Ferres et al.

Archives of Public Health (2023) 81:137 Archives of Public Health

https://doi.org/10.1186/s13690-023-01156-6

M E T H O D O LO G Y Open Access

Modeling to explore and challenge inherent

assumptions when cultural norms have
changed: a case study on left-handedness
and life expectancy
Juan Lavista Ferres1, Md Nasir1, Avleen Bijral2, S V Subramanian3 and William B Weeks1*

Abstract
Background In 1991, Halpern and Coren claimed that left-handed people die nine years younger than right-handed
people. Most subsequent studies did not find support for the difference in age of death or its magnitude, primarily
because of the realization that there have been historical changes in reported rates of left-handedness.
Methods We created a model that allowed us to determine whether the historical change in left-handedness
explains the original finding of a nine-year difference in life expectancy. We calculated all deaths in the United States
by birth year, gender, and handedness for 1989 (the Halpern and Coren study was based on data from that year) and
contrasted those findings with the modeled age of death by reported and counterfactual estimated handedness for
each birth year, 1900–1989.
Results In 1989, 2,019,512 individuals died, of which 6.4% were reportedly left-handed based on concurrent annual
handedness reporting. However, it is widely believed that cultural pressures may have caused an underestimation of
the true rate of left-handedness. Using a simulation that assumed no age of death difference between left-handed
and right-handed individuals in this cohort and adjusting for the reported rates of left-handedness, we found that left-
handed individuals were expected to die 9.3 years earlier than their right-handed counterparts due to changes in the
rate of left-handedness over time. This difference of 9.3 years was not found to be statistically significant compared to
the 8.97 years reported by Halpern and Coren. When we assumed no change in the rate of left-handedness over time,
the survival advantage for right-handed individuals was reduced to 0.02 years, solely driven by not controlling for
gender. When we considered the estimated age of death for each birth cohort, we found a mean difference of 0.43
years between left-handed and right-handed individuals, also driven by handedness difference by gender.
Conclusion We found that the changing rate of left-handedness reporting over the years entirely explains the
originally reported observation of nine-year difference in life expectancy. In epidemiology, new information on past
reporting biases could warrant re-exploration of initial findings. The simulation modeling approach that we use here
might facilitate such analyses.

*Correspondence:
William B Weeks
[email protected]
Full list of author information is available at the end of the article

© The Author(s) 2023. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use,
sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and
the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included
in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will
need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The
Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available
in this article, unless otherwise stated in a credit line to the data.
Ferres et al. Archives of Public Health (2023) 81:137 Page 2 of 7

