[R-sig-Geo] error using predict.sarlm

[Roger Bivand]

On Fri, 24 Apr 2009, evans324 at umn.edu wrote:

> Hi Roger,
>
>> Then the dak_test_data is a SpatialPolygonsDataFrame, for which there is no 
>> nrow() method, leading to problems. Please get around this by saying 
>> newdata=as(dak_test_data, "data.frame") to coerce it to data frame. I'll 
>> look at providing dim methods for Spatial*DataFrame objects other than 
>> SpatialPointsDataFrame, which are in sp already; nrow() just returns 
>> dim()[1] of the object.
>
> Thanks for your help so far. I tried your suggestion using the following 
> code:
>
> dak_test_vars.10nn.predict <- predict.sarlm(reg_vars_formula.lag, 
> newdata=as(dak_test_vars, "data.frame"), listw = 
> dak_test.10nn.listw,zero.policy=TRUE)
>
> Now I get a new error message: "Error in invIrW(listw, object$rho) %*% trend 
> : non-conformable arguments". I double checked the listw object and data 
> frame properties. They are as follows:
>
> nrow(as(dak_test_vars, "data.frame") = 2255
> length(dak_test.10nn.listw)
> [1] 3
> names(dak_test.10nn.listw)
> [1] "style"      "neighbours" "weights"   length(dak_test.10nn.listw[[2]])
> [1] 2255
> length(coords_test)
> [1] 4510

The lengths are OK now - the final one is just 2*2255, so the length of 
all the elements in a two-column matrix.

The problem is occurring in multiplying (I - rho W)^{-1} by the product of 
the newdata X values and the fitted regression coefficients - and suggests 
that something else is going wrong earlier. If you know how to run 
debug(), then do debug(predict.sarlm) and step through, checking the 
dimensions and values of B and X used to create trend. If this sounds 
scary, do:

save(reg_vars_formula.lag, dak_test_vars, dak_test.10nn.listw, 
file="predict_problem.RData")

and attach the saved file to me offlist, and I'll try to see what is going 
on.

Roger

>
> I am unsure of how to interpret this error. Do you have any advice? Thanks 
> for your help.
>
> Heather
>
>
>
>
>

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no