restorative services were substitutes based on the cross price effects, but these were very small in magnitude. It is possible that our general inability to identify significant price effects may have resulted from attenuation bias due to measurement error in using imputed prices for nondemanders. The CRE model can correct for measurement error processes that are time-invariant, but because we have no easy way of determining whether this is the case, we cannot rule out the possibility of attenuation bias. We also note again that earlier studies based on the 1977–1982 RAND HIE (Manning et al. 1986) and observational data from 1977 (Mueller and Monheit 1988) both found consumers to be insensitive to price over a wide-range of out-of-pocket price as measured by cost-sharing for dental care. More puzzling is the fact that the two studies found this price insensitivity over opposite ends of the price sharing spectrum. The RAND HIE study found effects of coverage only with the free care plan, with no differences in use between the 25%, 50%, 95%, and deductible plans, whereas the Mueller and Monheit study found that all plans regardless of cost-sharing levels increased dental care use equally over noncoverage. Better data on levels of dental coverage in the MEPS (not just whether a person reports coverage) might help resolve these puzzles.
After accounting for unobserved heterogeneity, we find that higher income is associated with statistically significant increases in the demand for preventive services and reductions in the demand for restorative care. If the additional preventive treatments most responsive to income are those for which preventive care is particularly efficacious, it may be the case that the demand for downstream services is reduced as a result. In fact, once recent study showed that having preventive dental care resulted in fewer visits for expensive nonpreventive procedures and reduced out-of-pocket payments (Moeller et al. 2010).
It is also important to note that nonprice/coverage factors play a strong role in determining use of dental services; in particular education and race and ethnicity. The large differences in use of preventive services by educational level conditional on family income suggest that preferences for and information about oral health may be a key determinant of preventive health service use. In addition, our results suggest that there are disparities in the use of dental services between the White and minority populations. Although less directly amenable to public policy interventions, such as increasing the number of Americans with dental coverage, preferences and information may be key routes to increasing preventive dental care use and decreasing chronic tooth decay and other oral health problems. Nonetheless, our results do suggest that the expansion of insurance offerings mandated under the ACA will likely increase the demand for dental services and preventive care in particular.
Another limitation of our modeling approach is that our method for accounting for endogeneity and measurement error is not fully general. We assume that the endogeneity of price, health insurance status, and self-reported measures of physical health is generated by correlation with a component of the residual that is time invariant. This is logical given the supposition that unobserved physical and oral health status and propensity to consume treatment are the primary confounders of these variables. However, it is also possible that there exists a correlation between the regressors with time-varying error components that we cannot capture. For example, some dental insurance plans have networks of preferred providers whose services are covered more generously than other providers. If there is positive selection into these types of plans along dimensions of health status that are not time invariant, it could bias the estimates of our own-price effects towards zero. Likewise, if the process generating differences between consumers' true out-of-pocket price and our imputed prices is not time invariant, then our parameter estimates still will be attenuated.
We identify the correlation between the random effect and included regressors using just two periods, such that the effect of regressors three periods removed is present in the model's residual. If the included regressors exhibit strong serial correlation, then some endogeneity bias may persist despite our attempt to ‘integrate out’ the random effect. Finally, the limited time dimension of our data precludes us from modeling dynamic patterns of demand between preventive and restorative services, which we view as a fruitful area of future research.