I'm making a model using a dataset with 290 observations, about 140 events, and five categorical variables in addition to one continuous variable (Days). There is also one interaction between one categorical variable and the continuous variable. Here are the results:
glm(formula = Quantified ~ List.Downlist + Days + Region + Vert.Invert.Plant +
New.Regs * Days, family = binomial(link = "logit"), data = df2)
Deviance Residuals:
Min 1Q Median 3Q Max
-2.8002 -0.9193 -0.5522 0.9584 2.8144
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -146.00183 49.36678 -2.957 0.00310 **
List.DownlistListing 0.95739 0.31218 3.067 0.00216 **
Days 0.04008 0.01370 2.925 0.00344 **
Region2 1.88350 0.54812 3.436 0.00059 ***
Region3 0.49631 0.71742 0.692 0.48906
Region4 0.53235 0.45920 1.159 0.24634
Region5 1.18442 0.72693 1.629 0.10324
Region6 1.66902 0.54333 3.072 0.00213 **
Region7 2.10745 1.04591 2.015 0.04391 *
Region8 0.92876 0.50820 1.828 0.06762 .
RegionForeign 0.36371 0.64169 0.567 0.57086
Vert.Invert.PlantPlant 0.93983 0.42213 2.226 0.02599 *
Vert.Invert.PlantVertebrate 1.17106 0.37779 3.100 0.00194 **
New.RegsBefore Regs 141.82501 49.27381 2.878 0.00400 **
Days:New.RegsBefore Regs -0.03918 0.01371 -2.858 0.00426 **
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for binomial family taken to be 1)
Null deviance: 399.19 on 290 degrees of freedom
Residual deviance: 324.46 on 276 degrees of freedom
AIC: 354.46
Number of Fisher Scoring iterations: 5
The coefficient for the variable "New.Regs" is huge, which gives me the sense that something is wrong. Perhaps overfitting? Or is overfitting not a problem because this is an explanatory and not a predictive model? Also, the New.Regs variable has two levels (Before/After). After has only 39 observations, with 22 of those being events.
The range for Days is 0-3810, and the observations for After all take place after day 3500.
Also, any advice on model evaluation in general would be much appreciated.
