I'm doing a survey in which I get public available data and re-run the statistics the authors did in the first place. I came across this article that ran nested mixed models with the lme4 package. When I ran the same models, I get the same estimates, with the same CI ranges, except for the intercept. I've tried scaling the continuous predictor, and it only made it worse. I tried fitting the model with REML being TRUE or FALSE, and it also didn't help. How and why can this be happening??
Hint: The R versions differ, and maybe the packages versions, but I don't think my intercept estimates should go from 13.39 (author's result) to 5.47 (my result).
Edit: The data set is quite large, so I'll post just the heading
Species Island Lizard_ID Headbob_duration_sec Background_speed_mm.per.sec Average_background_speed_mm.per.sec Habitat_light.log
1 Anolis_cooki Puerto_Rico P232 6.300000 8.933370 8.933370 2.082839
2 Anolis_cooki Puerto_Rico P233 2.133333 8.518135 9.331636 2.541373
3 Anolis_cooki Puerto_Rico P233 4.766667 10.076557 9.331636 2.541373
4 Anolis_cooki Puerto_Rico P233 7.133333 9.641858 9.331636 2.541373
5 Anolis_cooki Puerto_Rico P233 5.466667 9.347104 9.331636 2.541373
6 Anolis_cooki Puerto_Rico P233 6.000000 9.074527 9.331636 2.541373
Here is the model the authors (and I) did:
library(lme4)
modDHB<-lmer(Headbob_duration_sec~Average_background_speed_mm.per.sec+
Habitat_light.log+
Island+(1|Lizard_ID)+(Average_background_speed_mm.per.sec+Habitat_light.log|Species),
na.action=na.omit,control=lmerControl(optimizer="bobyqa"))
The model predicts the duration of a lizard behavior (head bob) as a function of background speed, environment light, and island the animal was observed. It uses lizard species as a random factor with lizard ID nested within species.
I'm not sure which optimizer the authors used, but I used bobyqa to prevent conversion problems.
Here is my output:
summary(modDHB)
Linear mixed model fit by REML ['lmerMod']
Formula: Headbob_duration_sec ~ Average_background_speed_mm.per.sec +
Habitat_light.log + Island + (1 | Lizard_ID) + (Average_background_speed_mm.per.sec +
Habitat_light.log | Species)
Control: lmerControl(optimizer = "bobyqa")
REML criterion at convergence: 16746
Scaled residuals:
Min 1Q Median 3Q Max
-3.4199 -0.3470 -0.0370 0.1199 9.7081
Random effects:
Groups Name Variance Std.Dev. Corr
Lizard_ID (Intercept) 1.412e+01 3.75754
Species (Intercept) 7.099e+00 2.66442
Average_background_speed_mm.per.sec 1.444e-04 0.01202 -1.00
Habitat_light.log 4.741e+00 2.17739 -0.74 0.74
Residual 4.081e+01 6.38801
Number of obs: 2490, groups: Lizard_ID, 429; Species, 16
Fixed effects:
Estimate Std. Error t value
(Intercept) 5.47746 1.48225 3.695
Average_background_speed_mm.per.sec 0.05186 0.06082 0.853
Habitat_light.log -1.41167 0.73028 -1.933
IslandPuerto_Rico 7.91151 1.27021 6.228
Correlation of Fixed Effects:
(Intr) A___.. Hbtt_.
Avrg_bc__.. -0.356
Hbtt_lght.l -0.597 -0.033
IslndPrt_Rc -0.628 0.009 0.065
convergence code: 0
boundary (singular) fit: see ?isSingular
As I said, everything is fine, except for the intercept estimate, which should be 13.39.
