Dear Farimah, 

Bootstrap methods have been proposed as a method for improving cutting scores for classifying observtions when local base rates for one's sample differ from the base rates for the sample upon which the original cutting scores were developed (Rorer & Dawes, 1982).

A related family of bootstrap methods involves iterative resampling of jackknife half-samples and cumulating a frequency distribution of the statistics under investigation (Stine, 1990).

Boottstrap procedures have been recommended for use in validity generalization research because of their ability to generate accurate estimates of the standard error of correlation coefficients (Switzer, Paese Drasgow, 1992).

Although bootstrap estimates may obverestimate positive predictive value when the population is low and may overstimate the base rate in applications involving a variable that has a low base rate for one of the response levels. 

The leave-one-out procedure represents an extreme variation of the one-sample jacknife.

The leave-one-out procedurereduce bias that occur when estimating model performance using a single sample and is widely used in time-series and discriminant analysis and gives an almost unbiased estimate of the confidence intervals for the true classification error rate. it is also robust over violation of the normality assumption and the relative way with which it is generalized to multicategory applications.

I hope that this help you in some way.

Have a nice day

Helena

Dear Farimah,

No one was right, the State of the Art about mediation is the potential approach of Imai et al. Please see some papers:

http://psycnet.apa.org/journals/met/15/4/309/

http://www.jstor.org/stable/41058997

http://journals.cambridge.org/abstract_S0003055411000414

The software R includes packages for mediation:

http://link.springer.com/chapter/10.1007/978-1-4419-1764-5_8

http://dspace.mit.edu/handle/1721.1/91154

Also, Stata has a package for mediation, but with fewer functions.

Good luck.

Andrés.

Farimah,

There are several approaches to mediation, and some are better than others. It depends on many things, but perhaps most important is the type of variables you have in the various parts of the model.

I am attaching a paper I recently wrote on mediation analysis, where we ran several simulations for different mixes of variable types for the intervention, mediator and outcome. As you will see, some models performed better than other.

All these methods can be applied in Stata and we refer the reader to those related commands.

I hope this helps

Ariel

Article Using mediation analysis to identify causal mechanisms in di...

Dear Farimah,

For my opinion, you can choose the causal step approach for mediator modeling and then you can use the bootstrapping method to justify your finding. Usually, both of these method will be provide almost similar finding. Maybe you can read the book " A handbook of SEM" written by Zainudin Awang. The new version will be launched next month. He will explain more about the mediator and second order model. 

Thank you all for your explanations and the attached files.

Wishing you all the best, 

My dear

..... to increase the robustness of our findings. 

Thank you for your reply dear Prof. Al-Atwi. Would you please share with me your idea regarding the function of Sobel test mentioned in the question?

Thank you,

Best regards,

Kindly find the detail explanation of Sobel test, Kenny mediation procedure, and Bootstrapping method

Article Making Sense of Mediating Analysis: A Marketing Perspective

Dear Farimah,

Good question, I think as what you mentioned that it is not frequently researchers use these test. Recently when bootstrappying added to recent versions of SPSS, researchers start to use it. In a recent published study, {attached} We c onducted moderation and mediation tests using multiple programs and software which included SEM, SPSS, AMOS, SOBEL test,and PROCESS Macrosfor SPSS. In order to confirm if a variable is making amoderation effect in the relationship between an independent variable and a dependent variable, the nature of this relationship must change once the moderator variable changes.

you can read more detail in the attached article.

Regards