[R-sig-ME] Modest questions about mixed-effect model and correlation structure

Laio Zimermann Oliveira |@|ozo||ve|r@ @end|ng |rom gm@||@com
Mon Aug 16 22:17:30 CEST 2021

Dear Dr. Ben Bolker,

I have a dataset of stem volume of individual trees gathered on 154 sample
plots in the subtropical Brazilian Atlantic Forest, where 1 to 20 trees had
their volume determined per plot; the total sample size is n = 1,192. I
need to fit a model to predict the stem volume (V) of ~30,000 trees
measured on 192 plots each with an area of 0.4 ha using diameter at breast
height (D) and stem height (H) as predictor variables. I fitted a
fixed-effect only power model with the following command:

M1 <- nlme(V ~ b0*(D^b1)*(H^b2), start=start, fixed=b0+b1+b2~1, groups=~g,
weights=varPower(form=~D), data=data)

My questions are:

1) I believe I should incorporate correlation among trees from the same
plot into the model. I don't mean spatial correlation. How could I do this?
The problem is that I have just a few trees on certain plots.

2) Suppose I fit M1 with random intercepts for each plot to relax the
assumption of residual independency (correct me if I shouldn't do it). Is
there a problem using just the fixed-effects part of the model to
predict stem volume of the trees on the 192 sample plots? I ask this
because data to fit the model were gathered on 154 among 192 plots, and I
need to consider all plots to generate population estimates of mean growing
stock volume per hectare.

Probably my questions are elementary, I'm sorry for that.

I appreciate your attention very much. Best regards.

	[[alternative HTML version deleted]]

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