fixed/random effects

[Pitchforkp]

Hi,

I was wondering if someone can enlighten me concerning the use of fixed effects/random effects models

I am researching the effect of 4 indep. variables on a depend. variable, data is unbalanced panel data over 10 years using 150 cross-sections. I've run some OLS regressions and i have some remarks/questions.
-I've used fixed effects since hausman p-value was very low (.00)
-using both fixed cros-sections and period effects greatly increases R2, but this is logically i assume. i was wondering how i know whether to use both fixed cross sections and fixed periods, i mean i suspect that there are significant cross-section and temporal effects, but how do i know that this is indeed the case?
-durbin watson value increased from .81 to 1.56 using fixed effects. so less autocorrelation in the residuals, which is good right?
-the coefficients carry the expected signs and significance whether or not i use fixed effects or not, although the values do differ, as i see it using both fixed cross-sections and periods is just like including dummy variables for t-1 cross sections and t-1 years, so not using this is just like having one big pooled ols regression?
-i've used an interaction variable testing the industry effect on one specific independ. variable. when doing this does it make sense to also include cross-section effects or just time-period effects? i've done both and using cross-sections means a lot of the variables lose their significance.
Sorry for the long story, but what it boils down to is if someone could tell me the added value of using these fixed effects and when not to use them

Regards,