I am interested in examining inventor features on their inventive performance using patent data. I have an unbalanced panel data of 7000 inventor-year observations on 3000 inventors over 15 years within one firm. The results of the inventor- and year-fixed effect model are very different from these of the random effect with the year-fixed effect only. The result of Hauseman test is significant, suggesting that I should use the fixed effect model rather than random effect to control inventor unobservable heterogeneity.
I guess the difference comes from the huge number of control variables – more than 3000 inventor dummies. As I have 3000 inventors and by average each inventor has only 3.5 observations, I am wondering which method I should choose and what I should do.