An Empirical Study of the Impact of the Choice of Persistence Models in Value Added Modeling upon Teacher Effect Estimates