Showing results for

- JMP User Community
- :
- Discussions
- :
- creating a mean table with p-vaue

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

May 9, 2017 1:58 PM
(747 views)

10 REPLIES

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

May 9, 2017 2:28 PM
(744 views)

If you go to

Analyze==>Screening==>Response Screening

You can enter in all of your Y variables, and you X variable(s) and it will generate a data table with your results, including the p-values

Jim

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Jul 5, 2017 11:00 PM
(332 views)

thanks for the suggestion. Is there a way to access the output table of Response Screening from jmp script?

I would like to iterate through the table and pick up all the row numbers that have a RSquared > 0.95.

This output table doesnt seem to be the current data table.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Jul 6, 2017 2:25 AM
(323 views)

Here is an example of how to do that taken from the Scripting Index

Help==>Scripting Index==>Response Screening==>Get PValues

```
Names Default To Here( 1 );
Open( "$Sample_Data/Probe.jmp" );
obj =
Response Screening(
X( :Process ),
Y( Eval( 8 :: 394 ) ),
Save Outlier Indicator
);
dtOut = obj << Get PValues ;
show( dtOut );
```

dtOut << Select Where( :RSquare > .95 );

I modified it very slightly, to add the pointer to the "PValues" data table which you can then use to do the selection of the RSquare values. I do recomment that you examine the FDR Logworth values. When doing so many tests, the FDR Logworth will help you determine which of the tests are real, and which are due to alpha slipping.

Jim

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Jul 11, 2017 7:35 PM
(259 views)

Actually my data set has about 6000 columns all with p-value < 0.01 and many with a significant FDR logworth. However, some of the columns correlate to other columns with RSquare > 0.95 (similar test results). So currenly I'm running response screening of all columns vs all columns, grouping those columns that correlate with RSquare > 0.95 and then keeping just 1 of the correlating columns per group. It does take a few hours (~1000 rows) to run the script though for a decent reduction in number of columns.

Would be great if there is a faster way to group correlating columns together and then just keep 1 column per group. Pls share if there's a better way to do this.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Jul 11, 2017 7:42 PM
(257 views)

You could create a correlation matrix using

Analyze==>Multivariate Methods==>Multivariate

or use Prinicpal Components analysis

Analyze==:Multivariate Methods==>Principal Components

Jim

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

Jul 12, 2017 4:42 AM
(249 views)

The procedure that you are looking for is called 'variable clustering.' The **Cluster Variables** command in the **Analyze** menu will launch the JMP platform for this purpose.

Please see **Help** > **Books** > **Multivariate Methods** > **Chapter 11: Cluster Variables** for more information.

Learn it once, use it forever!

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

May 9, 2017 2:49 PM
(740 views)

Use the **Oneway** platform. Start by selecting **Analyze** > **Fit Y by X**. I assume that you have a numeric variable stored in a data column using the continuous modeling type for the **Y** role. I also assume that you have a categorical variable in a data column using the nominal modeling type. Put that column in the **X** role.

Click the red triangle next to **Oneway** and select **Compare Means** > **Tukey-Kramer**. You will find the answer several different ways. The last report shows what you want. Remember that you can right-click on this last report and select **Make Into Data Table** if you need to do something else with results.

Learn it once, use it forever!

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

May 10, 2017 2:20 PM
(702 views)

Example: I have sample 1 (control) , sample 2 and sample 3. I recorded dimension data for length, width, and height.

I am comparing if sample 1 (control) is significantly different than sample 2 and 3 for each length, width and height.

So when I create the one way it is just for sample 1, 2 & 3 for length. How can I create a table with p-value for length, width and height. Do I need to go to the red triangle for each dimension, compare mean then create the table? This seems inefficient. Could I just use the tabulate function to create the table of means then in the data table add a formula to create the p-value?

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Get Direct Link
- Email to a Friend
- Report Inappropriate Content

May 10, 2017 2:35 PM
(696 views)

You can broadcast the Compare Means command across your report and then use Make Combined Data Table to make one data table from all the individual report tables.

Start with Fit Y by X using sample as X and length, width, height as Y.

Then, in the report, hold the Ctrl key (Windows) or Command key (Mac) down and choose **Compare Means -> All Pairs, Tukey HSD.**

Then in the report table that contains the p-Values, right click and choose **Make Combined Data Table**.

That gets you one data table with all the p-values.

-Jeff