I would like to examine the association between two survey questions asked to doctors. The first question asks whether they screened their patients for social needs in the past year (e.g., financial or housing needs for example). Doctors responded on a Likert-type scale of 0 (Never) to 5 (Yes, always). The second question asked whether the doctors believe social needs are important to their patient's health. This question was anchored on a scale of 0 (Disagree), 1 (Neutral), and 2 (Agree).
37 doctors were sampled. For the first question, only 2 or 3 doctors endorsed "Never" and "Yes, Rarely". For the second question, no doctor endorsed "Disagree"...and most doctors endorsed Agree.
While folks have recommended the Fisher Exact test, I was hesitant to use this method because the variables are ordinal (not nominal). I also contemplated examining the Spearman Rho or Kendall correlation, but I wasn't sure if the zero counts in some categories would bias (and resultantly, invalidate) the result.
I would appreciate any insights into the matter and thank you in advance.
Hello Dan,
From the description of your second variable, it sounds as if there won't be much of a relationship since there was so little variability. Both variables are ordinal strength; you can certainly use the Spearman rho.
Good luck with your work.
Hi, David Morse and everyone- I just wanted to hear what folks think about an ordinal regression modeling approach (ordered logit model). My concern, with this approach, is the zero cell counts (esp when you do cross-tabs between the survey items) and the very small cells (< 5). But I wanted to make sure I was covering all alternative approaches before running with the Spearman correlation. Thanks again!
- Badges
- Science topic
- Similar topics
- Condensed Matter Physics
- Zero