Applied data analysis/Consultancy

Well every now and then, you want to do something useful. As a member of the department of data-analysis in a growing faculty with lots of active researchers in various domains, it is inevitable that you get involved in the work of others.

Consultancy

People ask me (and my colleauges) `statistical questions' all the time. In case you are wondering, these are the type of questions that seem to come up most often:

• My reviewer wants me to report `effect sizes'. How do I proceed?
  
• My reviewer wants me to say something about `power'. How do I proceed?
  
• My reviewer wants me to do analysis X. How do I proceed?
  

It is no accident that every one of them is in the form `My reviewer wants me to do this or that'. Indeed, I found it an interesting observation that `statistical problems' are mostly brought up by a reviewer during the reviewing process of a paper, not by the researcher him/herself...

Applying `standard' methods to somebody else's data

Sometimes (and less frequent nowadays) a researcher can convince me to do the data-analytical part of his/her research myself. Many (but not all) of the papers in the `Applied papers' section in my Publication list fall into this category. The `standard' methods I have been applying most are:

• confirmatory factor analysis, testing for factor validity
• confirmatory factor analysis, testing for measurement invariance
• categorical data-analysis (especially logistic regression)

Applying `non-standard' methods to somebody else's data

One of the more interesting aspects of being involved in different types of research lines, is that you encounter some challenging problems where `standard' data-analytical methods are not appropriate.

This is a typical example:

• Verhofstadt, L.L., Buysse, A., Rosseel, Y. & Peene, O.J. (2006). Confirming the three-factor structure of the quality of relationships inventory within couples. Psychological Assessment, 18, 15-21.
  

Here, the problem was to establish measurement invariance across males and females for a questionnaire called the QRI (quality of relationships inventory). However, the males and females belonged to the same couple, and hence, the groups were not independent. This is an example of dyadic data, and traditionally, some type of multilevel modelling is used in this situation. In this paper, we opted to use a resampling strategy combined with traditional CFA-based measurement invariance tests.