Unit root test of logged variable in panel data

[Mia Vorster]

Hi everyone,

I have panel data on electric vehicles in the UK. I have logged the variable, and because there are several local authorities which had no electric vehicles at that point in time, Stata has generated missing values. So, when I try to test for a unit root Levin-Lin-Chu test it tells me I do not have strongly balanced data - (my data is strongly balanced).
How should I work around this?

Thanks!!

[Nick Cox]

I think there is an underlying question here which is what you intend to do later, as such a test is not in my experience ever the end goal of a project. The fact is that zero values in the data will frustrate many analyses based on logarithms. There are exceptions such as Poisson regression being feasible with zeros present as the underlying model for the mean function y = exp(Xb) can be estimated without transforming the outcome.