Subscribe Bookmark RSS Feed

How do I do T Test with data inside a column

chris_G_ttu

New Contributor

Joined:

Sep 13, 2017

Hi,

 

I do not know how to tell JMP to set up a t test the way I want. For example, I have two columns with my data. Column1 is A B C and column 2 is 1 2 3. I am trying to compare A1 to A2 and B1 to B2 etc. How can I do this in JMP? When I use the fit y by x and use the two exmaple columns it will give A vs B  A vs C etc. If I try to use a data filter  to filter 2 and 3 it will just show A1 vs B1. Do I need to make a column for each letter and number in order to get A1 vs A2 or is is there a simple way? My data set is pretty large so making columns for each letter and number would be difficult.  I am new to JMP and I have been trying to find a tuttorial to what I am trying to do. However I have not had much luck with my serach. If this has been answerd before then I would apperciate if some one could point in the right direction so that I can learn how to do what I am trying to do.

 

 

 

Thank you

3 REPLIES
Dan_Obermiller

Joined:

Apr 3, 2013

I think some more detail is needed. What are you measuring?

 

Maybe some general guidance will help (I hope). First, each column of your data table needs to be a variable. Each row of the table is an observation. You want to compare the means of different groups of data, correct? If so, then you need a column that identifies the group to which that particular observation belongs. You need another column that identifies the item that you are measuring. For a t-test, that item would need to be a continuous variable.

 

From your description, it sounds like you want three different comparisons: A1 to B1, A2 to B2, and A3 to B3. That would result in six column in your JMP table, structured something like this:

Capture.JPG

 

Then you can choose Fit Y by X. The Group variable goes in as the X. The measurement variable goes in the Y.

 

If your data is not structured this way JMP provides lots of ways to manipulate it to this format. Look at the Split and Stack commands under the Table menu. You can also ask for help in the community if you need help getting it into the proper format.

Dan Obermiller
chris_G_ttu

New Contributor

Joined:

Sep 13, 2017

Dan,

 

Thanks for the reply mybe this will help.Capture.PNG

 

this is an example of how my data is set up. I want to compare the mean tension from material A type 1 to material A type 2.

Capture2.PNG

I would like to have some graphs that have the x asis as A1 vs A2, B1 vs B2, A1 vs B1 etc.

 

Dan_Obermiller

Joined:

Apr 3, 2013

This does help. 


Because you have two different factors (Material and Type), you should really consider building a model rather than analyzing as a one-way ANOVA. I am also a little confused on why you are not interested in tests involving type 3 materials. Anyhow, here is one approach to making this happen (there are others).

 

This first part is just so that you can have the labelling as you requested. Create a new column (I called it Material-Type) that is based on a formula. You will use the Character function "Concatenate" to put together the Material and Type variables. As with your sample data, Type is numeric, so I needed to convert that to a character string using the Char function for the concatenation function to work. Your resulting formula would look like this:

Capture.JPG

 

Now use Fit Y by X, specify Tension as the Y and the newly created Material-Type column as the X. This will allow you to perform an ANOVA that compares ALL of the groups to each other. From the red pop-up menu choose Local Data Filter. Choose Material-Type as your filter variable and click Add. Hold the Ctrl key down and click on whichever two groups you wish to do the comparisons on. Then choose the Means/Anova/Pooled t from the Oneway Analysis of Tension By Material-Type red popup to get a t-test.

 

A big caution here is that you can conduct many different tests using this approach. Although each test may be at 95% confidence, the overall confidence level (often called the experimentwise confidence level) will be much lower than 95% because of the multiple tests. There are ways to perform the analysis that allows all of your desired comparisons that will control the overall confidence level.

Dan Obermiller