Programming the statistical procedures from SAS

Clustering people by location

Reply
Super Contributor
Posts: 333

Clustering people by location

We are trying to figure out a way to group people in our state by location (actually zipcode lat, long), to find out where reps should be to be efficiently close in order to visit these people. Does anyone have any hints or tricks that they can share? We know how many reps we need in each region but not necessarily where they should be located. We have tried clustering on lat and long but dont get equal sized clusters.

Just interested if anyone has any insight they can share!

Thanks!

EJ

Grand Advisor
Posts: 16,933

Re: Clustering people by location

Is that a cluster or minimization/optimization problem?

For example, minimize the distance between people and reps?

Super Contributor
Posts: 333

Re: Clustering people by location

We are actually trying to come up with a minimization of between people distances right now, we just dont know yet how it is going to turn out. We are manually writing the code for it so it taking some time to tweak and get it to  work.

Grand Advisor
Posts: 16,933

Re: Clustering people by location

Yeah...ArcGIS is nicer for that kind of analysis, if you have it. But its pricey like SAS Smiley Happy

Grand Advisor
Posts: 10,075

Re: Clustering people by location

By unequal cluster size to you mean the number of people to visit differs per rep? Unless you population is very homogenously distributed I would expect that given a geographic location for a rep.

Anyway, it may be helpful to show which clustering procedure code you used.

Respected Advisor
Posts: 4,606

Re: Clustering people by location

According to SAS doc, METHOD=COMPLETE in PROC CLUSTER gives you clusters biased toward equal cluster diameters and METHOD=WARD gives you clusters biased toward equal populations.

You will get more accurate distances if you translate lat-long coordinates into UTM X-Y coordinates (use a single zone, of course).

PG

PG
Ask a Question
Discussion stats
  • 5 replies
  • 177 views
  • 0 likes
  • 4 in conversation