topic Generate Spatial Weights Matrix Space time window - adjust direction of temporal data? in ArcGIS Pro Questions

Generate Spatial Weights Matrix Space time window - adjust direction of temporal data?

SJoseph — Wed, 28 Jul 2021 08:32:25 GMT

Hello!

I am running a spatial regression over time for my research, and I'm using ArcGIS Pro to generate the spatial weights matrix. My data is over 20 years, gathered every 2 years (1999, 2001, ... , 2019) and analyzes the relationship of sales from food stores (food stores are represented as points on the map, gathered at discrete moments in time). I cannot assume that the location of a future retailer has an influence on a past one (i.e. the location of retailer x in 2017 has no influence on retailer y in 2001 (even if their locations at collection are close to each other).

I am using the Space Time Window for the Conceptualization of Spatial Relationships parameter. To be safe, I just select the option so that their are positive weights associated only between neighboring retailers gathered in the same time period (ex. only 1999 shop points can have a relationship with each other). However, there may be some influence of a past retailer location on a later one (within a reasonable time period, such as a location from 2001 could have an influence on a location in 2003).

It appears there is no option to account for this potential unidirectional ("yesterday can exert influence on today’s pattern, but the inverse may not be possible"). Does anyone have a suggestion for this?

Re: Generate Spatial Weights Matrix Space time window - adjust direction of temporal data?

DanPatterson — Wed, 28 Jul 2021 22:26:59 GMT

I don't suppose that reversing the time sequence would be an easy option?

Re: Generate Spatial Weights Matrix Space time window - adjust direction of temporal data?

SJoseph — Mon, 02 Aug 2021 08:48:14 GMT

Hi Dan,

Thanks very much for your reply. What do you mean reversing the sequence?

One issue I have with the tool is that you aren't able to indicate a direction of the time sequence. My data is collected every two years, so if I put that as the time period it will indicate a relationship from the previous collection and the future (which logically doesn't make sense in my case because it is very unlikely that a future location will have an influence on the current (or at least one as strong as a previous one from the same time period).