Emerging Hot Spot Analysis Neighborhood Time Step

[TeriLandrum]

When using the Emerging Hot Spot Analysis (from within the Space Time Pattern Mining Tools in ArcGIS Pro 1.3.1) on 10 years worth of inner city crime data, I notice that I get significantly different results when I change the Neighborhood Time Step. For instance, when I use 1 year as my Neighborhood Time Step, large areas are coded as Diminishing Hot Spots but when I use 2 years many of these spots change to Persistent Hot Spots. When I use 3 years, the entire area is coded as a Persistent Hot Spot (I also have a larger concentration of Intensifying Hot Spots). I am unsure of how to interpret these results. Any help in interpreting these results as well as guidance on how to choose and justify a time step interval would be appreciated. FYI, there has been an overall reduction in the occurrence of crimes from 2006 (232,577) to 2015 (182,349) although some neighborhoods/census tracts have seen an increase.

[DanPatterson_Retired]

From here How Emerging Hot Spot Analysis works—Help | ArcGIS for Desktop 

the greater the time aggregation the more likely that 'persistent' will arise.  This will obviously affect your interpretation and the placement of the observations

What is a z-score? What is a p-value?—Help | ArcGIS for Desktop provides some of the statistical background

Perhaps it might be useful to look at some of the videos and links in the ...

Spatial Statistics Resources | ArcGIS Blog 

I wouldn't hone in on trying to pick an aggregation period.

[TeriLandrum]

Thank you for your reply, Dan. I assume that since Emerging Hot Spot Analysis is a tool for identifying trends that I should be able to do just that with some confidence or at least a logical explanation of why the trends I see are different depending on the length of the aggregation period. If I use the default Neighborhood Time Step of 1, I see a trend that would lead me to believe that a particular area of concern is a Diminishing hot spot. One thing I noticed when examining the Emerging Hot Spot results when using a time step of 3 is that 65% of the bins coded as Persistent Hot Spots have a Z-score that is less than zero. I do not know if this is coincidence or not but those are nearly the same bins that are coded as Diminishing Hot Spots when I use a time step of 1. This would lead me to believe/conclude that these particular bins are tipping towards a downward trend and that it would be valid to use the results from the 1 year aggregation period (time step of 1) to identify trends in my data. Agree or disagree?

[DanPatterson_Retired]

arrrrgh... I can sort of see what you are talking about without actually seeing it... so it is diminishing when it is a time step of one which would suggest that it would be persistent over the time step of 3 but diminishing at the same time... so I would agree.  Aggregation either over time or space is going to hide anything that may be a pattern.  Conversely, too detailed in either/both can have the same affect.  Since you have the data at the 1 year period, I would not aggregate as a step in describing trends, but to summarize over a period.  

To confuse things more, ArcGIS PRO seems to have a better description and options for similar analyses in their help

Emerging Hot Spot Analysis—ArcGIS Help | ArcGIS for Desktop 

and I can't quite find the equivalent in ArcMap, but that may reflect the differences in the capabilities.

I haven't gotten around to visualizing it using time-slider capabilities yet either 

This help topic says use ArcGlobe or ArcGIS Pro to actually 'see' what is going on

     Visualize Space Time Cube in 3D—Help | ArcGIS for Desktop 

so if you have either, then here is the methodology

   Visualizing the Space Time Cube—ArcGIS Help | ArcGIS for Desktop 

Which is in essence, old school overlay mapping with transparent overlays. 

[JamesMisencik]

Is it safe to say that the greater the aggregation, regardless of the type, the greater the likelihood of obtaining "persistent"? I'm having similar frustrations with EHSA, but with "neighborhood distance" rather than "neighborhood time step."

[DanPatterson_Retired]

James... My second sentence made a comment, but Teri hasn't returned to confirm findings, so I will defer until such time.