A new method of testing for a unit root in the INAR(1) model based on variances

We present a new method of testing for unit roots in the INAR(1) model based on estimated variances. We present detailed simulation evidence regarding the performance of the new test statistics that show that our method is more powerful than the Dickey-Fuller tests especially in nearly unit root circumstances. We evaluate the presence of a unit root in two empirical time series, namely, the number of schools for the blind, deaf, and the developmentally disabled people as well as the number of teachers in such schools. We find evidence of a unit root in either series.