Adam Blake posted on Wednesday, March 10, 2010 - 3:58 am
I was wondering what the appropriate way to calculate p-values from the standard errors generated with the bootstrapping option?
I am assuming I can use the estimate divided by the standard error to estimate a Z value. My concern is this approach does not always correspond with the sometimes asymmetric confidence intervals also generated with the bootstrapping option.
The current version of Mplus gives p-values. These are based on the z-score from taking the ratio of the parameter estimate to its standard error.
Adam Blake posted on Thursday, March 11, 2010 - 5:27 pm
So in the event that my 95% bootstrapped confidence interval for the unstandardized parameter does not include zero, but the probability derived from the Z value is >0.05, that parameter should still be deemed to be not significant?
You would need to decide which one is best for your situation. The bootstrapped confidence interval takes the non-normality of the parameter estimate distribution into account. It is not necessarily symmetric.