Table 4.

Comparison of Medical Expenditure Panel Survey (MEPS) and Claims-Based Utilization Regressions, 2001–2003 Matched Sample

	Logistic					Negative Binomial
	Any ED Visit					Ambulatory Days
	Claims		MEPS			Claims		MEPS
	Odds	SE	Odds	SE	Odds Ratio Difference p-Valuea	IRR	SE	IRR	SE	IRR Difference p-Valuea
Age (years)
<65	1.47	0.25**	0.97	0.19	0.027	0.829	0.079*	1.002	0.112	0.019
75–84	1.14	0.15	1.13	0.13	0.887	1.248	0.066***	1.222	0.062***	0.585
85+	1.43	0.29*	1.38	0.27	0.734	1.277	0.101***	1.276	0.128**	0.991
Nonwhite	1.01	0.11	0.88	0.12	0.181	0.750	0.050***	0.733	0.056***	0.714
Female	1.07	0.13	1.14	0.15	0.444	1.196	0.048***	1.214	0.062***	0.717
Married	0.73	0.10**	0.69	0.08***	0.435	1.026	0.053	1.000	0.060	0.515
Region
Midwest	0.78	0.12	0.76	0.13	0.776	0.946	0.047	0.985	0.074	0.499
South	0.73	0.09**	0.65	0.09***	0.214	0.944	0.051	0.981	0.061	0.516
West	0.74	0.12*	0.58	0.08***	0.118	0.983	0.058	1.087	0.087	0.126
MSA	0.91	0.11	0.82	0.10	0.183	1.300	0.072***	1.219	0.087***	0.167
Family income
100–199% FPL	1.00	0.13	0.98	0.18	0.876	1.081	0.063	0.921	0.057	0.005
≥200% FPL	0.73	0.10**	0.96	0.18	0.035	1.102	0.057*	1.105	0.065*	0.949
Education (years)
12	0.92	0.13	0.89	0.14	0.710	1.080	0.068	1.201	0.084***	0.031
>12	0.94	0.15	1.06	0.18	0.244	1.153	0.066***	1.282	0.076***	0.037
Perceived health status
Very good health	1.33	0.25	1.37	0.28	0.772	1.229	0.086***	1.290	0.094***	0.329
Good health	1.73	0.27***	1.65	0.31***	0.705	1.400	0.113***	1.538	0.136***	0.062
Fair health	2.10	0.39***	1.90	0.43***	0.518	1.793	0.159***	1.949	0.173***	0.128
Poor health	4.00	0.86***	3.34	0.85***	0.315	2.433	0.276***	2.519	0.305***	0.651
Cognitive limitation	1.21	0.15	1.27	0.22	0.696	0.919	0.068	0.999	0.085	0.134
Activity limitation	0.84	0.10	1.17	0.14	0.000	1.068	0.058	1.207	0.083***	0.007
Private insurance	0.94	0.12	0.98	0.12	0.716	1.087	0.049*	1.194	0.066***	0.063
Medicaid	1.63	0.27***	1.63	0.28***	0.960	1.332	0.100***	1.195	0.087**	0.152
α						0.863	0.036***	0.93	0.037***
Goodness of fit
Archer–Lemeshowb	F=1.58 (p=.12)		F=1.21 (p=.28)			—		—
Hosmer–Lemeshowc	F=1.39 (p=.19)		F=1.22 (p=.28)			F=0.22 (p=.99)		F=0.64 (p=.76)
N	4,045		4,045			4,045		4,045

Notes: All estimates were weighted using the propensity score-derived weight for the matched sample. All statistical tests adjusted for the clustered and stratified design of MEPS using the method of balanced repeated replication (BRR), which also accounts for lower levels of clustering at the household and individual level.

^ap-Value for difference in odds ratio and incidence rate ratios based on t-test of difference in coefficient estimates (Hosmer and Lemeshow 2000) estimated by BRR to account for correlation between pairs of coefficients estimated from the same stratified and clustered MEPS sample.

^bThis is an adaptation of the Hosmer–Lemeshow test by Archer and Lemeshow (2006) for complex surveys.

^cWe estimate a variant of the Hosmer–Lemeshow test as described by Manning, Basu, and Mullahy (2005), which is distributed as F (9,228) using Stata's SVY commands.

^*p<.10

^**p<.05

^***p<.01.

Reference groups are aged 65–74 years, white, not married, Northeast, non-MSA, <100% FPL, <12 years education, excellent health, no cognitive limitation, no activity limitation, no private insurance, no Medicaid, and survey year 2001.

ED, emergency department; FPL, federal poverty line; MSA, metropolitan statistical area.