I found this very interesting paper which I think provides some good guidance on testing the reliability of my formative measurement model - see earlier post.
The relevant information is on page 6 of the paper and I am quoting it below (because I know external links are unstable and should not be used for ongoing reference):
Extract from:
Phatcharee Toghaw Thongrattana, "Assessing reliability and validity of a measurement instrument for studying uncertain factors in Thai rice supply chain" (October 1, 2010). SBS HDR Student Conference. Paper 4. http://ro.uow.edu.au/sbshdr/2010/papers/4
*****Reliability test for Formative constructs***
As formative constructs composited of different aspects of a construct that their indicators are not necessary to correlate with each other (Diamantopoulos and Winklhofer, 2001). Straub, Boudreau et al. (2004, p.400) state that “it is not clear that reliability is a concept that applies well to formative constructs”. This statement is also supported by Diamantopoulos and Siguaw (2006, p.270) and Rossiter (2002, p.315) that no dimensionality and reliability test are performed on formative indicators because factorial unity in factor analysis and internal consistency are not relevant. Although, low item-to-total correlation should be dropped from measurement scales to increase internal consistency reliability for reflective measurement model because the scales are from the same content construct, the removal of measurement scales in formative measurement model can lead to change the empirical and conceptual meaning (MacKenzie et al., 2005). Andreev, Heart et al. (2009) conclude that construct reliability of formative should be performed by multicollinearity, test of indicator validity (path coefficients significance), and optionally, if appropriate, testretest (Petter et al., 2007).
*On the other hand, reflective constructs that multicollinearity among items in the same construct is desirable such as high Cronbach’s alpha, but reliability of formative construct in term of multicollinearity is not present because if multicollinearity is present, it means that indicators are tapping into the same aspect of the construct (Petter et al., 2007, p.641). Likewise, formative measurement model is based on a multi-regression that multicollinearity should not exist (Diamantopoulos and Winklhofer, 2001). Thus, reliability evaluation for formative constructs is to assess the assumption of no multicollinearity (Diamantopoulos and Siguaw, 2006). Variance Inflation Factor (VIF) is evaluated. There are some guidelines that can be applied:
• VIF is less than 3.3 that shows a excellent value (Diamantopoulos and Siguaw, 2006). • VIF is less than 10 that no collinearity is commonly accepted (Hair et al., 1995).
As collinearity also can be harmful effects to formative constructs, condition index is the standard diagnostics that measure the relative amount of variance associated with an eigenvalue. Its threshold value should be less than 30 to find no support for the existence of collinearity (Hair et al., 1995). If multicollinearity exists, Petter, Straub et al. (2007, p.642) recommended that at first, the model construct may have both formative and reflective measures. Secondly, the correlated measurement items can be removed, if content validity is not affected. Thirdly, the correlated measurement items can be collapsed into a composite index. Lastly, it can be converted into multidimensional construct.
My questions
- How can I perform VIF and Condition Index using SPSS?
- Is VIF and Condition Index suitable for assessing reliability of formative constructs?
