I am running a PLS model with a low number of observations ($n=50$). While several pieces of academic work argue that this sample size is appropriate to run this type of model, I am quite confused when it comes to the number of cases I take in my bootstrapping procedure.

I get some very strong path estimates for certain paths without a significant t-statistic (50 cases, 500 samples). However, when I put the number of cases above n (lets say 100 cases, 500 samples), the t-values are more in line with the path estimates.

  • $\begingroup$ That seems like a logical result to me. If you arbitrarily increase the bootstrap re-sample size, the estimates will be less variable. I presume such an arbitrary increasing of the bootstrap re-sample size is totally inappropriate, so what is your motivation for doing it at all? $\endgroup$ – Andy W May 29 '13 at 12:16

Bootstrapping typically involves sampling with replacement from your sample data. The size of each bootstrapped sample should be the same as your actual sample size (i.e., n=50). This allows you to study the sampling distribution of your estimators in your actual sample size.

As @Andy notes, you can arbitrarily increase your bootstrapped sample size (lets call it $b$). As $b$ approaches $\infty$ the standard error of your estimator will tend towards zero. All non-zero coefficients will by definition be statistically significantly different from zero. I remember reading a journal article where an author artificially inflated their sample size using bootstrapping and reported the resulting p-values. This is very wrong! It also may help to explain some people's naive mistrust in the concept of bootstrapping.

Bootstrapping is not a tool to make standard errors smaller. It is a tool to, among other things, give an estimate of standard error.

Your sample size is small (i.e., n =50) so presumably you have larger standard errors than you would like. This is the reality. Thus, you can accept the large standard errors or collect more data and increase your n.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.