hello everybody,
i ran a glm with poisson response and log link and i am trying to interpret the type I and III statistics... at the moment i was interested in disclosing some measure of overlapping between the predictors in explaining my response. Then i thought to work simultaneously at type I and type III chi square statistics and changing iteratively the order of predictors in the model. In this manner i wanted to measure the ratio of explained deviance of one predictor in the nested model containing only that predictor (chi square statistic in the type I for the model with predictor of interest as first predictor) and the explained deviance of the same predictor in the full model (chi square type III of predictor of interest in the full model). As i expected i found that the type III chi square is always less than the type I (ordered) one, meaning that when considering multiple predictors the marginal information of single ones somehow deteriorates due to interrelationships. Nevertheless i found that for ne predictor type III is greater than type I, how can interpret this ? Thanks for any help.
jacopo
ps. I understand i ve been quite qualitative in my description of the prblem thats why i dont expect u to give me quantitative and rigorous answers..any insights about my interpretation (probably wrong) of those chi square statistics will be of great help!