Dear Sir / Ms.,
I am using lme4 to analyze the effects of three treatment levels, three seed types, six sites on the removal percentage of seeds over 20 days. Each day, six sites were randomly chose, so the six sites' locations are different from another day's. There are total 10 seed for each combination (Treatment * Type) for oberving the removal number.
I think, Day should be a random effect and Site also a random effect. Then I use the following code:
glmer(cbind(Removal, 10-Removal) ~ Type + Treatment + (1 | Day / Site), family=binomial)
My quesiton is whether (1|Day/Site) correct for my experiment? Each day, there were six random sites, and we observed the removal number per 10 seeds for the combinations (Treatment * Type).
Thank you very much!
Best wishes,
Peijian Shi
Peijian Shi
2 messages · shi_peijian, Ben Bolker
2 days later
shi_peijian <shi_peijian at ...> writes:
Dear Sir / Ms.,
I am using lme4 to analyze the effects of three treatment levels, three seed types, six sites on the removal percentage of seeds over 20 days. Each day, six sites were randomly chose, so the six sites' locations are different from another day's. There are total 10 seed for each combination (Treatment * Type) for oberving the removal number.
I think, Day should be a random effect and Site also a random effect. Then
I use the following code:
glmer(cbind(Removal, 10-Removal) ~ Type + Treatment + (1 | Day / Site),
family=binomial)
My quesiton is whether (1|Day/Site) correct for my experiment? Each day, there were six random sites, and we observed the removal number per 10 seeds for the combinations (Treatment * Type).
This looks about right, if you have unique sites. If you wanted, you could look for time trends by adding Day as continuous fixed covariate; it's probably not going to make a very big difference, but might improve model fit (normality of conditional modes) slightly.