intra-class correlation coeff
I saw Andrew Tyne's reply and agree with what he said. Another way of approaching this is to realize that you only have 3 distinct levels of market. That's in the area where you probably are better off modeling it as a fixed-effects term, rather than a random-effects term. It may make sense logically to regard it as a random effect but practically it is difficult to estimate a variance from only a few observations. We can get an estimate but often it has very poor precision. I would suggest modeling it as a fixed effect and seeing if the term is significant there.
On Fri, Jan 23, 2009 at 4:33 AM, Metras, Raphaelle <rmetras at rvc.ac.uk> wrote:
Hello,
I am a very beginner with R and mixed-models, so please apologize if you
think my questions are naive.
I am fitting a glmer Poisson, with one variable as random effect
(market) and 2 variables as fixed effects.
My observations are clustered markets, there are 3 markets.
When looking at the variance of the random effect, and it is close to
zero (0.07484).
I would like to know if it is possible to extract the intra-class
correlation coefficient somehow, or if knowing the between market
variance (0.07484) is enough to say that there is almost no clustering.
Thank you very much, I copy the ouput below:
Generalized linear mixed model fit by the Laplace approximation
Formula: clear_bsk ~ dist_mkt + same_trader + offset(log(no_bsk)) + (1 |
market)
Data: essai
AIC BIC logLik deviance
55.9 63.39 -23.95 47.91
Random effects:
Groups Name Variance Std.Dev.
market (Intercept) 0.07484 0.27357
Number of obs: 48, groups: market, 3
Fixed effects:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -1.34246 0.32716 -4.103 4.07e-05 ***
dist_mkt -0.02948 0.01380 -2.137 0.032639 *
same_traderY 0.99278 0.27366 3.628 0.000286 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Correlation of Fixed Effects:
(Intr) dst_mk
dist_mkt -0.546
same_tradrY -0.656 0.282
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models