0
$\begingroup$

I am helping to review a manuscript and suffering from mental block on some data analysis. The data in question is from a cross-sectional survey, a health care provider KAP (knowledge, attitudes, and practices) on vector borne diseases (Zika, Chikungunya, Dengue, and West Nile Virus).

The question I am stuck on is formatted like so:

An image of the survey question that I am referencing, for clarity. image of question

Which of the following do you associate with illness caused by each virus? Check all that apply.

                     Dengue   Chikun    Zika    WNV    Don't know
Rash          
Fever
Cough
Arthralgia
Conjunctivitis
Myalgia
Retro-orbital pain
Headache
Hemorrhage
Altered mental status

For each disease and symptom, the author has coded a 1 for correct response, and 0 for incorrect response. Not all symptoms present with each disease, and thus leaving the check box empty is considered the correct response in these cases. The primary author is then computing a total correct for each participant (out of 10 possible correctly identified symptoms for each disease). Using the total correct scores, the author is running within-subjects ANOVA on mean correct scores for each disease group in order to identify any differences in levels of knowledge among the different diseases, and then running post-hoc paired t-tests.

In terms of hypothesis, it would be structured like this:

Ho: There are no significant differences in the average number of correctly identified symptoms/illnesses between selected vector borne diseases.

Ha: There is a significant difference in the average number of correctly identified symptoms/illnesses between selected vector borne diseases.

Post-hoc: If there are sig. difference, which diseases have significantly different scores?

I am hoping for your advice on how to best analyze this data as I am quite stumped. My initial thoughts were to code each response (0 - incorrect, 1 - correct) and create contingency tables for each symptom, and analyze with chi-square tests. Although I'm not sure how I would then locate potential differences between each disease.

I should note that the goal is to determine if any differences exist between diseases, not symptoms. Whether it is cumulative or individual to each symptom is not specified.

Is the author correct in running a within-subjects ANOVA, and simply needs to adjust to the proper post hoc test (Tukey?) to identify the differences in diseases?

I should also note that the author is using SPSS for analysis.

Unfortunately, I don't have access to the data at this time to run anything on my end. Any input and feedback is greatly appreciated!

Improve this question
$\endgroup$
2
  • $\begingroup$ With no clue what the authors' purpose is and no idea what data are going into their ANOVA it is difficult for me to imagine helpful advice coming from this question. For now, I'm voting 'unclear what you're asking' and hoping for some clarification. $\endgroup$
    – BruceET
    Commented Jul 29, 2019 at 22:32
  • $\begingroup$ Thanks for the response @BruceET. I added an image of the question from the survey that should hopefully provide some clarity on the data received (I couldn't think of a good way to format a check box/bubble). I have also added in more detail on the author's purpose, as well as a null and alt hypothesis. If you are able to revisit, that would be awesome. Thanks again. $\endgroup$
    – Theodore Efthemeou
    Commented Jul 29, 2019 at 23:21

1 Answer 1

Reset to default
0
$\begingroup$

If I am understanding this correctly, the goal is to compare the degree of knowledge of different diseases. Rather than ANOVA, I would use a count regression model, as each disease can only get integer values from 0 to 10 - with, hopefully, no 0s.

You could try Poisson regression to start, but negative binomial might be needed.

The "check all that apply" issue is removed by using the total correct.

Improve this answer
$\endgroup$
2
  • 1
    $\begingroup$ Thank you for the response Peter. I can see that count regression would be useful here, though I hadn't thought of it previously. I will suggest to the author, but I fear that modeling may be beyond the scope of his paper and statistical ability (health promotion student). $\endgroup$
    – Theodore Efthemeou
    Commented Jul 30, 2019 at 18:26
  • $\begingroup$ Yeah ..... that happens a lot. He might have to hire a consultant, but he probably won't want to. $\endgroup$
    – Peter Flom
    Commented Jul 30, 2019 at 20:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.