Thursday, August 25, 2011

A means of getting a better estimate of true poverty rankings

I have just read about a study that used four methods to rank order village families according to their poverty. The purpose of the study was to determine which of the three other methods best approximated the most expensive and "best" method.  (I assume that something like the Spearman Rank Correlation Coefficient was used to judge the approximation.)

 It occurs to me that none of the four methods can be taken to provide "the true" poverty ranking of villagers. Yet the researchers accepted the most expensive approach as defining the "truth" and judged the other methods according to their differences from its results.

One approach to improving on the "truthiness" of the ranking would be to add the rank orders from each indicator for each family, and reorder the families according to the sum of their previous rankings.

Perhaps a better way would be to weight the rank in each indicator by a weight proportional to the confidence one has in that indicator.

Indeed, one might iterate the procedure, weighing the ranking of each indicator according the Spearman's rho of that that ranking with respect to the weighted average ranking of the previous iteration.

It is hard to get at the truth!

No comments: