Reader Comments

Post a new comment on this article


Posted by TimHolland on 21 Feb 2008 at 20:31 GMT

A very interesting article, and certainly an important issue. I was surprised, however, by how ahistorical the analysis was for an issue such as this.

Taking a historical perspective, it is probably reasonable to assume that (1) areas having strong states in the past likely saw a reduction in linguistic / ethnic diversity (many languages being united under one leader and one language becoming dominant). The big hot spots of linguistic diversity (India, sub-Saharan Africa, etc) tend to have not had strong states historically (at least not along the nation-state boundaries as they are defined at present, and as they are defined in this analysis). You could also make the argument (pretty reasonably) that (2) a strong state in the present reduces instability. Take (1) and (2) together and you can potentially rephrase "present diversity is related to present instability" to "past state strength is related to present state strength" - a conclusion which isn't very remarkable.

Comments? Is there a way to determine if diversity and stability are causally related, or if instead they are simply both correlated to a third variable (state strength) rooted in the past? Is there a way of adding historical patterns to this analysis?

Also, I'm curious about the sample size. I can see that there was some data available for each of 212 countries, but how many countries had all data available?


RE: History?

DanielNettle replied to TimHolland on 25 Feb 2008 at 12:22 GMT

These are useful comments, Tim.

On your first point, the problem is that there isn't longitudinal data on linguistic diversity. If there were, we could test for the effect of strong states in the past in producing the contemporary pattern of diversity. Analysis elsewhere (e.g. in my book Linguistic Diversity, Oxford University Press 1999) suggest that the main determinants of linguistic diversity are physical geography and ecology rather than political factors, but it is a good point that requires further investigation.

On the sample size point, different variables have different numbers of values present, usually less than the 212 maximal set, but the method used of handling missing data does not delete countries for which some data are missing. There is not therefore a single sample size for each model.

Hope this is useful
Daniel Nettle