RE: st: spatial weights which also take into account time

[Simon Falck <simon.falck@abe.kth.se>]

Dear Ben,

If you are working on spatial econometrics and spatial panel models in particular, then I suggest you to take a look of the work of professor J. Paul Elhorst, who has made done considerable contributions in this field(s). His work covers both theory and useful applications, which typically are implemented using Matlab. Partly because the relative ease in Matlab to constrain spatial matrices, which slow down computation. Nevertheless there are several useful spatial econometric applications in Stata, as suggested by Nick.

Simon


-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Nick Cox
Sent: den 31 januari 2013 19:19
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: spatial weights which also take into account time

I doubt that you are missing anything because I said almost everything I had in mind. All that I was suggesting was that if you want spatial weights to be based only on recent data the calculation should be restricted by an -if- condition that selects observations that are recent.

You had a general question on weights in space and time and a specific question on -spwmatrix-, which is fine by me. But I didn't write and have never used the latter and, as said, don't advise specifically on it. For better advice, you may need to email P. Wilner Jeanty if he or some other expert doesn't speak up.

Nick

On Thu, Jan 31, 2013 at 6:08 PM, Ben Hoen <bhoen@lbl.gov> wrote:
> Thanks Nick,
>
> Though, I am not sure think I understand.  Are suggesting calculating 
> a knn weights matrix and interacting it with some expression that 
> would limit the data based on time?  Would this account for the fact 
> that each interaction is based on the target case's "time" (i.e., only 
> those nn that transacted within the previous 1 year from the target 
> cases date) and therefore would be a "moving window" of time for each transaction?
>
> I expect that I am missing some significant aspect of your response 
> b/c of my limited understanding of how to employ weights.
>
> Ben
>
> Ben Hoen
> LBNL
> Office: 845-758-1896
> Cell: 718-812-7589
>
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Nick Cox
> Sent: Thursday, January 31, 2013 12:05 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: Re: st: spatial weights which also take into account time
>
> A very general answer is
>
> 1. Many programs support weights of various kinds.
>
> 2. If you want your weights to depend on several variables, that's 
> often fine: you just have to compute them.
>
> 3. The syntax is typically [<weighttype> = <exp>] so that the <exp> 
> can depend on several variables. Often it would be a product or a 
> dividend.
>
> 4. The general syntax (literally -syntax-) for checking weights and 
> the specific ways that weights are handled within a program should be 
> considered separate matters. (At the worst end, I've seen user 
> programs support specification of weights but then ignore them, it 
> being supposed by novice programmers that it's sufficient to allow 
> weights via a -syntax- statement.)
>
> All that said, your specific problem sounds like a case for using an
> -if- restriction, not more complicated weights, but I can't comment 
> authoritatively on -spwmatrix-.
>
> Nick
>
> On Thu, Jan 31, 2013 at 4:51 PM, Ben Hoen <bhoen@lbl.gov> wrote:
>
>> I have been unsuccessful in my search this morning for a way to deal 
>> with both space and time (or for that matter any other variable other 
>> than
> space)
>> in developing a spatial weights matrix.  ALERT: I am a novice - 
>> slowly learning - about spatial models.
>>
>> The k nearest neighbors options in some user written programs (e.g., 
>> spwmatrix - SSC) allows me to limit the number of neighbors based on 
>> "nearness".  Alternatively I could use some cutoff distance (e.g., 1 
>> mile) to limit the collection of "neighbors".  But in addition to 
>> distance I
> would
>> like to use time to limit my data.  For example I have home 
>> transactions that I want to test for spatial autocorrelation between.  
>> I concern myself with BOTH the k (e.g., 5) nearest neighbors that 
>> ALSO transacted within
> the
>> previous year.  (I.e., using a transaction that occurred, for 
>> example, 10 years later does not make sense, even if it is 
>> "nearest".)
>>
>> Are there any ideas out there for how to limit the data based on both
> space
>> AND time in developing the weights matrix?
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/