Gayle Rubin, a professor of anthropology at Michigan and a leading scholar of L.G.B.T. studies, praises it as a pioneering attempt at "moral levelling," where the old prudish act of exposing deviants and curing them of deviance changed to the project of finding out what deviants did, and why it was, on inspection, usually no more deviant than what the rest of us did.
The stop criterion of the algorithm was a deviance change < 0.01.
Table 2 Generalized additive models (GAM) for the Poisson family with a log link Anova Residual deviance Change in deviance P-value 1) Null model 4569.7 - 2) Year 4541.7 27.98 <0.0001 3) Year + Age 3194.0 1347.74 <0.0001 4) Year + Age + Sex 2767.4 426.60 <0.0001 5) Year + Age + Sex + Area 2709.4 57.99 <0.0001 The models are evaluated with the response variable being the incidence of fractures.
Identified covariates were forced into the model and the observed QoL scores tested as to whether they also contributed to the prediction of missingness [ 14], as indicated by a reduction in deviance (change in -2*log likelihood).
The minimum adequate model from the analysis including shape variables eliminated effects of RW1, Patch and Treatment, and retained only a significant effect of the second shape axis, RW2 (deviance change = 18.86, df = 1, P < 0.001, b = -44.8 ± 13.1).
A GLM testing for an effect of treatment on survival rate among the E-ecotype snails confirmed significant variation among groups (deviance change = 16.35, df = 2, P < 0.001) with snails from the wave-exposed treatment showing significantly higher survival than the other two groups (wave vs. control: z = 2.33, P = 0.020, crab vs. control: z = 1.85, P = 0.064).
Survival of E-ecotype individuals was much higher than S-ecotype individuals (deviance change = 32.58, df = 1, P < 0.001), but varied among treatments: 7 of 9 snails from the wave-simulated treatment survived, 7 of 25 from the control treatment and 2 of 26 from the crab treatment.
Variables were added and removed based on significant improvement in the mean deviance after changes to the model.
In addition, we considered both the statistical and qualitative aspects of the observed geographic patterns, tempering interpretation of areas of sparse data by considering tests of global deviance and changes in the optimal span size along-side visual inspection of the patterns in our interpretation of results.
