BookmarkSubscribe
Choose Language Hide Translation Bar
dale_lehman
Community Trekker

Need help with Bootstrap

I am doing a proportional hazards model and trying to do a bootstrap for the risk ratios.  I can't provide my data, but since I keep getting the same error in both my data and a simple JMP example for the sample data library, I am attaching that example here.  I run the proportional hazards model, focusing on the risk ratios for two groups (which I set up at random) - a test group and a control group.  I click on the risk ratio estimates and attempt a bootstrap analysis - but I keep getting the error that it failed to converge.  I understand generally what the bootstrap sample is doing and that nonconvergence is a possibility, but since I get this error every time - and with any data I try it with - I suspect I am doing something wrong.  Please let me know if you can see a reason why my bootstrap keeps failing.  Thank you.

0 Kudos
1 ACCEPTED SOLUTION

Accepted Solutions

Re: Need help with Bootstrap

Hello, I'm not sure why it's failing to converge. Have you tried using the Fractional Weights option when you do the Bootstrap? That can help in situations where there aren't a lot of observations in particular cells. In this case, that would be cells in the Censor * group 2-by-2 contingency table.

Michael
Michael Crotty
Sr Statistical Writer
JMP Development
5 REPLIES 5
dale_lehman
Community Trekker

Re: Need help with Bootstrap

SORRY!  The file I attached was faulty - I created a random grouping (control/test group) that happened to contain no censored data so the hazard model was nonsense.  Attached is a revised one that works.  I still get the error that it could not converge, so I still have my original question about what I am doing wrong.  But the attached file at least has sensible data.

0 Kudos

Re: Need help with Bootstrap

Hello, I'm not sure why it's failing to converge. Have you tried using the Fractional Weights option when you do the Bootstrap? That can help in situations where there aren't a lot of observations in particular cells. In this case, that would be cells in the Censor * group 2-by-2 contingency table.

Michael
Michael Crotty
Sr Statistical Writer
JMP Development
Highlighted
dale_lehman
Community Trekker

Re: Need help with Bootstrap

The fractional weights works, at least on the sample data set I provided.  I don't completely understand the reason - the data I am really using is a much larger dataset so your explanation about sparse cells seems unlikely to me (though it applies to the small sample data set I provided).  I'll try it on my other data now, but for the time being your solution works!

0 Kudos
dale_lehman
Community Trekker

Re: Need help with Bootstrap

Fractional weights works very well for me - I had the same problem. When the event is rare enough, the bootstrap often does not converge.  The Fractional weights use all of the observations, so in my experience, it always converges.  It is a terrific feature!

Re: Need help with Bootstrap

Glad to hear that it's working for you!
Michael Crotty
Sr Statistical Writer
JMP Development
0 Kudos