Dear Marina,
The Poisson and negative binomial distributions assume counts =
non-negative integers. Your response variable is not integer. You'll need
the actual counts instead.
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 di 17 nov. 2020 om 11:20 schreef Marina Pastor <marinapastor using icm.csic.es
>:
>
> Hello,
> We are studying the jellyfish distribution and we would like to know
> if it depends of temperature and salinity. As our database are counts
> and we have a high amount of 0, but also other high numbers around
> 400, we used a Negative Binomial Generalised Lineal Model. As the
> filtered volume was not the same for all the counts, we incorporated
> this through an offset.
> Checking the residual distribution we realised we were violating the
> homogeneity assumption since the larger the temperature, the larger
> the variation.
> We first tried to apply different variance structures through
> varFixed, varPower and varExp functions (nlme package) depending on
> the temperature, but we did not manage.
> The reproducible example is below.
> - For varFixed(~Temp) we followed the R help example. Why we obtained
> this error?: Error in `$<-.data.frame`(`*tmp*`, VarFixedT, value =
> numeric(0)) : replacement has 0 rows, data has 10
> - For varPower(1, form =~Temp) and varExp(1, form =~ Temp), how to
> choose the best “value” before “form” in the formula (“1” in the
> example)?
> We would be very grateful if you could help us, we don’t know how to
> improve our model and is one of the requirements to publish our
> article. Many thanks for your time in advance.
> Best wishes,
> Marina Pastor
>
>
> Reproducible example:
> library("nlme")
> library("MASS")
>
> my.df <- data.frame (Jelly = c(1.13, 10.98, 0.00, 0.97, 0.62, 1.04,
> 0.00, 77.83, 4.12, 0.18), Temp = c(24.63, 24.61, 24.63, 25.64, 25.63,
> 26.22, 26.17, 25.34, 25.44, 25.09), Sal = c(37.16, 36.79, 38.06,
> 38.20, 38.15, 38.26, 38.25, 38.10, 38.07, 37.96), Vol = c(971.0,
> 965.5, 835.0, 823.0, 640.0, 1147.0, 1322.0, 912.0, 1018.0, 1095.0))
>
>
> my.df$VarFixedT <- varFixed(~Temp)
> GLMNB_W <- glm.nb(Jelly ~ Temp + Sal +
> offset(log(Vol)),
> data= my.df,
> weights = VarFixedT)
>
> my.df$VarPowerT <- varPower(1, form =~Temp)
> GLMNB_W <- glm.nb(Jelly ~ Temp + Sal +
> offset(log(Vol)),
> data= my.df,
> weights = VarPowerT)
> summary(GLMNB_W)
>
> my.df$VarExpT <- varExp(1, form =~ Temp)
> GLMNB_W <- glm.nb(Jelly ~ Temp + Sal +
> offset(log(Vol)),
> data= my.df,
> weights = VarExpT)
> summary(GLMNB_W)
>
> --
> Marina Pastor
> PhD Student
> Marine biology and oceanography department
> Institut de Ciències del Mar (ICM-CSIC)
> Spanish National Research Council
> Passeig Marítim de la Barceloneta 37-49, E-08003 Barcelona, Catalonia,
> Spain
> Phone: +34 932309500 (ext. 1113)
> E-mail: marinapastor using icm.csic.es
>
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
[[alternative HTML version deleted]]