[R-sig-ME] Significance of fixed effects. Kinship package (Marc Moragues)

[Pelle Ingvarsson]

Hi,

When I use the kinship package I fit two models, with and without the 
fixed effect you want to test. The compare the log-likelihoods of the 
two models, twice the difference in log-likelihood between models should 
be apprixmately chi-square with df equal to the difference in number of 
parameters in the two models.

So using your example:

aa<-lmekin(dta1[,j] ~ dta1[,k] + g1:g2:g3:g4,data = dta1, random = 
~1|geno, varlist = list(K), subset = Year==i)

aa2<-lmekin(dta1[,j] ~ g1:g2:g3:g4,data = dta1, random = ~1|geno, 
varlist = list(K), subset = Year==i)

X2<-2*(logLik(aa)-logLik(aa2))
df<-aa$df-aa2$df
p<-1-pchisq(X2,df)

would give you a chi-square value and the associated degrees of freedom. 
The p-value of the effect can then be calculated using pchisq.

-Pelle



> Message: 1
> Date: Mon, 9 Feb 2009 14:02:33 -0700
> From: Marc Moragues <marc.moragues at gmail.com>
> Subject: [R-sig-ME] Significance of fixed effects. Kinship package
> To: R Mixed Models <R-sig-mixed-models at r-project.org>
> Message-ID: <200902091402.33547.moragues at lamar.colostate.edu>
> Content-Type: text/plain
> 
> Hi,
> 
> Some time ago, I was pointed to use the kinship package to include the 
> variance/co-variance in a mixed model. My code is as follows and works well 
> (it does not give any error).
> 
>>     aa <- lmekin(dta1[,j] ~ dta1[,k] + g1:g2:g3:g4,data = dta1, random = ~ 
> 1|geno, varlist = list(K), subset = Year==i)
>> aa
> Linear mixed-effects kinship model fit by maximum likelihood
>   Data: dta1 
>   Subset: Year == i 
>   Log-likelihood = -581.048 
>   n= 192 
> 
> Fixed effects: dta1[, j] ~ dta1[, k] + g1:g2:g3:g4 
>                 Estimate  Std. Error    t value      Pr(>|t|)
> (Intercept)   80.5622757    1.468117 54.8745492 1.185858e-117
> dta1[, k]1    -0.5091199    1.448478 -0.3514860  7.256174e-01
> dta1[, k]na   -7.8414997    2.202464 -3.5603297  4.691413e-04
> g1:g2:g3:g4 -657.2792180 1337.706920 -0.4913477  6.237537e-01
> 
> Wald test of fixed effects =  18.29943 df =  3 p =  0.0003815272
> 
> Random effects: ~1 | geno 
>  Variance list: list(K) 
>                    geno     resid
> Standard Dev: 3.2863076 3.8496073
> % Variance:   0.4215502 0.5784498
> 
> 
> Now I would like to calculate the significance of dta[,k]. The anova function 
> does not work on objects of class lmekin. Any help will be very much 
> appreciated.
> 
> Marc.
> 
> 	[[alternative HTML version deleted]]
> 
> 

-- 
Pär K. Ingvarsson
Senior Researcher, Swedish Research Council
Associate Professor
Umeå Plant Science Centre
Department of Ecology and Environmental Science
Linneaus väg 6
Umeå University, SE-901 87 Umeå
tel. +46-(0)90-786-7414, fax. +46-(0)90-786-6705