Skip to content

Error in Spatial lag model

5 messages · Miluji Sb, Roger Bivand

Original
Miluji Sb
# 
I have a dataset as below (with NAs). I am trying to run a spatial lagged
model

##
E2_mod_pdssat <- as.data.frame(gcp_t_p[,c(2,3)])
names(E2_mod_pdssat) <- c("Lon", "Lat")
row.names(E2_mod_pdssat) <- NULL

coords_matrix  <- as.matrix(E2_mod_pdssat) # create coordinate matrix
IDs <- row.names(as(E2_mod_pdssat, "data.frame"))

spatial_knn2nb_obj <- knn2nb(knearneigh(coords_matrix, k=2, longlat=TRUE),
row.names=IDs)

# Now convert the nb to listw for use in spatial regression call.
# Spatial weights, illustrated with coding style "W" (row standardized)

spatial_knn2listw_obj <- nb2listw(spatial_knn2nb_obj, glist=NULL,
style="W",  zero.policy=TRUE)

E2_mod_pdssat_org_lon_lat <- E2_mod_pdssat_org[, c(2:3)] # The lon and lat
will be required
E2_mod_pdssat_org <- E2_mod_pdssat_org[, -c(2,3)]  # Remove the Lon, Lat
columns that are not required

E2_mod_pdssat_pdata_frame <- pdata.frame(E2_mod_pdssat_org) # Convert data
frame to pdata.frame

All goes well until here but when I try to convert the explanatory
variables in the pdata.frame to spatial lagged variables

E2_mod_pdssat_X_lagged <-
as.data.frame(cbind(slag(E2_mod_pdssat_pdata_frame$tmean, listw =
spatial_knn2listw_obj),
slag(E2_mod_pdssat_pdata_frame$tmeansq, listw = spatial_knn2listw_obj),
slag(E2_mod_pdssat_pdata_frame$pmean, listw = spatial_knn2listw_obj),
slag(E2_mod_pdssat_pdata_frame$pmeansq, listw = spatial_knn2listw_obj)))

I get the following error "Error in lag.listw(listw, xt) : object lengths
differ". What am I doing wrong? I have also tried removing the NAs with
complete.cases.

Any help will be greatly appreciated. Thank you.

Sincerely,

Milu

## Data

E2_mod_pdssat_org <- dput(head(temp,5))
structure(list(iso3 = structure(c(30L, 30L, 30L, 1L, 30L), .Label = c(".",
".M", "AFG", "AGO", "ALB", "ARE", "ARG", "ARM", "AUS", "AUT",
"AZE", "BDI", "BEL", "BEN", "BFA", "BGD", "BGR", "BHS", "BIH",
"BLR", "BLZ", "BOL", "BRA", "BRN", "BTN", "BWA", "CAF", "CAN",
"CHE", "CHL", "CHN", "CIV", "CMR", "COD", "COG", "COL", "CRI",
"CUB", "CYP", "CZE", "DEU", "DJI", "DNK", "DOM", "DZA", "ECU",
"EGY", "ERI", "ESH", "ESP", "EST", "ETH", "FIN", "FJI", "FLK",
"FRA", "GAB", "GBR", "GEO", "GHA", "GIN", "GNB", "GNQ", "GRC",
"GRL", "GTM", "GUF", "GUY", "HND", "HRV", "HTI", "HUN", "IDN",
"IND", "IRL", "IRN", "IRQ", "ISL", "ISR", "ITA", "JAM", "JOR",
"JPN", "KAZ", "KEN", "KGZ", "KHM", "KIR", "KOR", "KWT", "LAO",
"LBN", "LBR", "LBY", "LCA", "LKA", "LSO", "LTU", "LUX", "LVA",
"MAR", "MDA", "MDG", "MEX", "MKD", "MLI", "MMR", "MNE", "MNG",
"MOZ", "MRT", "MWI", "MYS", "NCL", "NER", "NGA", "NIC", "NLD",
"NOR", "NPL", "NZL", "OMN", "PAK", "PAN", "PER", "PHL", "PNG",
"POL", "PRI", "PRK", "PRT", "PRY", "QAT", "ROU", "RUS", "RWA",
"SAU", "SDN", "SEN", "SJM", "SLB", "SLE", "SLV", "SOM", "SRB",
"SUR", "SVK", "SVN", "SWE", "SWZ", "SYR", "TCD", "TGO", "THA",
"TJK", "TKM", "TLS", "TUN", "TUR", "TWN", "TZA", "UGA", "UKR",
"URY", "USA", "UZB", "VEN", "VNM", "VUT", "YEM", "ZAF", "ZMB",
"ZWE"), class = "factor"), lon = c(-69L, -68L, -72L, -71L, -70L
), lat = c(-55L, -55L, -54L, -54L, -54L), tmean = c(NA, NA,
4.55847677595631,
4.06015710382516, 3.71010245901642), pmean = c(NA, NA, 2.45569896697998,
1.6384220123291, 1.43925595283508), lngdp = c(-0.102380074560642,
-2.3195972442627, 0.0122363297268748, -0.483756244182587, -6.29744625091553
), tmeansq = c(NA, NA, 20.7797107696533, 16.4848766326904, 13.7648601531982
), pmeansq = c(NA, NA, 6.03045749664307, 2.68442678451538, 2.07145762443542
)), .Names = c("iso3", "lon", "lat", "tmean", "pmean", "lngdp",
"tmeansq", "pmeansq"), row.names = c("1", "2", "3", "4", "5"), class =
"data.frame")
Roger Bivand
#
On Sun, 20 Nov 2016, Miluji Sb wrote:

            

      
      
      
    

        
Did you examine this object? What did the print method tell you?

            

      
      
      
    

        
This object "_org" does not yet exist ...

            

      
      
      
    

        
So you omitted to say that the data are (unbalanced) panel data ...

            

      
      
      
    

        
You have defined weights for one object (possibly a cross section), and 
then try to use them on a different object with a different number of 
rows. If you really want to do this, you need a Kronecker product of the 
spatial weights to take the panel structure of your data into account. 
Further, I doubt strongly that you need to create WX manually, because 
spatial lags of all the required variables are created by model fitting 
functions in the splm package.

Roger

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Miluji Sb
    

    

      
      #
    

  


  

      
Dear Roger,

Thank you for your quick reply.

#
print(spatial_knn2listw_obj)

Characteristics of weights list object:
Neighbour list object:
Number of regions: 10927
Number of nonzero links: 21854
Percentage nonzero weights: 0.01830329
Average number of links: 2
Non-symmetric neighbours list

Weights style: W
Weights constants summary:
      n        nn    S0      S1    S2
W 10927 119399329 10927 10376.5 44660
#

The data is actually cross-sectional, is my mistake converting it
to pdata.frame? I have not tried the splm package, which is a mistake on my
part. I will try that now. Thanks.

Sincerely,

Milu

        
On Sun, Nov 20, 2016 at 8:29 PM, Roger Bivand <Roger.Bivand at nhh.no> wrote:

        

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

  
  


  


      

    

    

      
      

        


  

        

      Roger Bivand
    

    

      
      #
    

  


  

      
On Sun, 20 Nov 2016, Miluji Sb wrote:

        

            

      
      
      
    

        
If the data are not a spatial panel, then do not use splm, it is not 
relevant. If you have NAs in the response, you will need to address that 
once you can fit a model with complete cases. You may not have NAs in the 
covariates. I guess that the mistake was slag() and pdata.frame(), but 
your example is no help at all. Are you copying poor code from the 
internet? Why use k=2 (seems very ill-judged) - knn are intrinsically 
asymmetric, why 2?

Roger

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Miluji Sb
    

    

      
      #
    

  


  

      
Thanks again for your reply. k=4, I changed that to check what was throwing
the error.

Sorry, this is my first attempt at this - I am reading up and trying to
execute. I would appreciate any guidance. Thanks again.

Sincerely,

Milu

        
On Sun, Nov 20, 2016 at 9:05 PM, Roger Bivand <Roger.Bivand at nhh.no> wrote: