Hello,
I'm hoping to add some quantitative analysis to a thematic part of a larger study. I have not done any quantitative work for 10+ years so I'm a little rusty. I also don't have access to SPSS so am hoping for something workable in excel.
I have two groups of participants (n=32, n=18) who have each responded to survey and interview questions which have resulted in the emergence of a series of themes. I would like to analyse whether certain characteristics of the participants (e.g. gender, sexuality) make them more likely to give responses that map to certain themes. Is a chi-test the most appropriate / achievable approach to do so?
Thanks,
Sarah
Morning,
Try to use ANOVA. If you expect some interaction among your predictors, use "full factorial". Otherwise define your own model with direct effects only.
Best,
Eugene
Thanks for your response Eugene, I'll look into ANOVA. Can you please clarify direct effects only? Do you mean take aspects from the various statistical tests that best suit my application?
Imagine that you have 2 predictors (gender, position) and 1 outcome variable, eg. satisfaction. They are nominal/categorical. In SPSS you go to Analyze->General linear model -> Univariate. (if you have several outcome variables -> Mutlivariate). You add gender and position as "factors". (would they be ordinal or interval, you could add them as "covariates"). Above on the right of the window you see the button "Model". By default "full factorial" is set. It means, that SPSS will make following tests:
gender -> satisfaction
position -> satisfaction
gender*position -> satisfaction.
If you have more predictors, the number of tests will grow exponentially, since SPSS will try to calculate all possible moderations. For example, imagine that you have 1 more predictor - "country". Then the tests will look like this:
gender -> satisfaction
position -> satisfaction
country -> satisfaction
gender*position -> satisfaction.
gender*coutnry -> satisfaction
country*position -> satisfaction
gender*country*position -> satisfaction
So depending on your theory, you can also define your own model, where you test direct effects only, or in the event of too many predictors, only moderation effects which are of interest for your theory. For example, only:
gender -> satisfaction
position -> satisfaction
country -> satisfaction
gender*position -> satisfaction.
I hope it helped
Best,
Eugene
In Eugene's example, if your coding is whether or not "satisfaction" was mentioned (0/1), then a chi-sq test would be appropriate due to the use of two nominal-level variables. If you want to look at the frequency that each person mentioned satisfaction, then you would have an interval-level dependent variable, so ANOVA would be appropriate.
Either way, doing this kind of statistical analysis on qualitative data is controversial. From a quantitative point of view, this is due to the use of small, non-random samples. From a qualitative point of view, this is due to a departure from the goal of seeking an interpretation of meaning.
My own position lies somewhere in between both of those positions, since I think that examining basic patterns in qualitative data can be worthwhile, if it leads you to reanalyze the qualitative data to ask how and why the patterns arose. That does, however, raise the issue of how you recognize a pattern that is strong enough to be worth pursuing, and a statistical test might be one way to make such a choice.
Here is an article that I wrote on the subject some years ago:
Article Qualitative Content Analysis: A Guide to Paths Not Taken
Valuable suggestions have already been given by David so I would rather focus on second part of your question where you have asked for alternative of SPSS. Here is a wonderful online chi square calculator that I have used personally and found very easy-to-use and handy when having upto 10x10 table:
http://www.quantpsy.org/chisq/chisq.htm
It really depends on what detail you have in your response scale of your questionnaire. If you have a binary response are you satisfied yes no are you can do a Chi Square test to identify whether more women are more satisfied. If you have a response scale that has 3 or more levels then you may need something more advanced i.e. like a t test and if you want to look at interactions between gender, position and your dependent variables then you need to move to a more complex multivariate test. The type of test you chose depends on the question you want to interrogate. If you could outline the hypotheses first that you want to test then it would help you to identify the method by which you can determine whether it is due to chance or not.
I quite agree with the answers above and want to add that besides the Chi-square, Fishers exact test would suffice given the non parametric nature of the generated data and and sample size.
You can try it.
GOOD LUCK FROM CHRIS.
When vairables level of measurment are nominal, then chi square will be appropriate.
It is not parametric, the data could be skewed, sample size is small. Thus, used it when the parametric test cant be performed.
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology