[R-meta] network meta-analysis - include block (within-study) level

Wed Aug 9 00:50:14 CEST 2017

Sorry for that Wolfgang,

y = grain yield at plot (single value). Actually, plots were ~ 15m²,
however the grain weight was expressed in kg/10000 m² (1ha). 

x = is the leaf crop area damaged by a fungal disease (%)

Both are quantitative positive variables, single-point assessments,
and each row has the same plot values (block/treatment)

We expect that "x" has a negative effect on y. So we have interest on the
intercept (soybean yield in abscense of the disease) and how much yield it
is reduced by a unit increment of x. We also want to test the effect of
moderators of y~x.

I hope I have clarified your doubts.

Best,

Juan Edwards

On Tue, Aug 8, 2017 at 7:34 PM, Viechtbauer Wolfgang (SP) <
wolfgang.viechtbauer at maastrichtuniversity.nl> wrote:

> So is 'y' is the mean treatment yield here? Also, is that really the
> average of multiple measurements (e.g., if there is subsampling)? Or is 'y'
> just the single measurement (yield) for that particular block and
> treatment? I still do not quite understand what kind of data you have.
> Also, what is 'x'?
>
> Best,
> Wolfgang
>
> -----Original Message-----
> From: Juan Pablo Edwards Molina [mailto:edwardsmolina at gmail.com]
> Sent: Tuesday, August 08, 2017 23:26
> To: Viechtbauer Wolfgang (SP)
> Cc: r-sig-meta-analysis at r-project.org
> Subject: Re: [R-meta] network meta-analysis - include block (within-study)
> level
>
> Pretty close to that structure you say:  I have several treatments at
> each block (balanced experiments), actually different set of treatments
> across the k-trials (all trials have the untreated Check)
>
> This are a few lines of trial 3:
> 
> trt    trial bk  x    y
> Check  3     1   40   2493
> Check  3     2   45   2173
> Check  3     3   40   2628
> Check  3     4   40   2168
> Fox    3     1   35   3194
> Fox    3     2   30   2363
> Fox    3     3   35   2887
> Fox    3     4   30   3278
> NTX    3     1   40   2988
> NTX    3     2   35   2361
> NTX    3     3   35   2341
> NTX    3     4   35   3218
> 
> | Also, do you have the raw mean and variance (or SD) and sample size for
> each row of the dataset? It seems like you are first fitting some kind of
> ANOVA within each study, but | that might actually complicate things.
>
> Yes, I have the raw full dataset so I have the observation level values
> to calculate SD, means..
>
> Several authors from the Phytopathology area use ANOVA MSE :
>
> "...The within-study variance (V) for IND or DON for these fungicide
> trials is the residual variance (mean square error) from an analysis of
> variance (ANOVA) of the effects of treatment on disease or toxin. Where the
> original data were available, this variance was calculated directly from an
> ANOVA..."
>
> http://apsjournals.apsnet.org/doi/abs/10.1094/PHYTO-97-2-0211
>
> Juan
>
> On Tue, Aug 8, 2017 at 6:03 PM, Viechtbauer Wolfgang (SP) <
> wolfgang.viechtbauer at maastrichtuniversity.nl> wrote:
> Dear Juan,
>
> Could you show a bit of the data (structure)? In particular, does each
> block contain two treatments, so that the structure looks something like
> this?
>
> trial block treatment mean
> --------------------------
> 1     1     1         ...
> 1     1     2         ...
> 1     2     1         ...
> 1     2     2         ...
> 2     1     1         ...
> 2     1     2         ...
> 2     2     1         ...
> 2     2     2         ...
> 2     3     1         ...
> 2     3     2         ...
> ...
> 
> Also, do you have the raw mean and variance (or SD) and sample size for
> each row of the dataset? It seems like you are first fitting some kind of
> ANOVA within each study, but that might actually complicate things.
>
> Best,
> Wolfgang
>
> -----Original Message-----
> From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bo
> unces at r-project.org] On Behalf Of Juan Pablo Edwards Molina
> Sent: Tuesday, August 08, 2017 22:09
> To: r-sig-meta-analysis at r-project.org
> Subject: [R-meta] network meta-analysis - include block (within-study)
> level
>
> Dear list,
>
> I have a dataset containing crop field randomized block design experiments
> with observations at plot level (experimental unit), and I want to estimate
> the treatments grain yield difference relative to a untreated check.
>
> net1 <- rma.mv(yield, vi2, mods = ~ treatment, random = ~ treatment|
> trial,
>                          method="ML", struct="UN", data=df)
>
> where yield is the vector of mean treatments yield for vi2 is the vector of
> sampling variances obtained by:
>
> vi2 <- V_yield/n  (for each trial)
>
> (V_yield = MSE from anova)
>
> Do I need to include the block in the model? or using the experiment
> treatments means will obtain the same results? I suppose something like:
>
> net2 <- rma.mv(yield, vi2, mods = ~ treatment, random = ~ treatment|
> block|
> trial,
>                          method="ML", struct="UN", data=df)
>
> If the latter would be a better approach, how do I include the sampling
> variance?
>
> Thanks in advance,
>
> Juan Edwards
>

	[[alternative HTML version deleted]]