[R-sig-ME] strip plot design with pseudoreplication

Phillip Alday me @end|ng |rom ph||||p@|d@y@com
Wed Apr 6 04:19:00 CEST 2022

I am not commenting on whether the model makes sense, but you can use
the same nesting syntax in lme4 that you used in lme:


On 22/3/22 8:06 am, Javier Moreira wrote:
> hi, members of the group.
> I'm trying to find the way to express a model within R.
> My intention is to use a set of data from a wheat trial, harvested with GPS
> mapping. I'm comparing the standard model that use the average of the plot
> (experimental unit) vs. one that use every pseudo-replication (observation
> unit-harvest point).
> The design is a strip plot or split block, with 2 factors and 3 error
> terms. For the first step, I use the Agricolae package.
> model0_fijo = with(data21,strip.plot(BLOCK = BLOQUE,
>                    COL = AMBIENTE,
>                    ROW = TRATAMIENTO,
>                    Y = RTO))
> *data21, n=72
> and this produces an anova with 3 error terms (Ea,Eb,Ec) and the correct
> distribution of degrees of freedom (Ambiente, Ea/ Tratamiento,Eb/
> interacción, Ec).
> When I try to use the pseudo-replication, i first try to use lme(), and to
> take into account the correct grouping, it produces 2 different Error
> terms, and i get (Ambiente, Ea, and tratamamiento and interaction, Eb).
> This makes the design as it was a split plot (2 error terms).
>                 random=~1|BLOQUE/AMBIENTE/TRATAMIENTO,
>                 data=data1)
> *data1, n=3095
> According to a tutorial
> <https://vsni.co.uk/case-studies/dealing-with-pseudo-replication-in-linear-mixed-models>for
> the use of ASreml package i was able to use lmer() to account for the
> block:plot interaction (1|BLOQUE:parcela) and get the same result as the
> tutorial. But, this leaves the model with only 1 error term applied to
> every factor and interaction. So, it takes into account the
> pseudo-replication but not the correct error assign for the anova.
>         (1|BLOQUE)+(1|BLOQUE:AMBIENTE)+(1|BLOQUE:parcela),
>        data=data1)
> In the initial steps, to find a way with lmer() to get a 3 error model that
> accounts for pseudo-replication would be great.
> Following that, i also have to account for spatial correlation, and that
> isn't an option within lmer() so, i have to get the same model translate to
> lme().
> Thanks a lot for your help. This data analysis is for a on farm trial from
> my Msc thesis.
> best regards,
> Ing. Agr. Javier Moreira

More information about the R-sig-mixed-models mailing list