TABLE 2—

Manhattan Block Distance (in Miles) to the Nearest Supermarket Regressed on Neighborhood Characteristics^a With Ordinary Least Squares Regression and Moving Average Spatial Regression^b

	Model 1		Model 2
	OLS Coefficient (SE)	Spatial Coefficient (SE)	OLS b Coefficient (SE)	Spatial Coefficient (SE)
African American, %
1.98–62.63 (Medium)	0.086 (0.068)	0.092 (0.067)	−0.042 (0.096)	−0.028 (0.097)
63.11–98.43 (High)	0.273 (0.087)**	0.295 (0.088)***	−0.004 (0.316)	−0.002 (0.314)
Poor, %
5.03–17.20 (Medium)	0.010 (0.068)	−0.021 (0.068)	−0.048 (0.098)	−0.052 (0.097)
17.23–81.96 (High)	0.777 (0.092)***	0.703 (0.092)***	0.050 (0.238)	−0.190 (0.239)
Medium African American × medium poverty			0.115 (0.140)	0.078 (0.140)
Medium African American × high poverty			0.999 (0.268)***	1.125 (0.266)***
High African American × medium poverty			0.365 (0.340)	0.319 (0.337)
High African American × high poverty			0.956 (0.418)*	1.153 (0.391)**
P		<.001		.002
Adjusted R2	0.21		0.22
Log likelihood		−2699		−2691
Likelihood ratio test				15.83c**
Moran’s I statistic on residuals	0.008	< 0.001	0.018	0.004
Normal statistic	2.14*	0.37	4.31***	1.08

Note. OLS = ordinary least squares.

^aAdjusted for population density.

^bA binary 5-mile spatial weights matrix was used. This decision was based on exploratory analysis with an empirical semivariogram,54,55 which showed a 5-mile range for spatial autocorrelation for distance to the nearest supermarket. This distance conformed to available data on average travel distance to the supermarket.56 We also compared results with 2 other specifications. Results were similar based on a binary 3-mile spatial weights matrix. The interaction coefficients with a rook (neighborhoods sharing a common border) spatial weights matrix were in the correct direction but were not statistically significant. Neither alternative spatial model corrected for spatial autocorrelation. The moving average spatial regression technique is described elsewhere.57,58

^cSpatial regression model 1 compared with spatial regression model 2. Critical value for χ²=0.01, 4 df=13.28.

*P < .05; **P < .01; ***P < .001.