[R-sig-ME] working with log-transformed data

Austin Frank austin.frank at gmail.com
Fri Mar 23 19:59:22 CET 2007


If this question belongs on another list like R-help, please let me
know.  Thanks in advance for any responses.

I'm fitting a model of the form

m <- lmer( log(latency) ~ factor1 + factor2 +
                          cov1 + cov2 +
                          (1 | subject) + (1 | item) )

lmer reports a coefficient for factor1level1 of .04 and for
factor2level1 of -.10.  I'd like to convert these coefficients back
into the original units of the dependent variable, milliseconds.  Is
the following transformation appropriate for showing the expected
change in the dependent variable in its original units for each of the
fixed effects?

sapply(fixef(m), function (x) ((exp(x) - 1) * mean(latency)))

I also wondered whether I should use different transformations for the
balanced factors and the continuous covariates.  Do you think this is
an issue?  If so, any suggestions on an appropriate conversion for the

Thanks for your time,

Austin Frank
GPG Public Key (D7398C2F): http://aufrank.net/personal.asc

More information about the R-sig-mixed-models mailing list