Using Stata 13.1 in Windows 7 environment. I'm using a user written module (apc_ie) that conducts age-period-cohort analysis using glm. My model is a logitic regression that I wish use to predict probabilities for age, period and cohort, and graph. Looking at other papers in this field (e.g., http://digitalcommons.unl.edu/cgi/vi...ociologyfacpub) it appears that he first graph in Figure 4 is generated by:
And so on.
It would seem more appropriate to pursue a mariginal approach by computing the probabilities for each individual
where the values for each observation are substituted for A, B
and C, then averages computed each age, period and cohort group. The major flaw in this is that the first and last cohort are only associated with one age and period, so there's no variance. Unfortunately, the margins command does not work after this module, so I have to code the calculations. I'm just not sure what is the correct way of doing it, my way or what's done in the literature. I'd be interested in hearing opinions.
And so on.
It would seem more appropriate to pursue a mariginal approach by computing the probabilities for each individual
where the values for each observation are substituted for A, B
and C, then averages computed each age, period and cohort group. The major flaw in this is that the first and last cohort are only associated with one age and period, so there's no variance. Unfortunately, the margins command does not work after this module, so I have to code the calculations. I'm just not sure what is the correct way of doing it, my way or what's done in the literature. I'd be interested in hearing opinions.