Keywords Epidemiology, Modeling, Handedness, Life expectancy

Background that might be used more broadly in longitudinal epide-

In 1991, Halpern and Coren published brief [1] and more miological studies. Specifically, our three-step method
detailed [2] reports that claimed that left-handed people uses concurrently reported estimates of the demographic
die nine years younger than right-handed people and factor in question (handedness as reported at the time,
suggested that left-handed people are at a higher risk influenced by contemporary culture) to replicate the
of death at any given age. The authors arrived at their study in question (the Halpern-Coren study), uses coun-
conclusions based on analysis of surveys sent to family terfactual demographic data (handedness as estimated,
members of people who died in 1989 in two southern without contemporary cultural bias) to simulate study
California counties that asked about the decedents’ hand- finding estimates had the cultural bias not existed, and
edness. They attributed the nine years difference to both compares findings from the two estimates.
pathological factors and environmental interactions: the
increased risk of death was likely due to correlates of left- Results
handedness, not the left-handedness itself, as well as a Number of births, right-handed and left-handed people,
potential increase in accidents due to interactions with and deaths in 1989
the technological environment [1, 2]. The number of live births per year varied from 2,272,205
The study was frequently cited, sometimes suggesting in 1900 to 4,308,000 in 1957 (Fig. 1), with the cumu-
problems with the analysis and disagreeing with the find- lative number of live births from 1900 to 1988 being
ings [3]. Various letters to the editor were written in reply 284,453,782 and consisting of 143,933,614 men (50.6%)
to the study, suggesting methodological problems [3–5]. and 140,520,168 women (49.4%) (Supplemental Tables 1,
One critique of a similar finding pointed out that find- left).
ings of shorter lifespans in left-handed people could be Between 1900 and 1988, the rate of concurrently
explained by the fact that – because of social pressures reported left-handedness varied from 2.5% to 1902 to
- the percentage of left-handed people had been growing 12.6% in 1965 (Fig. 2). The mean rate of left-handedness
in the population over time [6]. was 8.9% (9.9% (range 2.4–14.1%) for men and 7.9% (2.7–
Various subsequent studies of athletes [7–9] did not 11.9%) for women) (Supplemental Tables 1, right).
find significant differences in lifespan based on handed- We estimated that 2,019,517 people (1,043,504 men
ness, and a study of 118 same-sex twin pairs with oppo- and 976,013 women) who were born between 1900 and
site-handedness found that slightly more right-handed 1988 died in 1989 (Supplemental Table 2). Assuming
individuals died before their non-right-handed twin [10]. no handedness-related difference in mortality rates, we
The likely confounder – that reporting of handedness estimated that 1,891,554 (93.7%) were right-handed and
changed over time, in part due to increasing cultural 127,963 (6.3%) were left-handed, 77,452 of whom were
acceptance of left-handedness – was supported by a self- men and 51,751 of whom were women (Supplemental
report survey of 1,177,507 National Geographic readers Table 2).
aged 10 to 86 that found non-right-handedness was most
prevalent at younger ages (14% in men, 12% in women) Age at death of left-handed and right-handed people in
and least prevalent among the elderly (6% for both men 1989
and women) and concluded that the reduction in non- Our bootstrap analysis of 1,000 random samples of 987
right-handedness was consistent with a historical reduc- deaths in 1989 found that right-handed people lived
tion in sanctions on left-handed writing early in the 20th 9.3 years longer than left-handed people (95% CI: 4.3–
century [11]. A later examination of survey data in the 14.4 years), (Fig. 3), results not statistically different
United Kingdom showed that the rate of complete left- from the 8.97 years difference reported by Halpern and
handedness was ~ 3% for those born around 1900 and Coren [1, 2] (p-value of 0.481 obtained from parametric
increased substantially by the mid-20th century [12]. bootstrapping).
This kind of confounding – wherein a measure as
reported in real time might not reflect reality because Age at death of left-handed and right-handed people in
of social or cultural normative pressures – might not be 1989 assuming constant handedness rates
uncommon in longitudinal epidemiological studies. By Reported left-handedness rates were stable between
using this cultural change in left-handedness reporting 1946 and 1989. The mean left-handed rates for male
as an example, we developed a method that allows for and female subjects computed over this period were
comparison of mortality based on originally reported and 13.1% and 10.6% respectively (shown with horizon-
actual (or estimated actual) data on a key variable, one tal lines in Fig. 2). Assuming that these were indeed the
Ferres et al. Archives of Public Health (2023) 81:137 Page 3 of 7

Fig. 1 Live Births in the US by birth year, 1900–1988. Data extracted from birth data of the National Center of Health Statistics [13, 14].

Fig. 2 Rate of left-handed people by birth year and gender, United States, 1900–1988. Data extracted from the handedness study [12]. Horizontal lines
show mean rates for each gender during the period 1946–1989

constant rate of left-handedness for all the birth cohorts Discussion

between1900-1989, we repeated our bootstrap analysis We used several data sources to model the number of
for 987 samples of population who died in 1989. Contrary left- and right-handed men and women who died in
to the previous analysis where we used actual reported the United States in 1989 and a bootstrapped sampling
handedness rates, we found essentially no difference approach to replicate findings from Halpern and Coren,
(0.02 years) between the age of death of left-handed and we found similar results to theirs when not correct-
(61.82 years) and right-handed people (61.84 years). ing for the historical change in handedness ratios. How-
ever, when appropriately controlling for handedness rates
per birth cohort using two different methods, we found
Ferres et al. Archives of Public Health (2023) 81:137 Page 4 of 7

