ARIMA modeling

[spiatkow]

Hello,

I used fixed effect, pooled time-series modeling to examine the impact of law dummy on crash fatalities rates across 50 states (1985-2016). I was suggested to use ARIMA modeling instead. Would ARIMA had to be estimated for each state separately (that is, one separate model for each state)? I have not seen ARIMA estimated across different units..
Can ARIMA model include control variables? Is it always a univariate model? What are pros and cons of ARIMA vs. fixed effect panel model?

Would yearly data be OK for such ARIMA model? Note that I have data from 1985 to 2016, which gives 32 cases only! Is 32 cases enough?
or is it mandatory to have monthly data for ARIMA modeling ?

Finally, how to detect seasonality - in both yearly and monthly datasets?

Thank you so much.
Sylwia

[startz]

Why did someone think that ARIMA modelling would be a good idea?

[spiatkow]

Good question. ARIMA is typically used in the lit. to examine the effects of law on crash fatalities... for individual states.
I would appreciate any help