Hello!
Maybe the community can help me with this question: I am preparing an analysis for a daily-diary study and most of the methodological literature as well as papers on study with similar designs apply a person-based centering before computing a MLM. Most of the time, authors specified that the predictors were centered: I was wondering if there is a rational on why only the predictors and not the outcome variable (also continuous) have to be centered in order to analyze changes on the daily level.
Thank you very much in advance,
Matteo
Hi Matteo,
if you center, do that only for the IV not the DV. But centering isn't necessary anyways, see
Antonakis, J., Bastardoz, N., & Rönkkö, M. (2019). On ignoring the random effects assumption in multilevel models: Review, critique, and recommendations. Organizational Research Methods, 24(2), 443–483. doi:10.1177/1094428119877457
Best,
Holger
By group mean centering and including the group means you are usually trying estimate both the between person and within person change; the standard RE estimate confounds, them see Article Fixed and Random effects models: making an informed choice
Hi Matteo Carmignola ,
first, let me post a quote from the Centering section in: Article Applications of multilevel modeling in psychological science...
"Keeping in mind Bryk and Raudenbush’s advice “That no single rule covers all cases,” we recommend the following, advice which is consist with that offered by Enders and Tofighi (2007). At level 1 continuous predictors should be entered group-mean centered, which makes the intercept similar to the original intercept (null model) which helps to maintain the stability of covariance matrices. Grand-mean centering at level 1 can be used when an analyst wants to adjust the level 1 intercept for level 2 differences in a predictor. At level 2 continuous predictors should be grand-mean centered. This makes the intercept the expected value for an observation at the mean of the level 2 predictor, which corresponds to the mean of the sample for the outcome and is a value that is easy to understand. Categorical predictors (discussed below) should be entered uncentered. This makes it easier to interpret results, i.e., what coefficients represent. A more detailed discussion of centering can be found in Nezlek (2011; pp. 13-18). "
To answer your question more directly - you could center the outcome but there is no need to, as it will only change the intercept and not the slopes.
- Badges
- Science topic