BookmarkSubscribe
Choose Language Hide Translation Bar
Highlighted
Mananahi
Community Trekker

Dunnett's Test Statistic

I have run an ANOVA in JMP with a Dunnett's post hoc test to compare 8 treatments to my control. In a publication I am working on I reported the p-values for each comparison to the control. A reviewer has asked me to include the test statistic along with the p-values. For some reason, a test statistic is not displayed along with the p-values in the default settings on the Dunnett's output within the fit model personality. I'm new to statistics so I'm not even certain what the test statistic is for a Dunnett's procedure.  Does anyone know what that test statistic would be and how I can get JMP to display it?

0 Kudos
2 ACCEPTED SOLUTIONS

Accepted Solutions
Thierry_S
Community Trekker

Re: Dunnett's Test Statistic

I believe what you are looking for is the Q value shown in the LSMeans Difference Dunnett table (equivalent to t statistics but for multiple comparison to a control value).
If this not what you are looking for, then I'm afraid that I will not be able to help you any further.
TS
Thierry R. Sornasse
0 Kudos
David_Burnham
Super User

Re: Dunnett's Test Statistic

I've put together some fake data to illustrate what is going on:

 

dunnetts.PNG

 

I've chosen the data such thast the contrast between c and the control (a) has a p-value of 0.05 i.e. the threshold at which we are assessing statistical significance.  

 

A t-statistic can be calculated as a the ratio of the difference to the standard error of the difference.  For the second contrast the statistic is 2.5 i.e. the Q statistic.  So whilst there is a statistic for each comparison, we can define a single statistic to that is the threshold for statistical significance.

 

You can check Wiki for references on the method although there is a tendency for Wiki to be overly abstract for statistical methods but you should easily be able to search for other explanations, e.g.

http://davidmlane.com/hyperstat/B112114.html

-Dave
0 Kudos
6 REPLIES 6
Thierry_S
Community Trekker

Re: Dunnett's Test Statistic

Hi, 

 

I think you may be able to get what the reviewer is asking by performing your test in the "Fit Model" analysis instead of the "Y by X".

 

Here is a simple example with dummy data (Y) for 3 categorical levels (X)

 

Go to Analysis > Fit Model and assign your Y and X as shown:

 Fit Model Simple ONEWAY.PNG

Click Run, and in the model result, go to GROUP effect > Dunnett's

Sub Menu Dunnet's in Fit Model Result.PNG

The resulting screen in giving you the full test statistics (LS Means Differences Dunnett's) as shown in this final screenshot:

Dunnet Test Statitics in Fit Model.jpg

That should satisfy the reviewer. If it does not, you should try to clarify what he/she means by Test Statistics.

 

TS

Thierry R. Sornasse
0 Kudos
Mananahi
Community Trekker

Re: Dunnett's Test Statistic

So I'm looking at the LS Means Differences Dunnett's table on your screenshot and it has columns for Level, -Level, Difference, Std Err Dif, Lower CL, Upper CL, and p-Value. Shouldn't there be one more column in there that says t-Statistic (or some other test statisctic) next to the p-value so that we know what was used to calculate that p-value. Something like that seems to be on most JMP outputs. That's what I think I am looking for, but maybe I just don't understand stats and am assuming that something exists. Do Dunnett's procedures not have test statistics?
0 Kudos
Thierry_S
Community Trekker

Re: Dunnett's Test Statistic

I believe what you are looking for is the Q value shown in the LSMeans Difference Dunnett table (equivalent to t statistics but for multiple comparison to a control value).
If this not what you are looking for, then I'm afraid that I will not be able to help you any further.
TS
Thierry R. Sornasse
0 Kudos
Mananahi
Community Trekker

Re: Dunnett's Test Statistic

I think that is what I'm looking for. thanks. I'm still a little confused though. Why is there only one statistic for the whole table. Most tables have an individual test statistic for each p-value. Can you or anyone enlighten me to why Dunnett's would be different?
0 Kudos
David_Burnham
Super User

Re: Dunnett's Test Statistic

I've put together some fake data to illustrate what is going on:

 

dunnetts.PNG

 

I've chosen the data such thast the contrast between c and the control (a) has a p-value of 0.05 i.e. the threshold at which we are assessing statistical significance.  

 

A t-statistic can be calculated as a the ratio of the difference to the standard error of the difference.  For the second contrast the statistic is 2.5 i.e. the Q statistic.  So whilst there is a statistic for each comparison, we can define a single statistic to that is the threshold for statistical significance.

 

You can check Wiki for references on the method although there is a tendency for Wiki to be overly abstract for statistical methods but you should easily be able to search for other explanations, e.g.

http://davidmlane.com/hyperstat/B112114.html

-Dave
0 Kudos
David_Burnham
Super User

Re: Dunnett's Test Statistic

Also I should have mentioned JMP's online help:

dunnetts2.PNG

https://www.jmp.com/support/help/14-2/effect-details.shtml#625536

 

-Dave
0 Kudos