1

I am sorry that my question repeats many others already available here. However I have read many of them (e.g. this, this and this) as well as other documents and I am still not sure about my model.

I have the following sampling protocol. I study the effect of crop species (e.g. wheat, barley, ...) and other parameters (e.g. humidity) on vegetation.

  • I have 4 different crops.
  • For each crop we randomly chose 5 plots (independant one to each other).
  • In each plot we sampled 5 microplots to measure species richness and humidity (1 measurement per microplot).
  • For some other measurements (e.g. abundance) we went one level further, with 25 squares in each microplot (1 measurement per square).

The 5 microplots within each plot (and the squares within each microplot) are not independant. So I should use random, nested effectsto avoid pseudoreplication.

# Microplot level
modAb = glmer(Richness ~ Crop + Humidity (1|Crop/Plot), family = poisson)
# Square level
modAb = glmer(Abundance ~ Crop + Humidity (1|Crop/Plot/Microplot), family = poisson)

But I have quite few observations, and I read that random effects in such situation were not very appropriate.

So is it the proper way to do it? Should I rather average my results on each plot?

Any confirmation / guidance will be much appreciated!

Community
  • 1
Nausi
  • 133
  • 8
  • Is `Crop` associated with a specific field / site / block? Otherwise, I don't understand why you group the random effects by it. In my (practical) experience, n = 5 can be sufficient for estimating a random effect, although you should keep in mind that you estimate a variance and that estimate is quite susceptible to outliers if n is small. – Roland May 02 '16 at 10:45
  • @Roland: each plot belongs to one crop only, but there are 5 plots per crop type. I thought I had to let the Crop in from other examples I read, but was quite unsure about it (it does not change the results). So is the correct formula `modAb = glmer(Richness ~ Crop + Humidity (1|Plot), family = poisson)` ? – Nausi May 02 '16 at 11:29
  • I believe so but might be missing something. It would be very unusual to have the same variable as a fixed effect and as a grouping factor of random effects. – Roland May 02 '16 at 12:37

0 Answers0