 |
 |
Dear AllStat I need your suggestions/ comments please I used GLIM (family-binomial, link-logit, probit) to see the exposure response relationship for the workers exposed to an chemical substance. Data is from a cohort study and had a very good fellow-up case record. As a first model I used GLIM with binomial family with logit link. The exposure- response relationship was found to be positive. Later I repeated the analysis with probit link. Here also I found the relationship was positive and even showing better fit than logit link (I mean p<0.05, <0.001 values of estimated betas (odds ratios) in probit were more statistically significant than in the logit link model). The next observation is that the 95% confidence intervals of the odds ratios were found narrow in probit than in logit models. I know from statistical point of view (both uses different tolerance distribution assumption)after all its a two different transformations . But I need somebody help, here how I can tell to non-statisticians that probit has additional features (if any?) than logit. Since many applications logit is commonly used than probit. Even when I presented these results, some researches are not aware about the probit transformation. Can any one suggest some references who had similar experience either in exposure response fitting or dose response fitting. Thank You in advance Dr. Venkata Putcha email: [log in to unmask] %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%