Table 2.

Goodness-of-Fit Indicators for Several Logistic Models Predicting High (Upper 10%) Expenditures in Year 2, MEPS Panels 1–4

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9	Model 10
Log-likelihood	−15,297	−14,093	−13,804	−14,556	−14,625	−13,596	−14,017	−13,694	−13,483	−13,845
Hosmer–Lemeshow test	4.29*	0.62	0.55	4.46*	1.62	0.80	0.65	0.29	0.74	0.58
Pseudo-R2	0.118	0.188	0.204	0.161	0.157	0.216	0.192	0.211	0.223	0.202
Correlation	0.310	0.414	0.419	0.385	0.380	0.441	0.420	0.433	0.452	0.436
c-statistic	0.765	0.813	0.827	0.796	0.793	0.833	0.815	0.830	0.836	0.821
BIC	−3,865	−6,165	−6,809	−5,238	−5,144	−7,115	−6,208	−6,920	−7,276	−6,487
Change in BIC	—	−2,300	−2,944	−1,373	−1,279	−306	+601	−111	−161	+628
Comparison model	—	1	1	1	1	3	3	3	6	6

^*p<0.05.

Model 1 (baseline): age, gender, insurance, panel.

Model 2: Model 1+number of chronic conditions.

Model 3: Model 1+DCG categories.

Model 4: Model 1+key condition indicators.

Model 5: Model 1+perceived health and functioning.

Model 6: Model 1+number of chronic conditions+DCG categories.

Model 7: Model 1+number of chronic conditions+key condition indicators.

Model 8: Model 1+DCG categories+key condition indicators.

Model 9: Model 1+number of chronic conditions+DCG categories+perceived health and functioning.

Model 10: Model 1+number of chronic conditions+key condition indicators+perceived health and functioning.

BIC, Bayesian information criterion; DCG, diagnostic cost group; MEPS, Medical Expenditure Panel Survey.