. 2016 Feb 23;127(600):445–494. doi: 10.1111/ecoj.12282

Table 6.

Choice Models with Different Choice Set Radii and Stratified by LSOA Population Growth

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Choice set radius						Population growth
	10 km	8 km	6 km	4 km	2 km	Urban 3 km Rural 7 km	Lower 4 quintiles	Top quintile
QOF 2006/7 total points	0.00222***	0.00217***	0.00217***	0.00230***	0.00215***	0.00219***	0.00208***	0.00285***
QOF 2006/7 total points	(0.00016)	(0.00015)	(0.00015)	(0.00016)	(0.00018)	(0.00016)	(0.00017)	(0.00042)
Distance	−0.7512***	−0.7717***	−0.7457***	−0.6715***	−2.0356	−0.6631***	−0.7506***	−0.7554***
Distance	(0.0163)	(0.0167)	(0.0273)	(0.1728)	(2.9421)	(0.0255)	(0.0184	(0.0337)
MRS	0.00296***	0.00281***	0.00291***	0.00342***	−0.00106	0.00331***	0.00277***	0.00377***
MRS	(0.00022)	(0.00021)	(0.00023)	(0.00091)	(0.00153)	(0.00027)	(0.00023	(0.00058)
McFadden R²	0.3991	0.3665	0.3151	0.2345	0.1081	0.2355	0.3938	0.4224
N LSOA	2,870	2,806	2,670	2,428	1,925	2,676	2,296	574
N practices	973	923	853	738	605	836	945	772
N patients	3,291,581	3,190,284	3,003,036	2,661,832	1,844,207	2,773,892	2,680,561	670,000
Percentage of East Midlands population choosing practice outside this choice set	0.21	1.69	4.51	11.00	32.35	8.56

Conditional logit models of choice of practice. Choice sets are all practices within stated distance of LSOA centroid and, if there are more than 30 such, the 30 nearest to the LSOA centroid. Choice set for the models stratified by LSOA population growth is all practice within 10 km, and if there are more than 30 such, the 30 nearest to the LSOA centroid. All models have the same covariates and cubic distance specification as Table 2. The distance coefficient is ${\hat{β}}_{d} + 2 {\hat{β}}_{d^{2}} \bar{d} + 3 {\hat{β}}_{d^{3}} \bar{d^{2}}$ where $\bar{d}$ and $\bar{d^{2}}$ are computed as the mean of the LSOA centroid to practice distance and squared distance for the sample with 10 km radius choice set for all models to ensure that the computed distance coefficient varies across models only because of variation in the linear, quadratic and cubic distance coefficients. MRS is the quality coefficient divided by the distance coefficient. Standard errors clustered at LSOA level are in parentheses. *p < 0.05, **p < 0.01, ***p < 0.001.