[R] Prediction on link- and responselevel for glm

Fredrik Lundgren fredrik.bg.lundgren at gmail.com
Mon Apr 2 19:31:08 CEST 2012


Dear R-ers,

We are doing prediction of risk for aortic aneurysm in screened  
subjects based on different riskfactors. When we do the prediction on  
the linklevel and transform the prediction and the limits of the 95%- 
confidence intervall (with exponentiation and then transformation from  
odds to probability) we get asymmetric intervalls which seem  
reasonable. When we use prediction on the responselevel (which is be  
on the probability level) the intervalls are completely symmetric. The  
figures are also different. Is something wrong with the prediction or  
with my use of the glm program?

Any help much appreciated!

Sincerely yours

Fredrik Lundgren

This is for an individual with certain riskfactors

Formation of confidence interval with linklevel (prob transforms from  
odds to probability)  fit = -4.7513497 and se.fit = 0.2474956
point value				lower limit								upper limit
prob(exp(-4.7513497)) 	prob(exp(-4.7513497-1.96*0.2474956))	 
prob(exp(-4.7513497+1.96*0.2474956))


Formation of confidence interval with responselevel  fit = 0.008566015  
and se.fit= 0.002101891

point value				lower limit								upper limit
0.008566015 	p0.008566015-1.96*0.002101891	0.008566015+1.96*0.002101891


########################

Fredrik Lundgren
fredrik.bg.lundgren at gmail.com

Engelbrektsgatan 31
582 21 Linköping
tel		013 - 47 30 117
mob 	0706 - 86 39 29

Sommarhus: Ljungnäs 158
380 30 Rockneby
0480 - 650 98



More information about the R-help mailing list