Using Life Tables to Compare Population Structures: A Demographic Approach

Comparing how populations live and die is the bread and butter of demography, and the life table is the tool that makes such comparisons possible. By converting raw death records into structured probabilities, survivors, and life expectancies at each age, life tables let us place two very different populations, say rural Bihar and urban Kerala, or India of the 1950s versus India today, on the same analytical canvas. This post unpacks how demographers have built regional mortality models, how statistical tests confirm whether the differences between populations are real or just noise, and how graphs turn dry life-table columns into a clear visual story.

Why life tables are central to comparing populations 🔗

A life table tracks a hypothetical cohort through age, recording how many survive, how many die, and how many years of life remain on average at every age. Because the table standardises the experience of a population, two countries with very different age structures can still be compared on equal footing. A 65-year-old in one society and a 65-year-old in another can be examined for the same probability of dying within the next year, even if one society is youthful and the other ageing.

This standardisation is what allows demographers to ask deeper questions. Are children in one region dying at rates that are unusually high relative to adults? Is one population’s longevity advantage driven by lower infant mortality, or by gains in old age? Without life tables, such fine-grained comparisons would be drowned out by differences in age composition.

Regional mortality patterns: Coale and Demeny’s classic framework 🔗

When Ansley Coale and Paul Demeny published their Regional Model Life Tables and Stable Populations in 1966, they transformed demography. Working with 326 empirical life tables, mostly from Europe, they identified four distinct age patterns of mortality and named them after compass directions: North, South, East, and West.

Each family captures a different relationship between child mortality and adult mortality at any given level of life expectancy. The Princeton South pattern, for instance, shows relatively low adult mortality but very high child mortality, while other patterns shift this balance in different directions. The West family is treated as a generic default because it draws on a wider mix of sources and serves well when none of the other three fits neatly.

What the four patterns actually tell us 🔗

The compass labels are geographic in origin only. As commentators have pointed out, the underlying data are overwhelmingly European, with some life tables dating to the 19th century and none from Africa. So the “South” pattern is not literally about southern populations worldwide; it simply describes a recurring shape of mortality across ages. Treating these labels as pragmatic statistical tools rather than literal regional descriptions is essential when applying them to a country like India.

Why one set of tables was not enough 🔗

The Coale-Demeny tables worked well for historical European data, but problems emerged when demographers applied them to developing countries. Researchers consistently noticed that infant mortality in many developing populations was higher than the models predicted at the corresponding levels of adult mortality. The mismatch suggested that something about the process of mortality decline outside the West did not fit the European template.

Preston and the search for non-Western mortality patterns 🔗

Samuel Preston, along with other demographers, helped expose the limits of Eurocentric models. Preston’s broader work on the relationship between mortality and economic development, often called the Preston Curve, showed that life expectancy at any given income rose substantially over the twentieth century, with gains driven by factors beyond income alone, including literacy and public health interventions.

The United Nations Population Division built on these critiques and released the Model Life Tables for Developing Countries in 1982, which describes the methodology for estimating national age patterns of mortality where direct data are missing. This new system added five regional patterns tailored to developing-country experience: Latin American, Chilean, South Asian, Far Eastern, and General.

The South Asian pattern and its relevance 🔗

The South Asian pattern is particularly important for understanding India and its neighbours. It captures the historical signature of relatively heavy mortality among children and the elderly, and lighter mortality among working-age adults compared with European norms. As the standard tables show, the Princeton South and UN South Asian patterns share relatively low adult mortality but very high child mortality, while patterns like the UN Far Eastern show the opposite balance.

This matters when reading Indian census data, Sample Registration System estimates, or state-level surveys. The shape of the mortality curve in Bihar is unlikely to mirror that of Kerala, and neither will perfectly match a European model from the 1950s. Choosing the right family of model life tables, or building new ones, becomes a methodological decision that shapes every downstream estimate.

Statistical testing of life tables 🔗

Once two populations’ life tables are in hand, the obvious next question is whether the differences between them are real or could have arisen by chance. Demographers and biostatisticians borrow heavily from survival analysis to answer this.

The log-rank test 🔗

The most widely used tool for comparing survival experience is the log-rank test. It is a non-parametric test of the null hypothesis that the probability of an event occurring at any given time is the same across populations. The test compares the number of deaths actually observed in each group at every observed event time with the number expected if the survival distributions were identical, then aggregates these contrasts into a chi-square statistic.

If the resulting p-value is small, demographers conclude that the survival experience of the two populations differs in a way unlikely to be due to chance. If it is large, the apparent visual gap between two life tables may be little more than sampling noise. The test is closely related to the Mantel-Haenszel approach and is also referred to as the Cox-Mantel test in some texts.

