Bootstraping the CPS from IPUMS using bsample

[Arnau Valladares-Esteban]

Hello,

I'm working with the basic CPS as provided by IPUMS.

I need to bootstrap the data to compute standard errors. As far as I understand, I can do it using bsample. I need to let bsample know that the CPS is a monthly survey. However, it's not clear to me if I need to do that by using the option strata() or cluster(). Any suggestion?

Also, I guess I need to use bsample with the option weights(). Does that mean that I do not need to use weights for my computations with the bootstrapped samples?

Any help is very much appreciated.

Thanks!

[Felix Bittmann]

I do not know the CPS data but you probably do not want to use bsample, unless you program something. I would start from your analysis part, what model do you intend to estimate in the end?

[Arnau Valladares-Esteban]

That's irrelevant. I know I have to bootstrap.

[Felix Bittmann]

In Stata you usually want to bootstrap directly in the analysis command. For example, when your model is OLS you would use:

regress y x1 x2 x3, vce(bootstrap)

[Arnau Valladares-Esteban]

I'm not running a regression after bootstrapping. I do something else that would be too long to explain.

[Weiwen Ng]

Many large public surveys need to be svyset, specifying a probability weight and a stratum variable at minimum. For the US Current Population Survey., it’s a bit more complicated. The suggested command appears to be:

svyset [iw=wtsupp], jkrweight(repwtp1-repwtp160, multiplier(.025)) /// vce(jackknife) mse

I offer this in case it helps someone answer the OP’s question.

[Felix Bittmann]

If the CPS is a panel study you want to sample entire histories of individuals, and here you need the cluster option, like:

bsample, cluster(person_id) idcluster(newid)

The important part is that you use the newly generated variable newid as the id for the following analyses to not count the observations incorrectly. Regarding the weight question, you do not need this if you are working with sampling weights.