I have an odd scenario I could use help with:

  • I was brought in to help with a study (after it was designed) that has the following characteristics:
  • People were given survey that deals with their current depression (has been validated elsewhere), after taking the survey you end up with a score out of 21 (higher score is more depressed).
  • They were also asked various other questions that might pertain to whether they are feeling that way, how many hours they work, how many close friends they have, etc.. mostly ordinal type data, some categorical and some numerical. The questions were asked in such a way that it referred to what has happened over the past few weeks.
  • We'd like to know if there is a relationship between those questions (hours they worked, what their job is like, do they have friends, etc..) and how depressed they are (we think there is)
  • So I'm thinking regression... but the only slight catch is that they were also asked a question PRIOR to these weeks about how they were feeling. Not the exact same tool as the current survey, but kind of a simpler "how depressed are you" type question.
  • So I'm assuming (maybe I shouldn't) that people who said they were more depressed before everything, would end up being more depressed after everything also (or at least it would affect their scores), and that should be taken into account when picking the analyses.

Thoughts? Also, I'll be using R for the analyses, and if you happen to know of any good examples in R that'd be great.

Thank you

  • $\begingroup$ Sounds like you will want to use an ordinal regression method treating their depression score as the outcome. If you think that it is reasonable to treat their score out of 21 as a continuous variable, you could use linear regression instead. Include the factors that you want to adjust for as covariates in your regression model. You can include their past depression score as a covariate, similar to ANCOVA. So essentially, your model might look something like... DEP. SCORE = PAST DEP. METRIC + HOURS WORKED + JOB TYPE. In R, mod = lm(score ~ prevscore + hourswork + jobtype); summary(mod) $\endgroup$
    – Emma Jean
    Jun 3, 2020 at 15:46
  • $\begingroup$ @EmmaJean Thank you. I guess i may have been overthinking it. So you're saying just treat that "pre" survey as just another covariate? Sounds good to me, especially given that I may have just learned from the other people involved that the "pre" question was actually asked after the fact. Like... "think back and tell us how depressed you were". $\endgroup$ Jun 3, 2020 at 16:00
  • $\begingroup$ I don't see why you couldn't include it as a covariate. If you had instead asked if you could calculate a change score that would be questionable but I think adjusting for an additional depression metric as a covariate makes sense, particularly if all subjects were asked. It would provide some information about their mental state in the past which may impact their current depression score despite the fact it may be on a different scale. $\endgroup$
    – Emma Jean
    Jun 3, 2020 at 16:02


Your Answer

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

Browse other questions tagged or ask your own question.