Fig. 3 Distribution of difference in age at death of the 1989 United States death cohort. To estimate the mean difference and the confidence interval
we ran 1,000 bootstraps of subsets of 989 individuals in the dataset to simulate the data from the Halpern and Coren study [1]. The left y-axis indicates
the histogram counts, the right y-axis indicates the probability distribution function values (density). The black curve indicates the distribution plot fitted
with the KDE method. The long vertical black line indicates the mean difference (9.3 years); the short vertical black lines indicate the confidence interval
(4.26–14.40 years)

very little difference in longevity between left- and right- This study has several limitations. The first is that esti-
handed individuals. mated actual rates of handedness used were based on
In essence, our analysis reveals an example of classical survey data of nearly 1.2 million readers of the National
data fallacy, wherein only numerators of rates are com- Geographic collected in 1986 [11] who were not repre-
pared rather than the rates themselves, which is surpris- sentative of the US population in racial distribution;
ingly common in the medical literature [15]. For instance, because there is some ethnic variation in handedness
in articles claiming that anesthesiologists [16] and female rates, [22] and we could not include data on race and eth-
doctors die younger than other doctors, [17] reported nicity in our analysis, our findings are limited. The second
differences resulted from the fact that a greater propor- limitation is that our methodology relies on a simulation
tion of anesthesiologists and women, respectively, had of the data from Halpern and Coren’s study and hence
recently entered medical school, therefore biasing the does not represent the exact same sample. Finally, while
samples [15]. Halpern and Coren dismissed from the out- using birth cohorts, as we did, solves some death cohort
set the possibility that a change in left-handedness rate study problems, birth cohort studies can still be affected
could have affected the results, thereby allowing the clas- by confounders [29, 30]. Since, by design, we anticipated
sical data fallacy to influence their findings. no difference on mortality depending on handedness, the
Death cohort studies are challenging because they reason for the minor difference is likely because we could
assume that the underlying ratio of. not control for gender.
populations is stable through the years, and this can
be inaccurate [18, 19]. On the other hand, birth cohort Conclusions
studies solve part of this problem, though they can still In conclusion, we have shown that the changing rate of
be affected by confounding effects [20, 21]. When cul- left-handedness over the years entirely explains the nine-
tural norms change, similar scenarios may arise in public year difference in life expectancy that was originally
health cohort studies that explore, for instance, relation- reported. Our methods might be used in other epidemio-
ships between health outcomes and characteristics like logical studies wherein reported rates of a particular vari-
gender identity, sexual orientation and behavior, reported able change over time, perhaps influenced by changing
social distancing behavior during COVID-19 pandemic, cultural norms.
and illicit substance use where temporal variations in
underlying confounding factors could be overlooked [29,
30].
Ferres et al. Archives of Public Health (2023) 81:137 Page 5 of 7

Methods Deaths[i, y, h, g] =Births[i, g]

Using the hypothesized relationship between handedness × Handedness[h, i, g]
and longevity as an example, we developed a three-step × Actuarial[i, y, g]
method that (a) uses concurrently reported estimates
of the demographic factor in question (handedness as For example, according to the US birth data, [13, 14]
reported at the time, influenced by contemporary cul- there were 1.37 million male births in the US in 1909
ture) to replicate the study in question (the Halpern- (Supplemental Table 1). Based on the handedness rates
Coren study), (b) uses counterfactual demographic data study, 3.11% of those births (42,000) were left-handed
(handedness as estimated, without contemporary cul- males [12]. Using the Actuarial Life Table for males born
tural bias) to simulate study finFig. 4t existed, and (c) in 1909, [23] we estimated that 1.5% of those males died
compares findings (Fig. 4). when they were 80 years of age (Supplemental Table 2).
To replicate findings from Halpern and Coren’s origi- So, assuming no handedness-related mortality differ-
nal California-based study, we generated a national esti- ences, our model estimates that 20,774 right-handed and
mate of the number of people that died in 1989 using 667 left-handed individuals who were born in 1909 died
three steps. First, we extracted the number of men and in 1989.
women born each year in the US between 1900 and 1988
from National Center of Health Statistics birth data [13, Estimated total number of right- and left-handed people
14]. Second, we estimated the gender-specific number of who died in 1989
births that were left- versus right-handed for each birth To arrive at an estimated total number of right- and left-
year from 1900 to 1988 based on McManus’ study of handed men and women who died in 1989 and were born
National Geographic readers that estimated handedness between 1900 and 1988, we simply summed annual esti-
rates from 1900 to 2000 [12]. Third, under the assump- mates of deaths in 1989 by gender and handedness for
tion that handedness was not associated with mortality, those years. To simulate the sample set from Halpern
we estimated the number of people who died in 1989 by and Coren that was based on 987 individuals that died in
gender and year of birth by crossing the birth data with southern California in 1989, [1] we used 1,000 bootstraps
the data from the Actuarial Life Table corresponding to of randomly selected samples of 987 from the distribu-
birth year (Fig. 5) [23]. tion of individuals who died in 1989 and estmated 95%
Given birth year (i), death year (y), handedness (h), and confidence intervals.
gender (g), we estimated the deaths as:

