[R-sig-ME] Single DV with multiple measures for time-varying IV?
Thierry Onkelinx
thierry@onkelinx @ending from inbo@be
Thu Nov 22 09:47:03 CET 2018
And even if the model would converge, then it still is wrong to do it. When
modelling the calf birth weight, then one calf = one observation. A calf is
born once.
In case you have plenlty of births, then you could try to model the
cortisol values and use that model to impute missing values. And then model
the birth weight using the augemented data. Make sure to use multiple
imputation to do so! Otherwise the confidence intervals will be too small.
See e.g. Rubin 1987 and Onkelinx et al 2017 (doi:10.1007/s10336-016-1404-9)
However given that you have only 60 observations, you have too few
observations to take each of the eight months into account.
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 using inbo.be
Havenlaan 88 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>
Op wo 21 nov. 2018 om 18:50 schreef Ben Bolker <bbolker using gmail.com>:
>
> Just for the record; I agree that it's almost definitely *not* going
> to work to have identical responses for different predictor values.
> Someone asked a similar question on StackOverflow recently:
>
> https://stackoverflow.com/questions/53034261/warning-lme4-model-failed-to-converge-with-maxgrad
>
> cheers
> Ben Bolker
>
> On 2018-11-21 10:28 a.m., Pero, Ellen wrote:
> > Thank you Bill and Thierry.
> >
> >
> > I don't yet have data in hand (cortisol samples await assay). However,
> this is what they will look like:
> >
> > cortisol
> > ---------------------------------
> > ID DV Month 1 Month 2 ... Month 8 dam age, sire age, calf
> birthdate
> > 1 ....
> > 2 ....
> > .. ....
> > 60 ....
> >
> > While I can simulate more data, my primary question is theoretical:
> >
> > Is it acceptable practice to share a single dependent response (DV: here
> calf mass (kg)) amongst multiple time-varying nested independent predictors
> (here, monthly cortisol) as long as I place a random effect to signify the
> individual I am nesting on (ID).
> >
> >
> >
> > ID DV cortisol, time, dam age, sire age, calf birthdate
> > 1 17 35 Month 1 4 3 140
> > 1 17 42 Month 2 4 3 140
> > ........................................................
> > 1 17 58 Month 8 4 3 140
> >
> > 2 19 30 Month 1 3 5 150
> > 2 19 33 Month 2 3 5 150
> > ........................................................
> > 2 19 42 Month 7 3 5 150
> >
> > ........................................................
> >
> > 60 14 51 Month 2 2 2 162
> > 60 14 58 Month 3 2 2 162
> > ........................................................
> > 60 14 70 Month 8 2 2 162
> >
> > From my digging, I don't think it is good practice. So, for now, I am
> planning to average repeated cortisol samples within an individual to
> produce an 'early' and 'late' value, and include both as covariates within
> a glm.
> >
> >
> > I appreciate your support and encouragement!
> >
> > El
> >
> >
> > Ellen Pero
> > PhD Student
> > Wildlife Biology Program
> > W.A. Franke College of Forestry and Conservation
> > University of Montana
> > 32 Campus Drive, FOR 318
> > Missoula, MT 59812
> >
> >
> >
> > ________________________________
> > From: Bill Poling <Bill.Poling using zelis.com>
> > Sent: Monday, November 19, 2018 4:26 AM
> > To: Pero, Ellen
> > Cc: Thierry Onkelinx; r-sig-mixed-models using r-project.org; Bill Poling
> > Subject: RE: [R-sig-ME] Single DV with multiple measures for
> time-varying IV?
> >
> > Hi Ellen.
> >
> > If the data frame is not too terribly large, a dput() would be useful.
> > See ?dput()
> > Or a str() would help as well
> > See ?str()
> > However, as Thierry suggests a subset of your data would be most helpful.
> >
> > I will be interested to follow this topic as I am teaching myself R and
> learning the various modeling methods and their purposes along the way.
> >
> > I think you will gain considerable support from this list relevant to
> your topic.
> >
> > Best regards.
> >
> > WHP
> >
> >
> > From: R-sig-mixed-models <r-sig-mixed-models-bounces using r-project.org> On
> Behalf Of Thierry Onkelinx via R-sig-mixed-models
> > Sent: Monday, November 19, 2018 4:01 AM
> > To: ellen.pero using umconnect.umt.edu
> > Cc: r-sig-mixed-models <r-sig-mixed-models using r-project.org>
> > Subject: Re: [R-sig-ME] Single DV with multiple measures for
> time-varying IV?
> >
> > Dear Ellen,
> >
> > An extract of your dataset or a small dummy dataset coverting the
> important
> > features of your data would make it much easier to answer your questions.
> > And please don't send HTML emails. Any HTML formating gets stripped which
> > can make your email very hard to read.
> >
> > 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
> > mailto:thierry.onkelinx using inbo.be
> > Havenlaan 88 bus 73, 1000 Brussel
> > http://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>
> >
> >
> > Op ma 12 nov. 2018 om 20:01 schreef Pero, Ellen <
> > mailto:ellen.pero using umconnect.umt.edu>:
> >
> >> Hi all:
> >>
> >> I have an analytical dilemma wherein I have a single DV with multiple
> >> categorical and continuous IVs (one of which is a continuous IV that has
> >> multiple measurements across time). I'm not sure the best way to model
> for
> >> this - though it's clearly a hierarchical situation so I thought this
> might
> >> be a good venue to pose the question.
> >>
> >> Specifically, I have 60 pregnant elk from which I took monthly cortisol
> >> samples across gestation (some missing values, so 5-8 samples/female
> across
> >> gestation). I'm interested in how those stress measurements across
> >> gestation (along with a range of other IVs that don't vary with time,
> e.g.,
> >> dam age, sire age, calf birthdate) influence the birth mass of each
> >> female's calf.
> >>
> >> Any suggestions on analysis for situations where a single DV is
> predicted
> >> by longitudinal measures of time-varying IV (along with non-varying
> IVs)?
> >>
> >> I'm new to this list and will spend some time familiarizing myself with
> it
> >> - but was eager to get my question out. Apologies if this isn't the
> right
> >> venue for my non-development related question. Please disregard if
> >> appropriate.
> >>
> >> I appreciate any thoughts/advice/suggestions!
> >> El
> >>
> >>
> >>
> >> Ellen Pero
> >> PhD Student
> >> Wildlife Biology Program
> >> W.A. Franke College of Forestry and Conservation
> >> University of Montana
> >> 32 Campus Drive, FOR 318
> >> Missoula, MT 59812
> >>
> >>
> >> [[alternative HTML version deleted]]
> >>
> >> _______________________________________________
> >> mailto:R-sig-mixed-models using r-project.org mailing list
> >> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
> >>
> >
> > [[alternative HTML version deleted]]
> >
> > _______________________________________________
> > mailto:R-sig-mixed-models using r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
> >
> > Confidentiality Notice This message is sent from Zelis. ...{{dropped:16}}
> >
> > _______________________________________________
> > R-sig-mixed-models using r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
> >
>
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
[[alternative HTML version deleted]]
More information about the R-sig-mixed-models
mailing list