Skip to content

Autologistic regression in R

7 messages · Mingke Li, Elias T Krainski, Bede-Fazekas Ákos +2 more

#
Hi,

I am new to autologistic regression and R. I do have questions when starting a project in which I believe autologistic regression (spdep package) is needed.

I have a point layer whose attribute table stores the values of the dependent variable (population of a kind of insect), all the independent variables (environmental factors), and the associated latitude and longitude. I hope to to fit an autologistic model to analyze which factors or combinations of factors have effects on the presence/absence of the insect (1 or 0).

I found other papers which applied autologistic regression in their study almost used a grid system and defined their window sizes. So, my question is do I have to convert my point layer into a grid system if I want to do this analysis with R?

Also, what should I consider when I generate the grid system? How to determine a proper cell size? How about the searching window (neighbourhood) size?

Many Thanks.

Erin
#
Hi,

Have you tried http://leg.ufpr.br/Rcitrus ?

Elias
On 15/11/17 04:46, Mingke Li wrote:
#
Hi Erin,

Although I'm not familiar with autologistic regression, the task you 
outlined seems to be a Species Distribution Modeling (SDM) problem for 
which several methods are available in R. So I'm not sure if you need 
only and exclusively autologistic regression. The SDM methods can solve 
the equation /pres = f(env1, env2, ...)/, where /pres /is 
presence/absence of the species and/envN/s are environmental predictors. 
See packages "/dismo/" and "/biomod2/", and also /vignette("sdm")/. Some 
of the methods need data.frame or SpatialPointsDataFrame, other ones 
need Raster* objects as input.

HTH,
?kos Bede-Fazekas
Hungarian Academy of Sciences

2017.11.15. 7:04 keltez?ssel, Elias T Krainski ?rta:

  
  
#
Hi Erin,

It is not quite clear to me what your data is. From your text I 
understand that you have a number of locations where you have measured 
the population of a specific insect (count variable?) together with 
independent/explanatory variables at these same locations. Is the 
"population" sometimes zero? Is it even restricted to be binary (0/1), 
which I guess would be required for logistic regression to make sense?

Cheers,
Ege
On 11/15/2017 02:46 AM, Mingke Li wrote:
#
Hi Ege,


Thanks for replying. Sorry for not clearfying the data.


The raw response data is the actural population counted at the specific sample points. Because my study object is spruce budworm which would cause forest defoliation, what we are interested in are those sample points where spruce budworm population is greater than 7 per brunch. Some of the sample points have zero population, and some have population fewer than 7 per brunch, but these samples don't attract us because spruce budworm are always there; fewer than 7 per brunch is not a problem for forests.


So what I understand is to set 7 as a threshold to transform the raw response data to 0 (population<7) and 1(population>=7). Also some papers point out the significance of encountering the spatial autocorrelation when dealing these species distribution problem, that is why I come across the autologistic regression.


This approach is so new to me, so I may have some misunderstanding. Thanks again.


Erin
#
OK. At least that helps people understand the basic properties of the data.
The sampling design could also be important for the choice of method. 
You should probably make this part a bit more clear and answer questions 
like:
- Did you make a sparse random sample of locations in a big area?
- Or did you sample a predefined small area extensively?
- Etc...
I can't really offer any specific help, but I would advise you to add a 
bit more detail in case it rings a bell for a helpful soul out there.
/Ege
On 11/15/2017 04:44 PM, Mingke Li wrote:
#
Hi Mingke Li,

Dorman et al. (2007) review different methods to account for spatial 
dependence in linear models, including the use of autocovariables, and 
provide R code for them.

HTH,

Marcelino

Dormann, et al. 2007. Methods to account for spatial autocorrelation in 
the analysis of species distributional data: a review. ? Ecography 30: 
609?628.

http://www.ecography.org/appendix/e5171



El 15/11/2017 a las 2:46, Mingke Li escribi?: