0

I am trying to conduct a temporal cluster analysis on a terrorism dataset, but I've hit a bit of a brick wall regarding the method.

The dataset spans between 01-01-2015 and 31-12-2019, and for my 3 target cities, the main data file consists of 1 daily ttset column, and 1 column for attack_count. The vast majority of the entries for attack_count will be 0 due to terrorist attacks being generally uncommon, but because of this, any cluster analysis I use tends to find clusters in the "0"s instead of the "1"s, "2"s or so on.

I have access to SPSS and Stata, but I'm not really sure what method of cluster analysis would be best for this dataset, as clustering is very unlikely to be present; ultimately making me wonder if my results are bad because there's no clustering, or because I've picked the wrong method.

Any advice would be greatly appreciated, thanks!

Example of dataset, there's three and they're all pretty barebones

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
MPH125
  • 1
  • 3
  • Can you show the data, some excerpt to see the variables and the values? – ttnphns Jul 23 '21 at 15:09
  • Included a picture now, it's pretty barebones stuff, so I'm not too sure what (if anything) would work with it – MPH125 Jul 23 '21 at 15:16
  • What exactly are you trying to cluster here? Is grouping some attacks together that are adjacent in time, or the type of attack, or ideology? – dimitriy Jul 23 '21 at 17:07
  • The only aim of this is to see if the attacks are clustered in time, there's no further distinction being made than that. So the data shown is all that's theoretically required to my (incredibly limited) knowledge. – MPH125 Jul 23 '21 at 17:22
  • "Clustering in time" maybe a point process model is the way to go? You could start by testing the null of a homogeneous Poisson process, see https://stats.stackexchange.com/questions/316844/testing-for-a-poisson-process – kjetil b halvorsen Jul 24 '21 at 02:10

0 Answers0