[R-sig-ME] GLMM - Prediction

[Ben Bolker]

This looks like a bug in lme4: https://github.com/lme4/lme4/issues/457
. A simple workaround is to specify some value for `ID`
(which will be ignored).

On Sat, Mar 3, 2018 at 7:06 PM, Cleber Iack <profiack at gmail.com> wrote:
> Dear
>
> First, thank you very much for your time in answering.
>
> I'm sending a code that I tried not to inform the ID and I couldn't,
>
> Displays the following error at the end
>
> "Error in eval (predvars, date, env): object ID not found"
>
> I'd appreciate it if you could give me some guidance
>
> Thanks
> ####
>
> library(plyr)
>
> library(dplyr)
>
> library(lme4)
>
> n = 300
>
> xx<-c("r1", "r2", "r3", "r4", "r5")
>
> xxx<-c("e1", "e2", "e3")
>
> p=0.3
>
> School = factor(sample(xxx, n, replace=TRUE), levels=xxx, ordered=FALSE)
>
> Rank = factor(sample(xx, n, replace=TRUE), levels=xx, ordered=FALSE)
>
>
>
> df1 <- data_frame(
>
>   ID = as.integer(runif(n, min = 1, max = n/7)),
>
>   xx1 = runif(n, min = 0, max = 10),
>
>   xx2 = runif(n, min = 0, max = 10),
>
>   xx3 = runif(n, min = 0, max = 10),
>
>   School = School,
>
>   Rank = Rank,
>
>   yx = as.factor(rbinom(n, size = 1, prob = p))
>
> )
>
> df1 = df1[order(df1$ID, decreasing=FALSE),]
>
> library(lme4)
>
> mm2 <- glmer(yx ~ xx1 + xx2 + xx3 + Rank +  (1 | ID) + (1 | School / Rank),
> data = df1,
>
>              family = "binomial",control = glmerControl(calc.derivs =
> FALSE))
>
>
>
> n11 <-  data.frame(School=factor("e1", levels =
> levels(df1$School),ordered=FALSE),
>
>                    Rank=factor("r1", levels =
> levels(df1$Rank),ordered=FALSE),
>
>                    xx1=8.58, xx2=8.75, xx3=7.92)
>
> predict(mm2, n11, type="response",re.form= ~(1 | School / Rank))
>
> ##
>
>
> 2018-03-02 9:46 GMT+00:00 Thierry Onkelinx <thierry.onkelinx at inbo.be>:
>
>> Dear Iack,
>>
>> Look at the re.form argument of ?lme4::predict.merMod. This allows you the
>> make predictions without the ID random effect.
>>
>> Best regards,
>>
>>
>> ir. Thierry Onkelinx
>> Statisticus / Statistician
>>
>> Vlaamse Overheid / Government of Flanders
>> INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE AND
>> FOREST
>> Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance
>> thierry.onkelinx at inbo.be
>> Havenlaan 88
>> <https://maps.google.com/?q=Havenlaan+88&entry=gmail&source=g> bus 73,
>> 1000 Brussel
>> www.inbo.be
>>
>> ////////////////////////////////////////////////////////////
>> ///////////////////////////////
>> To call in the statistician after the experiment is done may be no more
>> than asking him to perform a post-mortem examination: he may be able to say
>> what the experiment died of. ~ Sir Ronald Aylmer Fisher
>> The plural of anecdote is not data. ~ Roger Brinner
>> The combination of some data and an aching desire for an answer does not
>> ensure that a reasonable answer can be extracted from a given body of data.
>> ~ John Tukey
>> ////////////////////////////////////////////////////////////
>> ///////////////////////////////
>>
>> <https://www.inbo.be>
>>
>> 2018-03-02 1:09 GMT+01:00 Cleber Iack <profiack at gmail.com>:
>>
>>> Dear,
>>>
>>> I have n different companies with different characteristics and each
>>> company of this has the same positions
>>> I selected data from several members of each position of each Company, and
>>> there were behavioral variables ranging from 0 to 10, which to facilitate
>>> I
>>> will call x1, x2, x3, these variables were collected more than once for
>>> each person.
>>>
>>> The objective is to predict the probability of occurrence of a fact, and
>>> each person was also noted if this fact occurred or not. (0 or 1)
>>>
>>> Using
>>>
>>> formMod1= fato~  x1+x2+x3+(1 | company / position)
>>> Mod1 <- glmer( formMod1 , data = dadosord , family = binomial,
>>>                      control = glmerControl(optimizer="bobyqa"))
>>>
>>> This model I can make a prediction without "problems"
>>>
>>> n11 <-  data.frame(company =factor("M1", levels =
>>> levels(dadosord$company ),ordered=FALSE),
>>>                    position=factor("P1", levels =
>>> levels(dadosord$position),ordered=FALSE),
>>>                    x1=1, x2=7,x3=7)
>>> predict( Mod1 , n11, type="response")
>>>
>>> But I was worried because I have more than one observation of the
>>> characteristics per individual, if I would not have to put it also as
>>> random
>>> or somehow analyze this structure of possible correlation
>>>
>>> Thinking about the first case I did
>>>
>>> formMod2= fato~  x1+x2+x3+(1 | company / position) + (1 | ID)
>>> Mod2 <- glmer( formMod2 , data = dadosord , family = binomial,
>>>                      control = glmerControl(optimizer="bobyqa"))
>>>
>>> But in this case when trying to predict the probability of the fact, the
>>> function "predict" asks me who is the proposed id, which is out of
>>> interest, since I must study for a random person who has the same features
>>> x1, x2, x3, company and position.
>>>
>>> And the second form, if any, I do not know how I would do
>>>
>>> Thanks for the help of my friends.
>>>
>>> Iack
>>>
>>>         [[alternative HTML version deleted]]
>>>
>>> _______________________________________________
>>> R-sig-mixed-models at r-project.org mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>>
>>
>>
>
>         [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models