I am trying to replicate the derivation of the life expectancy formulas from Missov & Lenart (2013). They propose that the following:

Theorem 2. Life expectancy of the Gompertz–Makeham mortality distribution at age $x$ is: $$e_{GM}(x) = \frac{1}{b} e^\frac{a}{b} \left( \frac{a}{b} \right)^\frac{c}{b} \Gamma \left( -\frac{c}{b} , \frac{a}{b}e^{bx}\right) \qquad a,b > 0, \; c\geq 0$$ where $\Gamma(s,z) = \int_z^\infty t^{s-1} e^{-t} dt$ denotes the upper incomplete gamma function

Their proof goes as follows:

Proof. The individual force of mortality and the survival function for a Gompertz–Makeham homogeneous population is given by (1) and (2) for z = 1, i.e $$\mu_{GM}(x) = a ^{bx} + c \qquad \qquad S_{GM}(x) = \exp \left[ -\frac{a}{b} (e^{bx} - 1) -cx \right]$$ Remaining life expectancy at age $x$ can be calculated by $$e_{GM}(x) = \int_x^\infty S_{GM}(t, y) dt \tag{B1}$$

So far I can follow, except for the added $y$ argument to the survival function. Nevertheless, they continue:

Substituting $s = e^{bt}$ and denoting $a = \frac{a}{b}$ reduces (B.1) to $$e_{GM}(x) = \frac{e^a}{b} \int_{e^{bx}}^\infty e^{-as} q^{-\frac{c}{b}-1} dq$$

Here I have a problem: we seem to subsitute using $s$, but a variable $q$ appears. My intuition is that thefirst step here is to plug in the survival function as $\exp \left[ -\frac{a}{b} (e^{bt} - 1) -ct \right]$ and substitute for $a$, which allows us to draw $e^a$ out of the integral. But the we need to substitute for $t$, so what happens to $ct$? And how is $q$ defined?

To be complete, they the proof continues:

Finally we substitute $u = as$ and get $$e_{GM}(x) = \frac{e^a}{b}[a]^\frac{c}{b} \int_{ae^{bx}}^\infty e^{-u} u^{-\frac{c}{b}-1} du = \frac{1}{b} e^\frac{a}{b} \left( \frac{a}{b} \right)^\frac{c}{b} \Gamma \left( -\frac{c}{b} , \frac{a}{b}e^{bx}\right)$$


1 Answer 1


As happened previously, formulating the question and thinking about it helped to find a solution. Here it is:

We start from $$\begin{aligned} e(x) &= \int_x^\infty \exp \left[ -ct - \frac{a}{b} (e^{bt}-1) \right] \mathsf{d}t \\ &= \int_x^\infty \exp \left[ -ct - \frac{a}{b} e^{bt} + \frac{a}{b} \right] \mathsf{d}t \\ &= \exp\left( \frac{a}{b} \right) \int_x^\infty \exp \left[ -ct - \frac{a}{b} e^{bt} \right] \mathsf{d}t \end{aligned}$$

Now, define $q(t) = e^{bt}$, so that we can apply integration by substitution:

$$\frac{\mathsf{d} q}{\mathsf{d} t} = q'(t) = b e^{bt} \qquad \Leftrightarrow \qquad \mathsf{d} q = \mathsf{d} t \cdot q'(t)$$

This might be somewhat loose on notation, but it works (I'm not a mathematician). Nevertheless, we can continue: $$\begin{aligned} \exp\left( \frac{a}{b} \right) \int_x^\infty \exp \left[ -ct - \frac{a}{b} e^{bt} \right] \mathsf{d}t &= \exp\left( \frac{a}{b} \right) \int_x^\infty \exp \left[ -ct - \frac{a}{b} e^{bt} \right] \frac{q'(t)}{q'(t)}\mathsf{d}t \\ &= \exp\left( \frac{a}{b} \right) \int_x^\infty e^{-ct} \exp \left[-\frac{a}{b} e^{bt} \right] \frac{1}{b}e^{-bt} \;\; \mathsf{d} t \cdot q'(t) \\ &= \frac{1}{b}\exp\left( \frac{a}{b} \right) \int_x^\infty \exp \left[-\frac{a}{b} e^{bt} \right] \left(e^{bt} \right)^{-\frac{c}{b} - 1} \;\; \mathsf{d} t \cdot q'(t) \end{aligned}$$

Where we used the fact that $\left(e^{bt} \right)^{-\frac{c}{b} - 1} = e^{-bt - ct}$. We then get to the first step in the proof by substituting in $A = a/b$ (less confusing with a capital A) and the definition of $q(t)$, while also setting the lower bound to $q(x)$: $$\begin{aligned} \frac{e^A }{b} \int_{e^{bx}}^\infty e^{-Aq} q^{-\frac{c}{b} - 1} \;\; \mathsf{d} q \end{aligned}$$ So $s$ seems to have been $q$ all along. We can then continue, by defining $u(q) = Aq$ so that: $$\frac{\mathsf{d} u}{\mathsf{d} q} = u'(q) = A \qquad \Leftrightarrow \qquad \mathsf{d} u = \mathsf{d} q \cdot u'(q)$$

and perform a second round of integration by substitution.

$$\begin{aligned} \frac{e^A }{b} \int_{e^{bx}}^\infty e^{-Aq} q^{-\frac{c}{b} - 1} \;\; \mathsf{d} q &= \frac{e^A }{b} \int_{e^{bx}}^\infty e^{-Aq} q^{-\frac{c}{b} - 1} \frac{A^{\frac{c}{b}}}{A^{\frac{c}{b}}}\;\; \frac{u'(q)}{u'(q)}\mathsf{d} q \\ &= \frac{e^A }{b} A^{\frac{c}{b}} \int_{e^{bx}}^\infty e^{-Aq} q^{-\frac{c}{b} - 1} A^{-\frac{c}{b}} A^{-1}\;\; u'(q)\cdot\mathsf{d} q \\ &= \frac{e^A }{b} A^{\frac{c}{b}} \int_{e^{bx}}^\infty e^{-Aq} q^{-\frac{c}{b} - 1} A^{-\frac{c}{b} -1}\;\; u'(q)\cdot\mathsf{d} q \\ &= \frac{e^A }{b} A^{\frac{c}{b}} \int_{e^{bx}}^\infty e^{-Aq} u^{-\frac{c}{b} - 1} \;\; u'(q) \cdot\mathsf{d} q \\ &= \frac{e^A }{b} A^{\frac{c}{b}} \int_{Ae^{bx}}^\infty e^{-u} u^{-\frac{c}{b} - 1} \;\; \mathsf{d} u \end{aligned}$$ The last step to the proof is then simply to apply the definition of the upper-incomplete gamma function.


Your Answer

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

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