Draw a stratified random sample

[Tessa May]

Hi,

I am working with a survey where I have information at the individual and household levels. I see in the data each household for 5 periods. After the 5th period, the household changes but it's possible to keep having the same household and individual identifiers in the dataset because the survey recycles them. I want to create a variable that differentiates the households that are different but share the same identifier. To write that code I need to draw a stratified random sample because the dataset I currently have is too heavy for my computer to work with it. For that random sample, I need to keep all the observations that share the same identifier of household and person for all the periods in which it appears. Which command do you recommend?

Here is an example of my data:

HH	IN	ENT	YEAR	QR
1	1	1	2001	1
1	1	2	2001	2
1	1	3	2001	3
1	1	4	2001	4
1	1	5	2002	1
1	2	1	2001	1
1	2	2	2001	2
1	2	3	2001	3
1	2	4	2001	4
1	2	5	2002	1
2	1	1	2001	2
2	1	2	2001	3
2	1	3	2001	4
2	1	4	2002	1
2	1	5	2001	2

Thanks,

Tessa

[Erik Ruzek]

I personally have used Ben Jann's excellent gsample (findit gsample) and it can do what you are asking. You just need to specify the cluster variables in the options and decide whether you want to sample with replacement (the default) or without replacement.

[Tessa May]

Thank you Erik! I am a little confused with the Strata and Cluster options. If I want to draw all the observations with the same household ID, should I use strata(HH) or cluster(HH)?

[Mike Lacy]

My impression is that an observation in your data set is an individual-period, with collections (i.e., clusters) of individual-periods belonging to a household. You'd be taking a cluster sample, not a stratified sample, with households being the clusters.