[R] significance of random effect in mgcv gam

[Simon Wood]

 > Question. Am I correct that p = .126 above can be taken as the
 > p-value  associated with the random effect?

- Yes. See

http://biomet.oxfordjournals.org/content/100/4/1005.abstract

for details of the approximate test used.


On 03/12/13 20:09, William Shadish wrote:
> Dear R-helpers,
>
> I would like to test whether a random effect is significant when
> implemented with bs="re" in mgcv gam. For example, if I run:
>
> M3b <- gam(DVY  ~ s(SessIDX, fTX, bs = "re") + factor(TX),
>             data = PCP,
>             family = quasipoisson(link="log"), method="REML")
> summary(M3b,all.p=TRUE)
> gam.vcomp(M3b)
>
> I obtain the the following output:
>
>  > summary(M3b,all.p=TRUE)
>
> Family: quasipoisson
> Link function: log
>
> Formula:
> DVY ~ s(SessIDX, fTX, bs = "re") + factor(TX)
>
> Parametric coefficients:
>              Estimate Std. Error t value Pr(>|t|)
> (Intercept)   1.3282     0.2244   5.920 2.74e-07 ***
> factor(TX)1  -1.0546     0.7210  -1.463     0.15
> ---
> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
>
> Approximate significance of smooth terms:
>                   edf Ref.df     F p-value
> s(SessIDX,fTX) 1.052      2 1.138   0.126
>
> R-sq.(adj) =  0.388   Deviance explained = 39.5%
> REML score =  37.67  Scale est. = 1.4172    n = 54
>  > gam.vcomp(M3b)
>
> Standard deviations and 0.95 confidence intervals:
>
>                    std.dev      lower     upper
> s(SessIDX,fTX) 0.07842742 0.01095655 0.5613865
> scale          1.19029872 0.97816911 1.4484316
>
> Rank: 2/2
>
> Question. Am I correct that p = .126 above can be taken as the p-value
> associated with the random effect?
>
> Thanks.
>
> Will Shadish
>


-- 
Simon Wood, Mathematical Science, University of Bath BA2 7AY UK
+44 (0)1225 386603               http://people.bath.ac.uk/sw283