我怎样才能让 lme4::glmer 运行

Question

I have this csv file

Name	ID	Year	Tourist	Feed.bout
Anubis	1	2014	TRUE	0:01:17
Athena	2	2014	FALSE	0:01:53

I am trying to run a general linear mixed model with lme4 but I am struggling to get it to run.

My code is

library(tidyverse)
library(lme4)
library(lattice)

Feedbout=read.csv("Av_Fe_Bout3.csv")
head(Feedbout)

Feedbout$Tourist=as.logical(Feedbout$Tourist)
Av_Fe_Bout3 <- glmer(Feed.bout ~ Year + Tourist + Year:Tourist + (1|ID), family = poisson, data = Feedbout)
summary(Av_Fe_Bout3)

and I keep getting

Error in mkRespMod(fr, family = family) : 
  response must be numeric or factor
In addition: Warning message:
Some predictor variables are on very different scales: consider rescaling

I have tried it without a family and also with binomial.

If I change anything else the Feed.bout changes to NA s and I don't know how to change it back.

I think the problem is with Feed.bout but I'm not sure what to do.

I have also tried to run

read.zoo(file = "Av_Fe_Bout3.csv", format = "%H:%M:%S")

to try and change the format of Feed.bout however it just says

Error in read.csv.zoo(file = "Av_Fe_Bout3.csv", format = "%H:%M:%S") : 
  could not find function "read.csv.zoo"

Answer 1

Its hard to emulate what you've done, but here I will provide a worked example using a binary dataset to show you what I imagine may be the issue. First, I will load the requisite packages. Then I will set a random seed so you can try what I have here (and hopefully on your own data).

#### Load Libraries and Random Seed ####
library(lmerTest)
library(tidyverse)
set.seed(123)

From here, there is no need to convert the outcome into a factor or numeric value for this specific dataset , but here I do so just to show you how to do it. I also create a simulated value called fake to add to lme4 's Verbagg data, as there are no other numeric predictors in this dataset. As there are 7585 observations, I simulate as many below.

#### Convert Variable to Factor ####
VerbAgg <- VerbAgg %>% 
  mutate(r2 = factor(r2),
         fake = rnorm(n=7584))

Then I fit the model as always. In this case, I put the scale function around the two numeric predictors in the interaction term.The nAGQ argument is only run here to speed up the run, as this data is normally slow to fit.

#### Fit Data to Model ####
fit <- glmer(r2 
             ~ scale(Anger)*scale(fake)
             + (1|id) 
             + (1|item),
        family = binomial,
        data = VerbAgg,
        nAGQ=0L)

Finally from there you just summarize the model with summary(fit) , which gives you this:

Generalized linear mixed model fit by maximum likelihood (Adaptive
  Gauss-Hermite Quadrature, nAGQ = 0) [glmerMod]
 Family: binomial  ( logit )
Formula: r2 ~ scale(Anger) * scale(fake) + (1 | id) + (1 | item)
   Data: VerbAgg

     AIC      BIC   logLik deviance df.resid 
  8203.1   8244.7  -4095.5   8191.1     7578 

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-5.8761 -0.6234 -0.1787  0.6240 14.0824 

Random effects:
 Groups Name        Variance Std.Dev.
 id     (Intercept) 1.815    1.347   
 item   (Intercept) 1.275    1.129   
Number of obs: 7584, groups:  id, 316; item, 24

Fixed effects:
                          Estimate Std. Error z value Pr(>|z|)   
(Intercept)              -0.155244   0.244350  -0.635  0.52521   
scale(Anger)              0.261487   0.081500   3.208  0.00133 **
scale(fake)               0.001112   0.028515   0.039  0.96890   
scale(Anger):scale(fake)  0.026135   0.029000   0.901  0.36747   
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Correlation of Fixed Effects:
            (Intr) scl(A) scl(f)
scale(Angr)  0.000              
scale(fake)  0.002 -0.003       
scl(Ang):() -0.001  0.011  0.001