R help - Oct 2009 - lmer family=binomal p-values

Hi, 

If any one has time I need some help understanding the P-values given in the
lmer output.

Using AIC for model selection I find my minimal model is
FOLLOW~MOVERSTATUS+DISTANCE however it appears DISTANCE is not significant at
95% confidence, see output quoted below.
However, removing DISTANCE gives a higher AIC=433.5, therefore I will keep it
in, but am confused as to what is adds to the model.
Are the p values given in output to be taken lightly?  

Thanks
Krys
>model<-lmer(FOLLOW~MOVERSTATUS+DISTANCE+(1|DATE), REML=FALSE,
family=binomial,data=m1)
>model
#Generalized linear mixed model fit by the Laplace approximation 
#Formula: FOLLOW ~ MOVERSTATUS + DISTANCE + (1 | DATE) 
#  Data: m1 
#   AIC   BIC logLik deviance
#430.8 448.4 -211.4    422.8
# Random effects:
# Groups Name            Variance Std.Dev.
# DATE   (Intercept)     0.5957   0.77182 
# Number of obs: 611, groups: DATE, 122

#Fixed effects:
#                            Estimate    Std. Error   z value    Pr(>|z|)    
#(Intercept)              2.06881    0.31086     6.655      2.83e-11 ***
#MOVERSTATUSSUB   -0.82564    0.28118    -2.936     0.00332 ** 
#DISTANCE                0.02311    0.01298     1.780      0.07503 .  


--------------------------------------------
Krystyna Golabek
Room C79
School of Biological Sciences
University of Bristol
Woodland Road
Bristol
BS8 1UG
U.K. 

Email: bzkag@bristol.ac.uk

Office: 0117 9545945

Web: http://www.bio.bris.ac.uk/people/staff.cfm?key=1183

Pied Babbler Research Project
PO BOX 64
Van Zylsrus
Northern Cape
South Africa

Web: http://www.fitzpatrick.uct.ac.za/docs/piedbabb.html

Tel: +27 (0)53 781 0390
Email: mkproj@mweb.co.za
(Please put my name in the subject box when using the above 'mkproj'
email address)

--------------------------------------------





 		 	   		  
_________________________________________________________________
Chat to your friends for free on selected mobiles

	[[alternative HTML version deleted]]

Krystyna Golabek wrote:
> 
> 
> Using AIC for model selection I find my minimal model is
> FOLLOW~MOVERSTATUS+DISTANCE however it appears DISTANCE is not significant
> at 95% confidence, see output quoted below.
> However, removing DISTANCE gives a higher AIC=433.5, therefore I will keep
> it in, but am confused as to what is adds to the model. 
> 
> 
> Are the p values given in output to be taken lightly?  
> 
> 
> #Fixed effects:
> #                            Estimate    Std. Error   z value   
Pr(>|z|)
> #(Intercept)              2.06881    0.31086     6.655      2.83e-11 ***
> #MOVERSTATUSSUB   -0.82564    0.28118    -2.936     0.00332 ** 
> #DISTANCE                0.02311    0.01298     1.780      0.07503 .  
> 
> 
AIC based removal does not use the sacred p=0.05 criterion, but a penalized
criterion that with some simplifications is around p=.15. Which is good so,
because even that can lead to simplifications going too far. So keep it;
when you repeat the experiment identically, it is required.

(Frank Harrell: http://markmail.org/message/jdik3brpe22x7k7h)

What is wrong with p=0.075? Only VERY bad journals nowadays use the p=0.05
guillotine as publication cutoff.


Dieter


-- 
View this message in context:
http://www.nabble.com/lmer-family%3Dbinomal-p-values-tp25958695p25959790.html
Sent from the R help mailing list archive at Nabble.com.