Is it appropriate to use an OLS regression on a panel data if it yields statistically better results than a fixed effects model? OLS results provide statistically significant variables and it is an overall better fit model than fixed effects in which most independent variables lose significance?