11.11 Closing Thoughts

  • Interpretations of change are fundamentally restricted by the choice of model
  • The residualized change and difference score models will often lead to different inferences about the effect of the predictor of interest.
  • The assumptions each of these models make are untestable, and this makes it challenging to argue that either model is more appropriate than the other.
    • Dyads who report cohabiting prior to marriage will never be able to report what their relationship satisfaction after marriage would have been if they had not cohabited.
    • We will never know what level of relationship satisfaction would have been reported by dyads who did not cohabit before marriage if they had in fact chosen to live with their partners prior to getting married.

11.11.0.1 Random Assignment

  • When experimentatal procedures are followed, dyads would be randomly assigned to a grouping variable, and this would bring independence of the grouping variable with the pre-test scores.
  • Thus, proper experimentation prohibits preexisting groups to be formed at baseline.
  • Under such circumstances, the residualized change model and difference score models will arrive at the same inference, but the former has the advantage of having more power because it makes the correct assumption about the distribution of baseline scores.

11.11.0.2 Observational Studies

Which should be our choice when experimentation is not possible?

  • If we have different populations with different levels of the outcome at baseline, the residualized change model may not be appropriate.
  • For researchers going this route, a latent change score model may alleviate some concerns about reliability and measurement error.

11.11.0.3 Two-Occassion Change

If you found all of this to be inherently confusing it is worth thinking about the following quote: “Two waves of data are better than one, but maybe not much better.” (Rogosa, Brandt, and Zimowski 1982).

References

Rogosa, David, David Brandt, and Michele Zimowski. 1982. “A Growth Curve Approach to the Measurement of Change.” Psychological Bulletin 92 (3): 726–48. https://doi.org/10.1037/0033-2909.92.3.726.