A Matter of Life and Death
Can the Famous Really Postpone Death?
The distribution of death dates across the
year
Alisa Beck, Marcella Gift, Katie Miller
Basis for Project
• Case Study 6.3.2
• David Phillips’ study on postponing death
until after one’s birthday
• Theory of death dip/death rise
Questions to answer
• Do people postpone their death until after a
birthday?
• Is the distribution of death dates uniform
throughout the year?
• Is there a difference in distribution for
people who died in the 1920s vs 1990s?
• Can people postpone their death past
another special date? What date?
Sample
• 391 entries from two volumes of Who Was
Who in America
– Selected every other entry for a given number
of entries for each letter of the alphabet
• 39.1% from Volume I (1920s), 60.9% from
Volume XIII (1990s)
• 89.3% male, 10.7% female
Do people postpone death past
their birthday?
• Test of proportions to compare the number
of people dying in the month after their
birthday against the expected proportion
• Expected number of deaths in a given
month is 391/12=32.6
• Number of people dying in one month after
birthday is 38
Do people postpone death past
their birthday?
• Z=x-np/sqrt(np(1-p))
• Z=.99<1.64
• Therefore we cannot reject the null
hypothesis that the proportion of deaths in
the month after one’s birthday is 1/12.
• Phillips’ hypothesis does not hold for our
data.
Do people postpone death past
their birthday?
• Confidence interval for the mean difference
in the number of days between birth date
and death date
• Mean difference=6.84 days after birthday
• Range of -180 to 180
• 95% CI: (-3.57, 17.27)
• Therefore, the mean is not significantly
different from 0, so people are not more
likely to die after their birthday
Conclusion
• Our data does not support Phillips’
hypothesis
• Possible limitations
– Our people are not famous enough
Overall distribution by month
Is this distribution uniform?
Death Month
January
February
March
April
May
June
July
August
September
October
November
December
# died
49
35
45
30
27
32
36
19
30
26
33
29
% died
12.5%
9.0%
11.5%
7.7%
6.9%
8.2%
9.2%
4.9%
7.7%
6.6%
8.4%
7.4%
zstat
3.004
0.442
2.272
-0.473
-1.021
-0.107
0.625
-2.485
-0.473
-1.205
0.076
-0.656
Distribution by month and
volume
Is this distribution uniform?
• Unpaired test for two sample proportions
Overall distribution by season
Deaths per season by volume
Is this distribution uniform?
• Test for difference by volume:
• ANOVA for difference in seasons is not
significant (p=.07)
Implications
• People who died in the 1920s are more
likely to have died in the spring, while
people who died in the 1990s were more
likely to die in the winter.
• More people tend to die in winter...is this
because of postponement or other factors?
Can people postpone their death
dates?
• Dates we considered that would be
important to people
–
–
–
–
Birthday
Christmas
4th of July
New Year’s
• Expected number of deaths in any given
month is 391/12=32.6
Deaths in month before/after
each date
Date
Birthday
#deaths in
month before
35
#deaths in
month after
38
Christmas
34
48
New Year’s
29
49
July 4th
30
33
New Year’s
• The date with the greatest evidence of death
rise/death dip is New Year’s Day
• Test significance of date with z-test for
proportions
– H0: p=1/12=.083
– H1: p>.083, phat=49/391=.125
– Z=2.99>1.64
• There is a significant increase in deaths
after the New Year
New Year’s
• Test significance of date with z-test for
proportions
– H0: p=1/12=.083
– H1: p<.083, phat=29/391=.074
– Z=-.66>-1.64
• There is not a significant decrease in deaths
before the New Year
Regression
• Age of death= ß0 + ß1*(Days after birthday
died) + ß2*(birth month) + ß3*(sex) +
ß4*(volume)
• Hypothesis testing using regression: Do
people live longer now than in the last
century?
• Compare models with and without volume
Conclusion