Hi,

I am new to Stata and this forum. I am trying to find the effect of changes in fuel price on road traffic crashes in a panel of 28 EU countries over 14 years of monthly data (168 months for each country). My dependent variable is a monthly road traffic accident and my independent variables are GDP per capita, unemployment, fuel prices (petrol prices and diesel prices, used in two different models due to multicollinearity). My data has large T=168 and small N=28 (countries), also some non-stationary series. So far most of the literature in accident research is pointing towards the negative binomial model and after running the fixed and random negative binomial models, according to Hausman test, FE negative binomial model is a better fit for my data. My data is an unbalanced panel.

However, test for autocorrelation, xtserial rejects the null and there is serial correlation in the series. Unit root test xtunitroot using {xtunitroot ips varname, lags(1)}, however, providing independent variable with I(0) and some independent series I(1).

I am confused about whether to consider dynamic panel models for large T like ARDL PMG,MG, DFE and xtdcce2. Or I should I go ahead with negative binomial models?
Your help will be much appreciated. Thanks