currently I am trying to build a model as part of a research project assessing threats to ecosystems. I have used mixed effects models before but the random effect was contributing a lot to the variance then so it was easy to leave it in. I can't find anything on a threshold for a random effect in terms of variance to keep it in the model.
Here is the model:
library(lme4)
z <- lmer(Response ~ Threats + (1|Species))
Output:
> summary(z)
Linear mixed model fit by REML ['lmerMod']
Formula: Response ~ Threats + (1 | Species)
REML criterion at convergence: 2954.5
Scaled residuals:
Min 1Q Median 3Q Max
-4.1483 -0.4064 0.0134 0.4495 3.5343
Random effects:
Groups Name Variance Std.Dev.
Species (Intercept) 0.07566 0.2751
Residual 0.48394 0.6957
Number of obs: 1318, groups: Species, 532
Fixed effects:
Estimate Std. Error t value
(Intercept) 0.04292 0.03665 1.171
Threats1 -0.20438 0.05701 -3.585
Threats2 -0.17720 0.06601 -2.685
Threats3 -0.18473 0.06262 -2.950
ThreatsNULL 0.05391 0.10030 0.537
Correlation of Fixed Effects:
(Intr) d$T_T1 d$T_T2 d$T_T3
dt$Ttl_Thr1 -0.594
dt$Ttl_Thr2 -0.509 0.348
dt$Ttl_Thr3 -0.533 0.357 0.317
dt$Tt_TNULL -0.330 0.230 0.193 0.207
My understanding is that my random effect, Species, is contributing 0.07566 to the overall variance, but is there a minimum value I should otherwise exclude the random effects term? I am planning on adding more to the model so understanding now is crucial.