[R-sig-ME] model specification for continuous environmental vars

[Tim Howard]

Seth,
Thank you for the pointers and help. I've found the Atkins et al. tutorial and 
will go through it. 
 
Based on the earlier responses, I'm finding success with glmmadmb as 
well so I'm glad you are indicating the same. I'll continue down that path. 
 
Your question to test and model specification with the anova test also helps, thanks!
 
Tim

>>> "Seth Bigelow" <seth at swbigelow.net> 12/13/2013 2:36 PM >>>
Re: the plant count & summits discussion:

-I've had good results with glmmadmb & negative binomial distribution,
though the zero-inflation addition has not worked, probably because of a
small dataset.

- Is slope incorporated into the calculation of solar radiation? If so it
might be feasible to drop one of these terms to simplify the model

- There may be a quadratic relationship with elevation and plant count, it
would be wise to test for this (e.g., make models with and without quadratic
elevation term and use <- anova(linearmodel, quadraticmodel) to test.

- the David Atkins et al. tutorial on count regression, often mentioned on
this list, is highly recommended 

- one way of putting the questions would be, 'does the relationship between
plant count and elevation change from the first survey to the second?
Syntax:

glmmadmb(count~sample*elevation + solar.radiation + (1|summit), data=dat,
family="nbinom", zeroInflation=TRUE) 

...and test this vs. the no-interaction model.

--Seth 


-----Original Message-----
From: r-sig-mixed-models-bounces at r-project.org
[mailto:r-sig-mixed-models-bounces at r-project.org] On Behalf Of Tim Howard
Sent: Friday, December 13, 2013 1:03 PM
To: r-sig-mixed-models at r-project.org
Subject: Re: [R-sig-ME] model specification for continuous environmental
vars


>>>> Hans Ekbrand <hans.ekbrand at gmail.com> 12/13/2013 12:42 PM >>>
>On Fri, Dec 13, 2013 at 08:40:55AM -0500, Tim Howard wrote:
>> >From: Hans Ekbrand <hans.ekbrand at gmail.com> On Thu, Dec 12, 2013 at 
>> >01:39:04PM -0500, Tim Howard wrote:
>> >> List members -
>> >> I am learning a lot, quickly, but still have a way to go. I would 
>> >> greatly appreciate some help with model specification in glmer.
>> >> I can't find a good example that parallels what I've got.
>> >> 
>> >> My dataset consists of spatially-balanced random samples of rare 
>> >> plants within alpine summits. There were two sampling bouts (yr1 
>> >> and yr2) with yr2 collected 6 years after yr1. A new set of random 
>> >> plots were collected at each bout (e.g. new estimate of the 
>> >> population, not repeated measures). I would like to test the
difference in plant density from yr1 to yr2, overall.
>> >
>> >If you only want to test if there is a difference in plant density 
>> >between yr1 and yr2, then I don't think you should include the 
>> >environmental variables, since difference in the outcome that 
>> >relates to changes in the environmental variables between yr1 and 
>> >yr2 would be attributed to the enviromental variables and "hide" the 
>> >actual differences in outcome between yr1 and yr2.
>> >
>> >mod <- glmer(count ~ samp + (1|summit), data = dat, 
>> >family="poisson")
>> >
>> >would be more appropriate, I think.
>> >
>> >Inclusions of the envirmental variables should only be done if you 
>> >want to explain differences between yr1 and yr2, not for estimating 
>> >their size.
>> These are good points. My reasoning was that I need to control for 
>> these variables somehow. What if I have higher densities in yr2 but 
>> it is completely due to sampling -- that my plots happened to be at 
>> elevations where there are higher densities? I want to remove the 
>> effect of differences in elevation when testing for the differences
between yr1 and yr2.
>
>OK, I thought about real changes between yr1 and yr2, not artifacts due 
>to sampling. elevation and slope does not vary over time, but solar 
>radiation could vary over time, I guess.
>
>If your research question is about changes over time, then why did you 
>change the sampled areas between the measure points?

I am following the idea of Probabilistic Survey designs championed by USEPA
(authors are primarily Kincaid and Olsen, randomization method is GRTS using
R package spsurvey). The main idea is that you get an estimate of the target
population through a random sample at T1 and then another estimate of the
population through a separate random sample at T2. 

http://www.epa.gov/nheerl/arm/designpages/monitdesign/survey_overview.htm 

Perhaps that was a bad idea! :)  But, it actually would have been VERY hard
to visit the exact same locations again and again so this approach does make
sense in this context. 

Best,
Tim



_______________________________________________
R-sig-mixed-models at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models