# stepAIC - summary(binomial GLM) - insignificant variables

#### Audrey

##### New Member
Why stepAIC gives a model with unsignificant variables in the summary(model) ?

I would like to know what environmental variables allows to explain the presence of several species (binomial glm). I used a stepAIC procedure to select the best model.
But when I do a summary(model), some variables are not significant (according to pvalues).
How can I interpret these results ? Can one says that all the variables are usefull to explain the presence of the species but that only some of them are significant ?
Must I interpret coefficient estimates of unsignificant variables ?

Thank you very much for your help !

Here are the results of my summary(model) :

> summary(LiDAR_selection10[[1]])

Call:

glm(formula = dfSURVEY_presence_absence[, i] ~ Chm10 + Wetness10 +

Light.Bode10 + Chm10:Wetness10 + Chm10:Light.Bode10, family = binomial)

Deviance Residuals:

Min 1Q Median 3Q Max

-1.9818 -1.0054 0.5533 0.9494 1.7402

Coefficients:

Estimate Std. Error z value Pr(>|z|)

(Intercept) 27.243650 16.103729 1.692 0.0907 .

Chm10 -0.983108 0.498498 -1.972 0.0486 *

Wetness10 -1.159092 0.926263 -1.251 0.2108

Light.Bode10 -0.181315 0.096564 -1.878 0.0604 .

Chm10:Wetness10 0.042711 0.029435 1.451 0.1468

Chm10:Light.Bode10 0.006562 0.003038 2.160 0.0308 *

---

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

Null deviance: 84.635 on 62 degrees of freedom

Residual deviance: 72.251 on 57 degrees of freedom

AIC: 84.251

#### hlsmith

##### Less is more. Stay pure. Stay poor.
Interesting, well if you define model fit by stepAIC, I guess this is a possible outcome. If you define model by p value of terms that would be another approach. Stepwise approach are frown upon by many in that they automate the model building process and take away our own knowledge of the context for the model.

If you used the AICmodel you would have to mention those insignificant variables were controlled for, because if they weren't in the model the other terms may have been different.

#### Audrey

##### New Member
Thank you very much for your help ;

I'm sorry but what do you mean by "those insignificant variables were controlled for" ?

Last edited:

#### hlsmith

##### Less is more. Stay pure. Stay poor.
Variables that were apart of the model with pvalues > 0.05. So if they were in the model they were controlled for.

#### Audrey

##### New Member
I'm so sorry; I'm French and I don't understand what "controlled for" means ..
Does that mean that I must say that I test these variables but they were not significant ? That they do not explain my species presence absence ?