In estimating the model, use Hausman test to verify if there is a fixed or random effect in your model. If you get a random effect, then estimate using random effect regression.

Furthermore, on a random effect estimation, we live with unobserved heterogeneity by applying generalized least square and in some cases maximum likelihood techniques.

Change the default value of "random.method" argument to other options such as "amemiya" or "nerlove". That should help.