This question already has an answer here:

I'm working with a polylines shape. I'm searching for a tool that could help me to define neighbourhoods using the values of the polylines.

I have good results with Hot Spot Analysis (Getis-Ord Gi*), because it creates new values depending on the neighbourhood. The doubt is about the possible use of all the values. I don't need only the hot or cold spot polylines, I want to use all the network, included the ones classified with a P-value as a non significant level. ¿Is it correct or should I use other tool?

marked as duplicate by Jeffrey Evans, Midavalo May 18 '17 at 16:15

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

  • 1
    Please search the site and see previous discussions on the use of these statistics on network (linear) data. The underlying assumptions in how the statistic works, and is implemented, is not appropriate for a linear feature. You null is incorrect and will produce erroneous results. It is possible to modify the statistic but it would have to be at the code level. – Jeffrey Evans May 17 '17 at 17:32
  • Thank you Jeffrey Evans for the comment. I've edit the question in order to be more clear. Do you think the statistic is not working well with lines? – CarmenZ May 18 '17 at 12:35
  • Thank you again. The fact is that I'm not working with points through a Network, I'm just working with polylines. The tool Hot Spot Analysis (Getis-Ord Gi*) (not the optimized one) gives you the possibility to work with points or with lines. These lines have values of different indicators that I want to use to make groups. What I would like to know is if it's correct to use the P-value calculated as a new variable to construct my groups. – CarmenZ May 18 '17 at 14:52
  • 1
    You are completely missing the underlying point here. The statistic is not appropriate for linear features (events along lines or lines). Unless ESRI has come up with the one-dimensional derivative of the statistic, which I have not yet seen published, and completely re coded their tool, the results are incorrect. The statistic test against an expected spatial null that assumes area. Because of this, a linear dependency does not represent the correct spatial structure against the null. Have you tried a cluster analysis using something like k-means? I believe that this is what you are after. – Jeffrey Evans May 18 '17 at 15:53
  • I've tried the Grouping Analysis, before with k_nearest_neighbors and the solution didn't work for me, and now with no_spatial_constraint (which use the algorith of k-means). The result now is better and I can see the utility if I need to combine variables. But if I'm going to use just one, and this tool does not create new values for the variable depending on the neighbourhood, the result is no much better than making my own groups directly classifyng. Thank you again. – CarmenZ May 18 '17 at 17:06