FlowingData Forums » Data Visualization

Help Me Visualize This Data

Started 12 months ago by MGNute / 4 posts

  1. Hi All,

    I have some data with an awkward structure that I'm trying to find a good way to visualize, so I was wondering if people could help.

    In commercial insurance, we have two key metrics, Loss Ratio (losses / premium) and Retention (how likely the policy is to renew). Also, there are four key lines of insurance: 1) Package (property & liability, but they're always sold together so they're one line of business), 2) Auto, 3) Workers Comp, 4) Umbrella. What we want to illustrate is how much the loss ratio and retention are impacted given that an account has any particular subset of these lines of business. For example, if an account has just an auto policy, will they be worse on either metric than if they have an auto and package?

    So there's a natural structure to the data that I'm having a tough time showing graphically. One more key point: it's mostly interesting to compare the performance within a line of business rather than across lines of business. For example, we wouldn't want to compare the loss ratio of just-auto to that of just-workers-comp because they have much different grand means, but we would compare just-auto performance to the performance of auto when it's paired with comp or package.

    Anyone have any ideas on how to do this? I've been toying with some network-type graphs in my head but nothing seems to get at it. All we've got today is a big table that has the combo of lines-of-business on the left and performance of each line across the top, but this creates much wasted space (for example, the entry for auto performance on the row for just-package policies is always blank), and doesn't convey the conditional nature of it.

    Thanks in advance.

    Mike

  2. @MGNute - it might be helpful to post a sample of your data.

  3. I know this may sound off-the-wall, but have you calculated the "interference factors" as it were, for example, by running a GLM with interactions? If so, perhaps those factors would be the appropriate metrics to present; especially if normalized by "base class" so you could say that the addition of Auto to WC in/decreases retentions by 20% or so. Just brainstorming here.

  4. Sure, here's a sample (the numbers here are examples only).

    Line of Business Premium Retention
    Account Type Auto Package WC Umbrella Total
    Auto 80.0% -- -- -- 80.0%
    Package -- 82.0% -- -- 82.0%
    WC -- -- 81.0% -- 81.0%
    UMP -- -- -- 86.0% 86.0%
    Auto/Package 83.0% 85.0% -- -- 84.0%
    Auto/WC 84.0% -- 86.0% -- 85.0%
    Auto/Umb 75.0% -- -- 71.0% 73.0%
    Package/WC -- 84.0% 81.0% -- 82.5%
    Package/Umb -- 87.0% -- 83.0% 85.0%
    Auto/Package/WC 84.0% 89.0% 86.5% -- 86.5%
    Auto/Package/Umb 87.0% 82.0% -- 50.0% 65.0%
    Auto/WC/Umb 81.0% -- 83.0% 83.0% 82.0%
    Package/WC/Umb -- 84.0% 84.0% 80.0% 81.0%
    Auto/Package/WC/Umb 88.0% 88.0% 88.0% 88.0% 88.0%

    I do like Avi's idea, though I still wonder if there's a neat way to picture it.


Reply

You must log in to post.

About this Topic

Tags

No tags yet.