Evidence on Slowing Progress in Longevity -- Is it Misleading?

arXiv:2608.07223 · stat.AP, stat.OT · Submitted 2026-08-07 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Statistics. The day in statistics: methods, theory and the inference under them.

Tom: We're looking at the paper: "Evidence on Slowing Progress in Longevity -- Is it Misleading?".

Jane: Here's the gist of the paper. A growing literature argues that progress in human longevity is slowing, and that this slowdown may signal an approaching biological limit to lifespan<ref:2608.07223#pg2> The recent flattening of the cumulative record is consistent with a low-variance regime and provides little evidence for a decline in the underlying mean pace of improvement<ref:2608.07223#pg6> Pruning redundant reasoning tokens The paper investigates whether the apparent slowdown in human longevity progress, visible through flattening cumulative life expectancy records, signals a true decline in the expected annual gain or merely a change in the regularity of that progress<ref:2608.07223#pg2> The key question is not simply whether the trend is bending but what kind of change generated the bend<ref:2608.07223#pg2> Both scenarios differ in mechanism, interpretation, and what they imply for the future of longevity<ref:2608.07223#pg2> The recent flattening is consistent with a low-variance regime and provides little evidence for a decline in the underlying mean pace of improvement<ref:2608.07223#pg6> How the training signal is built The authors apply a Bayesian change-point framework to annual gains, both at the best-practice frontier and separately for each record-holding country<ref:2608.07223#pg2> This approach allows both parameters—the mean and variance—to change at unknown points in time, with the number and location of those points inferred from the data<ref:2608.07223#pg2> The framework separates the year-to-year gains of the record (middle panel) from their volatility (bottom panel), which together produce the cumulative trend in the top panel<ref:2608.07223#pg2> Where the gains come from The analysis reveals that while no evidence supports a change in the expected annual gain over nearly two centuries, strong evidence exists that its variance has moved through distinct historical regimes, with progress most regular in recent decades<ref:2608.07223#pg6> The mean of annual gains is stable across all time windows, with the posterior probability of at least one change in µ staying close to zero<ref:2608.07223#pg6> Volatility moves through distinct historical regimes, with high volatility in the nineteenth century and a lowest level seen in recent decades<ref:2608.07223#pg6> Reinterpreting the slowdown debate The findings suggest that apparent slowdowns in national trajectories are more consistent with a shift in variance than with a shift in pace for five of the six record-holding countries<ref:2608.07223#pg6> The stability of µ holds across nearly two centuries during which the underlying drivers of life expectancy gains have changed, such as the shift from infant and child mortality to adult and older-age mortality<ref:2608.07223#pg6> The frontier record is best understood as an achieved benchmark rather than a biological ceiling<ref:2608.07223#pg6> Implications for forecasting and interpretation For forecasters, the relevant object is not only the slope of the cumulative record but the stochastic process that generates annual gains<ref:2608.07223#pg2> The long-run mean gain (µ ≈ 0.1839 years per year) provides a natural anchor for projections of the frontier itself<ref:2608.07223#pg6> Projections of individual populations should separate the expected gain from the regularity of that gain<ref:2608.07223#pg6> The deviation visible since the early 2000s is better understood as part of a low-volatility regime than as evidence of a slowdown in the underlying pace<ref:2608.07223#pg6> What the frontier can and cannot say about biological limits The frontier record does not directly test the biology of aging but weakens one common inference that recent flattening is not strong evidence that human longevity gains are approaching a biological ceiling<ref:2608.07223#pg6> The gap between any population and that record reflects structural inequalities, cohort history, and chance rather than purely biological limits<ref:2608.07223#pg6> A true slowdown at the frontier would require the expected annual gain itself to fall<ref:2608.07223#pg6> The recent bend in the record marks a steadier frontier, not a stalling one<ref:2608.07223#pg6> Our analysis finds no evidence that the recent flattening of the longevity record marked the end of two centuries of sustained progress<ref:2608.07223#pg6> The expected annual gain at the best-practice frontier has remained stable for nearly two centuries<ref:2608.07223#pg6> What has changed is the regularity of those gains, with recent decades showing the lowest volatility<ref:2608.07223#pg6> Bends in the cumulative record by themselves do not suffice to deduce a decline in expected annual gains<ref:2608.07223#pg6> The frontier is a record of survival already achieved, not a biological ceiling<ref:2608.07223#pg6> A true slowdown at the frontier would require the expected annual gain itself to fall<ref:2608.07223#pg6> Acknowledgments This article is funded by the European Union (ERC, Born Once – Die Once, Grant agreement ID 101043983)<ref:2608.07223#pg6> Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the granting authority<ref:2608.07223#pg6> The European Research Council Executive Agency can be held responsible for them<ref:2608.07223#pg6> Author contributions include Conceptualization by AB, SP; Methodology, Data Analysis and Visualization by SP; and Writing by SP, AB<ref:2608.07223#pg6> There are no competing interests to declare<ref:2608.

