[R-sig-ME] sas to R

[Steve Hong]

Thank all of you for replying to me.

I tried lmer, lme, and SAS.  I was able to get outputs when I use
'lme' whereas no results from 'lmer'.  I don't know why.  Does anyone
know what the warning message mean?  Outputs from  'lme' were similar
with those from SAS.  Below is selected outputs from lmer, lme, and
SAS, FYI.

Thanks again,

Steve Hong

> fm.lmer <- lmer(y ~ trt + (1|trial/block/trt), data=df, na.action=na.omit)
Error: length(f1) == length(f2) is not TRUE
In addition: Warning messages:
1: In block:trial :
  numerical expression has 92 elements: only the first used
2: In block:trial :
  numerical expression has 92 elements: only the first used
3: In trt:(block:trial) :
  numerical expression has 92 elements: only the first used
4: In block:trial :
  numerical expression has 92 elements: only the first used
5: In block:trial :
  numerical expression has 92 elements: only the first used
> fm.lme <- lme(y ~ trt, random=(~1|trial/block/trt), data = df, na.action=na.omit)
> summary(fm.lme)
Linear mixed-effects model fit by REML
 Data: df
        AIC       BIC   logLik
  -85.22388 -60.68041 52.61194

Random effects:
 Formula: ~1 | trial
        (Intercept)
StdDev:   0.1112442

 Formula: ~1 | block %in% trial
         (Intercept)
StdDev: 1.449228e-06

 Formula: ~1 | trt %in% block %in% trial
        (Intercept)  Residual
StdDev:  0.07081356 0.1020226

Fixed effects: y ~ trt
                  Value  Std.Error DF    t-value p-value
(Intercept)  0.24428523 0.08793775 56  2.7779337  0.0074
trtau2      -0.00996643 0.05605221 25 -0.1778063  0.8603
trtberm     -0.12786905 0.05686903 25 -2.2484830  0.0336
trtls44      0.12326637 0.05478364 25  2.2500582  0.0335
trtsr10y5    0.02513355 0.05517460 25  0.4555275  0.6527
trtsr10y6    0.01932992 0.05478364 25  0.3528410  0.7272
 Correlation:
          (Intr) trtau2 trtbrm trtl44 trt105
trtau2    -0.314
trtberm   -0.309  0.486
trtls44   -0.321  0.504  0.497
trtsr10y5 -0.319  0.500  0.493  0.511
trtsr10y6 -0.321  0.504  0.497  0.515  0.511

Standardized Within-Group Residuals:
          Min            Q1           Med            Q3           Max
-2.614096e+00 -5.666986e-01 -9.727356e-05  4.692685e-01  2.410879e+00

Number of Observations: 92
Number of Groups:
                    trial          block %in% trial trt %in% block %in% trial
                        2                         6                        36
> anova(fm.lme)
            numDF denDF  F-value p-value
(Intercept)     1    56 9.907983  0.0026
trt             5    25 4.122070  0.0072


SAS code and outputs:
proc glimmix data=df;
model y=trt;
random trial block(trial) turf(block*turf);
run;

     Covariance Parameter Estimates

                                 Standard
Cov Parm             Estimate       Error

trial                 0.01237     0.01823
block(trial)                0           .
trt(trial*block)    0.005015    0.002546
Residual              0.01041    0.001963


        Type III Tests of Fixed Effects

              Num      Den
Effect         DF       DF    F Value    Pr > F

trt            5       25       4.12    0.0072



On Mon, Jun 25, 2012 at 10:25 AM, Kevin Wright <kw.stat at gmail.com> wrote:
>
> This could be similar to a multi-location RCB design were "trial" is
> location.  Since no distribution is specified, the distribution is
> assumed to be Gaussian.  Make sure that trial, block, trt are factors,
> this should be similar to SAS:
>
> lmer(y ~ trt + (1|trial/block/trt), data=df)
>
> > proc glimmix data=df;
> > class trial block trt;
> > model y=trt;
> > random trial block(trial) trt(block*trial);
>
> Kevin Wright