Bootstrapping t-values

[cringle]

Well,

this issue has been discussed in the SmartPLS(FAQ) section a while ago.

Please check
viewtopic.php?t=136

Best
Christian

[redlake]

Hi,

I have a question regarding the bootstrapping method:

what does it mean, if model parameters estimated with the PLS-algorithm differ widely from the results obtained using bootstrapping? say, a path coefficient is estimated in PLS as being .30, but in bootstrapping it turns out that its sample mean is .70. how can I interpret this result? do I have overly influential cases in my data, that distort the original PLS-solution?

thank you very much in advance

best regards


Matthias Rothensee

[stefanbehrens]

Matthias,

what you are describing sounds like an extreme case. Can you let us know some more parameters? E.g.

- original sample size
- bootstrap sample size
- number of bootstrap runs
- path estimate (original & bootstrap)
- bootstrap standard deviation


Cheers,
Stefan

[yusofahmad]

The bootstrapping output does not include such things as the AVE, R^2. In order for me to obtain these and also the significance tests for the model coefficients, do I need to run both separately-- the PLS algortithm and the bootstrapping procedure?
Any help here would be highly appreciated.

Thanks in advance,
Yusof

[boonchai]

In the bootstrapping results. I found that the standard diviation(SD) and standard error(SE) are the same value. I wonder how is the SE calculated? In my understanding, SE should smaller than SD. SE=SD/Square root(N)

Thank you very much.
Boonchai Kijsanayotin

[cringle]

Dear Boonchai,

these links may provide th information you are looking for:

http://www.uvm.edu/~dhowell/StatPages/R ... pping.html

http://www.stat.wisc.edu/~larget/math496/bootstrap.html

Best regards
Christian

[boonchai]

Thank you very much Christain.

Boonchai