Raw or standard score for longitudinal analyses of measures like CBCL scales?

There are many psychometric measures in ABCD like Child Behavior Checklist (CBCL), Adult Self-Report (ASR), etc. that include 2 versions of the scores: raw and standardized (e.g. t-score). For statistical analyses, we could either

  1. Use raw score and include age, sex as covariates
  2. Use standardized score.

Which is correct? In particular, when we are interested in longitudinal change over age?

I found studies using/recommending both #1 and #2. For example,

  • Achenbach (original developer of CBCL) recommends raw score for research unless sex is not controlled for or age difference is non-trivial.
  • King et al. (2018) says standardized score better reflects magnitude of change across all ages.
  • Barch et al. (2021) uses raw score rather than standardized score to better study developmental and sex differences.
  • Most, but not all, studies using ABCD data seem to use standardized score (t-score).

If your interest is in longitudinal change, I think raw scores are generally preferable. Since the t-scores are age-normed, they are less ideal for examining development. We have some forthcoming work examining trajectories of CBCL change in ABCD and we used raw scores. We also used raw scores in this study (not ABCD sample):

Brieant, A., King-Casas, B., & Kim-Spoon, J. (2022). Transactional relations between developmental trajectories of executive functioning and internalizing and externalizing symptomatology in adolescence. Development and Psychopathology , 34, 213-224.


I spoke to a few ABCD folks about your question and the consensus is to go with your option #1 and use the raw scores and include sex and age in your regression models. One colleague noted that raw scores are preferred bc ABCD might be a larger and more representative sample than was used to adjust for sex and age to generate the standardized scores. Hope this helps!

1 Like

That’s a very good point about ABCD probably being larger than the original sample for many metrics! Thank you for answering this question.