Results
Model output data were never normally distributed (Shapiro-Wilk – Diversity: W = 0.928, P < 0.001; Evenness:W = 0.781, P < 0.001) nor homoscedastic (Bartlett – Diversity: K2 = 105.69, df = 8,P < 0.001; Evenness: K2 = 651-64, df = 8, P < 0.001); thus nonparametric tests determined the effect of diet breadth on Shannon Diversity and Evenness. Shannon Diversity increased with diet breadth (Kruskal-Wallis:X2 = 974.05, df = 8, P < 0.001; Figure 1a) but Shannon Evenness was similar for every diet, regardless of number of food types (Kruskal-Wallis:X2 = 13.99, df = 8, P = 0.08; Figure 1b). One interesting observation was that variation in values of Evenness decreased with increasing diet breadth. Variability (measured as the coefficient of variation) decreased with increasing diet breadth (linear regression: F1,7 = 13.72, P = 0.008, r2 = 0.614). However, the relationship between variability in J ’ and diet breadth is non-linear: a reciprocal function provides a better fit between expected and observed values (F1,7 = 118.8, P < 0.001,r2 = 0.936). Another interesting observation from the null model analysis is that the value of J ’ under the assumption of equal preferences is constant (0.434294) regardless of the number of food types. Thus, the results can be more clearly visualized by using the standardized value (observed value/equal preferences value).