Survey research and design in psychology/Tutorials/ANOVA/ANCOVA


  View the accompanying screencasts: [1] [2]

In this scenario, the dependent variable of interest is the stress levels of university students in three courses (medicine, dentistry, podiatry). Students reported their stress levels at the beginning and the end of an academic year.

  1. Allen & Bennett: Chapter 10 (pp. 123-140)
    1. data_10_1.sav | sps
    2. data_10_2.sav | sps

The underlying research question is "Do stress levels differ between the three disciplines?" - or more precisely, "Are one or more of these disciplines more stressful than other(s)?".

Start off by conducting a one-way ANOVA using the end of year stress levels as the DV and course as the IV.

  1. Design
    1. DV = end stress
    2. IV = Course (medicine, dentistry, podiatry)

However, much of the variance in these stress levels is likely to be due to pre-existing (baseline) stress levels and it may be there are pre-existing differences in stress levels for the cohorts of students who choose medicine, podiatry, and dentistry. Therefore, an ANCOVA would be a preferred analysis (which would introduce beginning of year stress as a CV). In this way, variance associated with pre-existing stress levels (beginning of year stress) is removed from the DV (end of year stress). Any remaining differences in stress levels between the three groups are therefore not due to variations in pre-existing stress levels, but may instead be to do with the stressfulness of the course.

  1. Design
    1. DV = end stress
    2. CV = start stress
    3. IV = Course (medicine, dentistry, podiatry)
  2. SPSS - Analyze - General linear model - Univariate - first of all just run this as a t-test (but via ANOVA) without the covariate, to provide a baseline analysis (like running Step 1 of a MLR).
  3. SPSS - Analyze - General linear model - Univariate - enter DV, IV, and CV and choose options
  4. Note that the descriptives and graphs are not adjusted by the covariate so are misleading. The F test, p levels and effect size and power etc. are adjusted.
  5. Results indicate a significant difference in stress levels between the three groups with and without the covariate. However, the group differences are less strong in the ANCOVA.

See also edit