Skip to content

LMER: applying a random term to only one level of a factor

2 messages · Aislinn Pearson, Bob OHara

Original
Aislinn Pearson
# 
Hi,

I've tried googling this but haven't been very successful. Essentially, I'd like to know what is the most statistically valid way of dealing with a random term which doesn't apply to every level of fixed-effect factor.

I have a mixed effect model that looks like this

Disease level <- weight + Flown +(1|DateFlown)

Either I flew my insects on a flight mill (which can be thought of as a 'treadmill' for flying insects) or I didn't, hence flown is a two level factor (Yes or No) and I want to understand how this affects the amount of disease in my insect. To get as many replicates as I could on a single day, I had two different banks of flight mills (A & B), each bank containing 16 individual insect treadmills. The insects were randomly assigned to one of the two sets of 16 flight mills. Previous studies tell me there are differences between these two sets of flight mills, so I would like to account for them as a random term in my model.

However, I can't run this in LMER. When I tried I got the error;

Error in `contrasts<-`(`*tmp*`, value = contr.funs[1 + isOF[nn]]) :
  contrasts can be applied only to factors with 2 or more levels

Which I imagine means that one of my factors (i.e. Flown) doesn't include any levels for the random term mill set (i.e. for all unflown insects the value in the mill set column is NA)

Is it possible to include this form of experimental design in LMER (the package I know best) or, alternatively, nlme (which I am a lot less accustomed to using)?

I would be really grateful if anyone has any insight.

Many thanks

Rothamsted Research is a company limited by guarantee, registered in England at Harpenden, Hertfordshire, AL5 2JQ under the registration number 2393175 and a not for profit charity number 802038.
Bob OHara
#
On 06/04/16 20:12, Aislinn Pearson wrote:
As a practical matter, it's not worth setting a level with two levels as 
random: you don't gain anything in the analysis and the variance 
component is really poorly estimated. In practice, this might actually 
make things cleaner, as you will have to look a bit more at the flight 
mill effects, so you should get a better feel for what's happening.
I can think of two ways of doing this: either set up a factor with three 
levels (Flight mill A, Flight mill B, Not Flown) or set the Not Flown to 
one of the flight mill levels. The first way feels less confusing, but 
you might have to set up some contrasts to estimate the differences. But 
hopefully your insects will cooperate nicely and make the difference 
between the flight ills will be much smaller than between flight mills 
and not flown.

HTH

Bob