Longevity, Density Dependence, and the Demo-graphic Origins of Agriculture: Why the Neolithic Demographic Transition Cannot Be the Cause of What It Accompanies

1. Introduction

The Neolithic Revolution, also called the Agricultural Revolution, marked the transition in human history from small nomadic bands of subsistence foragers to larger, more permanent, settlements of farmers. During the millennia prior to the Neolithic, growth rates, constrained by the density-dependent pressures of their subsistence economy, are calculated to have been less than 0.01 per cent; following the development of agriculture around ten thousand years ago these rates soared to 0.1 per cent (Coale 1974; Freedman and Berelson 1974).

A satisfactory explanation for our ancestors’ decision to ‘settle down on the farm’ has eluded archaeologists (Shennan and Sear, 2021), anthropologists (Cowgill, 1975), mathematicians (Kabir, Mimura and Tsai, 2018) and economists (Weisdorf, 2005) for decades. A recurring component of the attempted explanations is that the transition to agriculture occurred spurred on by increased population pressure (i.e., population size/population density). Unfortunately, this important anthropological question has become something of a “chicken or the egg” causality dilemma. Did population growth spur the adoption of agriculture or did the adoption of agriculture facilitate population growth? Missing from these scenarios is an explanation for the increased population pressure. What changed? Already under density-dependent pressure for millennia, what changed for those subsistence foragers that further increased population pressure?

A second issue, one less often discussed, is the need to explain the timing of the adoption of agriculture. Plants that could be cultivated were available to foragers for long before anyone cultivated them, and the people who eventually did so were, on most skeletal and stature evidence, worse off for it in the short run (ref). Explanations that make agriculture an opportunity — a better technology, a warmer Holocene, a richer resource base — have to say why the opportunity was declined for so long. Explanations that make it a response to stress have to say what the stress was and, have to do so without attributing it to a population increase that is itself usually attributed to the adoption of agriculture.

More recently, Bocquet-Appell (2002, 2008a, 2008b, 2011a, 2011b) has attempted to explain the agricultural revolution by establishing a theoretical footing in demographic transition theory. By analogy to the Demographic Transition Model (DTM) and in contrast, to what he calls the Contemporary Demographic Transition (CDT), Bocquet-Appel asserts that the Neolithic Demographic Transition (NDT) was led by increased fertility. Specifically, Bocquet-Appel (2008a: 35) argues that "[t]he part of the signal of the NDT which is demographically identifiable... ...shows that a baby-boom occurred and, beyond that, with the onset of the change in the economic system (i.e., agriculture), a fertility transition towards high values also occurred.”

The NDT is the most widely reproduced regularity in the archaeological record of the transition (cf., Bellwood and Oxenham, 2008). Bocquet-Appel's juvenility index — the proportion of skeletons aged 5–19 among those aged 5 and over — rises from a pre-agricultural level near the stationary expectation to a markedly higher level in early farming cemeteries, and does so in sequences from western Europe, the Levant, Mesoamerica and North America that are historically independent of one another (Bocquet-Appel 2002; Bocquet-Appel and Bar-Yosef 2008; Bocquet-Appel 2011). The index responds mainly to fertility, so the standard reading is a fertility increase of roughly one additional birth per woman, released by the shortening of birth intervals that sedentism, weaning foods and a higher-calorie staple allow.

The problem with using this as an explanation of the transition is that the index is measured on cemeteries, and cemeteries date to the transition. Whatever the NDT records, it records it at the same time as the appearance of farming, not before. A variable that moves at the same moment as the thing it is supposed to explain does not distinguish cause from consequence, and in this case, there is a more straightforward reading in which it is a consequence: people took up farming. Farming let them have more children.

Here I argue that the demographic signal is real but that its causal arrow runs earlier and in the opposite direction (i.e., a decrease in fertility), and that the NDT is the visible release of a constraint that had been tightening for thousands of years before any cultivation. The argument has two parts. The first is a model, set out in Section 2, in which rising adult survivorship in a density-dependent economy forces compensating fertility decline, and the compensation overshoots. The second is methodological, set out in Section 3: the same density-dependence assumption that drives the model also identifies the demographic evidence for its first step, because it removes the growth-rate confound that otherwise makes age-at-death distributions uninterpretable as longevity.

Section 4 and Section 5 set out what the existing evidence already supports and what it does not, Section 6 gives the predictions that separate this model from its rivals, and Section 7 gives a research design for the test that matters most: whether the NDT signal is present in sedentary pre-agricultural populations ahead of the agricultural front.

2. The Model

The model has six steps. I state them plainly first and defend them afterwards.

Step 2.1.

Adult survivorship rose slowly through the Paleolithic. This is the empirical premise, and Section 4 assesses how well it is supported. The rise is in adult survival, not in the survival of children, and the evidence that bears on it most directly is the ratio of older to younger adults among the dead.

Step 2.2.

