[R-sig-ME] lmer nesting
cerru006 @end|ng |rom umn@edu
Fri Mar 11 15:13:05 CET 2022
Dear list members,
I have a conceptual doubt about mixed modeling for the following data.
Soybean grain samples were collected all around North and South
America during 2012-2013-2014 cropping seasons, and analysed for
several grain quality biochemical features. Hierarchical sampling
structure: The American continent* was divided into three regions: US,
Brazil (including Paraguay) and The Pampas (including Argentina and
Uruguay). These three main regions were subdivided into sub_regions:
north, south, central. Each subregion was integrated by all states or
provinces in each subregion.
The main objective is to analyze variation and variance for regions,
subregions and years, in particular: 1) to assess differences among
regions in each quality variable; 2) to have an idea of the stability
of this pattern. compare the proportion of the variation explained by
region and proportion of the variation explained by year; 3) to
compare variability within a region (among subregions) against
variability among regions.
Our first attempt was fitting a mixed model where region, subregion
and state were fixed effects and year was a random one: "mod".
I basically wonder if this model syntax is ok to address our
objectives (in this case for protein) or should I modify the nesting
So far I have generated a nested variable: subregion = sub_region
nested in a region and state = state_proc nested in the subregion.
mod<- lmer(prot_db ~ region + subregion + state +(1|year) +
(1|region:year) + (1|subregion:year), data = df)
#VCA extracts variances fixed and random
anovaVCA(form= prot_db~ region + (year) + region:(year)+ subregion +
subregion:(year)+ state , Data=df)
Sorry if it was too long, I'm pretty newby in modeling so any help is
*The American continent produces 90% of the soybean globally, and
represents 60% of the protein that the world consumes each year.
Environment and genetics affect composition (quality) and there is a
need to determine spatials patterns. The study is oriented to end
users, where to buy soybean for different end uses.
More information about the R-sig-mixed-models