The heterogeneous impact of insurance on health care demand among young adults: a panel data analysis^*

* An earlier version of this paper circulated as a Western Kentucky University working paper, cited as Zimmer [21] in the reference list.

ABSTRACT

Success of the recently implemented Affordable Care Act hinges on previously uninsured young adults enrolling in coverage. How will increased coverage, in turn, affect health care utilization? This paper applies variable coefficient panel models to estimate the impact of insurance on health care utilization among young adults. The econometric setup, which accommodates nonlinear usage measures, attempts to address the potential endogeneity of insurance status. The main finding is that, for approximately one-fifth of young adults, insurance does not substantially alter health care consumption. On the other hand, another one-fifth of young adults have large moral hazard effects. Among that group, insurance increases the probability of having a routine checkup by 71–120%, relative to mean probabilities, and insurance increases the number of curative-based doctor office visits by 67–181%, relative to the mean number of visits.

KEYWORDS:

• Random slopes
• random effects
• treatment effects
• average partial effects
• adverse selection

JEL CODES:

• I13
• I18
• C23

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

* An earlier version of this paper circulated as a Western Kentucky University working paper, cited as Zimmer [21] in the reference list.

1 Numbers in this paragraph come from the 2010 wave of the Medical Expenditure Panel Survey.

2 This paper does not consider subjects interviewed as part of the NLSY ‘oversample’ of minority and low-income populations.

3 Estimations in this paper do not employ sample weights for several reasons. First, when selecting a specific subpopulation, such as 18–29 year olds, it is not obvious how to adjust the provided weights to reflect such sample constructions. Second, and more importantly, this paper seeks to interpret its empirical findings as structural, and therefore, unweighted estimations are asymptotically consistent, regardless of the survey's weighting structure [7, p. 820].

4 This paper does not jointly model the two measures of utilization, despite efficiency gains that would come from such an exercise. The main reason is that, in addition to the model already being somewhat complex, the main parameter estimates, reported below, do not seem to suffer from low efficiency.

5 An an alternative approach, this paper removes all within-person averages from Equation (3(3) $f(y_{it}|c_i+{\delta }_i{I}_{it}+{\mathbf{X}}_{it}^{\mathrm{\prime }}\beta +m{\overline{I}}_{i\cdot }+{\overline{\mathbf{X}}}_{i\cdot }^{{}^{\mathrm{\prime }}}\mathbf{m}+(I_{it}{\overline{\mathbf{Z}}}_{i\cdot }{)}^{\mathrm{\prime }}\alpha ),$(3) ) and replaced them with ‘initial conditions’ of all control variables – that is, their values during each person's first period in the survey [12, p. 16]. Results from that alternative specification were very similar to those reported below.