Bootstrapping stopped without output

Forscherin_SB · Post by **Forscherin_SB** » Wed Feb 08, 2012 7:00 am

Hello!

Yesterday I wanted like to test my model by running the bootstrapping. Because it is a very complex model I started bootstrapping (sample: 5.000) before going home to have the results this morning.
Unfortunately this morning I couldn't read any results. It seems to be that nothing has been calculated by night. The model is displayed as usual, only with the path coefficients and I only can have a look at the output of the pls algorithm.

What can I do to avoid this? Could there be a problem with the cache of smartpls so that the results of the bootstrapping are deleted after some hours?

Before running the bootstrapping I turned the standby-modus of the computer of.

As the bootstrapping with a sample of 5.000 with my model takes more hours than a normal workday has, I have no possibility to test it whithin a workday.

Thanks for help!

vonbergh · Post by **vonbergh** » Sat Mar 10, 2012 8:26 pm

Hi Cati,

Bootstrapping will require more power from your computer for each of the following:
- the larger the sample,
- the more complex the model (more IV's, DV's and paths)
- the larger the number of iterations

I can suggest the following:
- close all unnecessary applications
- close screen saver
- allow more priority to the SmartPLS program (e.g. realtime or high) (ctrl+alt+del -> processes -> select smartpls proces -> right mouse click -> set priority -> realtime)
- first run the bootstrap for a low number of iterations (e.g. 50 or 100), record the results and increase your number of iterations step by step.

Should others have additional tips, then I am curious to learn these.

Good luck,
Dennis

Taufik · Post by **Taufik** » Sun Mar 25, 2012 10:54 am

I have the same problem when run the model with 5 IV and 4 moderators. The bootstrapping procedure stop at n=4539. I tried to run the same analysis with different PC with i3 and i7 processor but the same problem occurred. I also tried set the priority to "real-time" and close other applications as suggested by Dennis, but it's not solve the problem.

Since Hair et al., 2011 suggested minimum of resampling is 5,000, how to report the result?

Hengkov · Post by **Hengkov** » Sun Mar 25, 2012 9:30 pm

Hi Taufik,
Some literature suggest number of resampling 200-1000 is reasonable for correct standar error estimate.
Note: must be grather than original sample.

For recomendation Hair et al. (2011) minimum number of resampling 5.000 => computer program is error with large complex model and difficult open the report HTML analysis (My experience).
Regards,
Hengky

Taufik · Post by **Taufik** » Mon Mar 26, 2012 4:51 am

Tq Sarjana, so far i didn't find literature suggested n=100-1000. Karim(2011) use 1000 samples for bootstrap but no justification given.

Hengkov · Post by **Hengkov** » Mon Mar 26, 2012 9:45 pm

Hi Taufik,
Bootstrap procedure used all original samples for resampling.
Thus, number of bootstrap samples must be grather than original sample.
For example, you original sample 375, you can used number of resampling 500-1000 is enough.
See Efron, B., and Tibshirani, R.J. 1993. An introduction to the bootstrap. New York: Chapman & Hall.
Efron/Rogosa/ Tibshirani. 2004. Resampling method of estimation. In Smelser and Baltes (Eds). International Encyclopedia of the social and behavioral sciences, New York: Elsevier.
For report analysis see Chin, W. W. 2010. How to write up and report PLS analyses. Handbook PLS. pp. 650-690, Berlin: Spinger
Regards,
Hengky