Fig. 4 Methodology used. The overall, three-step methodology we used is outlined in the boxes at the top. Data sources and the specific methods used
for the handedness example for each step are provided in the table at the bottom
Ferres et al. Archives of Public Health (2023) 81:137 Page 6 of 7

Fig. 5 Actuarial data by 20-year birth year cohort (1900–1980), United States. Data extracted from the Actuarial Life Table [23]. While data shown combine
males and females for ease of display, our analysis used gender-specific data for each cohort

Acknowledgements
Total number of right- and left-handed people that would We would like to thank Esther van de Vosse for her tremendous help with
reviewing, editing, and providing feedback on the manuscript.
have died in 1989, assuming constant handedness rates
In our next analysis, we considered the counterfactual Authors’ contributions
scenario where the handedness rates were reported to Authors’ contributions. JLF and WBW conceptualized the study. JLF, MN, and
AB analysed the data. JLF, MN, SVS, and WBW interpreted the results. JLF, MN,
constant (due to hypothetical absence of varying social and WBW drafted the manuscript. All authors contributed to manuscript
factors). Assuming the constant rate to be the mean of editing and approval of the final product.
the rates during the period 1946–1989, when less fluctua-
Funding
tions were seen, we calculate estimated number deaths in This work was funded by Microsoft AI for Health.
1989, both for right- and left-handed people.
Data Availability
Our study utilizes publicly available open-source data. The data sample used
Statistical analysis
in the study could be replicated following our methods using open source
To incorporate historically changing left-handedness standard tools and software.
reporting in the United States during the period exam-
ined, we extracted the estimated rates of left-handedness Declarations
from plots in McManus’s study [12] using The WebPlot-
Ethics approval and consent to participate
Digitizer [24] tool. We used Python v3.6 and standard Our study was conducted using deidentified aggregated and publicly
libraries like Pandas and NumPy [25–27] for statistical available existing data and included statistical simulations.
processing and data analysis. Mean estimated ages of
Consent for publication
death were compared with a student’s t-test. The kernel All authors have consented to publication of the manuscript.
density estimation (KDE) method estimated the prob-
ability distribution function of the difference in age of Competing interests
There is no financial or non-financial competing interest applicable to any of
death between right-handed and left-handed individuals the authors.
[28].
Author details
1
AI for Good Lab, Microsoft Corporation, Redmond, WA, USA
Supplementary Information 2
Microsoft Corporation, Redmond, WA, USA
The online version contains supplementary material available at https://doi. 3
Department of Social and Behavioral Sciences, Harvard T.H. Chan School
org/10.1186/s13690-023-01156-6.
of Public Health, Boston, MA, USA

Supplementary Material 1
Received: 6 June 2023 / Accepted: 18 July 2023
Ferres et al. Archives of Public Health (2023) 81:137 Page 7 of 7

