A Scottish question

The Scots are an innovative bunch, including the inventor of the telephone and the discoverer of penicillin.  Not all of our innovations have been positive, however.  Human welfare did not advance with the invention of the deep-fried Mars bar, for example.

With such dietary crimes it comes as no surprise that the Scots are known for having the shortest life expectancy within the four countries of the United Kingdom.  This even extends to select sub-groups, such as the portfolio of pension annuitants behind Table 1.

Table 1.  Time lived between ages 60 and 95 for males in United Kingdom (curtate life expectancies for holders of pension annuities).

U.K.
Region
Male life
expectancy
(years)
Scotland 20.9
Wales 21.5
England 22.3
Northern Ireland 22.4

The question is whether the shorter life expectancy shown in Table 1 is due to being Scottish, living in Scotland, unhealthy Scottish lifestyles, or something else entirely.  At this stage I should of course declare an interest: as a person of Scottish ancestry living in Scotland, I very much hope that being Scottish or being in Scotland is not the cause!  And no, I have never seen a deep-fried Mars bar, let alone eaten one.

We will use the data behind Table 1 to fit a series of survival models allowing for age, gender and time trends, together with UK region and geodemographic profile.  These latter two factors are both determined by the pensioner's postcode, and we are restricting ourselves to pensioners with a valid UK residential postcode.  The results are shown in Table 2.

Table 2.  Selected models using Perks' Law of mortality for a portfolio of pension annuitants.

Model AIC Improvement
over Model 1
1. Age*Gender+Time 385,036 n/a
2. Age*(Gender+Region)+Time 384,680 356
3. Age*(Gender+Region+Acorn)+Time 383,435 1,601
4. Age*(Gender+Acorn)+Time 383,564 1,472

Fortunately for non-statisticians, you don't need to understand either the model syntax or what an AIC is to work out what is going on.  Most obviously, geodemographic group (Acorn) is much more powerful at explaining mortality variation than region: Model 4 is a much bigger improvement than Model 2.  Furthermore, the only difference between Models 1 and 2, and Models 3 and 4,  is the region factor, yet the difference from including the region factor falls from 356 in the former case to just 129 (=1601-1472) in the latter. This leads us to the following two conclusions:

  1. Socio-economic group or lifestyle is a much more important rating factor for longevity than region.
  2. More than half of the apparent regional differences are in fact due to varying socio-economic mix around the UK.

These conclusions are of interest for insurers using postcodes to price annuities.  If you are using geodemographic profiles, then you are correctly allowing for socio-economic or lifestyle differences in longevity.  However, if you are just using region as a rating factor then you are missing an important trick.  Indeed, you may be losing money as a result: if your competitors use geodemographic profiles, then conclusion (2) above suggests that you are being selected against.

Written by: Stephen Richards
Publication Date:
Last Updated:

Geodemographics in Longevitas

Longevitas users can control the geodemographic profiler used in the Deduplication tab in the Configuration area. The Upload Processing section contains a drop-down option list for available profilers. Options for UK data include Mosaic, Acorn, P2, Health Acorn, FSS, CAMEO and Personicx.

A variety of other options exists for territories outside the UK, such as the USA, Canada and the Netherlands. Note that each profiler requires a separate licence from the owner: Experian for Mosaic and FSS, CACI for Acorn and Health Acorn, Beacon Dodsworth for P2, Eurodirect for CAMEO and Acxiom for Personicx.

Previous posts

Out for the count

In an earlier post we described a problem when fitting GLMs for qx over multiple years.  The key mistake is to divide up the period over which the individual was observed in a model for individual mortality. 
Tags: Filter information matrix by tag: survival models, Filter information matrix by tag: force of mortality, Filter information matrix by tag: GLM, Filter information matrix by tag: missing data

Stabilising projections

With many stochastic models of mortality, projections of future mortality rates are done using a time series.  In a landmark paper, Currie, Durban and Eilers (2004) introduced the idea of using P-splines as an alternative means of generating a forecast.
Tags: Filter information matrix by tag: P-splines, Filter information matrix by tag: time series, Filter information matrix by tag: mortality projections, Filter information matrix by tag: CMI

Add new comment

Restricted HTML

  • Allowed HTML tags: <a href hreflang> <em> <strong> <cite> <blockquote cite> <code> <ul type> <ol start type> <li> <dl> <dt> <dd> <h2 id> <h3 id> <h4 id> <h5 id> <h6 id>
  • Lines and paragraphs break automatically.
  • Web page addresses and email addresses turn into links automatically.