My question is when I compare the frequency distribution of people's answers based on their education levels (for example) through chi-squared test, I got a warning that says (for instance) 7 cells have less than count five; minimum expected count is; 43. I got this warning for almost my all questions - demographic characteristics comparisons.
So can i ignore this warning and go along with Pearson's Chi Square value and its corresponding p value?
Or I will take into consideration Fisher's exact test value?
Thank you.
Sarah, you do not have a 2x2 table, you have a 4x2 table, with two **really** tiny and underrepresented groups (upper class, n=3; and upper lower, n = 7). It would clean up your analysis to just use the two groups you have signal from - the upper middle and the lower middle. You can either exclude your extremes or you can fold them into their adjoining groups. Given your chi-square, I don't think that will bump you into significance, but at least you will have been spared the effort of talking about groups that are too small to provide signal in your data.
Principle 1 - more information is better. Following this principle, you would examine both Pearson and Fisher and see if your story changes. If it does, you would do more at the exploratory level to discuss why.
Principle 2 - the outcome matters more than your predictors when it comes to this error. Do you have fewer than you need in the frequencies for your predictor or your outcome? For the predictor, you might be able to fold together groups (example: I have a project which involves education levels, but few identified as Ph.D. It was possible to fold them into the post-masters group without much loss of fidelity in the analysis). For the outcome, you may need to do a bit more exploratory testing to determine if you have too many groups.
Of course you have to respect the Fisher's Exact Chi Square for this situation.
Sarah, you said you got a warning message that says (something like): 7 cells have less than count five; minimum expected count is 43. Surely the warning listed the number of cells that have expected counts < 5, did it not? And if 7 cells have expected counts < 5, the minimum expected count must also be less than 5.
What are the dimensions of your contingency table (i.e., how many rows and columns)? Can you share the observed counts with us?
Note that for tables larger than 2x2, it is okay for some expected counts to be < 5. I have summarized the assumptions on the page linked below.
HTH.
https://sites.google.com/a/lakeheadu.ca/bweaver/Home/statistics/notes/chisqr_assumptions
Thank you Julia and Ibrahim.:)
Thx Bruce Weaver.. The dimension of contingency table is 2*2. I am attaching output generated by SPSS for comparing socio economic status of Schizophrenia and Depression patients. There you can find Observed frequency and chi square value table. Will be obliged if you will tell me which values I have to take into consideration.
Sarah, you do not have a 2x2 table, you have a 4x2 table, with two **really** tiny and underrepresented groups (upper class, n=3; and upper lower, n = 7). It would clean up your analysis to just use the two groups you have signal from - the upper middle and the lower middle. You can either exclude your extremes or you can fold them into their adjoining groups. Given your chi-square, I don't think that will bump you into significance, but at least you will have been spared the effort of talking about groups that are too small to provide signal in your data.
This article might help:
http://www.angelfire.com/wv/bwhomedir/notes/categorical.pdf
Thanks for the plug, Arpita Dutta. Those notes (and others) are now available on a newer Google Sites page:
- https://sites.google.com/a/lakeheadu.ca/bweaver/Home/statistics/notes
I wrote that chapter many years ago before I knew about the N-1 Chi-square, which I would not typically recommend in place of the Fisher-Irwin test (aka., Fisher's exact test) for 2x2 tables with low expected counts. (The exception would be where all marginal totals are fixed in advance, where the Fisher-Irwin test is the one to use.) My more up-to-date thoughts about all of that can be seen here:
- https://sites.google.com/a/lakeheadu.ca/bweaver/Home/statistics/notes/chisqr_assumptions
Cheers,
Bruce
Good day everyone, I have a question:
Can I use ordinal regression, Crosstabulation, and Chi-square in Spss if I have only 25 samples/observations?
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS