[R-sig-ME] Exponent random effect in nlmer

[Thierry Onkelinx]

Hi Tim,

AFAIK nlmer requires the fixed and random effects to be additive. The model
to be used _after_ this this summation can be non linear.

Best regards,

ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek / Research Institute for Nature and
Forest
team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance
Kliniekstraat 25
1070 Anderlecht
Belgium

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

2016-10-11 12:30 GMT+02:00 Cole, Tim <tim.cole op ucl.ac.uk>:

> Dear Thierry,
>
> Thanks very much for your speedy response.
>
> I agree my model looks odd, but it has a theoretical basis which I'd
> prefer not to spell out at this stage. Suffice to say that
> • -Inf < y < Inf
> • 0 < E(y) < 1
> • there is a subject random effect.
>
> For these reasons the usual models and/or transformations won't work,
> whereas my proposed exponent random effect ought to. I just need to fit it,
> to see if I'm right!
>
> Best wishes,
> Tim
> ---
> Tim.cole op ucl.ac.uk Phone +44(0)20 7905 2666 Fax +44(0)20 7905 2381
> Population, Policy and Practice Programme
> UCL Great Ormond Street Institute of Child Health, London WC1N 1EH, UK
>
>
> From: Thierry Onkelinx <thierry.onkelinx op inbo.be>
> Date: Tuesday, 11 October 2016 11:06
> To: Tim Cole <tim.cole op ucl.ac.uk>
> Cc: "r-sig-mixed-models op r-project.org" <r-sig-mixed-models op r-project.org>
> Subject: Re: [R-sig-ME] Exponent random effect in nlmer
>
> Dear Tim,
>
> y centred on 0 and a valid range (0, 1) seems to be conflicting
> statements.
>
> Here a some solutions depending on y
>
> - y stems from a binomial process
>      - use a binomial glmm.
> - y is continuous and you are willing to transform y
>     - 0 < y <  1
>         - apply a logit transformation on y. lmer(plogis(y) ~ f + (1 |
> id) )
>     - 0 <= y < 1
>         - apply a log transformation on y. lmer(log(y) ~ f + (1 | id) )
>     - 0 < y <= 1
>         - apply a log transformation on 1 - y. lmer(log(1 - y) ~ f + (1 |
> id) )
> - y is continuous are not willing to transform y
>    - use a beta regression with 0 and/or 1 inflation in case you have 0 or
> 1 in the data. Have a look at the gamlss package to fit this model.
>
> Best regards,
>
>
> ir. Thierry Onkelinx
> Instituut voor natuur- en bosonderzoek / Research Institute for Nature and
> Forest
> team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance
> Kliniekstraat 25
> 1070 Anderlecht
> Belgium
>
> 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
>
> 2016-10-11 11:29 GMT+02:00 Cole, Tim <tim.cole op ucl.ac.uk>:
>
>> I have a model of the form
>>   m1 <- lmer(y ~ f + (1 | id) )
>> where y is a continuous variable centred on zero, f is a unordered factor
>> with coefficients b such 0 < b < 1, and there is a signficant random
>> subject intercept.
>>
>> The random intercept can lead to predicted values outside the valid range
>> (0, 1). For this reason I'd like to reformulate the model as
>> m2 <- nlmer(y ~ (f - 1) ^ exp(1 | id) )   (using a invalid but I hope
>> obvious notation), where the random effect is now a power centred on 1.
>> This would constrain the fitted values to be within c(0, 1).
>>
>> My question is: can this be done in nlmer, and if so how? Please can
>> someone point me in the right direction?
>>
>> Thanks,
>> Tim Cole
>> ---
>> Tim.cole op ucl.ac.uk<mailto:Tim.cole op ucl.ac.uk> Phone +44(0)20 7905 2666
>> Fax +44(0)20 7905 2381
>> Population, Policy and Practice Programme
>> UCL Great Ormond Street Institute of Child Health, London WC1N 1EH, UK
>>
>>
>>         [[alternative HTML version deleted]]
>>
>> _______________________________________________
>> R-sig-mixed-models op r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>
>
>

	[[alternative HTML version deleted]]