I'm working on a database with matching pairs (1 to 5/6 matching) and am trying to obtain the Odds Ratio (95% confidence interval) for my variables using conditional logistic regression.
G1_LR<-glm(formula=MCM.POSS~factor(AED1TRIM_EXP)+factor(BMI)+factor(FOLIC_ACID),family=binomial(),data=G1)
summary(clogit((MCM.POSS)~factor(AED1TRIM_EXP)+factor(BMI)+factor(SMOKING)+factor(FOLIC_ACID)+strata(BABY_PATID),data=G1))**
Thereafter, I found that that two of my predictor variables had a quasi-complete separation so I decided to try the penalized maximum likelihood estimation method (logistf).
G1_edited<-logistf(formula=MCM.POSS~factor(AED1TRIM_EXP)+factor(BMI)+factor(SMOKING)+factor(FOLIC_ACID),family=binomial(),data=G1)
However, two problems emerged. Firstly, the odds ratio(95% confidence interval obtained from the logistf command (OR = 0.9) differed widely from the clogit and glm command (OR = 3.0). Secondly, I tried but failed to include the strata() function (like what I did in the clogit command to account for matching pairs in my cond. logistic regression) to the logistf command so I suspect that the OR = 0.9 obtained is incorrect because it did not compare my cases to its respective 5/6 controls.
Is there an explanation for the differences in OR obtained and is there a way for me to include the strata() function to my logistf command? Thank you very much in advance!
PS: I'm new to R so please let me know if the problem wasn't clearly expressed or if you need any further clarification.