Clustering by two variables if one is missing

[em lowthian]

Hi all,

I've had a look at the manual and I can't seem to find what I need...

I'm looking to cluster by mpid (mothers id) as my population of interest is children, and that's the closest thing I have to household. However, I do not want to exclude households who only live with their father - as mpid would be missing. Ideally I need a model that will cluster by mpid and if not then fpid (fathers id).

Something like:

xtreg selfesteem ypsex dvage divorcemo emotabusemo physpunishtmo mwbmo, re, vce(cluster mpid [missing] fpid)

[Clyde Schechter]

How about this:

gen cluster_variable = mpid
replace cluster_variable = fpid if missing(cluster_variable)
xtreg ...., vce(cluster cluster_variable)

[em lowthian]

Genius, I'll try it out now.

Many thanks

Em

[em lowthian]

Hi Clyde,

It has somewhat worked - but because mpid is a unique ID (has duplications as it is the mothers id, so multiple adolescents can have the same mpid) it seems to have added on a number to the end see code.

This is just for:

g clusterhh = mpid

 
mpid
clusterhh

68014287
68014288

68028567
68028568

68037407
68037408

68041487
68041488

68041487
68041488

68046927
68046928

68046927
68046928

68056451
68056448

Do you know how I can stop it doing this?

[Clyde Schechter]

The most likely problem I can see here is a precision issue. When you -g clusterhh = mpid-, by default clusterhh is generated as a float. It looks to me as if the ids have too many digits to be stored exactly in a float, so "rounding" error is occurring. (Float can only hold up to 7 digits without loss of precision.) If I'm right, you can fix this by changing it to -gen long clusterhh = mpid-. (The -replace- statement will not require any modification.)

[em lowthian]

Thanks Clyde I'll give that a go -

[em lowthian]

Hi Clyde

Sorry for the late response, I got caught up in some xtoverid stuff! Yes that worked perfectly thanks and noted.

Emily