Nominal data and before-after significance

#1
For my Master's thesis I asked two different groups (experimental and control) to name the most important cause of homelessness (7 choices - drug/alcohol addiction, mental illness, lack of affordable housing, family conflict, own choice, eviction/foreclosure, unemployment). Then after the experiment, the two groups were asked the same question.

The results show a difference in the number of people who ranked each cause the number one cause, but I'm uncertain as to whether the difference is significant. What statistical test is best to determine if the within and between group differences are significant?

Thanks in advance for any help - I haven't been able to find anything relating to this in any stats book.
 
#2
Hi,

I am not a statistician but I hope this can help you. If I am not mistaken, you can use any of the following tests to check for differences in populations:

- Median test.
- Mann-Whitney U test.
- Kolmogorov-Smirnov test

I think you are dealing with non-parametric tests and ordinal data (ranking data are considered ordinal data) Anyway, I think it would be better for you to wait for some other hints before cracking on with the analysis.

Best of luck!
 
#5
On second thought...

Thanks, but the data isn't ordinal and it's not independent. It's nominal and the two sets of data are dependent in that the results are from a before and after test. Yes, the respondents were asked to rate which was the most important cause, but the various categories given are not themselves ranked (you can't really say mental illness is "higher" or "lower" than "unemployment"). I have a gut feeling that the results are significant (those saying drug/alcohol abuse was the primary cause of homelessness rose from 23.5 to 35.3 while those who said mental illness was the primary cause dropped from 26.5 to 8.8 percent), but I need a test that'll show that. Sorry if my first post didn't make that clear. I really appreciate any help on this.


If I am not mistaken, you can use any of the following tests to check for differences in populations:

- Median test.
- Mann-Whitney U test.
- Kolmogorov-Smirnov test

I think you are dealing with non-parametric tests and ordinal data (ranking data are considered ordinal data) Anyway, I think it would be better for you to wait for some other hints before cracking on with the analysis.

Best of luck!
 
#6
Hey Westcoast!!!

Sorry about that. I really thought the data would be ordinal. If the data are nominal and the samples are related you can use 'McNemar test'. I think it would be great if you could have a look at this book:

'Nonparametric statistics for the behavioural science'.
Author: Sidney Siegel.

I have found the book in plenty of libraries and it is probably at your university. At the end of the book it specifies the kinds of tests you can carry out depending on the type of data you have.

I hope it helps.
Best of luck.