Appendix B: Results

Table B1 presents the results of the geographic regression discontinuity analyses comparing nail salon inspection outcomes in Connecticut and New York. The coefficients represent the expected difference in an inspection outcome between a nail salon in New York and a nail salon in Connecticut. For example, the coefficient of 0.049 in the violation rate model indicates that an inspection in New York would be expected to result in a violation rate about 4 percentage points higher than an inspection in Connecticut. The confidence intervals, shown in parentheses adjacent to coefficients, reflect a range of values within which the “true” difference likely falls. For example, the coefficient of 0.049 has a confidence interval ranging from 0.025 to 0.074, indicating that the true difference in the expected violation rate could be about 2 percentage points on the lower end or about 7 percentage points on the upper end.

Table B1. Regression Discontinuity Results for Nail Salon Inspection Outcomes in Connecticut & New York

	Coefficient (95% CI)
	Conventional	Robust
Violation Z-Score
Model	0.662 (0.370, 0.953)	0.697 (0.342, 1.051)
Model + Covariates	0.671 (0.429, 0.913)	0.691 (0.395, 0.987)
Violation Rate
Model	0.049 (0.025, 0.074)	0.048 (0.019, 0.077)
Model + Covariates	0.049 (0.029, 0.068)	0.047 (0.023, 0.070)

Note. All coefficients are statistically significant at the 5% level. CI: Confidence Interval.

Table B2 reports descriptive statistics for inspection outcomes for firms in Connecticut and New York within the bandwidth around the border.

Table B2. Descriptive Statistics for Nail Salon Inspection Outcomes in Connecticut & New York

	Connecticut	New York
Total Observations	690	1,458
Bandwidth	18.295	18.295
Effective Observations	320	334
Mean Violation Z-Score	-0.166	0.067
SD Violation Z-Score	0.897	1.03
Mean Violation Rate	0.018	0.052
SD Violation Rate	0.032	0.095

Note. Bandwidth units are miles. Bias bandwidth is 33.055. Inspections on Long Island were excluded. SD: Standard deviation.

Figure B1 is a visualization of the discontinuity in violations at the border of Connecticut (left side of chart) and New York (right side of chart). The geographic regression discontinuity design here involves estimating the trends in inspection outcomes in both states as the distance to the border decreases. The blue dots in the figure represent average outcomes among subsets of observations with similar distances to the border. The red lines represent trends in those average outcomes. The treatment effect is the difference between the intercepts of the two trendlines.

Figure B1. Regression Discontinuity Plot for Connecticut and New York Nail Salon Inspection Outcomes

Table B3 shows the results of the geographic regression discontinuity analyses comparing barbershop inspection outcomes in Alabama and Mississippi.

Table B3. Regression Discontinuity Results for Barbershop Inspection Outcomes in Alabama & Mississippi

	Coefficient (95% CI)
	Conventional	Robust
Model	-0.076 (-0.134, -0.017)	-0.083 (-0.158, -0.009)
Model + Covariates	-0.069 (-0.120, -0.019)	-0.078 (-0.145, -0.011)

Note. All coefficients are statistically significant at the 5% level. CI: Confidence Interval.

Table B4 reports descriptive statistics of inspection outcomes for firms in Alabama and Mississippi within the bandwidth around the border.

Table B4. Descriptive Statistics for Barbershop Inspection Outcomes in Alabama & Mississippi

	Alabama	Mississippi
Total Observations	896	2,322
Bandwidth	33.298	33.298
Effective Observations	81	478
Percent Passed	97.5	95.4

Note. Bandwidth units are miles. Bias bandwidth is 68.255.

Figure B2 is a visualization of the discontinuity in inspection outcomes at the border of Alabama (left side of chart) and Mississippi (right side of chart).

Figure B2. Regression Discontinuity Plot for Alabama and Mississippi Barbershop Inspection Outcomes

Tables B5 and B6 show the results of the sensitivity analyses. Table B5 reports only the results for the violation z-score dependent variable, but results were comparable for the violation rate dependent variable. Overall, the magnitudes of coefficients changed—which is to be expected given that regression discontinuity design model estimates are most influenced by observations closest to the cutoff—but substantive conclusions did not. ¹

Table B5. Regression Discontinuity Model of Nail Salon Violations (Z-Score) in Connecticut & New York with Donut Hole Approach

	Coefficient (95% CI)
	Conventional	Robust
Full
Model	0.662 (0.370, 0.953)	0.697 (0.342, 1.051)
Model + Covariates	0.671 (0.429, 0.913)	0.691 (0.395, 0.987)
Donut Radius = 1 Mile
Model	0.545 (0.220, 0.870)	0.590 (0.201, 0.980)
Model + Covariates	0.689 (0.439, 0.939)	0.728 (0.418, 1.037)
Donut Radius = 2 Miles
Model	0.476 (0.055, 0.898)	0.541 (0.025, 1.056)
Model + Covariates	0.559 (0.202, 0.917)	0.589 (0.120, 1.057)
Donut Radius = 3 Miles
Modelns	0.377 (-0.081, 0.835)	0.430 (-0.140, 1.000)
Model + Covariatesns	0.398 (0.008, 0.788)	0.386 (-0.144, 0.916)

Note. All coefficients are statistically significant at the 5% level, except the two indicated with “ns.” CI: Confidence Interval.

Table B6. Regression Discontinuity Model of Barbershop Inspection Outcomes in Alabama & Mississippi with Donut Hole Approach

	Coefficient (95% CI)
	Conventional	Robust
Full
Model	-0.076 (-0.134, -0.017)	-0.083 (-0.158, -0.009)
Model + Covariates	-0.069 (-0.120, -0.019)	-0.078 (-0.145, -0.011)
Donut Radius = 5 Miles
Model	-0.090 (-0.152, -0.027)	-0.098 (-0.178, -0.018)
Model + Covariates	-0.084 (-0.138, -0.030)	-0.093 (-0.164, -0.022)
Donut Radius = 6 Miles
Model	-0.097 (-0.163, -0.031)	-0.108 (-0.191, -0.024)
Model + Covariates	-0.092 (-0.148, -0.036)	-0.102 (-0.175, -0.029)
Donut Radius = 7 Miles
Model	-0.105 (-0.175, -0.035)	-0.118 (-0.205, -0.031)
Model + Covariates	-0.098 (-0.157, -0.040)	-0.109 (-0.184, -0.034)

Note. All coefficients are statistically significant at the 5% level. CI: Confidence Interval.