Hi,
Is there a way to provide starting values for the fixed effects in
lmer()? I'd like to fit the following model, which requires starting
values in the glm.fit() part of the code:
lmer(dbh.sum ~ Treatment + (1|Site), nets, gaussian("log"),
subset=Treatment!="sforest" & iocTreat!="forest")
I tried tinkering with the code but I couldn't figure out the
namespace problems I ran into. I also saw an earlier R-help post with
a reply mentioning a possible lmer-initialize function that I couldn't
find. I can fit the model without the random effect using glm:
Call:
glm(formula = dbh.sum ~ Treatment, family = gaussian("log"),
data = nets, subset = Treatment != "sforest" & iocTreat !
"forest", start = c(5, 5, 5))
Deviance Residuals:
Min 1Q Median 3Q Max
-270.55 -72.32 -36.25 40.92 793.02
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 5.87873 0.04536 129.603 < 2e-16 ***
Treatmentioc -1.82540 0.40063 -4.556 9.83e-06 ***
Treatmentshade -0.88166 0.11856 -7.436 4.69e-12 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05
'.' 0.1 ' ' 1
(Dispersion parameter for gaussian family taken to be 18392.07)
Null deviance: 5763318 on 174 degrees of freedom
Residual deviance: 3163431 on 172 degrees of freedom
AIC: 2220.0
Number of Fisher Scoring iterations: 10
Also, are there plans to create a predict function for lmer? Thanks.
Richard
--
Richard Chandler, PhD student
Department of Natural Resources Conservation
UMass Amherst
(413)545-1237