Hello everyone, I am a PhD student from strategy area. I encountered a problem in model choice.
My dependent variable is "status". It is Bonacich’s (1987) power centrality, which accounts for the status of the affiliations in addition to connections. Y is in [0,1], at each endpoint, there would be a number of Zero and one. When Y equals to 0, that means the focal firm doesn't have affiliation within the time window. In fact, the reason why the Y is in [0,1] is the Bonacich’s (1987) power centrality is normalized. for example, if the original value is a, it would be normalized using the formula: (a- min)/(max-min). The max and min means the max and min in this year group.

My data structure is the panel data. in this case, if I intend to consider within difference and hence control the firm-fixed effect, could I use xtgee y x i.year, fam(bin) link(logit) corr(uns) vce(robust)? I am not sure whether fractional logit would be more suitable than xttobit. Anyone could give me some ideas about this.

P.s, I did run xtgee and the error hint is "convergence not achieved". Moreover, in year 2013, if the Y won't equal to 0. it would be omitted.(I am not sure whether the omission is caused by the endpoint.)

Thanks very much!

Best,
Christine