R help - Sep 2008 - Testing this significance of a factor in a mixed-model "ANCOVA"

R-users -

I must preface this question by saying that I'm a relative newbie to  
both R and mixed-modeling.

I'm using lme fit an ANCOVA-like model.  My data consist of bone  
length measurements for a developmental series of two capuchin monkey  
species.  I'm interested in whether the rate of bone length scaling  
to body mass (i.e., growth) differs between species.

My call to lme looks like this:

model=lme(lghumlen ~ species + lgbmkg, random= ~ lgbmkg | id,  
correlation=corCAR1(), weights=varExp())

where lghumlen is the log of humerus length and lgbmkg is the log of  
body mass.

If I ask for a summary of my model, I get this from the table of  
fixed effects:

Fixed effects: lghumlen ~ species + lgbmkg
                         Value  Std.Error  DF   t-value p-value
(Intercept)          3.567226 0.02059541 540 173.20495   0.000
speciesCebus apella -0.096348 0.04523323  16  -2.13004   0.049
lgbmkg               0.532745 0.01804411 540  29.52459   0.000

So the difference between species looks to be barely significant.  I  
get the same result using the fit.contrast function from the gtools  
package:

fit.contrast(model,"species",coeff=c(1,-1))

                      Estimate Std. Error  t-value   Pr(>|t|)
species c=( 1 -1 ) 0.09634857 0.04523308 2.130046 0.04903713

However, if I use the anova function to summary the model, I get a  
totally different result:

anova(model)

             numDF denDF  F-value p-value
(Intercept)     1   540 49336.06  <.0001
species         1    16     0.45  0.5115
lgbmkg          1   540   871.70  <.0001

So the difference between species is not significant at all.

What gives?

Thanks,

Jesse

______________________________________________
Jesse W. Young, Ph.D.
Postdoctoral Research Fellow
Department of Anthropology
University of Texas at Austin
1 University Station C3200
Austin, TX  78712
Phone: (512) 232-6380

Email: jesse.young@mail.utexas.edu
Web page: https://webspace.utexas.edu/youngjw3/www/






	[[alternative HTML version deleted]]

R-users -

I must preface this question by saying that I'm a relative newbie to  
both R and mixed-modeling.

I'm using lme fit an ANCOVA-like model.  My data consist of bone  
length measurements for a developmental series of two capuchin monkey  
species.  I'm interested in whether the rate of bone length scaling  
to body mass (i.e., growth) differs between species.

My call to lme looks like this:

model=lme(lghumlen ~ species + lgbmkg, random= ~ lgbmkg | id,  
correlation=corCAR1(), weights=varExp())

where lghumlen is the log of humerus length and lgbmkg is the log of  
body mass.

If I ask for a summary of my model, I get this from the table of  
fixed effects:

Fixed effects: lghumlen ~ species + lgbmkg
                         Value  Std.Error  DF   t-value p-value
(Intercept)          3.567226 0.02059541 540 173.20495   0.000
speciesCebus apella -0.096348 0.04523323  16  -2.13004   0.049
lgbmkg               0.532745 0.01804411 540  29.52459   0.000

So the difference between species looks to be barely significant.  I  
get the same result using the fit.contrast function from the gtools  
package:

fit.contrast(model,"species",coeff=c(1,-1))

                      Estimate Std. Error  t-value   Pr(>|t|)
species c=( 1 -1 ) 0.09634857 0.04523308 2.130046 0.04903713

However, if I use the anova function to summary the model, I get a  
totally different result:

anova(model)

             numDF denDF  F-value p-value
(Intercept)     1   540 49336.06  <.0001
species         1    16     0.45  0.5115
lgbmkg          1   540   871.70  <.0001

So the difference between species is not significant at all.

What gives?

Thanks,

Jesse

______________________________________________
Jesse W. Young, Ph.D.
Postdoctoral Research Fellow
Department of Anthropology
University of Texas at Austin
1 University Station C3200
Austin, TX  78712
Phone: (512) 232-6380

Email: jesse.young@mail.utexas.edu
Web page: https://webspace.utexas.edu/youngjw3/www/



______________________________________________
Jesse W. Young, Ph.D.
Postdoctoral Research Fellow
Department of Anthropology
University of Texas at Austin
1 University Station C3200
Austin, TX  78712
Phone: (512) 232-6380

Email: jesse.young@mail.utexas.edu
Web page: https://webspace.utexas.edu/youngjw3/www/







	[[alternative HTML version deleted]]