We did a survey asking about the behaviour of people concerning Single-Use Plastics before and during the COVID-19 pandemic. We have two set's of corresponding responses "before" (the outbreak of the pandemic) and "after" (the outbreak). Depending on the question there were from 3 to 6 possible responses. We gathered also demographic data of the respondents (age, place of residence, education).
An example pair of questions:
Before the outbreak: Were you bringing your own cup to coffee shop Yes/No/Occasionally;
After the outbreak: Were you bringing your own cup to the coffee shop Yes/No/Occasionally.
We want to statistically test:
a) whether there is a statistically significant difference of frequency of responses between corresponding questions (before the outbreak and after the outbreak)
b) whether the change of frequencies of responses is significantly different in different demographical groups
Questions:
1. Can we use crosstabulation in SPSS, applying McNemar-Bowker's test to test the difference of frequencies (before/after)?
2. Can we do the same adding age/education etc. as a layer to check if differences among groups are significant?
If not, anyone has any suggestion how to conduct the analysis? I prefer to do it in SPSS (or if it is not possible in R)
Hello Augustyn,
If your analysis is question by question, and you simply want to know how consistent the responses are using the self-report of before/after as the dimension of interest, consider trying Cohen's kappa index. It's pretty simple, is available in SPSS (via the descriptives/crosstabs subprogram), and has a very useful interpretation: The degree of agreement beyond that expected by chance.
If you're more interested in how much explanatory power variables such as age, locale, etc. are with respect to whether judgments stay the same (=1) or differ for a case before/after (=0), you could use logistic regression. Be sure to include interaction terms for your IVs of interest unless you have a good rationale for excluding them from your model.
Good luck with your work.
Yes you can McNemar test for paired variables of two possible (yes/ no) responses to see a before-after change. It has been used it for seeing change in diabetic foot care before and after health education (Article Effectiveness of foot care education among people with type ...
). Similarly Bowker's test will do multiple McNemar test if you are seeing before-after change for 3 possible outcomes (yes/ no/ occasionally) type of responses. Although response options for such questions could be framed better as Frequently or always / Occasionally/ Hardly or never.However, i think using it for age/education is not appropriate. Because you will have to categorize the continuous variables and it would result in loss of information.
You can use logistic regression with response to question converted to binary (such as willing to bring your own cup - yes/ no or occasionally) as independent variable and before/after as dependent variables. Of course, age category and gender can be used as layers (or levels) using multilevel modeling approach, to see whether the behavioral response to outbreak is different among genders and young/old individuals.
Hello Augustyn Mikos,
Just in case you still want to use McNemar-Bowker's test in SPSS, you might find helpful the link below from IBM troubleshooting support.
https://www.ibm.com/support/pages/bowkers-test-symmetry-spss
Good luck with your project.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS