[R-sig-ME] why does my inter-annual SD = 0?

Tue Oct 12 10:36:40 CEST 2010

Dear Scott,

Thanks for sharing your data. That makes things a lot easier.

IMHO you have several problems with your model. First you have only 4
years. Estimating a variance based on only 4 levels will give you very
unreliable estimates. Therefore it is better to treat them as a fixed
effect.
Another possible problem might be the structure of your nested random
effects. I noticed that many levels of WB have only one level of site,
and many levels of site have only one levels of zone. In such a case
there is competition between the nested effects. I'm not sure how lmer
handles that. If it were just a few cases I guess it will not be a big
issue. But your data has quite a lot of this things. A possible solution
is to use (1|WB:site:zone) as random effect.

HTH,

Thierry

------------------------------------------------------------------------
----
ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek
team Biometrie & Kwaliteitszorg
Gaverstraat 4
9500 Geraardsbergen
Belgium

Research Institute for Nature and Forest
team Biometrics & Quality Assurance
Gaverstraat 4
9500 Geraardsbergen
Belgium

tel. + 32 54/436 185
Thierry.Onkelinx at inbo.be
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

> -----Oorspronkelijk bericht-----
> Van: r-sig-mixed-models-bounces at r-project.org 
> [mailto:r-sig-mixed-models-bounces at r-project.org] Namens Scott Bennett
> Verzonden: maandag 11 oktober 2010 19:39
> Aan: r-sig-mixed-models at r-project.org
> Onderwerp: Re: [R-sig-ME] why does my inter-annual SD = 0?
> 
> 
> 
> Hi all, thanks for the quick replies! I have attached the 
> data, if it helps for you to look at it directly. The end 
> product that I would like to achieve is an estimate of the 
> variance associated with each factor (depth, surveyor, year, 
> water body, site and zone), to then model the probability of 
> misclassifying the health status of a water body, based on 
> the variability associated with each of the respective 
> factors (An uncertainty analysis of our seagrass index). The 
> index we are using POMI_14 (Posidonia oceanica multivariate 
> index, Romero et al. 2008) is comprised of 14 metrics relates 
> to the health and status of /P. oceanica./ There are only 4 
> years of data. The 2005, 2006 and 2008 data is from annual 
> sampling of  30 seagrass meadows (sites) sampled at a single 
> depth along the Catalan coast, Spain. The 2002 data is of a 
> subset of those sites, but includes replication between 2 
> discrete depths (5m and 15m) and among three discrete zones 
> nested within each site. The sites are nested within 
> 'water-bodies'. A water body represents an area of coastal 
> water (15 - 50 km in length) which has been classified based 
> on its exposure to water quality pressures. The surveyor 
> factor, is only from the 2008 series, where we calculated 
> POMI based on two separate surveyors. Needless to say the 
> design is unbalanced.  
> 
> In short the data looks like this:  
> 
> 'data.frame':   231 obs. of  8 variables: 
>  $ year    : Factor w/ 4 levels "2002","2005",..: 4 4 4 4 4 4 
> 4 4 4 4 ...
>  $ WB      : Factor w/ 17 levels "1","2","3","4",..: 1 2 3 3 
> 3 4 5 6 7 8 ...
>  $ Site    : Factor w/ 30 levels "Balis    ","Cadaques ",..: 
> 22 19 27 7 9 2 16 21 20 15 ...
>  $ Zone    : Factor w/ 3 levels "a","b","c":..: 1 1 1 1 1 1 1 
> 1 1 1 ...
>  $ Depth   : Factor w/ 2 levels "p","s": 1 1 1 1 1 1 1 1 1 1 ...
>  $ surveyor: Factor w/ 2 levels "1","2": 1 1 1 1 1 1 1 1 1 1 ...
>  $ POMI_14 : num  0.781 0.633 0.717 0.936 0.86 ...
>  $ POMI_9  : num  0.803 0.67 0.745 0.942 0.873 ... 
> 
> I hope this makes things clearer. Any help will be greatly 
> appreciated.  
> 
> Kind regards  
> 
> Scott Bennett  
> 
> > On 10-10-11 11:32 AM, Scott Bennett wrote:
> >>
> >>
> >> Hi,
> >>
> >> I am applying a mixed model to calculate the variance 
> components of 
> >> different factors in our seagrass data. The model i was 
> using looks 
> >> something like:
> >>
> >> POMI14_vc <- lmer(POMI_14 ~ Depth + surveyor + (1|region/site/zone)
> >> + (1|year), data = P_oceanica)
> >>
> >> When I apply this model, however, year comes out with SD = 
> 0. Year, 
> >> in this data set signifies inter-annual variation (in the health 
> >> status of seagrass meadows), of which there is a 
> considerable amount. 
> >> That makes me believe that there is is a feature of the 
> model which 
> >> is 'absorbing' the inter-annual variation.
> >>
> >> Can you suggest why this may be occuring? What 
> modifiations could i 
> >> use to fix this?
> >>
> >> kind regards
> >>
> >> Scott Bennett
> >
> >    Hard to say for sure without seeing the data.
> >    How many years do you have?  Are Depth and surveyor well 
> > distributed across years?
> >    What happens if you treat year as a fixed effect and 
> calculate the 
> > among-year variance on the basis of the fixed effect estimates?
> >
> >
> >
> >
> >         [[alternative HTML version deleted]]
> >
> > _______________________________________________
> > R-sig-mixed-models at r-project.org mailing list 
> > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
> >
> 
>  
> 