Island life

We have written extensively about the use of postcodes and geodemographics for mortality modelling.  Two peer-reviewed papers recently presented to the Institute of Actuaries in London have testified to the power of geodemographics when applied to pensioner mortality: Richards (2008) and Madrigal et al (2009).

One feature of standard U.K. postcode profilers is that they typically exclude the crown dependencies, which are not legally part of the United Kingdom.  This makes it impossible to assign a geodemographic type, despite the fact that crown dependencies often have a postcode system which follows the same hierarchical structure as that of the UK.  Postcodes for crown dependencies are therefore often lumped in with the "unrecognised" category after profiling.

The largest crown dependencies are Guernsey (postcodes beginning GY), Jersey (postcodes beginning JE) and the Isle of Man (postcodes beginning with IM).  It is possible to enhance the standard postcode profilers to include crown-dependency postcodes.  While this doesn't enable a geodemographic type to be assigned, it does allow separation of crown dependencies from the "unrecognised" category for modelling.  This can lead to some unexpected insights, as shown in Table 1.

Table 1.  Selected parameters from a model of regional variation of U.K. and Crown Dependencies (Guernsey, Jersey and the Isle of Man).  The baseline for the intercept is Guernsey, and a Perks survival model has been fitted for ages 60-95 over the years 2000-2006.

Table 1 shows the association of island residence on mortality.  Guernsey is the baseline, so residents of the Isle of Man do not have statistically significant differences compared with Guernsey.  The surprise is that residents of Jersey have mortality which is statistically different from people on Guernsey.  Although the numbers of lives and deaths are small, the p-value for the difference is 3.33%, which is rather low to be a pure chance result.

I don't have any special insights into the Channel Islands, but I'm sure the residents will happily argue over the respect merits of their island.  One thing I can confirm is that it apears to be unrelated to differences in income: adding pension size enhances the overall model, but the p-value for the difference between Guernsey and Jersey actually becomes more significant at 2.32%.