|
|
Pseudorandom Number Algorithm |
|
Message/Author |
|
|
Sorry, I don't think this is entirely the best topic for this post, but it seemed as applicable as any other... I was wondering what pseudo-random number algorithm is implemented in Mplus (specifically 4.2, if it matters). I ask because I've noticed differences in bootstrap results when using the same seed (0) between Mplus and R, for example (R seems to use Mersenne-Twister by default..). Thank you for your time. |
|
|
Regardless of the seed or the method, the bootstrap standard errors should be the same as long as a sufficient number of bootstraps have been used. This may be a large number if there are outliers. Try 10000 bootstraps. |
|
|
Hi Linda: Thank you very much for your reply. Moreso than the standard errors, I'm interested in the choice of endpoints from bootstrapping and comparability between different software programs. The method by which the bootstrap samples are generated will play a role in the bootstrap sampling distribution, thus influencing the confidence interval. With a sufficient number of bootstraps, I agree answers will tend to be similar, but these things matter at the margin and this is what I'm exploring. If this information is available, can you let me know what the algorithm is? If not then regardless I appreciate your time. |
|
|
Following is all of the information we have on our bootstrapping: http://statmodel.com/download/Resampling_Methods5.pdf when the bootstrapping is for survey data. Everything else is standard and it is from Efron's Bootstrap book http://www.amazon.com/Introduction-Bootstrap-Monographs-Statistics-Probability/dp/0412042312 or these two articles http://smr.sagepub.com/content/21/2/205.short http://www.tandfonline.com/doi/abs/10.1207/S15327906MBR3703_3 |
|
Back to top |
|
|