The DHS Program User Forum: Weighting data » Combining sampling strata categories

Home » Data » Weighting data » Combining sampling strata categories

Show: Today's Messages :: Show Polls :: Message Navigator

Combining sampling strata categories [message #7081]

Thu, 20 August 2015 13:01

Tracy
Messages: 1
Registered: August 2015

Member

Hello,

I am having difficulty with running a linear regression once I have completed the svyset command. All the F and P values are marked as missing in the regression output in STATA. In my svyset command I have b_strata as my strata variable and I found that this gives me 1037 strata, a large number of which have only 1 primary sampling unit. This appears to be the problem as when I remove the strata element of my svy weighting command, then the regression output includes F and P values, the only issue is it is not representative of the general population. Is it acceptable to combine sampling categories, or will this lead to the estimates I generate still being non-representative of the general population? Or should I use an extra command during svyset such as singleunit(certainty)?

This is the error message I get:

Note: 2 strata omitted because they contain no population members.
Note: missing standard errors because of stratum with single sampling unit.
Thanks you for your help,

Tracy

Report message to a moderator

Re: Combining sampling strata categories [message #7087 is a reply to message #7081]

Fri, 21 August 2015 09:10

Bridgette-DHS
Messages: 3230
Registered: February 2013

Senior Member

Following is a response from Senior DHS Stata Specialist, Tom Pullum:

Hi Tracy-- You are right that there could not possibly be 1037 strata. If this number is greater than 60 or so then you should be suspicious. This kind of error will indeed cause a regression to fail.

What survey and file are you using? In some less recent surveys the stratum variable is misidentified.

Report message to a moderator

Previous Topic:	How to use All Women Factor in SPSS
Next Topic:	Weighting for subgroups vs. entire population

Goto Forum:

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Thu Oct 23 07:10:21 Coordinated Universal Time 2025