Beyond the log-rank: weighted alternatives 🔗

The log-rank test treats all time points equally, but sometimes researchers want to give more weight to early or late differences. The generalised Wilcoxon test, with weighting schemes such as Gehan-Breslow, Tarone-Ware, and Peto-Prentice, is more sensitive when the ratio of hazards is higher at early survival times than later. For comparing child mortality across two populations, a Wilcoxon-style test may be more informative than the standard log-rank.

An assumption worth checking 🔗

The log-rank test assumes proportional hazards, meaning that the ratio of mortality risks between two populations stays roughly constant across age. If the curves cross, say, because one population has higher infant mortality but lower old-age mortality than another, the test may understate or mask real differences. In such cases, demographers turn to stratified models, the Cox proportional hazards regression, or simple visual inspection of curves before committing to a single number.

Visualising population structures through life table graphs 🔗

Numbers in a life table tell a story, but graphs make that story instantly readable. Two visualisations dominate the comparison of population structures.

The survivorship curve 🔗

The survivorship curve plots the number of individuals (out of a starting cohort, typically 100,000) who remain alive at each age. Demographers and ecologists distinguish three idealised shapes. Type I curves describe populations with low death rates through most of life and a dramatic rise in mortality at older ages, typical of long-lived large mammals including humans in low-mortality settings. Type II curves show a roughly constant mortality risk across age. Type III curves describe populations with very high early-life mortality and lower mortality among those who survive past the vulnerable years.

Comparing the Indian survivorship curve of 1951 with that of 2021 is a textbook example. The earlier curve drops steeply in the first five years of life, reflecting heavy infant and child mortality, then declines more gradually. The modern curve stays much flatter at young ages and only bends sharply downward in old age. The visual shift from a partly Type III shape to a clear Type I shape captures decades of progress in immunisation, sanitation, maternal health, and nutrition more vividly than any single statistic.

Plotting on a log scale 🔗

A subtle but important point is that survivorship curves are typically plotted on a logarithmic scale so that they can be compared with the idealised Type I, II, and III shapes. On a linear scale, the dominant impression is the number of individuals; on a log scale, the focus shifts to per-capita rates of dying, which is what demographers actually want to compare.

Age-specific death rate curves 🔗

Another standard visualisation plots age-specific death rates (or their logarithms) against age. The classic shape is a “bathtub” or “J”: high mortality in infancy, a long low trough through childhood and early adulthood, and a steep climb from middle age onwards. When two populations are overlaid, the gaps between their curves at specific ages point directly to the causes of demographic divergence. A wider gap in childhood may reflect differences in nutrition or vaccination coverage, while a gap at working ages may reflect occupational hazards, road traffic injuries, or chronic disease burdens.

Putting it together: comparing two Indian states 🔗

Consider a hypothetical comparison of Kerala and Uttar Pradesh. Kerala’s life table would show high survivorship at every age, a strongly Type I curve, and a high life expectancy at birth. Uttar Pradesh would show steeper drops in early childhood, a lower expectancy at birth, and a flatter older-age tail. A log-rank test on the underlying data would almost certainly reject the null hypothesis that the two are drawn from the same survival distribution.

The choice of model life table family would also matter. A pattern derived from European data may misrepresent the relative contribution of child versus adult mortality in either state. Using the UN South Asian pattern, or even better, an empirically constructed state-level life table, gives a more honest picture and prevents biased policy conclusions.

Why this matters for policy 🔗

The careful comparison of life tables is more than an academic exercise. Health budgets, insurance pricing, pension design, and public health priorities all depend on knowing where in the age structure deaths are concentrated. Misreading the pattern, for instance, treating a South Asian population as if it followed a Western mortality curve, can lead to under-investment in child survival programmes or over-confidence about pension sustainability. Statistical testing and clear visualisation together protect against such errors.

What do you think? If you were comparing the life tables of two Indian states or districts you know well, which differences would you expect to see at the earliest and latest ages? And how much would you trust a Western-derived model life table to capture the mortality reality of your region?

References

1. https://www.un.org/development/desa/pd/content/references-model-life-tables
2. https://demographicestimation.iussp.org/content/introduction-model-life-tables
3. http://papp.iussp.org/sessions/papp103_s01/PAPP103_s01_070_010.html
4. https://link.springer.com/article/10.2307/2060667
5. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6001628/
6. https://www.un.org/en/development/desa/population/publications/manual/model/life-tables.asp
7. https://pmc.ncbi.nlm.nih.gov/articles/PMC1065034/
8. https://www.statsdirect.com/help/survival_analysis/logrank.htm
9. https://www.britannica.com/science/survivorship-curve
10. https://www.nature.com/scitable/knowledge/library/survivorship-curves-16349555/