3 Questions about using point data (not raster data) to calculate Getis-Ord Gi*...?

[peifenkuo]

Hi,

I am trying to find the hotspots of combining two types of data, crimes and crashes.

My first question is how to use point data to calculate Getis-Ord Gi*.

My point data is positive skewed and spread out, so I can not a good fix distance band which makes the corner points have 8 neighbors but the middle points do not includes all features. Please see the image for its kernel density map.

For this kind of data, is it better that I aggregate it as grid data for hot spot analysis or any other methods? 

The second question is about the different results by using kernel density and Gi*.

Is kernel density used to focus on the area with high crash/crime risk, and Gi* is used to identify the data cluster pattern? So, if there is an area with extremely high value, but its neighbors are all low value, will Gi* identify it as a hot spot?

The third question is about using unbiased kernel density in Network.

Does ArcGIS 10 or other software programs provide this function?

Thank you very much. 🙂



pei-fen

[LaurenRosenshein]

Hi Pei-Fen,

The first thing I want to make sure I understand is what kind of field you're using for the Hot Spot Analysis using the Getis-Ord Gi* statistics.  The Hot Spot Analysis tool requires that you have an analysis field, so if you don't have one, you'll have to start by aggregating your incident data.  There are a number of ways to do that (including the grid option that you mentioned, which may be a good one), all discussed here. 

Once you've aggregated your data, then you do have to move towards defining your spatial relationships.  One thing that you might want to consider, since it looks like your data is all along a road network, is creating a Network Spatial Weights Matrix, and using the K Nearest Neighbors conceptualization, and choosing 8 neighbors.  That option would 1) take into consideration the network relationships of your features, and 2)ensure every feature had 8 neighbors.  Even if you don't want to use a network, creating a normal Spatial Weights Matrix, and using the k nearest neighbors option, might be good. 

As far as the difference between a kernel density analysis and the Getis-Ord Gi* statistics, they are really answering two different questions.  A kernel density analysis will give you an idea of how many features are in a given area, making it easy to find places with lots of features and areas with fewer features.  The hot spot analysis tool is looking at the value of each feature and its neighboring features, comparing that to the values for global dataset, and figuring out if the difference between the two is significant.  When the values are significantly higher for a neighborhood than for the global dataset, then its a hotspot.  If you have one very high value surrounded by much lower values, then it will depend greatly on the global average whether or not that is considered a hot spot or not (If you're interested in finding outliers, you may want to consider the Cluster and Outlier Analysis tool).  When the values are significantly lower for a neighborhood than for the global dataset, then its a cold spot.  Then, what's red on the map and what's blue on the map is based on statistical significance, as opposed to a density surface which can be relatively subjective.  You may want to check out this video: Performing Proper Density Analysis.

[ReneeGerasimtchouk]

This is a great description of the two, thank you.

I recently inherited mapping sewer overflows in hopes of identifying where to send inspectors to restaurants and other FOG (Food, Oil, Grease) locations for inspection. The person who did this previously identified all the sewer segments with incidents of 1 or greater, isolated just those segments, found the midpoint, and then did a count field (some segments have multiple incidents). He then used Kernel Density to come up with a raster image but in running the Hot Spot Analysis, and showing the incident count numbers on each sewer line, the resulting raster doesn't match up.

What I think is needed is 1) clustering of sewer incidents and 2) problem areas - sewers sections that have had multiple sewer incidents. Would this be two different analysis?

The result would be a map showing the areas of concern and then a spatial query of FOG areas to send our inspectors out.

Any help/suggestions appreciated, I've been reading and watching the tutorials but am a bit confused as to what to use or to combine analysis.

(I'm just getting back to using ArcGIS after using GeoMedia for the last 9 years)

Renee