The DHS Program User Forum: Weighting data » Multilevel GLM poisson analysis

Home » Data » Weighting data » Multilevel GLM poisson analysis (Declaring the DHS data for this analysis)

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

Multilevel GLM poisson analysis [message #30295]

Wed, 30 October 2024 06:45

David
Messages: 6
Registered: October 2024

Member

Hi,
I have followed the DHS guidelines to set the svyset via https://www.dhsprogram.com/pubs/pdf/MR27/MR27.pdf.
Yet, I still have challenges as i cannot excute the model.
Regards!

Report message to a moderator

Re: Multilevel GLM poisson analysis [message #30305 is a reply to message #30295]

Thu, 31 October 2024 08:12

Bridgette-DHS
Messages: 3230
Registered: February 2013

Senior Member

Following is a response from Senior DHS staff member, Tom Pullum:

We may be able to help if you tell us what survey you are using, what is your svyset command, what is your estimation command, and what you mean by "cannot execute". Are you getting an error statement?

Does your estimation command work if you remove "svy: "? My general strategy to solve a problem like this is to start with something simple, and then add complexity in stages. In that way you can find what part of the svyset command is causing trouble.

Report message to a moderator

Re: Multilevel GLM poisson analysis [message #30308]

Thu, 31 October 2024 16:20

David
Messages: 6
Registered: October 2024

Member

Thank Bridgette,

I am using 2022 GDHS and i have followed https://www.dhsprogram.com/pubs/pdf/MR27/MR27.pdf

However, without "svy," the model "meglm devar i.in_var||v001:, family(poisson) link(log)" works without an error.
Fortunately, adding "svy," I could now run it; the issue now at hand is "convergence not achieved.". So if there is a way to resolve that, I will be grateful.

Report message to a moderator

Re: Multilevel GLM poisson analysis [message #30316 is a reply to message #30308]

Fri, 01 November 2024 11:37

Bridgette-DHS
Messages: 3230
Registered: February 2013

Senior Member

Following is a response from Senior DHS staff member, Tom Pullum:

You can add the following option (after the comma): "iterate(50)".

Usually when there is a convergence problem it is because the estimation procedure is oscillating between two nearby values of the convergence criterion. It would be a good idea to repeat with limits other than 50, such as 51 or 100 to get a better diagnosis of what's wrong. If you don't see any difference between the coefficients for those alternatives, then you should be ok.

Report message to a moderator

Re: Multilevel GLM poisson analysis [message #30317 is a reply to message #30316]

Fri, 01 November 2024 12:24

David
Messages: 6
Registered: October 2024

Member

Thank you Tom Pullum,

Rightly, as you suggested, I have added "inter (200) and even (500)". Yet it still persist.
Then, I again added "difficult" or "technique (nr)" and the situation continue.

Infact gentle following your procedure, I thought I will rejoice but here comes the serious one. Still looking forward for any other alternatives.
Regards!

Report message to a moderator

Re: Multilevel GLM poisson analysis [message #30321 is a reply to message #30317]

Mon, 04 November 2024 11:46

Bridgette-DHS
Messages: 3230
Registered: February 2013

Senior Member

Following is a response from Senior DHS staff member, Tom Pullum:

Is there a typo in your question? I suggested "iterate(50)", not "inter (200)". In my experience this option is like putting on the brakes abruptly. It stops the iterative procedure when it is looping endlessly. Instead of 50 you could try 2.

One strategy would be to repeat the procedure with less complexity. You could reduce the svyset command by dropping the strata or the multilevel parts (the most complex components). You could reduce the number of variables on the right hand side of the model. I suspect that you have many categorical variables on the right hand side and some (many?) combinations of covariates are empty (have no observations). You may need to reduce the number of coefficients in the model.

Report message to a moderator

Re: Multilevel GLM poisson analysis [message #30324 is a reply to message #30321]

Mon, 04 November 2024 12:05

David
Messages: 6
Registered: October 2024

Member

Reaspectfully, Sir, it's a typo error "iter" (iteration)

Report message to a moderator

Previous Topic:	weighting data in regression analysis
Next Topic:	Multilevel weights level 2

Goto Forum:

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Tue Oct 14 02:17:47 Coordinated Universal Time 2025