Lags and diffs in panel model

Maybe someone could explain what is the reason of using lags or diffs in panel model?

I also have some difficulties while interpretating the results of panel models which were got using different methods.

I know that one can decide which panel modeling method (within or random) is better using Haussman test, but how should i know if simple linear model is not the best if pooling model gives effiecient coefficients (I made this assumption while looking at starred coefficients and R squred in model)?

Thank you in advance!