topic Re: Polychoric correlation assumptions in Statistical Procedures

Polychoric correlation assumptions

DavidB1 — Sat, 29 Feb 2020 03:29:07 GMT

I am performing a polychoric correlation to use in Principal Component Analysis. The data are mostly Likert scale (4 levels), with a sample size of 68. For some questions in the survey, there were only a few responses (1-4 out of 68) for certain numbers. I know for Chi-Square tests (ordinal or otherwise), you have to combine some cells together first when that happens (if the expected value will be less than 5). Do I also need to combine cells first for polychoric correlation?

Re: Polychoric correlation assumptions

Ksharp — Sat, 29 Feb 2020 04:24:18 GMT

Try PROC PRINQUAL

Re: Polychoric correlation assumptions

DavidB1 — Sat, 29 Feb 2020 04:29:07 GMT

Thanks. My question is, do I need to worry that some cells only have a few values (like n=3 for "Strongly Agree") when I do polychoric correlation? With Chi-Square Tests, you have to be careful that the expected value for any cell doesn't go below 5. Does polychoric correlation have a similar requirement?