Ho Jung S. Yoo. Year. Local population size in a flightless insect: importance of patch structure dependent mortality. Ecological Monographs VOL: pp-pp.

Appendix C. Method of model selection for the logistic and Poisson regressions.

The “complete” logistic regression model for each mortality factor took the general form

,

(C.1)

where p_j = probability an individual died due to factor j, X₁ = density of moth stage specific to factor j, X₂ = patch size, and X₃ = local patch density. The fecundity data were analyzed by multiple Poisson regression (SAS 2001, PROC Genmod). The complete model took the form

,

(C.2)

where = number of eggs per female. For each mortality factor and fecundity, Atkinson’s Q was calculated for the complete model plus all seven possible subsets of explanatory variables, including the null model.

,

(C.3)

where a = 4 (constant), q = number of explanatory variables in the model being tested, = scaling parameter for the complete model. Only observations with no missing data for all three explanatory variables were used. The “best” model among the eight most closely fit two criteria: (1) it minimized Q and (2) the individual plots of mortality against each explanatory variable retained in step 1 confirmed a strong relationship. If none of the models fit both criteria, the null model was selected as the best model.

LITERATURE CITED

SAS Institute. 2001. The SAS System. Version 8. Cary, North Carolina, USA.