by Gösta Hägglund
Research Report 1994:6
Department of Statistics, Stockholm University, S-106 91 Stockholm, Sweden
The specification of the reference variables in the FABIN - methods is studied. First some theoretical aspects are considered. These are based on the presentation in Hägglund (1993). Here upper bounds for the number of possible parameter (or estimation) sets under different assumptions is discussed. This is clearly associated to the reference variables. Moreover, some extreme cases are described. Then, some real, performed factor analyses are viewed. There, the possibilities of choosing appropriate reference variables on substantive grounds is discussed. These analyses together with some previously performed analyses serve as the basis for a numerical study. There it is concluded that the FABIN - methods seem to have a certain degree of robustness against less appropriate specifications of the reference variables. However, using one or more inappropriate reference variables make the pattern difficult to interpret. Connected to this is factor correlations becoming fairly large.
Key words: factor analysis, instrumental variables, reference variables.
Close this Window
Last update: 990916/CE