Page 53 - IJPS-1-1

Giambattista Salinari and Gustavo De Santis

WLS as opposed to OLS (ordinary LS) reduces the risk of bias due to heteroscedasticity
(Horiuchi and Coale, 1990; Horiuchi and Wilmoth, 1998).
3. Results

Our first major finding was a decline in the age for the onset of mortality acceleration (Sa-
linari and De Santis, 2015). Figure 2 presented the log-probability of death for two groups
of cohorts: 1895 (i.e., born in 1895–1904) and 1910 (i.e., born in 1910–1919) together
with the estimates of their threshold ages. Note that in all the panels of Figure 2, two dif-
ferent phases in the evolution of the death probabilities emerged: (1) when they are ap-
proximately stationary, (2) when they increase exponentially (i.e., log-linearly). In other
words, both conditions for the correct identification of the onset of mortality acceleration
are satisfied.
The estimated age at the beginning of mortality acceleration is close to 50 years in the
first cohorts, born between 1895 and 1904. It progressively declined to about 40 years in
the most recent cohorts, born in 1910 or later with a large variation of about ten years

Figure 2. The onset of mortality acceleration in the cohorts born from 1895–1904 and 1910–1919.
Source: Authors’ elaborations on HMD data.

International Journal of Population Studies | 2015, Volume 1, Issue 1 47