[R-sig-ME] Not the correct model?

[Seth W. Bigelow]

A couple of questions:

Are you sure that Substrate should be modeled as a random factor? It seems
that you used 12 substrates (citric acid, etc) as some kind of standard
protocol for testing respiration, rather than selecting the 12 substrates
from among a population  of possible substrates.

Are you sure that Pesticides should be nested within block? Have you tried a
simpler model, in which Pesticides are not nested within Block but are only
treated as fixed effects, and then used the usual model adequacy checking
methods (examination of residuals, etc) to determine whether nesting is
warranted? 

--Seth


-----Original Message-----
From: r-sig-mixed-models-bounces at r-project.org
[mailto:r-sig-mixed-models-bounces at r-project.org] On Behalf Of GIMENEZ
ANALIA VERONICA
Sent: Thursday, January 17, 2013 12:35 PM
To: r-sig-mixed-models at r-project.org
Subject: [R-sig-ME] Not the correct model?

anyone who can help me with this?
I have one fixed (PESTICIDES) factor nested in (5) block factor and one
random factor (SUBSTRATES) nested in the fixed factor. I think it4s the
correct model. (Four pesticides (including control) were randomly assign
into the five blocks and then a soil sample were taken and treated with 12
substrates to evaluate respiration rate).
Firs, I fit my model with lme() function like this
> lme09<-lme (RESPIRACION~PESTICIDA, random=~1| BLOQUE/PESTICIDA, 
> weights =
varIdent(form = ~1|SUSTRATO), data=RESP09, na.action=na.omit,
control=lmeControl(maxIter=200, msMaxIter=200, niterEM=100))
> summary(lme09)
Linear mixed-effects model fit by REML
 Data: RESP09
       AIC      BIC    logLik
  111.7775 166.4314 -38.88874

Random effects:
 Formula: ~1 | BLOQUE
         (Intercept)
StdDev: 3.086521e-08

 Formula: ~1 | PESTICIDA %in% BLOQUE
        (Intercept)     Residual
StdDev:   0.1366234 7.911512e-07

Variance function:
 Structure: Different standard deviations per stratum
 Formula: ~1 | SUSTRATO
 Parameter estimates:
AC CITRICO    AC GLUT  AC MALICO    AC OXAL     ASPARR    FENILAL
 GLUCOSA  HISTIDINA
       1.0   704328.3   675541.5   186809.3   380757.2   410564.8
370571.6   441234.7
    MANOSA     LISINA   ARGININA
  338306.9   257493.9   330352.4
Fixed effects: RESPIRACION ~ PESTICIDA
                     Value  Std.Error  DF   t-value p-value
(Intercept)     0.13914000 0.06109982 168  2.277257  0.0240
PESTICIDACIP    0.02539722 0.08762818  12  0.289829  0.7769
PESTICIDAGLIF  -0.14447975 0.09004731  12 -1.604487  0.1346 PESTICIDASEC 1
-0.29229772 0.08990939  12 -3.251026  0.0069
 Correlation:
               (Intr) PESTICIDAC PESTICIDAG
PESTICIDACIP   -0.697
PESTICIDAGLIF  -0.679  0.473
PESTICIDASEC 1 -0.680  0.474      0.461

Standardized Within-Group Residuals:
          Min            Q1           Med            Q3           Max
-1.983260e+00 -5.103333e-01  1.970778e-06  7.050516e-01  2.783527e+00

Number of Observations: 188
Number of Groups:
               BLOQUE PESTICIDA %in% BLOQUE
                    5                    20

When I tried to make a barchart with the mean and standard errors of the
fixed effects (I used the model without the intercept) I realized that these
ES are not the correct ones because I had a non-significant p-value and the
ES didn4t reflect that. Then I tried to get the confidence intervals of the
fixed effects mean and I got these error message:

> intervals (lme09)

Error in intervals.lme(lme09) :

  cannot get confidence intervals on var-cov components: Non-positive
definite approximate variance-covariance

I googled what this error means and I found that these message could be an
advise that the model may be wrong. At this point, I don4t know how to fit
my model correctly.

Then I tried to fit the model with lmer() function but I don4t know if the
syntax is the correct one and when I run the lmer model i get a warning
message

> lmer093<-lmer (RESPIRACION~PESTICIDA+(0+SUSTRATO|BLOQUE/PESTICIDA),
data=RESP09, na.action=na.omit)

Warning message:

In mer_finalize(ans) : iteration limit reached without convergence (9)

I think this is the same message as before with the lme model and so, my
model is wrong.

So, I couldn4t get the confidence intervals to make a barchart and now I 4m
in doubt with the model.

I4m doing something wrong. I tried other models but the degrees of freedoms
were pseudoreplicated.

Thanks

--
Lic. Analma V. Giminez
IFEVA-CONICET
Facultad de Agronomma.
Av. San Martmn 4453 C1417DSE
Bs. As. Argentina

	[[alternative HTML version deleted]]