Hi Blaze:
Modeling a general method factor--although a very common
procedure--would not be at all a good option. It is simply a myth that
an unmeasured factor can "remove" unwanted variance. See:
Richardson, H. A., Simmering, M. J., & Sturman, M. C. (2009). A Tale of
Three Perspectives: Examining Post Hoc Statistical Techniques for
Detection and Correction of Common Method Variance. Organizational
Research Methods, 12(4), 762-800.
Abstract: Many researchers who use same-source data face concerns about
common method variance (CMV). Although post hoc statistical detection
and correction techniques for CMV have been proposed, there is a lack of
empirical evidence regarding their efficacy. Because of disagreement
among scholars regarding the likelihood and nature of CMV in self-report
data, the current study evaluates three post hoc strategies and the
strategy of doing nothing within three sets of assumptions about CMV:
that CMV does not exist, that CMV exists and has equal effects across
constructs, and that CMV exists and has unequal effects across
constructs. The implications of using each strategy within each of the
three assumptions are examined empirically using 691,200 simulated data
sets varying factors such as the amount of true variance and the amount
and nature of CMV modeled. Based on analyses of these data, potential
benefits and likely risks of using the different techniques are detailed.
One thing you could do is to model the fixed effects of the individuals,
which will remove any unobserved variance due to the individual
correlating with y and the x's.
HTH,
John Antonakis
________________________________________
Prof. John Antonakis, Associate Dean
Faculty of Business and Economics
Department of Organizational Behavior
University of Lausanne
Internef #618
CH-1015 Lausanne-Dorigny
Switzerland
Tel ++41 (0)21 692-3438
Fax ++41 (0)21 692-3305
Home page:
http://www.hec.unil.ch/people/jantonakis
________________________________________
On 30.08.2010 19:18, Blaze Aylmer wrote:
> Dear Members,
>
> I would like to some views on the above topic.
>
> I expect to collect data from 110 individuals over three measurement
> occasions and would like to test for common method variance. The intended
> approach is post hoc, in that a general latent method factor will be used
> to examine loadings from a construct A with 6 items.
>
> I would like to ask the following question.
>
> Should I permit all the factor loadings to load on the general method
> factor across all three time periods or should there be a method factor for
> each time period? If anyone knows of a paper which demonstrates this or an
> alternative could the post the reference please?
>
> Thank you,
>
> Blaze Aylmer