Misspecification and estimation effects in the Lagrange multiplier tests for heteroskedasticity