Mixed linear model with nested and interaction term
Hi Heng-An, it is good practice to reply also to the mailing list in Cc. Is this dataset fully balanced now? What is the number of levels of each of the factors and what is the number of total observations now? Why is the SAS ANOVA now only containing the fixed effect, while before it also contained the random effects? Did you change any other model statements? It is really strange, that the estimates differ. Could you please report also the estimated variances? See summary(model) in R and the beginning of the SAS output. It should be 4 variances (one for each random effecct and the residual variance). One more thing, will you get same results for Satterthwaite ddf? Best regards, Samuel
On 07/05/18 21:42, Lin, Heng-An wrote:
Hi Samuel, Thanks for the suggestion, that's really helpful. so I tried to use the code below in r to see the difference with smaller and balanced data set here is what I got in R
anova(model_Test, type="3", ddf="Kenward-Roger")
Type III Analysis of Variance Table with Kenward-Roger's method
Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
Treatment 60.219 15.055 4 4 0.8347 0.5674
and in SAS (with type3 and KR method)
df Sum Sq F value p-value
Treatment 4 78.9246 0.81 0.5801
They seems more similar for F and P value, but the Sum sq still different...not sure why
Sorry for sending repeating email.
Thanks for your time again.
________________________________________
?: Samuel Knapp [samuel.knapp at tum.de]
????: 2018?5?7? ?? 03:39
?: Lin, Heng-An
??: r-sig-mixed-models-request at r-project.org
??: Re: [R-sig-ME] Mixed linear model with nested and interaction term
Hi Heng-An,
The SAS model specification seems to be in accordance with the lmer
model. However, I'm slightly surprised, that the SAS ANOVA also contains
the random effects.
Looking at the degrees of freedom, I get the impression that your data
is not fully orthogonal, e.g. that you don't have the same number of
reps at each location or some missing values. If you dont't have fully
orthogonal and balanced data (i.e. n obs = n treatments * n locations *
n replications), there could be differences between different SS methods
(Type 1, 2 or 3) or df methods (Kenward-Roger, Satterthwaite).
In order to test that, you could
1) create an orthogonal and balanced subset of your data and compare
results.
2) Test other df methods in R with the lmerTest package
library(lmerTest)
anova(model_MW, ddf="Satterthwaite")
# or
anova(model_MW, ddf="Kenward-Roger")
3) Test other SS methods in R, also implemented now in the lmerTest package
anova(model_MW, type =1) # or type=2 or type=3
4) Test the same settings in SAS using the method statement in the proc
mixed call, and the ddfm statement in the model line.
Additionally, to just simply playing around, also get familiar with the
different concepts (The SAS documentation on proc mixed has quite a good
overview). This shall rather be an overview on how to specify different
options.
I'd be interested, if you get similar results....
Best regards,
Samuel
On 04/05/18 22:15, r-sig-mixed-models-request at r-project.org wrote:
Date: Fri, 4 May 2018 20:15:30 +0000
From: "Lin, Heng-An" <henganl2 at illinois.edu>
To: Ben Bolker <bbolker at gmail.com>
Cc: "r-sig-mixed-models at r-project.org"
<r-sig-mixed-models at r-project.org>
Subject: Re: [R-sig-ME] Mixed linear model with nested and interaction
term
Message-ID:
<14FC5E11B32FB84FB0D08FE2F4127AFE2B53751C at CHIMBX6.ad.uillinois.edu>
Content-Type: text/plain; charset="utf-8"
** Sorry I didn't notice that the format of the previous email was off, so I just send the same email again
Here is my SAS syntax and output :
proc mixed data=A method=type3; class Location Block Treatment;
model Yield= Treatment/ddfm=kr;
random Location Location*Treatment Block(Location);
run;quit;
Source Df Sum_of_squares F_value
Treatment 4 46.196951 0.41
Location 2 4670.0979652 44.74
Location*Treatment 8 224.44332 1.66
Block (Location) 9 369.782487 2.43
Residual 34 574.051330
And here is R output:
anova(model_MW)
Analysis of Variance Table
Df Sum Sq Mean Sq F value
Treatment 4 34.847 8.7118 0.5085
I am not sure why the sum of square, and the F- value are different.
Maybe is because I use type III in SAS and in lmer is using REML?
I would also like to check the sum of square of other factors as SAS did, is there any way could do this in lmer?
I am really new to this, Thanks for your time!
Heng-An
________________________________________
?q: Ben Bolker [bbolker at gmail.com]
?H????: 2018?~5??4?? ?U?? 01:39
??: Lin, Heng-An
???: r-sig-mixed-models at r-project.org
?D??: Re: [R-sig-ME] Mixed linear model with nested and interaction term
This seems like a reasonable model specification. Can you show us
the results you're getting from R and SAS, and your SAS syntax (some
people here understand that language), so that we can see what looks
different? (It would help if you also wrote a few sentences about
what you see as the important differences between the results.)
On Fri, May 4, 2018 at 2:30 PM, Lin, Heng-An <henganl2 at illinois.edu> wrote:
Hi all,
I am analyzing my data with following model,
model1 <- lmer(Yield~Treatment+(1|Location)+(1|Location:Treatment)+(1|Location:Block), data=A)
in here, I want to set an random interaction term (Location*treatment) and an random nested term (block nested within location).
But I couldn't get similar ANOVA results when I compare the output with SAS porc mixed output.
So, I think i might make some mistake in the model in R...
Can anyone give me some suggestion?
Thanks in advance!
Heng-An
[[alternative HTML version deleted]]
_______________________________________________
R-sig-mixed-models at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
[[alternative HTML version deleted]]
------------------------------
Subject: Digest Footer
_______________________________________________
R-sig-mixed-models mailing list
R-sig-mixed-models at r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
------------------------------
End of R-sig-mixed-models Digest, Vol 137, Issue 6
**************************************************