topic Fewer boostrap samples than requested in Discussions

Fewer boostrap samples than requested

tMinnx — Thu, 18 Apr 2024 07:47:31 GMT

Hello. I am trying to use the bootstrap function on a parameter table. I asked for 1,000 samples but the results are fewer than 1,000. After manually checking the data table, I noticed some BootIDs were skipped. Is it supposed to happen? I appreciate any help you can provide.

Re: Fewer boostrap samples than requested

Ross_Metusalem — Mon, 22 Apr 2024 22:09:32 GMT

Does your data have missing values? It could be that some of the bootstrap samples end up containing rows with missing values such that the parameters you are bootstrapping cannot be estimated. These samples are then not reported in the final table.

You could try excluding rows with missing values and running the bootstrap again to see if you get the expected number of rows / BootIDs.

To illustrate the behavior, you can try boostrapping the summary statistics table in the Distribution platform for a continuous column with a relatively large proportion of missing values. (Ensure the Summary Statistics table contains both N and N Missing.) You'll see that all rows / BootIDs are reported, but there can be some where the summary statistics are missing values except for N and N Missing, where N will equal zero and N Missing will equal the desired sample size. If you then bootstrap the quantiles table in the same Distribution report, your results may skip some BootIDs altogether.

Re: Fewer boostrap samples than requested

tMinnx — Wed, 24 Apr 2024 02:49:18 GMT

Thank you very much. Yes, I have missing values in my column where parameters are estimated. That must be the reason.