We estimate two sets of models to identify the potential effects of small-group regulation on employer coverage among adults in worker families, reflecting the firm size definitions available in the 1996-1998 CPS. First, we estimate coverage among adults in families of workers in firms of under 100 workers. Although workers in firms of more than 50 workers (in some states more than 25) were outside the scope of states’ small-group statutes, the 1996-1998 CPS public use data do not allow us to isolate workers in firms of 50 or fewer employees. Thus, we estimate potential impacts on workers in firms of less than 100, and anticipate weaker results than we might find were we able to isolate the target population more precisely. Second, we estimate coverage among adults in families of workers in firms of under 25; all states’ reform statutes applied to workers in these firm sizes.
For both populations we estimate three alternative specifications of a linear fixed-effects model, as follow:
(1) ERCOVist = f (GMARKETst, SOCIOECONist, STATEs, YEARt),
- ERCOVist is a dummy variable indicating whether individual i has employer coverage either from his/her own employer or as a dependent;
- GMARKETst is a vector of variables which varies by state and year, and describe the number of insurers that sold group major medical coverage, group market concentration, the group market shares of HMOs and commercial insurers, and the average loss ratio of group commercial insurers in the state;
- SOCIOECONist is a vector of variables that include the sociodemographic characteristics of the individual and the employment characteristics of the individual’s greatest family earner; and
- STATE and YEAR are vectors of dummy variables controlling for the state and year.
(2) ERCOVist = f (GMARKETAst, GREGULATIONst, SOCIOECONist, STATEs, YEARt),
where ERCOV, SOCIOECON, STATE and YEAR are as defined above, and
- GMARKETAst is a vector of variables which varies by state and year, and includes only the variables that in our earlier work (Chollet et al., 2000) were insensitive to access regulation in the group market (HMO and commercial insurer market share, and average commercial insurer loss ratios); and
- GREGULATIONst is a vector of categorical and continuous variables identifying whether in its small-group market the state has implemented guaranteed issue (of some or all products, and separately, all-products), the maximum waiting period for coverage of preexisting conditions, and the narrowness of bands on rating for health and for age (if any), and the narrowness of the composite rate band (if any).
(3) ERCOVist = f (GMARKETst,GREGULATIONst ,SOCIOECONist, STATEs, YEARt),
where all vectors are as defined above. Table 1 includes definitions of all of the variables included in the specifications.
The inclusion of the GMARKET and GMARKETA variables in this analysis warrants additional explanation. Differences among states in the structure of their group (and individual) health insurance markets are striking.8 Large-population states characteristically have many fewer insurers per capita — and much larger premium volume per insurer — than small population states. Moreover, commercial insurers in either market have characteristically low average premium volume. The literature estimating economies of scale in health insurance, while scant, uniformly suggests that insurers experience increasing economies of scale (i.e., declining marginal cost) over an extensive range of production, and only the largest insurers experience even constant economies of scale. Thus, all else being equal, one might expect markets with greater concentration to have lower average costs of production, lower prices, and greater coverage. Conversely, more insurers, all else being equal, may foster greater price competition. These competing hypotheses underlie our expectation that measures of market structure may affect (via unobserved prices for insurance) coverage in the small-group market, and also in the individual market.
The alternative specifications of our employer-coverage model differ only in whether GMARKET (a full vector of market variables) or GMARKETA (a partial vector) was entered, and whether small-group regulation variables (GREGULATION) were considered. The first specification of our model omits regulation variables to test solely for the potential influence of market structure on insurance prices (which are unobserved) and on employer coverage. The second specification includes a full array of regulatory variables, but omits market variables that are sensitive to regulation.9 We anticipate that the results of this specification will most closely approximate those of earlier research which did not control for market structure. The third and final specification includes the full array of both market and regulatory variables. We argue that this is the most appropriate specification for estimating the immediate impacts of regulation, controlling for the intervening effects of market structure.