Standardized residuals in two-way tables

[#1]

I should compute the standardized residuals for two-way (as well as multiway) tables - option CROSSLIST(STDRES) of PROC FREQ in SAS, if this helps. I didn't find any options, nor a suitable script. Any suggestion? Thank you.

[#2]

You can analyze two-way contingency tables using Analyze > Fit Y by X (launches the Contingency platform) or by using Analyze > Fit Model (launches Nominal Logistic platform or Ordinal Logistic platform, depending on modeling type of response variable). Unfortunately, none of these analyses provide the residual that you want.

I am also unaware of any script for such a result.

[#3]

thank you :(

[#4]

The computation is simple so I scripted it for you. It is a simple script without any extra features. It will launch the Contingency platform with your selected response and factor and then add a new report at the bottom with the standardized residuals.

I would appreciate it if you would independently verify the results, perhaps running your example through SAS first and then JMP.

[#5]

Thank you for your script.

Actually there are two formulas for the std res: adjusted and not adjusted. The one you used was the adjusted one (which is fine), but there should be an error in the way pidot# and pdotj# are calculated. Unfortunately I can't really understand the script, hence I can not fix it. I simply added a row with the (slightly simpler) alternative formula, just to check where the problem is - in case you are still interested.

[#6]

This formula is from the SAS documentation for the FREQ procedure:

I didn't see mention of an adjusted or not adjusted residual, only this formula. You modification of the script is fine, but you can comment out the line with the adjusted formula. You are computing the residual twice.

What do you ;mean "there should be an error in the way pidot# and pdotj# are calculated?"

Why do you prefer the un-adjusted version?

[#7 · Accepted solution]

I discovered what you meant by the error and corrected the computation. Also, the launch dialog now provides a choice between the adjusted or the not adjusted computation, with not adjusted as the default choice.

Thanks for pointing out the error. That is why I asked for your review! I was taking a short cut and missed the important distinction in the counting process.

[#8]

Now it's fine, I will definitely adopt it. Thank you again

[#9]

Actually, I mistakenly got the computation of the not adjusted and adjusted backward. I fixed it. I also added the option analysis Freq role in case your data is pre-summarized. My apologies.

Use the new script here.

[#10]

Got it, thanks. I hope it will be useful for somone else.