Dear Julia,
Perhaps this is helpful too:
"The assessment of a GLMM is notoriously difficult (e.g., Gelman and
Hill, 2007); here the optimal GLMM was compared against a null model
containing only listener group and speaking style as fixed predictors,
excluding phonological features and acoustic measures in its fixed part.
The standardized residual errors of the optimal GLMM are indeed smaller
than those of that null model, for nearly all talkers in both speaking
styles, as illustrated in Fig. 7. The optimal model predicts the
listeners’ responses significantly better than the null model
(likelihood ratio test, chi2 =4711, df =22, p < 0.0001), although the
relative reduction of standardized residual error (an evaluation measure
somewhat comparable to the proportion of variance explained) is only
small at 0.052 relative to the null model."
Cited from:
Ferguson, S.H. & Quené, H. (2014). Acoustic correlates of vowel
intelligibility in clear and conversational speech for young
normal-hearing and elderly hearing-impaired listeners. Journal of the
Acoustical Society of America, 135 (6), 3570-3584.
[http://dx.doi.org/10.1121/1.4874596
<http://scitation.aip.org/content/asa/journal/jasa/135/6/10.1121/1.4874596>,
PMCID:PMC4048446 <http://www.ncbi.nlm.nih.gov/pubmed/?term=24907820>]
See also:
Gelman, A., & Hill, J. (2007). Data Analysis Using Regression and
Multilevel/Hierarchical Models. Cambridge: Cambridge University Press.
Snijders, T., & Bosker, R. (1999). Multilevel Analysis: An introduction
to basic and advanced multilevel modeling. London: Sage.
The latter book (first ed, 1999) has a good explanation of the "relative
reduction of standardized residual error" mentioned in the quote, but I
don't have my copy at hand to check chapter or page numbers.
Hope this helps! Best, Hugo
Alex Fine <mailto:abf...@gmail.com>
26 January 2016 at 16:56
Hey Julia,
You might find Florian Jaeger's blog post from a few years ago
helpful:
https://hlplab.wordpress.com/2009/08/29/nagelkerke-and-coxsnell-pseudo-r2-for-mixed-logit-models/
Here is more discussion with tons of references:
http://statisticalhorizons.com/r2logistic
Seems like the best you can do is pick a measure that sort of makes
sense and explain why it only sort of makes sense.
Alex
--
Alex Fine
Ph. (336) 302-3251
web: http://abfine.github.io/
Julia Strand <mailto:jstr...@carleton.edu>
26 January 2016 at 16:26
Hi Ling-R,
I'm creating a series of models using lme4 and a categorical DV to
model the effect of various lexical variables on word recognition
accuracy, and then comparing those models using the anova() function.
A reviewer has asked me to include more discussion of how good the
overall fit of the models are of the data. That is, do the models
account for the vast majority or only a small fraction of the total
variance in word recognition accuracy? I know the challenges of trying
to use pseudo R^2 with categorical DVs and am not trying to explain
how much variance each predictor accounts for, rather, I'm looking for
a way to contextualize the absolute explanatory power of the full model.
Any thoughts?
Thanks,
Julia
Julia Strand, PhD
Assistant Professor of Psychology
Carleton College
One North College Street
Northfield, Minnesota 55057
507-222-5637 (office)
Schedule an appointment <http://juliastrand.youcanbook.me>
--
--
Dr Hugo Quené | onderwijsdirecteur Undergraduate School | Dept Talen Literatuur
en Communicatie | Utrecht inst of Linguistics OTS | Universiteit Utrecht |
Trans 10 | kamer1.43 | 3512 JK Utrecht | The Netherlands |+31 30 253 6070
|h.qu...@uu.nl |www.uu.nl/gw/medewerkers/HQuene |www.hugoquene.nl |
uu.academia.edu/HugoQuene