Table 2.

Are outliers acting more like their ethnic group, or their network group?

Regression	Coefficient (SE)
A. Regression with outlier A as the base, who is E-A with a majority of ties to the K-A network
E-A network	−0.171 (0.030)*
K-A network	0.017 (0.055)
V-A network	−0.047 (0.044)
B. Regression with outlier B as the base, who is K-A with a majority of ties to the E-A network
E-A network	−0.018 (0.066)
K-A network	0.169 (0.060)*
V-A network	0.200 (0.066)*
C. Regression with outlier C as the base, who is K-A and has ties split between the K-A network, the V-A network, and other nonfishers
E-A network	−0.144 (0.067)*
K-A network	0.045 (0.057)
V-A network	0.073 (0.042)
D. Regression with outlier D as the base, who is K-A and has a majority of ties to the V-A network
E-A network	−0.244 (0.051)*
K-A network	−0.057 (0.044)
V-A network	−0.028 (0.040)

Values shown are coefficients (and SEs) from four negative binomial regressions (A–D). The dependent variable is shark per fishing set in Hawaii’s tuna longline fishery from 2008 to 2012 (n = 12,062). Network variables account for observed homophilous groupings along ethnic lines; outliers designate circled nodes in Fig. 2, which are independently tested to determine whether their rates of shark bycatch are significantly different from their ethnic or network group. Controls include target species catch, vessel length, number of hooks, set location, soak time, temperature, type of bait, seasonality, lunar variability, and annual variability (see Table S3 for variable descriptions and Table S5 for full model results). SEs are clustered to account for multiple observations of 120 individual fishers. In each model (A–D), the network groups in question are bold.

^*Significance at <0.05.