hello
I need a clarification.
in logistic regression, saturated model having all combinations and
interactions of variables should be constructed in this way ? :
> rsat2=glm(cbind(landp,landa) ~
as.factor(rlito)*as.factor(rslp)*as.factor(rasp)*as.factor(rplc)*as.factor(rwi),familybinomial(link=logit),data=rdf46)
then
using stepAIC to eliminate some models having high AIC values.
> stpAIC_r2=stepAIC(rsat2,list(lower=~1,upper=formula(rsat2)),trace=F)
But obtained final model looks very intricate and I couldn't interpret it.
> stpAIC_r2$anova
Do I have to consider last line (13) ?
1
2 -
as.factor(rlito):as.factor(rslp):as.factor(rasp):as.factor(rplc):as.factor(rwi)
3 -
as.factor(rslp):as.factor(rasp):as.factor(rplc):as.factor(rwi)
4 -
as.factor(rlito):as.factor(rslp):as.factor(rasp):as.factor(rwi)
5 -
as.factor(rlito):as.factor(rslp):as.factor(rplc):as.factor(rwi)
6 -
as.factor(rlito):as.factor(rslp):as.factor(rwi)
7 -
as.factor(rlito):as.factor(rasp):as.factor(rplc):as.factor(rwi)
8 -
as.factor(rlito):as.factor(rplc):as.factor(rwi)
9 -
as.factor(rlito):as.factor(rslp):as.factor(rasp):as.factor(rplc)
10 -
as.factor(rslp):as.factor(rasp):as.factor(rplc)
11 -
as.factor(rlito):as.factor(rslp):as.factor(rplc)
12 -
as.factor(rlito):as.factor(rasp):as.factor(rplc)
13 -
as.factor(rlito):as.factor(rslp):as.factor(rasp)
Df Deviance Resid. Df Resid. Dev AIC
1 NA NA 191 458.3157 3310.353
2 1 0.3148536 192 458.6305 3308.668
3 24 31.8915731 216 490.5221 3292.560
4 7 1.3523216 223 491.8744 3279.912
5 13 13.1002785 236 504.9747 3267.012
6 19 23.7270352 255 528.7017 3252.740
7 7 4.8455686 262 533.5473 3243.585
8 9 5.7980351 271 539.3453 3231.383
9 8 7.9086994 279 547.2540 3223.292
10 40 50.8028626 319 598.0569 3194.095
11 17 12.8919748 336 610.9489 3172.987
12 14 8.1175215 350 619.0664 3153.104
13 18 27.9371216 368 647.0035 3145.041
I will be appreciate if you explain.
regards
Ahmet Temiz
--
Open WebMail Project (http://openwebmail.org)
--
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.
