I would like to find geographic clusters of zip codes where each zip code has a measurement of a person's age (integer), sex (categorical/binary), and if they recovered from a certain medical condition (categorical/binary). Each zip code could have multiple people in it.

Will having multiple records associated with each geographic unit pose a problem for a cluster or hotspot analysis? Are there methods for dealing with this situation? Aggregating the data for each zip code (average or counts) doesn't seem appropriate.

I'm not tied to any specific software.

  • Hey there, I did health related socioeconomic census study at my university. Also used cluster analysis. Could you post an example (screenshot) of your database and how it's laid out ?
    – GISHuman
    Jun 4 '14 at 18:59
  • Unfortunately, I cannot post a screenshot of the data, per HIPAA requirements. Perhaps I can describe the attribute table better. There is a column for a unique id for each patient, sex, age, and outcome (recovery or not), and zip code of residence. So the geographic unit is zip code, but there could be multiple people in each zip code. Jun 4 '14 at 19:02
  • I've tried to answer the best I can for now, I will update at a later time with the specific terms (my mind is a bit hazy). If you could provide a fake spreadsheet (with made up postal codes/ IDS) it might help you get some additional answers
    – GISHuman
    Jun 4 '14 at 19:13
  • Are you looking for clusters of cases or clusters of zip codes? It's understood you won't get a more specific case location than at the zip level. But if I understand correctly you have individual case records that have a zip code. I might suggest creating random points within each zip to 'locate' your records and then use those (with the notation they have been randomized to the zip level) to perform your cluster/hotspot analysis. Assuming your study extents are big enough to provide meaningful results anyway.
    – Chris W
    Jun 5 '14 at 5:42

You can have a look at some of the techniques developed by Stan Openshaw and his team at Leeds University in the 1990's. They include GAM which was developed for exactly the sort of problem you are describing. There is a large literature discussing this and other methods.

I have code that implements several of these methods at https://github.com/ianturton/spatial-cluster-detection that may help you get started on your analysis.


There are definitely problems with using such large unit of measurement (postal codes) for health related/cluster analysis. In laymen's terms the area is too big and you're aggregating up. This doesn't allow for variations that may be present at a smaller scale. If possible you should try to get a finer scale for your data (look into dissemination areas or DE zones). Although, there are many studies that use postal code or even census block information. Please look at studies which use the census so that you can have a better idea of the pros/cons of aggregated data. I can provide these resources at a later time (unfortunately not at home right now should you need them). There is also issues as to how to boundaries are disseminated. How did they choose these arbitrary boundaries? There are other methods which take into account landuse, natural habitat and rivers, etc.

As for having multiple records for a given postal code, you're going to need to do additional statistical analysis so that each postal code is comparing against the same things (ex. by percentage).

If you could update your answer with the specific clustering technique I could definitely try and tailor my answer.

I definitely recommend additional reading on the topic as there are numerous spatial analysis techniques for health related data, especially clusters. I will definitely try to give a better answer with links when possible, hopefully this will help you out a little!

  • 1
    I'm looking for suggestions for methods, so I'm open to ideas about which clustering technique will work best. I did not design the study and the researcher wasn't approved for identifying the patients at smaller geographic units, so while I agree that the size of zip codes poses a problem I cannot change that for now. Jun 4 '14 at 19:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.