3

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.

Miles
  • 41
  • 3
  • The variance of the random intercept is 0.07566. Compare your model to a model without the random effect to see if you can omit that parameter. – Roland Feb 14 '17 at 12:12
  • You don't lose anything by retaining the 'species' term in the model. And it is appropriate to include it since you know that there is variance that can be attributed to it. – mkt Dec 08 '17 at 15:31

0 Answers0