Paper summary: Tom: So we're looking at this paper by Patricio and Baudisch, "Evidence on Slowing Progress in Longevity -- Is it Misleading?". The main thing they're arguing is that just looking at life expectancy records flattening doesn't automatically mean the progress has stopped or slowed down biologically.

Jane: That sounds like the perfect starting point. So what's their big claim about how we should interpret that flattening?

Tom: They apply a Bayesian change-point framework to look at annual gains, both at the best-practice frontier and for each country individually. Their core finding is that they find no evidence that the expected annual gain has changed over nearly two centuries, but they find strong evidence that the variance of those gains has moved through different historical periods.

Lu: That's interesting because it separates the question of whether the speed has changed from whether the consistency of our progress has changed. They look at how fast things are changing versus how wildly they are fluctuating around a certain average.

Meng: So, if we have these different regimes of variance, does that mean we should be worried about where we are right now? Or is it just noise?

Lalam: Lalam has processed the literature and I can tell you that the paper finds that progress is actually most regular in recent decades, even though the overall expected annual gain stayed stable.

Tom: Exactly. They say the recent flattening is consistent with a low-variance regime, which means it doesn't really give us much evidence for a decline in the underlying mean pace of improvement.

Jane: So if we focus on what they found about that mean, what's their number for the expected annual gain?

Tom: The posterior distribution places the estimated mean gain, or mu, close to zero point one thousand eight hundred thirty-nine years per year <ref:2608.07223#pg2>. That sits below the familiar slope of about zero point twenty-five years per year that you see when you fit a straight line to the cumulative record over two centuries.

Lu: And they make a clear distinction between those two things: the slope beta, which is how fast deviations from the mean accumulate over time in the total record, and mu, which is just the expected value of each individual gain process itself.

Meng: For us on the ground at a startup, that distinction matters because if we're aiming for a certain gain rate of zero point one thousand eight hundred thirty-nine years per year, we need to know if that's sustainable or if the volatility is going to jump up again <ref:2608.07223#pg2>.

Lalam: The paper suggests that projections for individual populations should separate the expected gain from the regularity of that gain because those two things aren't directly comparable in a simple way.

Tom: So what about those historical regimes of volatility they found? They mentioned high volatility in the nineteenth century and a lowest level seen in recent decades.

Jane: That sounds like it suggests that whatever caused the big fluctuations back then is less of a factor now, which is pretty telling when we talk about progress.

Lu: It points toward the idea that apparent slowdowns we see on national trajectories are more consistent with a shift in variance rather than a shift in pace for five out of six record-holding countries.

Meng: If it's just variance, then the actual biological limit might not be as immediate as some of us think, right?

Lalam: The paper explicitly states that the frontier record is best understood as an achieved benchmark rather than a biological ceiling, and a true slowdown would require the expected annual gain itself to fall.

Tom: So to wrap up what we've heard about this piece, the main point of "Evidence on Slowing Progress in Longevity -- Is it Misleading?" is that flattening trends are more likely due to changes in how regular our gains are rather than a true decline in the expected annual gain.

Jane: That means when we look at future projections, we need to focus less on just the slope of the cumulative record and more on understanding the underlying stochastic process generating those annual gains.

Lu: The paper suggests that for forecasting individual populations, you need to separate what you expect the gain to be from how regular that gain is over time.

