A common practice in International Business (IB) research is to collapse multiple dimensions of distance (or any other higher order construct) into a single formative index.
- The term ‘formative‘ implies that the various dimensions are added together in some manner in order to create this single index.
- This is in contrast to ‘reflective‘ indices, which is what we used to create our measures such as linguistic distance (Lang Dist) and religious distance (Relig Dist). With reflective indices, one is attempting to identify a single underlying factor common to all the indicators (i.e. you are counting on the indicators being highly correlated to each other).
Within the IB literature, the Kogut and Singh (JIBS, 1988) National Cultural Distance Index is the most common example of a formative index. However, as Berry et al. (JIBS, 2010) correctly point out, most techniques for combining items into a formative index – such as euclidean distance – assume that the underlying dimensions are orthogonal. That is not the case with our distance dimensions (the Pearson Correlations range from 0.011 to 0.415).
As a result, in order to create a single formative index of psychic distance stimuli from our five dimensions, we take a two stage approach.
- First, we use the Mahalanobis Distance approach to combine our five distance dimensions – Ind Dev Dist, Edu Dist, Dem Dist, Lang Dist and Relig Dist – into a single index. This approach controls for correlations amongst the underlying dimensions.
- Second, the resulting factor is then re-scaled such that the value of zero (0) represents ‘no difference’, and the value of ten (10) represents the highest possible score (i.e. that all five underlying dimensions have a score of ten)
The attached spreadsheet at the bottom of the page supplies this formative index – Psy Dist Mahal for 22,350 country pairs and across 3 distinct time periods (1995, 2005 and 2015).
A WORD OF CAUTION
I must admit that I include a formative index in this webpage with some trepidation. While such an index is simple and easy to use, it it also not without some limitations. In particular, using such an index implicitly implies that all the underlying dimensions are relevant and contributing to any observed relationships. That may not necessarily be true, and this indeed is one of the many criticisms of the Kogut & Singh index. In numerous instances, it has been demonstrated that only one or two of the Hofstede dimensions underlying the Kogut & Singh index are actually driving the observed relationship. Now, if one is using such an index purely as a control variable, this is not a problem. However, if one is using such an index as one of the main independent variables in a study, we strongly suggest that, as a robustness check, you test each of the five dimensions separately.