Page 1 of 1
The results of bootstrapping
Posted: Fri Aug 29, 2014 12:34 pm
by lihu15402
Hi, all! I am trying to test a multilevel model using polynominal regression method. I need MLwiN report the results of regressions. For example, if the sample size of bootstrap is set 100, I need MLwiN give me a table including the coeffecients of these 100 regressions. Can MLwiN provide this function? As my known, SPSS and SYSTAT have this function.
A big thanks!
Re: The results of bootstrapping
Posted: Fri Aug 29, 2014 5:26 pm
by ChrisCharlton
I'm not sure whether this is exactly what you want, but if you switch the estimation method to bootstrap and run the model then the results of each replicate for the current set are stored stacked into column c1090.
Re: The results of bootstrapping
Posted: Sat Aug 30, 2014 9:15 am
by lihu15402
ChrisCharlton wrote:I'm not sure whether this is exactly what you want, but if you switch the estimation method to bootstrap and run the model then the results of each replicate for the current set are stored stacked into column c1090.
Thanks a lot! I found some data in column c1090 and some columns next to c1090. But, I am confused because, I set the replicate size to 20, after bootstrap, there were 100 numbers in this line.
My model is simple as follows:
Yil=b0j+b1Xij+b2X2+eij
b0j=b0+ u0j
What I need are the values of b0, b1, b2 of these replicates. I guess there would be three columns, for example, if the replicate size is set to 20, the three columns will respectively list 20 values of bo, 20 values of b1 and 20 values of b2.
Hope for your answer!
Re: The results of bootstrapping
Posted: Sat Aug 30, 2014 10:15 am
by lihu15402
Dear Mr. Chris,
I think I have found the answer in your answers to other users, such as:
"The parameter chains are stored by default in c1090 as a sequence of successive sets, one for each iteration. Within each set the parameters are stored with the fixed parameters first, followed by the random parameters. The order of these should match that in the equations window. You can extract a chain for each parameter by using the SPLIt command on this column."
Do I understand right? And, could you tell me how to use the Split command to extract the parameter chains I need?
Re: The results of bootstrapping
Posted: Sat Aug 30, 2014 3:40 pm
by ChrisCharlton
Yes, this is correct. The model results can be split into a column for each parameter in the same way as for MCMC chains. For an example of this see section 4.9 (for commands) or 22.4 (using the GUI) of the MCMC guide (
http://www.bris.ac.uk/cmm/software/mlwi ... mc-web.pdf).
For your model with five parameters and a set size of 20 you would want something like the following (assuming c300-c305 are available):
Code: Select all
code 5 1 20 c300
split c1090 c300 c301-c305
name c301 'beta0' c302 'beta1' c303 'beta2' c304 'lev2var' c305 'lev1var'
Re: The results of bootstrapping
Posted: Sun Aug 31, 2014 3:05 am
by lihu15402
Mr. Chris,
I got it. Thank you very much!
Best,
Hu