1

I have a very basic question; maybe a bit too basic to find a helpful response googeling it.

I am calcultating multi-level-models using the lmer function using this code:

lmer(H1_rirs, data= df_long_cl, REML = T)

Am I right in assuming that the retrieved coefficients are unstandardized? If yes, is there an easy way to standardize them?

Best,

Carolin

Carolin
  • 23
  • 4
  • How is `H1_rirs` defined? – Roland Dec 04 '19 at 09:29
  • Hi Roland, it is defined like this: `H1_rirs<-commresc ~ 1 + cvar1c + (1 + cvar1c | ID_new) ` both variables are person-mean-centred, although I guess I should not mean-centre my outcome (it functions as a mediator in the bigger model) – Carolin Dec 04 '19 at 10:11

2 Answers2

1

Yes, by default, no standardizing is applied. If you like to get standardized coefficients, one way would be to standardize the data before fitting your model. There is a robust implementation of such a function in the effectsize-package. Or you can do some post-hoc standardization (also effectsize-package). The latter yields different results, the most accurate would be standardizing the data before model fitting.

Daniel
  • 7,252
  • 6
  • 26
  • 38
0

Adding to what @daniel suggested with the effectsize package, there is also the "pseudo-standardized" coefficient (Hoffman, 2015) where the response and the predictor are standardized based on the level of prediction:

library(lme4)

m <- lmer(Reaction ~ Days + (Days | Subject), data = sleepstudy)

effectsize::standardize_parameters(m, method = "pseudo")
#> # Standardization method: pseudo
#> 
#> Parameter   | Coefficient (std.) |       95% CI
#> -----------------------------------------------
#> (Intercept) |               0.00 | [0.00, 0.00]
#> Days        |               0.68 | [0.48, 0.88]

Created on 2021-06-07 by the reprex package (v2.0.0)

MSB
  • 86
  • 4