What concept would I use to develop a spatial regression model for misaligned Insurance Claims and Policy data? For example, consider a situation where I have 1000 points that represent policies, and 100 points that represent claims. While some of the claims are located at the same spatial location as some of the policies, most are not. In addition, I have 20 predictor variables associated with the policy points, and 2 predictor variables associated with the claim points.

How can I develop a spatial regression where Y represents Claim $ amount, and X represents all of the predictor variables associated with the policies?

The image below is a toy example of this situation, where blue points represent policies and red triangles represent claims. Triangles either align perfectly with the circles, or they do not align with any circles. I'm assuming I have complete information about the policies (circles), so there are no missing circles.

Blue circles = Policies, Red triangles = Claims

  • $\begingroup$ Questions about programs/coding are off-topic here, so I'd recommend that you edit this to focus on your statistical question. $\endgroup$ – mkt - Reinstate Monica Aug 20 '19 at 13:31
  • $\begingroup$ Oh sorry! I'll try and fix it. $\endgroup$ – Kristaps Aug 20 '19 at 13:32
  • $\begingroup$ what do you want to predict? $\endgroup$ – user31264 Aug 20 '19 at 13:38
  • $\begingroup$ I want to predict Claim amount ($) at the claim points $\endgroup$ – Kristaps Aug 20 '19 at 13:40
  • 1
    $\begingroup$ @mkt Gotcha - thank you mkt $\endgroup$ – Kristaps Aug 20 '19 at 14:13

(Turning my comments into an answer)

This problem is probably best solved by a spatial interpolation approach such as kriging. An alternative approach would be to select the nearest by some distance metric and treat that as the true location. In the first approach, some avoidable error is introduced by using information from more distant points if the nearest point is the correct one. In the second approach, results will be perfect if the nearest point is the correct one. But for the subset of points for which this is not true, this approach will produce results that are worse, possibly by a substantial amount.


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.