regular t-test VS Welch's t: white paper

[gianmarco]

Hi,
I came accross this interesting white paper.
Here is the link.
It deals with the topic of the assumptions of traditional t-test and of its comparison with the Welch's t. The pros and cons of each are evaluated.
I found it in the site of Minitab program. Nonetheless, I think it can prove useful to a more general audience.

Hope it helps
Best Regards
Gm

[Dason]

Oh man. We did a lab/homework that was investigating essentially everything they went over in that paper in my Stat 601 course. I probably still have my R code lying around if anybody wants to conduct similar simulation studies for themselves in R.

[victorxstc]

Thanks for sharing this nice paper with us

[GretaGarbo]

I probably still have my R code lying around if anybody wants to conduct similar simulation studies for themselves in R.

I think it would be really great if Dason shared his code, or part of it, for this.

This (area) seems to be one of those that are most frequently asked about. “My data is not normally distributed….”. Those who have, say 12 and 17 observations that are slightly skewed could check themselves how sensitive it is.

edit:
Thanks Gianmarco for bringing this link.

[Dason]

Actually I just checked and apparently I don't have that code anymore. The server I used for my 601 homework crashed and there were adequate backups made on that one apparently. Since I only used that server for 601 I forgot that it had crashed since I had backups of most other stuff. I could probably recreate most of it fairly easily though.

[victorxstc]

GM, I couldn't find time yet to read it. Just some pages. Could you please tell me according to it, which is better? I am interested particularly because Dason had asked me this.

When doing a t-test do you default to assuming equal variances or unequal variances?

[gianmarco]

Victorxstc,
you are welcome.
You could also read another article I posted some time ago... If I manage to find the thread I will surely point it out to you replying to this thread.
As to Dason's question, I would suggest to you to assume unequal variance....just to keep on the safe ;-)

Gm

Edit: here is the link to the article: GD Ruxton, "The unequal variance t-test is an underused alternative to Student’s t-test and the Mann–Whitney U test".

[victorxstc]

Hey GM

Thanks for the second paper

The title itself is indicative of its advantage over the routine t test So I will switch to using it for usual cases

[victorxstc]

That second paper was great I will use Welch t test only (or mostly) from now on