On the identification of a single-factor model with correlated residuals

SUMMARY A necessary and sufficient condition for the identification of a single-factor model with correlated residuals is derived by studying the zero elements of their concentration matrix. In particular the new condition is expressed in terms of the complementary graph of the residuals. This graphical condition can be simply checked by means of an efficient algorithm. An example is also given showing that sometimes the single-factor model with correlated residuals can be used to overcome the problem of non-identification of a larger factor model.