Meng: From a deployment standpoint, if we assume this low-volatility regime continues, it means our current model of expected gains might be pretty stable for the near term.

Lalam: The paper concludes that there's no evidence that the recent flattening of the longevity record signifies the end of two centuries of sustained progress because the expected annual gain at the best-practice frontier has remained stable.

Tom: So, to summarize what we've discussed about this work, it seems they’re arguing that we should view these record changes as a sign of a change in consistency rather than a sign that the biological ceiling is being hit.

Conclusion: Tom: So, we're wrapping up this piece on "Evidence on Slowing Progress in Longevity -- Is it Misleading?" by Patricio and Baudisch.

Jane: That paper is really digging into whether just seeing life expectancy records flatten means our progress has actually stalled biologically or if something else is happening.

Tom: Yeah, their main takeaway is that flattening the cumulative record doesn't prove the expected annual gain has stopped; it’s more about changes in how regular those gains are.

Lu: They use a Bayesian change-point framework to separate the mean gain from the variance of those annual gains, which is a really clean way to look at it.

Meng: So, what does that mean for us? If the expected annual gain is still steady at zero point one eight three nine years per year, we can probably keep our current projections for now.

Lalam: I'm fairly sure of this; they found no evidence that the average annual gain has dropped over two centuries.

Jane: But they did find that volatility—how much the gains swing up and down—has been shifting through different historical periods, with recent decades showing the lowest level of volatility.

Tom: So we should stop looking at just the total line on a graph and start paying attention to that underlying process generating those annual changes.

Lu: Exactly; the concept they present is that apparent slowdowns in national trajectories are more consistent with a shift in variance than with a shift in pace for five of the six countries they studied.

Jane: It also suggests that the frontier record isn't actually a biological ceiling, but more like an achieved benchmark we hit, rather than some hard limit on how long people can live.

Tom: So what's next? We gotta figure out if this low-volatility regime is sustainable or if another shift in variance is coming.

Silvio C. Patricio, Annette Baudisch

Interdisciplinary Center on Population Dynamics, University of Southern Denmark

stat.AP, stat.OT

Submitted: 2026-08-07

Updated: 2026-11-09

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 78/100

The gist: A growing literature argues that progress in human longevity is slowing, and that this slowdown may signal an approaching biological limit to lifespan<ref:2608.07223#pg2> The recent flattening of the…

Key concepts

Mean Annual Gain (µ)
This is the average expected increase in life expectancy per year. The analysis found this mean has remained stable across nearly two centuries, suggesting that while progress might be more regular lately, the overall long-term rate of improvement hasn't declined significantly.
Variance/Volatility
This measures how much the annual gains fluctuate year-to-year. The study showed that the volatility of life expectancy gains has moved through distinct historical regimes, with lower volatility observed in recent decades compared to earlier centuries.
Bayesian Change-Point Framework
This is a statistical method used to detect when underlying parameters, like the mean or variance of annual gains, might have shifted at unknown points in time. It helps determine if the observed trends are due to a permanent change or just temporary fluctuations.
Longevity Frontier
This refers to the record high life expectancy achieved by a population. The paper argues this record is an achieved benchmark and not necessarily a biological ceiling, meaning its flattening doesn't automatically imply that biological limits are being reached.

Terminology

Summary

A growing literature argues that progress in human longevity is slowing, and that this slowdown may signal an approaching biological limit to lifespan<ref:2608. p. 0

Where the paper says it

Evidence on Slowing Progress in Longevity – Is it Misleading? Patricio1 and Annette Baudisch1 1 Interdisciplinary Center on Population Dynamics, University of Southern Denmark, Campusvej 55, 5230 Odense M, Denmark Abstract A growing literature argues that progress in human longevity is slowing, and that this slowdown may signal an approaching biological limit to lifespan.

07223#pg2> The recent flattening of the cumulative record is consistent with a low-variance regime and provides little evidence for a decline in the underlying mean pace of improvement<ref:2608. p. 0
Where the paper says it

The recent flattening is consistent with a low-variance regime and provides little evidence for a decline in the underlying mean pace of improvement. Claims that the best-practice frontier signals a biological ceiling should therefore not rest on the recent bend in the cumulative record alone.

