Table 2

Results of the empty multilevel logistic models and empty spatial logistic models for healthcare utilisation, France, 1998 and 2000

No regular primary care physicianHigh percentage of specialist consultations
*p<0.05; **p<0.01; ***p<0.001 (p values are two sided). †The multilevel model parameters were estimated by the Markov chain Monte Carlo method (MLwiN). The Wald test was used for the area level variance. To compute the Moran’s I, we used the area level residuals of 67% of the municipalities (n = 2167) for the variable regarding regular primary care physicians, and the residuals of 73% of the municipalities (n = 3227) for the variable regarding specialty care use (the other municipalities had no adjacent municipality in the sample). All broad area residuals were used to compute the Moran’s I. Based on the assumption of normality for the area level residuals, the Moran’s I is normal under the null hypothesis, with a mean approximately equal to 0 and a known variance. We computed a two tailed p value for the Moran’s I. Scaled deviances come from multilevel models estimated using the glimmix macro. ‡Spatial model parameters were estimated with glimmix. The Wald Z test was used for the covariance parameters. The parameter σ2 is the partial sill, σ12 is the nugget effect, and three times the parameter ρ is the range of the model (the distance beyond which the correlation is less than 5% of the correlation at distance 0).
Municipality level multilevel model†
Area level variance σu2 (SE)0.382 (0.133)**0.175 (0.059)**
Moran’s I for area residuals (SE)0.33 (0.02)***0.20 (0.02)***
Scaled deviance4738.98841.2