Table 3

Sensitivity Analyses, Average Marginal Effects Reported: Summary of Key Findings Compared with Base Probit Models

Marginal Effect on Water-Sharing ArrangementsMarginal Effect on First Nations Water-Sharing Arrangementsa
First Nations water systemsBase probit model (N=145)−0.438***
(0.086)
Sensitivity 1: bivariate probit estimation (N=145)−0.523***
(0.050)
Sensitivity 2: feasible distance Sample (N=95)−0.382***
(0.074)
Sensitivity 3: addition of FNLMA variable (N=145)b−0.439***
(0.093)
Municipal water systemsBase probit model (N=565)0.034
(0.027)
Sensitivity 1: bivariate probit estimation (N=565)0.026
(0.040)
Sensitivity 2: feasible distance sample (N=551)0.031
(0.025)
Full sample with interaction effectBase probit model (N=710)−0.075**−0.331***
(0.030)(0.086)
Sensitivity 1: bivariate probit estimation (N=710)−0.101**−0.348***
(0.050)(0.091)
Sensitivity 2: feasible distance sample (N=646)−0.042*−0.335***
(0.024)(0.097)

Note: Standard errors are in parentheses and are clustered by census subdivision (i.e., community housing the water system). The dependent variable equals 1 if the drinking water advisory was in effect at some point during the study period (2009–2010), 0 otherwise.

  • a Stata does not generate marginal effects for interaction terms (in our case, i.WSA##i.FN) using the standard “margins” command. The marginal effect for this interaction term was calculated separately using the following command: margins WSA, dydx(FN) pwcompare(effects). This marginal effect compares First Nations water systems supplied through water-sharing arrangements to First Nations water systems that are independently supplied.

  • b Marginal effect from probit model specification reported; result remains consistent when the model is run as a bivariate probit. Full probit and bivariate probit results are provided in Appendix Table A2.3.

  • * p < 0.1;

  • ** p < 0.05;

  • *** p < 0.01.