Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • ANCOVA or MANCOVA? Help is MUCH appreciated

    Hi all,

    I am undertaking a thesis and it has been a long while since I've been at university (and using Stata!) and am looking for some clarification:

    My experiment is a within-subjects design whereby participants watch 4 consecutive videos (2 different conditions) and rate the videos on 3 measures after each video (Attitudes towards help-seeking, Trivialisation and Stigma - all continuous). Participants also undertake a big 5 measure of personality (which includes 5+1 continuous predictors)

    The experiment is really: 6x personality IV's x 2x condition IV levels x 3 DV measures.

    My plan was to run 3 separate ANCOVA's (as I am only really interested in 3 of the personality measures) controlling for gender and source credibility but am wondering if this is the best analysis to use.

    As a mature-aged student, this is confusing the heck out of me and for some reason I'm unable to wrap my head around the interaction syntax.

    Any help is greatly appreciated. Thanks in advance

  • #2
    Correct me if I don't have this right, but I assume that your research question has to do with exploring some kind of association between each of the three personality measures and each of the three rating types at each of the two conditions. Is that the interaction whose syntax is giving you trouble?

    Regardless, I don't recommend ANCOVA with Stata when you have these repeated measures (video-presentation conditions and types of attitude rating), primarily because if your research question is about the associations involving the personality measures, which will be between-subjects continuous covariate in the ANOVA model, the test statistics (if those are what you're going to use to judge the strength of association) will not be quite right. I suspect that the analogous MANCOVA model will have the same problem.

    So, with Stata, I'd recommend using mixed for this, with an unstructured covariance pattern for the residuals at the lowest level (study participant and session), which mimics a MANOVA (MANCOVA in your case).

    Comment


    • #3
      Originally posted by Joseph Coveney View Post
      Correct me if I don't have this right, but I assume that your research question has to do with exploring some kind of association between each of the three personality measures and each of the three rating types at each of the two conditions. Is that the interaction whose syntax is giving you trouble?

      Regardless, I don't recommend ANCOVA with Stata when you have these repeated measures (video-presentation conditions and types of attitude rating), primarily because if your research question is about the associations involving the personality measures, which will be between-subjects continuous covariate in the ANOVA model, the test statistics (if those are what you're going to use to judge the strength of association) will not be quite right. I suspect that the analogous MANCOVA model will have the same problem.

      So, with Stata, I'd recommend using mixed for this, with an unstructured covariance pattern for the residuals at the lowest level (study participant and session), which mimics a MANOVA (MANCOVA in your case).

      Hi Joseph,

      My research is exploring how different types of tiktok videos (videos made by medical professionals vs non-professionals) influence help-seeking behaviour, trivialisation and stigmatization of mental illnesses, and whether this relationship is moderated by personality variables.

      Each participant will be viewing both types of condition at the same time point (4 total videos (2 prof and 2 non prof)). It is technically a within-subjects, repeated measures design, correct?

      I have been advised against a MANCOVA just due to complexity but am wholeheartedly open to your suggestion.

      My plan was to look at:

      Main effects of type of video content on trivialisation
      Main effects of type of video content on help seeking behaviour
      Interaction between type of video content and trivialisation, which is hypothesized to be moderate by neuroticism.

      Could this not be 3 separate ancovas?

      Thanks again
      Last edited by Sam Dreyfus; 19 Apr 2023, 20:25.

      Comment


      • #4
        Originally posted by Sam Dreyfus View Post
        Could this not be 3 separate ancovas?
        Seems as if the first two could be paired Student's t-tests (although I'd be inclined to at least include dotplots), with professional-versus-nonprofessional as the grouping variable and rating score as the outcome variable.

        The third, "Interaction between type of video content and trivialisation", seems perplexing in that you're posing the research question as an interaction, not of two predictors (explanatory variables), but rather of a predictor and the response (an interaction of an explanatory variable and the outcome variable).

        Comment

        Working...
        X