The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Data » Dataset use in Stata » Calculating ICC
Calculating ICC [message #22544] Tue, 30 March 2021 09:36 Go to next message
Ghose is currently offline  Ghose
Messages: 11
Registered: May 2018

I am working with Benin IR data to explore the factors associated with educational achievement (v106). My supervisor suggested that I check ICC value first since the data are clustered, and then choose between -logit- and -melogit- depending on whether the ICC is above the threshold of 0.15. Now it seems ICC can be calculated both before:

. loneway v106, v021

Intraclass Asy.
correlation S.E. [95% Conf. Interval]
0.25242 0.01320 0.22655 0.27829


and after melogit:

melogit v106, || v021:

esttat ICC

Level | ICC Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------ ------------------
v021 | .3073749 .0158921 .2771332 .3393674


As you can see, these two methods produce dissimilar results. Despite the necessity of using -melogit- being obvious, I'd like to know the right course of action to follow: before/after/both?

Thank you.

Dr Ghose

[Updated on: Tue, 30 March 2021 09:37]

Report message to a moderator

Re: Calculating ICC [message #22557 is a reply to message #22544] Wed, 31 March 2021 16:33 Go to previous message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 2277
Registered: February 2013
Senior Member

Following is a response from DHS Research & Data Analysis Director, Tom Pullum:

I really don't know why the two procedures give you such different estimates, but since you will be proceeding with melogit, I recommend that you use the ICC estimate coming from melogit rather than from loneway.
Previous Topic: Myers blended index
Next Topic: Nigeria 2008 Strata for Svyset
Goto Forum:

Current Time: Sun May 16 12:04:39 Coordinated Universal Time 2021