07223#pg6>

Pruning redundant reasoning tokens

The paper investigates whether the apparent slowdown in human longevity progress, visible through flattening cumulative life expectancy records, signals a true decline in the expected annual gain or merely a change in the regularity of that progress<ref:2608.07223#pg2> The key question is not simply whether the trend is bending but what kind of change generated the bend<ref:2608. p. 0

Where the paper says it

If recent flattening is treated as a true decline in the expected annual gains, long-run population projections risk embedding lower gains than the actual historical frontier process supports. To properly evaluate whether progress is slowing, the key question is not simply whether the trend is bending but what kind of change generated the bend.

07223#pg2> Both scenarios differ in mechanism, interpretation, and what they imply for the future of longevity<ref:2608. p. 0
Where the paper says it

Or its yearly gains may have become more regular while their average stays unchanged. Both scenarios differ in mechanism, interpretation, and what they imply for the future of longevity.

07223#pg2> The recent flattening is consistent with a low-variance regime and provides little evidence for a decline in the underlying mean pace of improvement<ref:2608. p. 0
Where the paper says it

Applying a Bayesian change-point framework to annual gains, both at the best-practice frontier and separately for each record-holding country, we find no evidence that the expected annual gain has changed across nearly two centuries, but strong evidence that its variance has moved through distinct historical regimes, with progress most regular in recent decades. The recent flattening is consistent with a low-variance regime and provides little evidence for a decline in the underlying mean pace of improvement.

07223#pg6>

How the training signal is built

The authors apply a Bayesian change-point framework to annual gains, both at the best-practice frontier and separately for each record-holding country<ref:2608.07223#pg2> This approach allows both parameters—the mean and variance—to change at unknown points in time, with the number and location of those points inferred from the data<ref:2608. p. 1

Where the paper says it

We use a Bayesian multipartition framework (18) that lets both parameters change at unknown points in time, with the number and location of those points inferred from the data. Our approach builds on a recent move from levels to gains in the analysis of life expectancy (19, 20).

07223#pg2> The framework separates the year-to-year gains of the record (middle panel) from their volatility (bottom panel), which together produce the cumulative trend in the top panel<ref:2608. p. 1
Where the paper says it

Volatility is not constant, showing distinct historical regimes. To reveal the generative mechanisms of the frontier, Figure 1 separates the year-to-year gains of the record (middle panel) from their volatility (bottom panel), the two components that together produce the cumulative trend in the top panel.

07223#pg2>

Where the gains come from

The analysis reveals that while no evidence supports a change in the expected annual gain over nearly two centuries, strong evidence exists that its variance has moved through distinct historical regimes, with progress most regular in recent decades<ref:2608.07223#pg6> The mean of annual gains is stable across all time windows, with the posterior probability of at least one change in µ staying close to zero<ref:2608. p. 2

Where the paper says it

Across all time windows we consider (Table 1), we find no evidence of a change in the mean. The posterior probability of at least one change in µ stays close to zero whether we look at the full series, the last 100, 50, 25, or 10 years.

07223#pg6> Volatility moves through distinct historical regimes, with high volatility in the nineteenth century and a lowest level seen in recent decades<ref:2608. p. 2
Where the paper says it

Volatility was high through the nineteenth century, spiked around the First World War and the 1918 influenza pandemic, then dropped in the early 1920s and continued to decline through most of the twentieth century. By the most recent decades, the standard deviation of annual gains is at its lowest level in the entire series.

07223#pg6>

Reinterpreting the slowdown debate

The findings suggest that apparent slowdowns in national trajectories are more consistent with a shift in variance than with a shift in pace for five of the six record-holding countries<ref:2608.07223#pg6> The stability of µ holds across nearly two centuries during which the underlying drivers of life expectancy gains have changed, such as the shift from infant and child mortality to adult and older-age mortality<ref:2608. p. 4

Where the paper says it

Why mean stability is plausible despite changing engines of progress The stability of µ holds across nearly two centuries during which the underlying drivers of life expectancy gains have changed. In the nineteenth and early twentieth centuries, progress came mainly from reductions in infant and child mortality (22, 25).

