Hi there,
I have a confusion regarding the appropriate test I should choose to study media consumption pattern between two populations.
I would like to compare people from Punjab and Kerala in terms of their reliance on newspapers for COVID-19 related information.
One variable is state and it has two options.
The second variable is a rating on a 1 to 5 agreement scale for the following statement. The statement is "I read newspaper to get COVID-19 related information"
I think I could use Chi-square test for homogeneity or Chi-square test for association.
But there is a problem. The distribution of the Age of respondents is very different in the samples from Kerala and Punjab. The sample from Kerala is mostly young. So, I think this may create a problem in the analysis. The difference in the sample may be actually due to Age not the difference in State. But we are not sure.
In this case, is it appropriate to perform an ordinal logistic regression with both Age and State as predictors and Rating as the dependent variable?
Or should I just do the chi-square test by ignoring Age?
Or is there a better way to solve this issue?
Thanks in advance.
NB:
Sampling: The questionnaire was created using Google Form and the link was distributed through various social media groups.
Are the age distributions in your samples similar to the distributions in the populations of the two places (e.g., are people in Kerala younger?). Were the sample techniques you used for each area the same, and what were they? Were they drawn from existing lists? Were the response rates similar?
If these were conveniences samples and the age difference in the samples is just an artefact of the sampling being different, then there is likely no statistical procedure available (i.e., stats aren't magic). So, more details are necessary before anyone should answer. (Though likely people will throw answer out).
Thank you Daniel Wright for your answer.
The questionnaire was created using Google Form and the link was distributed through various social media groups. The sampling was done in both area at the same time.
The actual age distributions in both populations are more or less the same. The difference in the sample could be due to the sampling procedure. Due to our sampling technique, I believe our population is now only social media users in both states, not the entire people.
You told that If the age difference between the samples are due to problems with the sampling, then there is no statistical analysis can be done with that. I have a doubt in that. Is it possible to ensure that our two samples are similar in all dimensions? There will be hundreds of things we need to consider. Like education level (or even the literacy level), gender, internet penetration in the state, etc.
Another doubt is can't we use Age as a control variable here?
Or if do a Logistic regression with Age and State as the predictors, can't we know the effect of both Age and State separately?
There are some options that you can select based on your data
1. Two-way ANOVA and using the plots of effect size
2. Chi square for the age ranges separately and integrately
3. Fischer's exact test for each age range
Finally, I suggest you compare these three methods and the others recommended by respected professors and select the best based on presumptions. You can use more than one test to support the results.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS