Sensory Analysis: Simple Discrimination Tests

[#1]

This Add-In creates an easy interface to some of the most common sensory tests: Triangle Test, Duo-Trio Test, Paired Comparison, and 2-Out-Of-5 tests.
Each of these is a chi-squared tests of proportions, with a given pair of probabilities for a "right" answer under the null hypothesis that the samples are not different. An example test for the Triangle Test and the original script are included with the Add-In.

The data table should contain a binary column to be tested. This column must have the Modeling Type Nominal. For best results with the Triangle Test and 2-Out-Of-5 Tests, assign a value label with Yes/No or Correct/Incorrect to ensure that the porportions are assigned properly. 

The JMP documentation shows an example of the Test of Proportions.

[paulp]

Adding sensory tests is a nice feature.  I use triangle tests a lot. How do I access the example for the Triangle Test?  I have the add-in installed, and can open it. However, I don't see where the sample data is installed -- I looked in the Sample Data Library,is that where it's located?   Thanks, Paul

[melinda_thielba]

Hi, Paul! My apologies. I didn't realize that you can't access a file packaged with an AddIn. 

I've attached the file to the original post instead. Please make sure you have the current version. I uploaded a fresh copy yesterday. Please also let me know if you have questions or problems!

[melinda_thielba]

And apparently there is a way to get to the sample data I packaged with the AddIn.

In JMP,

1) Go to View->AddIns.

2) Click the name of the Addin in the window that appears

3) Click the hyperlink for the addin Home folder. 

[paulp]

Hello Melinda, thank you for your reply and for pointing me to the data.

I have a question --what was the alternative hypothesis used? I'm thinking that it should be that the proportion of Correct Answers > 0.3333, because that is evidence that there is truly a difference between products A and B. The triangle test is a discriminatory test, so we care about whether panelist choose the correct answer more than would be expected by chance.

[cid:image001.png@01D2D0B0.86FD89B0]

When I ran the test with the one-sided alternative hypothesis, I got a slightly different probability than was in the Distribution script supplied with the sample data. Not sure if this 2-sided Chi-Square test was intended to represent a triangle test.

[cid:image002.png@01D2D0B0.86FD89B0]


Regards, Paul

[melinda_thielba]

Hi, Paul. 

The Chi-Squared is always a one-sided test. StatsExchange over on StackOverflow has a nice discussion on this.

So, following the guidelines on SensorySociety.org, the stated p-values are correct.  

One could argue that we should perform a two-sided test instead. JMP offers that as an exact binomial test. 

That test is not as well-known, and I haven't seen it recommended in the sensory analysis literature. I feel like I'm in a bit of a bind when it comes to reporting p-values for the user. The software is written to spec, but maybe it's not what a statistician would recommend generally. 

What's your opinion?