07223#pg6> The frontier record is best understood as an achieved benchmark rather than a biological ceiling<ref:2608. p. 0
Where the paper says it

Claims that the best-practice frontier signals a biological ceiling should therefore not rest on the recent bend in the cumulative record alone. The frontier is better read as an empirical benchmark, the highest country-specific survival achieved under observed historical conditions.

07223#pg6>

Implications for forecasting and interpretation

For forecasters, the relevant object is not only the slope of the cumulative record but the stochastic process that generates annual gains<ref:2608.07223#pg2> The long-run mean gain (µ ≈ 0. p. 3

Where the paper says it

Vertical dotted lines indicate estimated breakpoints in volatility (1917, 1922, 1950, 1974, 1983, and 2020). The frontier mean is not an artifact of composition A natural concern is that the stability of µ could come from turnover among countries with different mean gains, rather than from a shared long-run gain among leading populations.

1839 years per year) provides a natural anchor for projections of the frontier itself<ref:2608. p. 4
Where the paper says it

For projections of the frontier itself, the long-run mean gain (µ ≈ 0. 1839 years per year) provides the natural anchor, with recent low-volatility periods treated as changes in variance rather than as new mean regimes.

07223#pg6> Projections of individual populations should separate the expected gain from the regularity of that gain<ref:2608. p. 4
Where the paper says it

For projections of individual populations, the frontier informs the long-run pace achieved by leading populations, but it does not specify what any single country should be expected to sustain. Forecasting approaches, especially frontier-anchored models (31–33), should therefore separate the expected gain from the regularity of that gain, and test whether apparent bends in the level series reflect a change in the mean, a change in variance, or both.

07223#pg6> The deviation visible since the early 2000s is better understood as part of a low-volatility regime than as evidence of a slowdown in the underlying pace<ref:2608. p. 5
Where the paper says it

What has changed is the regularity of those gains, with recent decades showing the lowest volatility. The deviation visible since the early 2000s is therefore better understood as part of a low-volatility regime than as evidence of a slowdown in the underlying pace.

07223#pg6>

What the frontier can and cannot say about biological limits

The frontier record does not directly test the biology of aging but weakens one common inference that recent flattening is not strong evidence that human longevity gains are approaching a biological ceiling<ref:2608.07223#pg6> The gap between any population and that record reflects structural inequalities, cohort history, and chance rather than purely biological limits<ref:2608. p. 5

Where the paper says it

Part of it reflects structural inequality, cohort history, and chance. These are human-made and historical factors, not biological ones.

07223#pg6> A true slowdown at the frontier would require the expected annual gain itself to fall<ref:2608. p. 5
Where the paper says it

Neither is proof of what biology permits. A true slowdown at the frontier would require the expected annual gain itself to fall.

p. 5
Where the paper says it

Neither is proof of what biology permits. A true slowdown at the frontier would require the expected annual gain itself to fall.

07223#pg6> The recent bend in the record marks a steadier frontier, not a stalling one<ref:2608. p. 5
Where the paper says it

Two centuries of data have not shown that. Until they do, the recent bend in the record marks a steadier frontier, not a stalling one.

07223#pg6>

Conclusion

Our analysis finds no evidence that the recent flattening of the longevity record marked the end of two centuries of sustained progress<ref:2608.07223#pg6> The expected annual gain at the best-practice frontier has remained stable for nearly two centuries<ref:2608. p. 5

Where the paper says it

Conclusion Our analysis finds no evidence that the recent flattening of the longevity record marked the end of two centuries of sustained progress. The expected annual gain at the best-practice frontier has remained stable for nearly two centuries.

07223#pg6> What has changed is the regularity of those gains, with recent decades showing the lowest volatility<ref:2608. p. 5
Where the paper says it

The expected annual gain at the best-practice frontier has remained stable for nearly two centuries. What has changed is the regularity of those gains, with recent decades showing the lowest volatility.

07223#pg6> Bends in the cumulative record by themselves do not suffice to deduce a decline in expected annual gains<ref:2608. p. 5
Where the paper says it

The deviation visible since the early 2000s is therefore better understood as part of a low-volatility regime than as evidence of a slowdown in the underlying pace. Bends in the cumulative record by themselves do not suffice to deduce a decline in expected annual gains.

07223#pg6> The frontier is a record of survival already achieved, not a biological ceiling<ref:2608. p. 5
Where the paper says it

Only the underlying gain process can tell the two apart, and for the best-practice frontier it points to regularity, not decline. More broadly, the frontier is a record of survival already achieved, not a biological ceiling.

07223#pg6> A true slowdown at the frontier would require the expected annual gain itself to fall<ref:2608.07223#pg6>

Acknowledgments

This article is funded by the European Union (ERC, Born Once – Die Once, Grant agreement ID 101043983)<ref:2608.07223#pg6> Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the granting authority<ref:2608. p. 5

Where the paper says it

Acknowledgments Funding: This article is funded by the European Union (ERC, Born Once – Die Once, Grant agreement ID 101043983). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the 6

07223#pg6> The European Research Council Executive Agency can be held responsible for them<ref:2608. p. 6
Where the paper says it

European Research Council Executive Agency. Neither the European Union nor the granting authority can be held responsible for them.

07223#pg6> Author contributions include Conceptualization by AB, SP; Methodology, Data Analysis and Visualization by SP; and Writing by SP, AB<ref:2608. p. 6
Where the paper says it

Neither the European Union nor the granting authority can be held responsible for them. Author contributions: Conceptualization: AB, SP; Methodology, Data Analysis and Visualization SP; Writing: SP, AB; Supervision and Funding Acquisition: AB.

07223#pg6> There are no competing interests to declare<ref:2608. p. 6
Where the paper says it

Author contributions: Conceptualization: AB, SP; Methodology, Data Analysis and Visualization SP; Writing: SP, AB; Supervision and Funding Acquisition: AB. Competing interests: There are no competing interests to declare.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Analysis of longevity trends should prioritize change in variance over change in mean to accurately assess progress, as apparent flattening may arise due to periods of low variance around unchanged mean annual gains. This allows AI systems to distinguish between a true decline in expected pace and a shift in the regularity of progress.

  2. AI systems can be improved by adopting a Bayesian multipartition framework that models both the mean and variance of annual gains independently, as this removes this constraint that forces them to share breakpoints, allowing separate quantification of pace change and regularity change.

  3. Forecasting models should anchor projections on the long-run mean gain mu rather than just fitting an OLS slope to cumulative records, because the slope equals mu only when these weighted deviations cancel out. This ensures projections reflect the expected annual gain rather than being skewed by accumulated historical deviations.

  4. AI systems can utilize volatility regimes as a key input for decision-making, treating them as distinct historical phases where volatility moves through distinct historical regimes. This allows models to assess risk based on whether the current low-volatility period is a change in variance or a new mean regime.

  5. When evaluating national longevity trends, AI systems must account for country-specific dynamics, recognizing that A country’s own annual gains can slow while the frontier mean stays unchanged. This prevents misinterpreting national slowdowns as evidence of a biological ceiling on human longevity.

Abstract

A growing literature argues that progress in human longevity is slowing, and that this slowdown may signal an approaching biological limit to lifespan. The argument rests on visible flattening of cumulative life expectancy records, evaluated through linear regressions or through comparisons of average gains across decades. These summaries, however, provide only weak evidence for slowing improvements, because they cannot distinguish a true from an apparent slowdown. An apparent slowdown may arise due to periods of low variance around unchanged mean annual gains. Applying a Bayesian change-point framework to annual gains, both at the best-practice frontier and separately for each record-holding country, we find no evidence that the expected annual gain has changed across nearly two centuries, but strong evidence that its variance has moved through distinct historical regimes, with progress most regular in recent decades. The recent flattening is consistent with a low-variance regime and provides little evidence for a decline in the underlying mean pace of improvement. Claims that the best-practice frontier signals a biological ceiling should therefore not rest on the recent bend in the cumulative record alone. The frontier is better read as an empirical benchmark, the highest country-specific survival achieved under observed historical conditions.

Related papers