Hunting and gathering is strongly density-dependent, so the intrinsic rate of increase is held near zero over the long run. The support for this is arithmetic rather than ethnographic: sustained positive r over Paleolithic timescales produces population sizes that the archaeological record excludes by many orders of magnitude, and sustained negative r produces extinction. Whatever the mechanism — territory, resource depression, conflict, disease, or deliberate limitation — the long-run outcome is near-zero growth, and the mechanism operates through density.

Step 2.3.

‡ Rising survivorship in a density-dependent economy must be absorbed by falling fertility. This step is forced rather than merely plausible, because of what the survivorship measure actually is. The older/younger adult ratio with its boundary at the eruption of the third molar and at twice that age is, in a stationary population, l(30)/l(15) divided by l(15) − l(30) [in a life table cohort, l(15) = the number of individuals surviving to age 15; l(30) = the number of individuals surviving to age 30]. Its rise is therefore a rise in survival across the female reproductive span. A woman who is more likely to survive from 15 to 30 spends more of her life at risk of conception. Holding the net reproduction rate at 1 then requires total fertility to fall by very nearly the reciprocal of the survivorship gain. This is not an assumption added to the model; it is what the measured quantity means.

Step 2.4.

The compensation overshoots, pushing r below zero. This is the model's substantive claim and the one that does the causal work. Fertility reduction in foragers is not a dial that can be set to a target. It operates through birth spacing, lactation, mobility costs, and the timing of weaning, all of which respond to the same conditions that are driving the survivorship change, and none of which has a feedback loop that reads the current value of r and stops. A system that adjusts fertility in response to density and survivorship, with lags and without a set point, will not settle exactly at replacement. Once it undershoots, the population declines, and the strategy that produced the decline is the same strategy that has been working for tens of thousands of years.

Step 2.5.

Agriculture is adopted as a rescue, conditional on two further requirements. The reason to adopt agriculture, on this account, is to return r to zero or above. That requires, first, that a population recognizes what the solution is — an intensification of an existing relationship with a plant or animal, not an invention from nothing — and second, that dependable, harvestable, local food resources be present to intensify. Neither condition is automatic, which is why the transition is patchy in space and staggered in time even where the demographic pressure is general. Outside places like the Northwest Coast, where wild resources were dependable enough to support a terminal sedentary forager economy indefinitely, local wild resources are a bridge to agriculture rather than a destination.

Step 2.6.

Sedentism releases the fertility constraint and denser settlement then raises mortality, demanding still higher fertility again. The NDT is the signature of this release. But the release is not free: aggregation raises exposure to infectious disease and to sanitation failure, so mortality rises, and the fertility increase must run further than the release alone would predict in order to hold the new equilibrium. The Neolithic demographic regime is therefore high-fertility and high-mortality, and the skeletal health decline that accompanies the transition is not an anomaly requiring separate explanation but part of the same process.

Two qualifications apply throughout. First, band membership is fluid . A band that fails demographically does not usually vanish; rather, its members merge with neighbors. The unit at which "extinction" occurs is therefore larger than the band. The relevant demographic unit is a regional metapopulation whose scale I will not attempt and do not need to fix here Second, the overshoot does not need to be large or fast. A rate of decrease of −0.002/yr is invisible in any archaeological proxy and still halves a population in about 350 years.

3. Density Dependence Identifies the Survivorship Signal

The objection that has kept paleodemographic age-at-death ratios out of arguments about longevity is that a death distribution is not a survivorship distribution. In a stable population, deaths at age x are weighted by exp(−r(x − 15)) relative to the stationary case, so a growing population buries proportionally more young adults and a declining one proportionally more old ones. The same ratio of older to younger dead can be produced by long lives with high growth or short lives with low growth. This is the substance of the general critique of the method, and it is why the older/younger adult ratio computed on a death assemblage — the "dead-OY" ratio — varies across primates by two orders of magnitude while the corresponding ratio among the living is nearly invariant (Hawkes and O’Connell, 2005).

The critique is correct as stated, and it is fatal to any attempt to read the ratio as longevity without an independent constraint on r. What Step 2 of the model supplies is exactly such a constraint. If hunting and gathering is density-dependent enough to hold r near zero over the long run (cf., then r ≈ 0 is not an assumption smuggled in to rescue the measurement — it is a prediction of the model being tested, and the measurement inherits it. Under r = 0 the ratio reduces to

dead-OY(15/30) = l(30) / (l(15) − l(30)),

one equation in one unknown. Juvenile mortality cancels out of it entirely, which matters because juvenile representation is the part of a skeletal assemblage least likely to survive and most likely to be biased.

This is the pivot of my assertion. The model's central assumption is also the required condition for its own first premise. That is a strong testable position: if r is not near zero, the identification fails and the survivorship reading goes with it. Section 4.4 checks what happens under the model's own prediction of mildly negative r.

A second index is available on the same material. The juvenility index 15P5 — deaths at 5–19 over deaths at 5 and above — correlates strongly with the crude birth rate (r² ≈ 0.96) and less strongly with the intrinsic growth rate (r² ≈ 0.875), with 0.18 as the stationary expectation. It responds mainly to fertility. The older/younger adult ratio responds to both. Computing both on the same assemblages therefore gives two equations in two unknowns and allows fertility and mortality to be apportioned rather than assumed. Section 7 builds the pre-agricultural test around this.

