Somers' D for survey data/bootstrapping

[Ryan Kuo]

I'm trying to test for the significance of differences in an ordinal measure across two groups in enterprise survey data. For reference, I have the following variables:

• invest_plan: 0 - Withdraw investment, 1 - Reduce Investment, 2 - Maintain investment, 3 - Expand investment
• sector_num:0 - Manufacturing, 1 - Services

My data consist of 20 strata (Countrysector, e.g. 11 to denote Manufacturing in Nigeria), and I have a svyweight (survey weight) variable that accounts for 2 things: 1) differences in the likelihood of being sampled within each strata (driven by some complications in our sampling frame) and 2) desire to give each strata equal weight (somewhat arbitrary feature of the types of calculations we want to make).

I'm looking to use Somers' D because I'd like to test for differences (accounting for ordinality) in invest_plan between manufacturing and services firms, but I'm not too familiar with how to accomplish this via bootstrapping.

As I understand it, I'll need to create my own replicate weights (which have not been provided). Can somebody provide guidance on how to correctly implement this in Stata?

Alternatively, am I better off just doing an ordered logit like the following?

svyset [weight=svyweight], strata(Countrysector)
svy: ologit invest_plan sector_num

[Ryan Kuo]

Related question: I've been reading up a bit more on ordered logit (proportional odds) regressions being generalizations of Wilcoxon-Mann-Whitney. Therefore, the univariate ordered logit specification in my first post should give me the same p-values as if I applied Wilcoxon-Mann-Whitney but accounted for survey design, right?

I was originally trying to just do the Wilcoxon-Mann-Whitney test but had switched to Somers' D because Stata does not allow us to implement Wilcoxon-Mann-Whitney with svy.