VIF cut-off value in Marketing

[spoeko]

G'day all,

the literatur frequentely suggests a Variance Inflation Factor cut-off value of 10. This cut-off value comes from econometrics and doesn't seem appropriate in marketing studies.
Instead a cut-off value of approximately 1,5 might be more suitable for behavioural science (as for exampel suggested within the PLS Workshop in Lüneburg/Germany this year).

Does anyone know literature that deals with this question and gives recommendations?

Thanks very much
sebastian

[jjsailors]

Sebastian,

I haven't thought about this very deeply, but given that VIF is an indicator of data characteristics that cause problems with the errors associated with the estimates, I'm not at all certain why there would be different criteria in one domain versus another?

In other words, if data that show a VIF of 1.5 are not problematic from an estimation standpoint in econometrics, why would they be in marketing? The formulas that lead to the estimates and the associated error terms are the same and do vary across domains; it doesn't seem like we should have different cut-offs.

John

[schroer]

Hi Sebastian, hi John,

you might want to have a look at:

Craney, T. A., & Surles, J. G. (2002). Model-dependent Variance Inflation Factor cutoff values. Quality Engineering, 14, 391-403.

The models, ideas and formula presented there are quite straightforward and would lead me to be cautious about VIFs as low as 1.34 in one measurement model—which made perfect sense in that case theoretically.

I'd be really interested in ideas or other sources about model-dependet VIF cutoff values, because a one-size-fits-all approach does not seem appropriate to me.

Best wishes,

Joachim

[AndrewG]

There is also a conference paper by Andreev et al (2009) which refers to an earlier article by Diamantopolous and Siguaw (2006) and recommends a VIF tolerance level of <3.0. Note: there is a typographical error actually showing it as <3.3.

Andreev, Pavel; Heart, Tsipi; Maoz, Hanan; and Pliskin, Nava, "Validating Formative Partial Least Squares (PLS) Models:
Methodological Review and Empirical Illustration" (2009). ICIS 2009 Proceedings. Paper 193.
http://aisel.aisnet.org/icis2009/193