The two-parameter model must be identified by both a location and a scale restriction. The former can be the same restriction as above, that is, μ = 0. The latter can be the additional restriction that the variance of the latent distribution is equal to one, that is, the model is identified by assuming a standard normal distribution, N(0,1), for the latent ability parameters θj. Alternatively one fixes one of the discrimination parameters to unity. Generally, however, this identification solution is not advisable, because the standard errors of the parameters blow up if the discrimination parameter chosen for the identification is poorly identified.