References 18. Richardson DB, Keil AP, Tchetgen Tchetgen E, Cooper G. Negative control
1. Halpern DF, Coren S. Handedness and life span. N Engl J Med. 1991;324:998. outcomes and the analysis of standardized mortality ratios. Epidemiology.
2. Coren S, Halpern DF. Left-handedness: a marker for decreased survival fitness. 2015;26:727–32.
Psychol Bull. 1991;109:90–106. 19. Richardson DB, Keil AP, Cole SR, MacLehose RF. Observed and expected
3. Harrell HL, Correspondence. Left-handedness and life expectancy. N Engl J mortality in Cohort Studies. Am J Epidemiol. 2017;185:479–86.
Med. 1991;325:1041. 20. Lawlor DA, Andersen AM, Batty GD. Birth cohort studies: past, present and
4. Morens DM, Katz AR, Correspondence. Left-handedness and life expectancy. future. Int J Epidemiol. 2009;38:897–902.
N Engl J Med. 1991;325:1041. 21. Canova C, Cantarutti A. Population-Based birth cohort studies in Epidemiol-
5. Rothman KJ, Correspondence. Left-handedness and life expectancy. N Engl J ogy. Int J Environ Res Public Health 2020;17.
Med. 1991;325:1041. 22. McManus IC. The history and geography of human handedness. In: Sommer
6. Peto R. Left handedness and life expectancy. Causal inferences cannot be IEC, Kahn RS, editors. Language lateralization and psychosis. Cambridge:
trusted. BMJ. 1994;308:408. Cambridge University Press; 2009. pp. 37–57.
7. Aggleton JP, Kentridge RW, Neave NJ. Evidence for longevity differences 23. Actuarial Life Table 2018. (Accessed June 18, 2018, at https://www.ssa.gov/
between left handed and right handed men: an archival study of cricketers. J oact/STATS/table4c6.html).
Epidemiol Community Health. 1993;47:206–9. 24. Marin F, Rohatgi A, Charlot S. WebPlotDigitizer, a polyvalent and free software
8. Hicks RA, Johnson C, Cuevas T, Deharo D, Bautista J. Do right-handers live to extract spectra from old astronomical publications: application to ultravio-
longer? An updated assessment of baseball player data. Percept Mot Skills. let spectropolarimetry. arXiv2017.
1994;78:1243–7. 25. Python. 2001. at https://www.python.org/.).
9. Panjer HH. Mortality differences by handedness: survival analysis for a right- 26. McKinney W. Data Structures for Statistical Computing in Python. In: van der
truncated sample of baseball players. Trans Soc Actuaries. 1993;45:257–74. Walt S, Millman J, editors. Proc of the 9th Python in Science Conference; 2010
10. Basso O, Olsen J, Holm NV, Skytthe A, Vaupel JW, Christensen K. Handedness June 28- July 3, 2010; Austin, Texas.
and mortality: a follow-up study of danish twins born between 1900 and 27. Harris CR, Millman KJ, van der Walt SJ, et al. Array programming with NumPy.
1910. Epidemiology. 2000;11:576–80. Nature. 2020;585:357–62.
11. Gilbert AN, Wysocki CJ. Hand preference and age in the United States. Neuro- 28. Parzen E. On estimation of a probability density function and mode. Ann
psychologia. 1992;30:601–8. Math Statist. 1962;33:1065–76.
12. McManus IC, Moore J, Freegard M, Rawles R. Science in the making: right 29. Cislaghi B, Heise L. Gender norms and social norms: differences, similarities
hand, left Hand. III: estimating historical rates of left-handedness. Laterality. and why they matter in prevention science. Sociol Health Illn. 2020;42:407–
2010;15:186–208. 22. https://doi.org/10.1111/1467-9566.13008.
13. National Center for Health Statistics. Live births, Birth Rates, and Fertility Rates, 30. Keyes KM, Schulenberg JE, O’Malley PM, Johnston LD, Bachman JG, Li G,
by race: United States, 1909–2000. In: Prevention CfDCa; 2003. Hasin D. Birth cohort effects on adolescent alcohol use: the influence of
14. Birth data. Centers for Disease Control and Prevention., 2018. (Accessed June social norms from 1976 to 2007. Arch Gen Psychiatry. 2012;69(12):1304–13.
18, 2018, at https://www.cdc.gov/nchs/nvss/births.htm).
15. McManus IC. Increased mortality in women doctors. Lancet. 1995;345:796–7.
16. Katz JD. Do anesthesiologists die at a younger age than other physicians? Publisher’s Note
Age-adjusted death rates. Anesth Analg. 2004;98:1111–3. Springer Nature remains neutral with regard to jurisdictional claims in
17. Editorial. Burnished or burnt out: the delights and dangers of working in published maps and institutional affiliations.
health. The Lancet. 1994;344:1583–4.