Correct for optimism in prediction modelling using bootstrapping (bsample command), is this the correct approach?

[Linda Velde]

Dear statalist users,

I have a question about the bsample command in STATA. I want to use this command to obtain the optimism-corrected performance estimates (R squared and adjusted R squared) for a prediction model using linear regression. I plan to calculate this as:
Optimism= (adj) R squared_bootstrapped - (adj) R squared_ original sample
Optimism corrected values of (adj) R squared= original sample (adj) R squared value minus optimism
To this end, I first ran the linear regressions in the original sample (using the ‘regress’ command) to obtain the ‘(adj) R squared_ original sample’ and then used the bsample command where after I ran the same linear regressions to obtain the ‘(adj) R squared_bootstrapped’.
Could someone indicate whether this is the correct approach?

And should the bootstrapped R squared value always be lower than the original value? Or can it either be higher or lower and you just take the difference to calculate the optimism (so that the optimism-corrected value is always lower than the original value)?

Thank you in advance your you help!

[Phil Bromiley]

Welcome to Stata list. You didn't get a quick answer. You will increase your chances of useful answer by following the FAQ on asking questions-provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

I'm afraid I don't work with these specific tools. It's not clear to me why bsample is the right approach rather than bootstrap or bstat. I suggest you try reposting following the FAQ on asking questions and see if you can get someone who knows these issues better than I do to help you.