IJ obtained inspection data from local health agencies in Connecticut and from the Department of State in New York. The full dataset for Connecticut and New York comprises 2,594 inspections across 1,988 firms for 2017 and 2018, years during which New York licensed manicurists and Connecticut did not. However, I excluded firms on Long Island from the analysis because they are separated by the Long Island Sound. ¹  This reduced the sample to 2,148 inspections across 1,604 firms. Figure 6 displays the geographic location of inspections, with the color of the dots representing the distance to the Connecticut/New York border. Dots with similar colors are assumed to reflect more similar businesses and business environments, whereas dots with different colors are assumed to reflect less similar businesses and business environments. As described above, inspections of businesses closer to the border (i.e., the blue dots) receive greater weight in my analysis. Despite differences in the states’ inspection regimes, their forms are similar in that they list possible violations and require inspectors to identify actual violations. Thus, health and safety violations can be distinguished from other types of violations (like licensing violations) in the data, quantified, and compared across states.

There are more possible health and safety violations in Connecticut (roughly 30 to 40, depending on the locality) than in New York (seven), so comparing the raw count of violations per inspection would be misleading. I therefore created two standardized variables that account for the different numbers of possible violations. The first outcome variable I created by transforming the count of violations into standard deviation scores, often called “z-scores,” specific to each state. A score of zero for a given inspection would mean the inspection resulted in the average number of health and safety violations per inspection for the state, while a positive or negative score would mean the inspection resulted in an above or below average number of health and safety violations for the state. The second outcome variable I created by dividing the number of violations by the number of possible violations (i.e., the rate of violations per possible violation). Higher values indicate an inspection resulted in a higher rate of violations.

Figure 6. Locations of Connecticut and New York Nail Salon Inspections

The analysis weights nail salons closer to the border more heavily, as they, and the locations in which they operate, are assumed to be more similar

My analytical strategy started with simple comparisons of inspection outcomes and then proceeded to the regression discontinuity analysis. First, I calculated descriptive statistics, including the average violation z-score and the average rate of violations for nail salon inspections in Connecticut and New York. I calculated the same statistics restricted to nail salons within the bandwidth around the border—the businesses and business environments assumed to be most similar. Second, I estimated the relationship between the salons’ distance to the border and inspection outcomes in each state. The expected outcome of an inspection for a nail salon in Connecticut compared to New York is estimated as the difference between the predicted outcome in Connecticut and the predicted outcome in New York for a hypothetical nail salon that is located on the border. For both the simple comparisons and the more sophisticated analyses, I used both the violation z-score and violation rate as outcome variables.

Whether licensing is the cause of any differences in inspection outcomes depends on the extent to which businesses on either side of the border are essentially similar but for licensing conditions (i.e., randomly distributed within the bandwidth around the border). I therefore conducted tests to assess the validity of the study design. First, I examined whether census block groups near the border were similar in population size, percentage of the population with a bachelor’s degree, and median household income; I reran my analysis adjusting for these characteristics. Second, to account for the possibility that some business owners might have chosen to set up shop in Connecticut rather than New York precisely to avoid New York’s license, I reran the analysis excluding those businesses closest to the border.