How serious is it to ignore warnings in Incremental Spatial Autocorrelation analysis?

846
3
07-17-2018 07:51 AM
SiobhanDancey
New Contributor

Good Afternoon, 

  I am running some Incremental Spatial Autocorrelation analysis on some disease outbreak data for a series of different years. The results are intuitively what I would expect in terms of the distance where I am seeing a peak in each analysis. However, I get ' Warning 001532: * At least one distance increment resulted in features with no neighbours which may invalidate the significance of the corresponding results.' How serious is this if the results look okay? Can it be over looked or does this mean that my output is meaningless? I attempted to create a spatial weights matrix as that was suggested but I'm not sure how I incorporate that into the analysis. 

Thank you for any help or advice!

Siobhán

0 Kudos
3 Replies
ChrisDonohue__GISP
MVP Alum

Tagging:  https://community.esri.com/community/gis/analysis/spatial-statistics for greater exposure.

Chris Donohue, GISP

RyanRuthart1
New Contributor II

Spatial autocorrelation evaluates a set of features and an associated attribute to determine whether the pattern expressed is clustered, dispersed, or random. If the feature does not have any neighbors within the distance band it cannot compare that features values to its neighbors. As you increase the distance threshold the lone features may eventually get neighbors, but it will also increase the number of neighbors for features which are close together.

 

By using generate spatial weights matrix (Generate Spatial Weights Matrix—ArcGIS Pro | ArcGIS Desktop ) you have more control over what features are considered as neighbors. You can use the Conceptualization of Spatial Relationships - K nearest neighbors option so that every feature is given the same number of neighbors and no feature will have 0 neighbors. The spatial weights matrix can be used as input to the Spatial Autocorrelation tool (Spatial Autocorrelation (Global Moran's I)—ArcGIS Pro | ArcGIS Desktop ) but not the incremental version (which increases the distance to evaluate its effect).

 

There may be distance bands for which every feature has neighbors, that is indicated in the tool messages output. You can also identify the features with no neighbors to further investigate the impact they have in your data set.