Using Summarized data to perform Chi Square testsI got this question from a customer who is transitioning from Minitab. Has anyone run into something like this? Any suggestions are appreciated.
Three challenges/questions that I have are regarding categorical data analysis and chi-square testing.
(1) I’ve noticed that if I have two-way tables, as shown in Table 1, I cannot input this into JMP directly unless I create a column format table (Table 2). This seems a bit cumbersome as it takes quite a bit of time to convert from one to the other… please advise if there is such capability or feature in JMP that can help with this:
Table 1
Table 2
(2) If I complete a chi-squared test on the data above and get a p-value that is below my alpha level of significance, how do I know which sample(s)/group(s) is/are significantly different from the rest? With ANOVA, JMP has the Each Pair Student’s t-test and Tukey’s HSD test to tell the differences between each group/sample. How is this done with multiple categorical data sets? Minitab, for instance, provides the p-value as well as table/charts of expected vs observed to be able to tell the groups apart, per Figure 1 below:
Figure 1
(3) Challenge #2 above was for multiple categorical data sets with 2 outcomes. What if I have 3 or more outcomes? For instance, see Table 3 below where it compares 5 groups (Area) and 3 outcomes (invoice error types). Minitab allows one to complete a Chi-squared test for association and illustrates the comparison of expected vs actual observations for each group per Figure 2 below. How can this be done in JMP? And how does the data have be inputted into the JMP data table to conduct the analysis? Can I use a 2-way table like in Minitab?
Table #3
Figure #2
Re: Using Summarized data to perform Chi Square testsThe Categorical platform should be helpful here.
(1) The user can create a new col of Actual Fail frequencies. Then in the Categorical platform, select the Multiple tab and put False and Actual Fail cols together as Response Frequency cols. Down below, put LOBP as the X, Grouping Category and Total as the Sample Size.
(2) In the report, select Top Red Triangle > Test Response Homogeneity for the association test, then Compare Each Sample for the follow-up pairwise tests. I believe the p-values for the pairwise tests are uncorrected even if FDR is turned on under Test Options. The user may need a walkthrough of interpreting the comparison letters. IMO they're a little confusing compared to the top right table in that Minitab output (but it might just be me).
(3) The same multiple response format as in (2) should work again here, just with all three response cols together as Response Frequency cols and a new Total col for Sample Size. But in the example I wonder if the invoice error types are mutually exclusive, because if not then the LR or Pearson tests aren’t appropriate. The three error types could be analyzed separately or perhaps somebody knows of an appropriate alternative analysis.
