[R-sig-ME] lme4

[Ben Bolker]

(please keep r-sig-mixed-models at r-project.org in the cc: list ... I've
been busy the last couple of days, and there's a good chance that
someone else will pop in and answer questions ...)

On Fri, Jul 14, 2017 at 11:48 PM, Dan Selechnik <danselechnik at gmail.com> wrote:
> Hi Ben,
>
> Thank you very much for the reply, but I'm still not sure what exactly is
> the fix here? Every sample has a single ID. Is this my issue...?

 Yes, that's right.

  If you have a single observation per level of the grouping factor
("observation-level random effects") in a *linear* mixed models, the
random-effect variance would be completely confounded with the
residual variance, so there's no point in using a LMM. (However,
observation-level random effects can be useful in GLMMs with binomial
or Poisson response distributions, as a way of measuring
overdispersion.)

>
> Cheers,
> Dan
>
> On Sat, Jul 15, 2017 at 12:10 PM, Ben Bolker <bbolker at gmail.com> wrote:
>>
>>
>>
>> On 17-07-14 09:26 PM, Dan Selechnik wrote:
>> > Hello,
>> >
>> > My name is Dan and I'm a PhD student in Australia. I was hoping that I
>> > could ask you for some help with using lme4. I have a dataset in which I
>> > have PC1 as a response variable. Population, treatment, RBC, and
>> > population*treatment are my explanatory variables. ID is a random
>> > factor.
>> > (I have attached the CSV file here)...
>> >
>> > I am trying to run a power analysis, and first to fit my data using
>> > lmer.
>> >
>> > First I read my data into R:
>> > pc1=read.csv("R-PowerAnalysis.csv", header=TRUE)
>> >
>> > Then I attempt to fit:
>> > fm1=lmer(pc1$PC1 ~ pc1$RBC + pc1$Population + pc1$Treatment +
>> > pc1$Population*pc1$Treatment + (1|pc1$ID), data=pc1, REML=FALSE)
>>
>>    A small point, but in general you should *not* use pc1$ in specifying
>> your formula: instead,
>>
>> fm1=lmer(PC1 ~ RBC + Population*Treatment + (1|ID), data=pc1,
>>     REML=FALSE)
>>
>> (also, the * operator includes both the main effects of Population and
>> Treatment and their interaction). But that should be tangential to your
>> problem.
>>
>> >
>> > However, this fails, returning the message:
>> > Error: number of levels of each grouping factor must be < number of
>> > observations
>> >
>> > My number of populations and treatments is much less than my number of
>> > observations, so I am not sure why I am getting this error...
>>
>>   That's not your problem.  lme4 is referring to the number of levels of
>> the *grouping factor*, which is ID (not Population or Treatment).  Your
>> ID variable must contain a single observation per group (cheating and
>> looking at the data you sent me offline, I can see that's true).
>>
>> If you had sent the results of summary(pc1), we could have guessed this:
>> ID is coded as an integer so we don't know for sure that it consists of
>> the values 1..20, but since the min is 1 and the max is 20 and mean is
>> 10.5, we can guess that that's the case ...
>>
>>        ID        Population         Treatment       RBC
>>  Min.   : 1.00   QLD:10     LPS-Injection:10   Min.   :-113.00
>>  1st Qu.: 5.75   WA :10     PBS-Injection:10   1st Qu.:  22.00
>>  Median :10.50                                 Median :  49.00
>>  Mean   :10.50                                 Mean   :  53.55
>>  3rd Qu.:15.25                                 3rd Qu.: 107.00
>>  Max.   :20.00                                 Max.   : 181.00
>>       PC1
>>  Min.   :-4.5411
>>  1st Qu.: 0.1017
>>  Median : 1.1470
>>  Mean   : 0.6258
>>  3rd Qu.: 1.9251
>>  Max.   : 3.2004
>>
>>
>> Also, when I
>> > run this, it works fine:
>> > fm1=lm(pc1$PC1 ~ pc1$RBC + pc1$Population + pc1$Treatment +
>> > pc1$Population*pc1$Treatment + (1|pc1$ID), data=pc1)
>>
>> If you look at the results of this model:
>>
>> Coefficients:
>>                         (Intercept)                                  RBC
>>                             1.47759                              0.01318
>>                        PopulationWA               TreatmentPBS-Injection
>>                            -1.38234                             -1.85619
>>                          1 | IDTRUE  PopulationWA:TreatmentPBS-Injection
>>                                  NA                              0.24754
>>
>> you can see that something funny is happening to the (1|ID) term ...
>>
>> >
>> > I was hoping I could ask for your assistance in figuring out what may be
>> > the problem. Thank you very much.
>> >
>> > Cheers,
>> > Dan
>> > _______________________________________________
>> > R-sig-mixed-models at r-project.org mailing list
>> > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>> >
>>
>> _______________________________________________
>> R-sig-mixed-models at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>