Skip to content

error in errorsarlm

5 messages · elaine kuo, Nikhil Kaza, Roger Bivand

Original
Nikhil Kaza
# 
example(errorsarlm)

Nikhil Kaza
Asst. Professor,
City and Regional Planning
University of North Carolina

nikhil.list at gmail.com
On Aug 1, 2010, at 7:57 AM, elaine kuo wrote:

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Roger Bivand
    

    

      
      #
    

  


  

      
On Sun, 1 Aug 2010, Nikhil Kaza wrote:

        

            

      
      
      
    

        
Right!

Or any other available dataset that mirrors the application case, which 
seems to be ecological, but crucially has point or cell support in decimal 
degrees with very large differences in distances between observations. You 
are right that packages include lots of example data sets, and that they 
are the place to start. In addition, the case is large, but the example 
should not be, so that all the other problems can be eliminated first 
before trying to scale up to moderate N (about 4000 if I recall 
correctly).

The main problem is that the spatial process is conceptualised as 
contagious in distance, and that the distance threshold is chosen to 
ensure that all observations have at least one neighbour. This doesn't 
seem helpful as many observations then have "too many" neighbours to make 
sense. If the coordinates of the observations can be projected to the 
plane, a graph-based neighbour scheme can be used, which represents the 
spatial process as contagious in contiguity (neighbouring observations are 
neighbours irrespective of distance).

So the first step is to think through how "neighbouring" observations can 
influence each other in terms appropriate to the subject domain.

Roger

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    
10 days later

    

      
      

        
      

    

    

      
      

        


  

        

      Roger Bivand
    

    

      
      #
    

  


  

      
On Thu, 12 Aug 2010, elaine kuo wrote:

        

            

      
      
      
    

        
Or more recently Bivand et al. (2008) Applied Spatial Data Analysis with R 
(Springer), ch. 9-10, (see www.asdar-book.org).

I am concerned that you still do not know what you are doing!

You continue to use lat as your x dimension and lon as your y dimension, 
this is very confused and most likely wrong. In addition, because you do 
not work on this thread regularly (gaps of days or weeks), it is difficult 
to tell whether you are still working in decimal degrees, in which case 
your whole analysis is compromised - you should use cbind(lon, lat) then 
longlat=TRUE in all the neighbour finding steps to use Great Circle 
distances.

As I said at the beiginning of the thread, your use of arguments in 
functions is sloppy, and the reason why things now work is because you 
have changed some arguments. Whether the changes were appropriate or not 
is a very different question.

Roger