How to compare categorical variables

[#1]

Hi everyone,

/* Using JMP 14.1.0  */

my question today is simple, however all the similar topics I found only responded to more complex topics, so here I am.

I am comparing categorical variables from different process conditions, say process one and process two. For each process, the items are tested and given a categorical class (e.g, Good, Bad, Medium, etc). For a particular class, I would like to know whether the total percentage of items are statistically different. E.g. is the ratio of 'Good' items for process one statistically different from the ratio of 'Good' items for process two. Both processes have a similar sample size around 300.

I tried using the fit Y by X platform, plotting class by process. In the the different analyses available, I'm guessing the one I need is the "Analysis of correspondences" (using jmp in a different language, might not be exact translation) which gives me a graph plotting 'c1' by 'c2'. I'm guessing the c1 value for each class gives me an information as to the whether they are statistically different from one process to the other but not sure which way it goes.

Any insights ?

[#2 · Accepted solution]

So you have a multinomial categorical response, Outcome = {Good, Bad, Medium, et cetera), and a categorical factor, Process = {one, two}. The response should use the ordinal modeling type. The factor should use the nominal modeling type. Starting with Analyze > Fit Y by X, assign Outcome to Y role and Process to X role. This set up will launch the Contingency platform. The default analysis provides the Pearson and the Likelihood Ratio chi square tests assuming no association.

The correspondence analysis is a graphical analysis to help interpret the association, if the tests indicate that there is a significant association. It is a form of 'biplot,' which allows two different variables (Outcome, Process) to be plotted together. The origin is based on the centroid and the two dimensions maximize the differences in the distances between the rows and columns of the contingency table. The ONLY aspect that is important and reliable is direction away from the origin. The distances are not comparable. So if a row level and a column level are away from the origin in the same direction, there is a strong positive association between these two levels. If they are away from the origin in opposite directions, then there is a strong negative association between them. The correspondence analysis does not provide a hypothesis test. Use the chi square tests for such decisions.

[#3]

Hi, I don't have an answer, but a further question on the same subject.

We are testing sikcness response of different varieties of olive trees. we have categorical response (level 0, level 1, level 2....) and 10 different varieties.

We obtained frequencies of the level of infection for each variety. Is there a way to test the differencies between all varieties (if level 1 for variety 1 is different from variety 2, variety 3...) or do we have to test varieties couple by couple?

Thanks

[#4 · Accepted solution]

From a statistics and modeling point of view, your situation is exactly the same. Be sure to change the modeling type of your response to Ordinal and add the Value Order column property to establish the right order. Your factor (variety) should use the Nominal modeling type.

[#5]

 hello, ı have a prolem too,

ı try to do categorical consumer research via jmp, I have 8 different row and 10 different column. I want to see both significance in each column for each row and significance between rows  for each column. I mean there are letter that show significance  for each column only but I want to see the letters both under each column and each row.  How can I see all them in that page ?