bootstrap sample 5000

[hussein]

I am writing to inquire about bootstrapping sample size.
I am preparing an explanation to my supervisor who asked whether my 500 bootstrapping sample was enough for my analysis.

The explanation I have prepared so far is:

Hair et al. (2014 p132) stated that the bootstrap samples must be at least larger than the number of valid observations in the original data set but recommended 5000. However, running the recommended 5000 with the current research's complex model resulted in indeterminacy problem i.e. SmartPLS stopped producing results after running. Thus, 500 bootstrap samples were used, and the number of cases were 333, which was identical to observations in the sample as recommended by Garson (2016; Hair et al., 2014). Additionally, the researcher deemed 500 bootstrap samples adequate as it found support from a study by Deng et al. (2013) who found that the number of bootstrap replicates, ranging from 500 to 2000, had little effect on either bootstrap standard error or confidence interval.

I would appreciate any pointers or input to my above justification.
Does anyone else run into PLS3 stopping with complex models or am I the only one. And if so, I wonder what I am doing wrong!

[jmbecker]

Usually, there is no big difference whether you use 500 or 5000 bootstrap sample, in terms of the substantial implications drawn from your PLS model.
It will only matter if you have effects that are on the edge of being significant (e.g., around a p-value of 0.05). In addition, a larger number of bootstrapping samples is advised when using confidence intervals, because the estimates of the confidence intervals are getting more robust and stable as the number of bootstrapping samples increases.
Hence, I usually advise people to use as many bootstrapping samples as possible with your computer. If it is possible to calculate 1,000 than do that instead of just 500.

That the algorithm stops with bootstrapping samples larger than 500 can also be a problem of insufficient memory (https://www.smartpls.com/faq/outofmemor ... t-can-i-do). You should increase the available memory for SmartPLS and you might also use just the basic bootstrapping and not the complete bootstrapping to save computational power, if you are not interested in the advanced bootstrapping results (i.e., model fit, etc.).