I am currently analysing my thesis data which is looking at counts of different bat species across a number of sites, using the predictors: canopy gap fraction, canopy height, clutter index and woodland connectivity.
When I run hurdle models on individual species I see slightly different significant predictors (I expected this since different species utilise different landscape features).
However, when I run a GLM with quasi-poisson family on all of the bats lumped together regardless of species (data is not zero-inflated but there is overdispersion, quasipoisson fits the data better than negative binomial model), none of the predictors are significant.
Call: glm(formula = Total_Bats ~ Tree_Height + Clutter_Index + Canopy_Cover + Connectivity, family = quasipoisson, data = speciesbats) Deviance Residuals: Min 1Q Median 3Q Max -9.3579 -4.6243 -0.9103 2.7429 17.1388 Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 3.768092 0.212199 17.757 <2e-16 *** Tree_Height 0.041352 0.022548 1.834 0.0685 . Clutter_Index 0.214933 0.543463 0.395 0.6930 Canopy_Cover -0.007780 0.008568 -0.908 0.3652 Connectivity -0.129322 0.077727 -1.664 0.0981 . --- Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 (Dispersion parameter for quasipoisson family taken to be 28.46795) Null deviance: 4883.0 on 167 degrees of freedom Residual deviance: 4574.2 on 163 degrees of freedom AIC: NA Number of Fisher Scoring iterations: 5
Is it reasonable to make the interpretation that since different predictors are of different importance to different species, that these effects are hidden when all bats are put together? This doesn't seem right as I would still have expected woodland connectivity to have been a positive predictor for all bat species. Or is it an indication of a problem with my model?