[R-sig-Geo] Spatial Lags Excluding Neighbors' Missing Attribute Values

[Christopher Moore]

Dear Roger,

I am glad to hear that the NAOK argument will be implemented in the next 
release. Will it be implemented in functions besides lag.listw, such as 
moran.plot?

I agree that "missing data is missing, really," which motivated me to 
develop an ad hoc alternative to treating NA as zero, one that uses all 
available information in a manner similar to pairwise deletion. I can not 
make a case for including that approach in the spdep package (which is 
great, by the way). However, I do think the approach is worthy of further 
methodological inquiry, perhaps as a complement to multiple imputation on 
the attribute side.

Thanks,
Chris

P.S. My apologies for double posting to the list. I mistakenly used an 
uninformative subject line in the first posting.

-- 
Christopher Moore, M.P.P.
Doctoral Student
Quantitative Methods in Education
University of Minnesota
moor0554 at umn.edu
http://www.tc.umn.edu/~moor0554/


On Jan 1 2009, r-sig-geo-request at stat.math.ethz.ch wrote:
>Date: Wed, 31 Dec 2008 12:36:57 +0100 (CET)
>From: Roger Bivand <Roger.Bivand at nhh.no>
>Subject: Re: [R-sig-Geo] Spatial Lags Excluding Neighbors' Missing
>	Attribute Values
>To: Christopher Moore <moor0554 at umn.edu>
>Cc: r-sig-geo at stat.math.ethz.ch
>Message-ID: <alpine.LRH.2.00.0812311223420.5052 at reclus.nhh.no>
>Content-Type: TEXT/PLAIN; format=flowed; charset=US-ASCII
>
>On Tue, 30 Dec 2008, Christopher Moore wrote:
>
>> Greetings,
>>
>> I posted a request for assistance on 12/27/08: 
>> <https://stat.ethz.ch/pipermail/r-sig-geo/2008-December/004733.html>.
>>
>> I developed a satisfactory solution, which is posted below for others 
>> who wish to similarly exclude neighbors' missing attribute values from W 
>> style spatial lags.
>>
>
>Well, this is quite a detailed question, and this isn't the best time of 
>year to get rapid response. I don't think that your solution is 
>satisfactory - and I don't think that there is any viable alternative to 
>either imputing to fill all the missing values (I wouldn't do that), or 
>subsetting out incomplete cases and using the same subset on the "nb" 
>object (there are subset methods). What happens now is that in the 
>compiled code in "lagw" the sum is not incremented when the value is not 
>finite:
>
> 		sum = 0.0;
> 		for (j=0; j<INTEGER_POINTER(card)[i]; j++) {
> 		    k = INTEGER_POINTER(VECTOR_ELT(nb, i))[j];
> 		    wt = NUMERIC_POINTER(VECTOR_ELT(weights, i))[j];
> 		    tmp = NUMERIC_POINTER(x)[k-ROFFSET];
> 		    if (R_FINITE(tmp)) sum += tmp * wt;
> 		}
> 		NUMERIC_POINTER(ans)[i] = sum;
>
>in src/lagw.c. In fact the NAOK argument is not being used, and should be, 
>so in the next release you will either get an error, or, for NAOK=TRUE, 
>you'll get an NA for the lagged value - that was the intention. If you can 
>make a case for a third choice, essentially doing what you describe, I can 
>consider it, but being conservative, missing data is missing, really, 
>isn't it?
>
>Best wishes,
>
>Roger
>