5
$\begingroup$

This is more of a historical related question. Edmond Halley in 1693 published an actuarial table showing mortality rates. The tables show yearly statistics for ages 1-84 and a summary statistic for ages 84-100. I have searched hard for the missing data, but with no luck (See also Geoffrey Heywood's article).

So this question has two parts:

  • Can anyone point me to a different source showing the original paper?
  • What would be the best model to generate the missing statistics between these years?
Improve this question
$\endgroup$
1
  • $\begingroup$ As I see the table and the graph (I don't know if it's the right one, but suppose that, it is linked from ) the data for age 84+ are insufficient (the graph line starts to bend). They cleary know why they cut it there. I think I wouldn't try to extrapolate in this case. $\endgroup$
    – Tomas
    Oct 21, 2012 at 10:54

2 Answers 2

Reset to default
1
$\begingroup$

Gompertz (1825) - log hazard linear in age - and Makeham (1860) models (adding a constant to that exponentially increasing hazard) were used pretty successfully for mortality in their time, at least at adult ages. They don't work as well now (but mortality has changed a lot).

My first inclination would be to try simple models like those for mortality, and check them for fit. The logit of $q_x$ should behave similarly to log-hazard. If you need the whole life table in a functional form, perhaps a Heligman-Pollard model or if just a good fit is needed something using splines (as is often used these days to model mortality).

There are whole books on modelling mortality; it's not a subject you can properly cover in ten lines.

Improve this answer
$\endgroup$
2
$\begingroup$

Maybe a sensible approach could be the Brass logit model. You calibrate $ln(p^{target}_t)=a+b*ln(p^{base}_t)$ being target and base an incomplete and a complete table. Then you can complete the missing $p_t^{target}$ thanks to the calibrated model.

Improve this answer
$\endgroup$
2
  • $\begingroup$ You have not explained what you mean with $p^{target}_t$ and $p^{base}_t$. In the question the O.P. has only one mortality rate per year ($p_t$). $\endgroup$
    – Tomas
    Oct 21, 2012 at 10:37
  • $\begingroup$ p^{target}s come from the yearly statistics, p^{base} come from another life table. The Brass logit model assumes that the log - linear relationship holds between two life table. It could provide a sensible approximation. $\endgroup$
    – Giorgio Spedicato
    Oct 29, 2012 at 8:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.