Forecasting Techniques for Demographics

Tomorrow, we will discuss in our cours forecasting tools for demographics. But first, we will see basic static tools. Before playing with longitudinal dataset, let us use “standard” life tables. Some (French) datasets are available on the INED website, but let us use the most popular ones, the TV8890 and TD8890. Those two tables can be downloaded from wikipedia,

url="https://fr.wikipedia.org/wiki/Table_de_mortalit%C3%A9"
download.file(url,"mortalite.html")
library(XML)
tables=readHTMLTable("mortalite.html")

The code to get the dataset is the following (here, we should be careful since there is a space in the number, as a separator for thousands)

TV8890_0=tables[[2]]
a1=as.numeric(as.character(TV8890_0[,1]))
a2=as.numeric(as.character(TV8890_0[,3]))
espace=function(x) gsub("[[:space:]]", "", x)
b1=espace(as.character(TV8890_0[,2]))
b2=espace(as.character(TV8890_0[,4]))
TV8890=data.frame(x=c(a1,a2),lx=as.numeric(c(b1,b2))

One can also read the second life table (note that there is a typo in wikipedia since here, the two tables are exactly the same)

TD8890_0=tables[[1]]
a1=as.numeric(as.character(TD8890_0[,1]))
a2=as.numeric(as.character(TD8890_0[,3]))
b1=espace(as.character(TD8890_0[,2]))
b2=espace(as.character(TD8890_0[,4]))
TD8890=data.frame(x=c(a1,a2),lx=as.numeric(c(b1,b2)))

It is possible to use that survival function to compute some sort to life expectancy at birth

sum(TV8890$lx)/100000-1
[1] 72.01518

One can visualize the survival probability (up to a 100,000 scaling constant)

plot(TV8890,type="l")

or the death probability, i.e. the probability to die at some specific age x, given that you did reach age x, also called the force of mortality

n=nrow(TV8890)
px=(TV8890$lx[1:(n-1)]-TV8890$lx[2:n])/
TV8890$lx[1:(n-1)]
x=TV8890$x[1:(n-1)]
plot(x,px,type="l",xlab="age")

A more popular visualization is obtained with a log scale for the probability

plot(x,px,type="l",log="y")

Finally, we can compute the density of the age at death

pbx=TV8890$lx[1:(n-1)]*px/100000
plot(x,pbx,type="l")

that can also be used to compute life expectancy

sum(x*pbx)
[1] 72.01518

That is for the static case. For longitudinal tables, we can use those from the Human Mortality Database. For instance, for France, we can get information of the number of death at age x during year t, as well as the exposure (number of people alive). For France, there are datasets available here

url="http://freakonometrics.free.fr/FranceDeaths_1x1.txt"
download.file(url,"FRD.txt")
url="http://freakonometrics.free.fr/FranceExposures_1x1.txt"
download.file(url,"FRE.txt")

and for Canada

url="http://freakonometrics.free.fr/CanadaDeaths_1x1.txt"
download.file(url,"CAD.txt")
url="http://freakonometrics.free.fr/CanadaExposures_1x1.txt"
download.file(url,"CAE.txt")

The following code can be use to read those files.

FRD=read.table("FRD.txt",skip = 3,header=TRUE)
tail(FRD)
      Year  Age Female  Male  Total
22195 2015  105 297.96 35.86 333.82
22196 2015  106 182.95 20.39 203.34
22197 2015  107 104.87 10.50 115.37
22198 2015  108  57.27  5.07  62.34
22199 2015  109  31.93  2.59  34.52
22200 2015 110+  33.03  1.61  34.64

OpenEdition suggests that you cite this post as follows:
Arthur Charpentier (January 29, 2018). Forecasting Techniques for Demographics. Freakonometrics. Retrieved October 8, 2024 from https://doi.org/10.58079/ov9p


One thought on “Forecasting Techniques for Demographics”

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.