My colleagues and I are trying to replicate the CWI methodology described in this publication: http://www.dhsprogram.com/pubs/pdf/MR9/MR9.pdf and we were wondering if someone could explain rational behind using a logistical regression to determine asset cut-points as described on page 9. This method works great when asset ownership is relatively low, but seems to produces unrealistic cut-points in surveys where the vast majority of the population owns the asset (e.g., if 98% of the sample owns a TV, the wealth associated with a theoretical sample where only 50% own a TV results in a cut-point below the mean wealth for the whole sample). Could one determine asset cut-points by taking the mean wealth of respondents that own the asset, similar to the methodology used to calculate the cut-points at different levels of unmet basic needs?

Thank you for your time,

-Alex Miller]]>

Quote:

]]>

"The values of the unmet basic needs is not the mean but the point on the wealth index where it changes in level. While the mean wealth index value of each of the other assets could be used, we use the median value which is more representative in a sample since it is less affected by outliers. If the distribution is approximately normal, there is little difference between the mean and the median. In a country where almost all have TVs, the mean value of the index for those with TVs will be still be very low. Using the mean would have a different value for the CWI but I suspect not very different. To do so, would mean that the baseline values (Vietnam, 2000) would also need to be redone."

Regards,

Shea