[R] random effect question and glm

[Michael Dewey]

At 06:28 26/11/2006, you wrote:
>Below is the output for p5.random.p,p5.random.p1 and m0
>My question is
>in p5.random.p, variance for P is 5e-10.
>But in p5.random.p1,variance for P is 0.039293.
>Why they are so different?

Please do as the posting guide asks and reply to the list, not just 
the individual.
a - I might not know the answer
b - the discussion might help others

You give very brief details of the underlying problem so it is hard 
to know what information will help you most.

Remember, if a computer estimates a non-negative quantity as very 
small perhaps it is really zero.

I think you might benefit from reading Pinheiro and Bates, again see 
the posting list.

>Is that wrong to write Y~P+(1|P) if I consider P as random effect?

I suppose terminology differs but I would have said the model with 
Y~1+(1|P) was a random intercept model

>Also in p5.random.p and m0, it seems that there are little 
>difference in estimate for P. Why?
>
>thanks,
>
>Aimin Yan
>
> > p5.random.p <- 
> lmer(Y~P+(1|P),data=p5,family=binomial,control=list(usePQL=FALSE,msV=1))
> > summary(p5.random.p)
>Generalized linear mixed model fit using Laplace
>Formula: Y ~ P + (1 | P)
>    Data: p5
>  Family: binomial(logit link)
>   AIC  BIC logLik deviance
>  1423 1452 -705.4     1411
>Random effects:
>  Groups Name        Variance Std.Dev.
>  P      (Intercept) 5e-10    2.2361e-05
>number of obs: 1030, groups: P, 5
>
>Estimated scale (compare to  1 )  0.9999938
>
>Fixed effects:
>              Estimate Std. Error z value Pr(>|z|)
>(Intercept)  0.153404   0.160599  0.9552   0.3395
>P8ABP       -0.422636   0.202181 -2.0904   0.0366 *
>P8adh        0.009634   0.194826  0.0495   0.9606
>P9pap        0.108536   0.218875  0.4959   0.6200
>P9RNT        0.376122   0.271957  1.3830   0.1667
>---
>Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
>
>Correlation of Fixed Effects:
>       (Intr) P8ABP  P8adh  P9pap
>P8ABP -0.794
>P8adh -0.824  0.655
>P9pap -0.734  0.583  0.605
>P9RNT -0.591  0.469  0.487  0.433
> > p5.random.p1 <- 
> lmer(Y~1+(1|P),data=p5,family=binomial,control=list(usePQL=FALSE,msV=1))
> > summary(p5.random.p1)
>Generalized linear mixed model fit using Laplace
>Formula: Y ~ 1 + (1 | P)
>    Data: p5
>  Family: binomial(logit link)
>   AIC  BIC logLik deviance
>  1425 1435 -710.6     1421
>Random effects:
>  Groups Name        Variance Std.Dev.
>  P      (Intercept) 0.039293 0.19823
>number of obs: 1030, groups: P, 5
>
>Estimated scale (compare to  1 )  0.9984311
>
>Fixed effects:
>             Estimate Std. Error z value Pr(>|z|)
>(Intercept)   0.1362     0.1109   1.228    0.219
>
> > m0<-glm(Y~P,data=p5,family=binomial(logit))
> > summary(m0)
>
>Call:
>glm(formula = Y ~ P, family = binomial(logit), data = p5)
>
>Deviance Residuals:
>     Min       1Q   Median       3Q      Max
>-1.4086  -1.2476   0.9626   1.1088   1.2933
>
>Coefficients:
>              Estimate Std. Error z value Pr(>|z|)
>(Intercept)  0.154151   0.160604   0.960   0.3371
>P8ABP       -0.422415   0.202180  -2.089   0.0367 *
>P8adh        0.009355   0.194831   0.048   0.9617
>P9pap        0.108214   0.218881   0.494   0.6210
>P9RNT        0.374693   0.271945   1.378   0.1683
>---
>Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
>
>(Dispersion parameter for binomial family taken to be 1)
>
>     Null deviance: 1425.5  on 1029  degrees of freedom
>Residual deviance: 1410.8  on 1025  degrees of freedom
>AIC: 1420.8
>
>Number of Fisher Scoring iterations: 4
>
>
>At 06:13 AM 11/24/2006, you wrote:
>>At 21:52 23/11/2006, Aimin Yan wrote:
>>>consider p as random effect with 5 levels, what is difference between these
>>>two models?
>>>
>>>  > p5.random.p <- lmer(Y
>>>~p+(1|p),data=p5,family=binomial,control=list(usePQL=FALSE,msV=1))
>>>  > p5.random.p1 <- lmer(Y
>>>~1+(1|p),data=p5,family=binomial,control=list(usePQL=FALSE,msV=1))
>>
>>Well, try them and see. Then if you cannot understand the output tell us
>>a) what you found
>>b) how it differed from what you expected
>>
>>>in addtion, I try these two models, it seems they are same.
>>>what is the difference between these two model. Is this a cell means model?
>>>m00 <- glm(sc ~aa-1,data = p5)
>>>m000 <- glm(sc ~1+aa-1,data = p5)
>>
>>See above
>>
>>
>>>thanks,
>>>
>>>Aimin Yan
>>
>>Michael Dewey
>>http://www.aghmed.fsnet.co.uk
>
>
>
>
>
>--
>No virus found in this incoming message.
>Checked by AVG Free Edition.
>Version: 7.5.431 / Virus Database: 268.14.16/551 - Release Date: 
>25/11/2006 10:55

Michael Dewey
http://www.aghmed.fsnet.co.uk