Jackknife xtqreg

[David Barker]

Using the Grunfeld data, this works:

bootstrap, reps(50) cluster(company) idcluster(comp): xtqreg invest mvalue kstock i.comp#c.time, quantile(.25) i(company)

But if I use jackknife instead of bootstrap it does not work. What needs to be different for jackknife? I took out reps and changed i(company) to the jackknife equivalent.

[David Barker]

In other words, this works:

bootstrap,cluster(company) idcluster(comp): xtqreg invest mvalue i.comp#c.year, q(.25) i(company)

but this does not work:

jackknife,cluster(company) idcluster(comp): xtqreg invest mvalue i.comp#c.year, q(.25) i(company)

Why doesn't the jackknife version work? The error is:

insufficient observations to compute jackknife standard errors

[Joao Santos Silva]

Dear David Baker

Typing

jackknife, noisily cluster(company) idcluster(comp): xtqreg invest mvalue i.comp#c.year, q(.25) i(company)

will give you a clue.

Best wishes,

Joao

[David Barker]

Thanks for the suggestion. I did that, and the error message for each sample is:

collinearity in replicate sample is not the same as the full sample, posting
missing values

The odd thing is that it works just fine for bootstrap instead of jackknife.

[Joao Santos Silva]

Dear David Barker,

Your model contains a trend for each firm and when you do the jackknife each firm is dropped in turn and therefore it is not possible to estimate all the parameters estimated with the full sample. With the bootstrap, the number of firms in the sample is always the same, although not always the same firms are in the sample, and therefore all parameters can be estimated. Anyway, I wonder why you were trying to use the jackknife in this context.

Best wishes,

Joao

[David Barker]

That makes sense. I was thinking that jackknife might automatically drop the appropriate all-zero variable for each firm that is dropped.

I am trying to find clusters that influence a particular regression coefficient. Jackknife drops one cluster at a time and reestimates the regression coefficients, so I was hoping to check all of those to see if the coefficient was dramatically different when some cluster is dropped.

[David Barker]

Is it possible to use jackknife for this or do I need to write a program to drop each cluster, run xtqreg and save the results each time?

[Joao Santos Silva]

You have to write a loop to do that.

[David Barker]

Got it, thank you!