6

In a semi-controlled experiment, seeds were sown in about 20 boxes, associated with 3 types of soil. A few months later, the morphological characteristics of the shoots were measured. We would like to study the effect of the type of Soil on these Morphological characteristics but it appears that there is some variability in these characteristics between shoots sown in different boxes with the same type of soil.

Is it correct to consider this data as nested and analyse it by means of a linear mixed model like:

Morphological characteristics ~ Soil + 1|Box

Or is there any other solution to deal with this Box variable, which is a confounding variable because each Box is only associated with one type of Soil.

Robert Long
  • 53,316
  • 10
  • 84
  • 148
Circus pygargus
  • 419
  • 3
  • 18
  • You're talking about nested vs crossed random effects. See [this question](https://stats.stackexchange.com/questions/228800/crossed-vs-nested-random-effects-how-do-they-differ-and-how-are-they-specified). – Joe Jun 12 '19 at 14:37
  • 1
    @Joe I don't think so. There are only 3 levels of `Soil` and they are specifically interested in the fixed effect of it. – Robert Long Jun 12 '19 at 14:56

1 Answers1

4

I am not sure why you say that Box is a confounder. To be a true confounder, it should be a cause, or a proxy for a cause, of both the outcome (Morphological characteristics) and the exposure (Soil type).

Anyway, from the information given, it seems that you need to account for possible correlations within each box, and a mixed effects model, with random intercepts for Box would be appropriate, as per the example in the OP, provided that there are sufficient shoots per box.

Robert Long
  • 53,316
  • 10
  • 84
  • 148
  • There are approximately 30 shoots per box so it should not be a problem. I thought Box would be a confounder because it has a significative effect on the Morphological explained variables (ANOVAs performed within each type of Soil) and because each Box is associated with a single type of Soil, thus the variables Box and Soil are sort of linked. – Circus pygargus Jun 12 '19 at 16:07