4. What the Upper Paleolithic Evidence Supports

4.1. The Series

Overlooked in or undervalued and omitted from other discussions of the Neolithic transition is the work of Caspari and Lee (2004, 2006). In their study of 768 dental individuals representing four hominid groups from the Early Pleistocene through the Early Upper Paleolithic, Caspari and Lee (2004: 10895) found a trend of “increased adult survivorship over the course of human evolution” with “a dramatic increase in longevity in the modern humans of the Early Upper Paleolithic.”

These findings, in and of themselves, are not new. It is a common belief that during the Neolithic there was rapid population growth among humans (Hole et al. 1971; Hassan 1973; Dumond 1975). According to Acsadi and Nemeskeri "the fundamental factor in the changed demo graphic conditions was a considerable lowering in the mortality level, i.e., the lengthening of life spans" (1970:215). However, as we have shown above, the "lengthening of life spans," which is, in fact, an increase in the mean age at death, does not necessarily imply a lowering in the mortality level. Although the mortality rates may have dropped, data on the mean age at death pro vide no evidence for this. Similar conclusions about decreased mortality as a result of "lengthened life spans" have been drawn by Hassan (1981), Weiss (1973), Petersen (1975), and other studies estimating life tables from skeletal samples. What is new is the discovery that “[in the Upper Paleolithic] for the first time there are a larger number of older adults than younger adults in the death distribution” (2004: 10898; emphasis added).

The older/younger adult ratio computed on fossil hominin dental samples rises across four groups: australopithecines, early Homo, Neandertals, and the Early Upper Paleolithic, with the published values and the resampled confidence assessment reproduced in the accompanying data table (Table 1). The step from Neandertals (0.39) to the Early Upper Paleolithic (2.08) is much larger than any earlier.

4.2. The Survivorship Reading

Under r = 0, dead-OY 0.39 (cf., Table 1, Neanderthals) corresponds to l(30)/l(15) = 0.281 and dead-OY 2.08 (cf., Table 1, EUP moderns) to l(30)/l(15) = 0.675: a 2.40-fold increase in survival across the female reproductive span. Holding the net reproduction rate at 1 across that change requires total fertility to fall by a factor of about 0.42. That figure is approximate in two ways. Mean maternal age is nearer 27 than 30, so the exact weighting differs; and if the Neandertal value is depressed by taphonomy (Section 4.5), 2.40-fold is an upper bound on the true survivorship gain and 0.42 an upper bound on the required fertility decline.

4.3. The Fertility Bound

The alternative reading is that the rise is not survivorship at all but a fall in growth rate — a population declining more steeply buries proportionally more old adults. Holding mortality at a fixed Gompertz–Makeham schedule with e(15) = 25 years [life expectancy at age 15 equals 25 years; 25 years not age 25] and asking what change in r alone reproduces the 0.39 → 2.08 rise gives Δr = −0.130/yr. That is a net reproduction multiplier of 0.026 per generation and a population halving time of 5.3 years.

4.4. The Model's Own Negative Growth Does Not Undermine the Reading

Step 4 predicts r < 0, which is precisely the direction that inflates the dead-OY ratio, so the model appears to threaten its own measurement. Taking the schedule that reproduces the EUP value at r = 0 and re-solving at negative growth rates: at r = −0.010/yr the ratio is inflated by 21%, and the implied e(15) falls from 21.0 to 19.5 years — a shift of 1.5 years. At r = −0.002/yr, the rate consistent with a slow multi-millennial decline, the effect is negligible. The overshoot the model predicts is far too gentle to manufacture the observed rise.

4.5. The Taphonomic Floor, and What the Series Does Not Support

The honest counterweight is that the low end of the series is not interpretable. Across adult life expectancies from 8 to 45 years, the reachable range of dead-OY on this hazard family is 0.149 to 9.420. The australopithecine value of 0.12 lies below any plausible mortality schedule — no survivorship curve produces it. Early Homo at 0.25 and Neandertals at 0.39 imply e(15) of 9.4 and 10.8 years respectively, below every ethnographically recorded forager population and below wild chimpanzees. The Early Upper Paleolithic value of 2.08 implies 21.0 years, which is unremarkable.

The natural reading is that older individuals are systematically under-recovered in the earlier assemblages, through some combination of preservation, sample size, and the difficulty of aging worn dentitions at the old end. This partly vindicates the preservation objection to the method, even though the growth-rate objection fails. The consequence for the argument is that the Upper Paleolithic threshold is the robust signal and the archaic gradient is not. I claim a threshold; a critical build up that caused an unstoppable change. I do not claim a smooth australopithecine-to-modern trend.

5. Why Concurrency Is Not Quite What It Looks Like

The NDT's concurrency with agriculture is usually displayed by aligning sites on their distance in time from the local agricultural front — each sequence is re-zeroed on its own transition, then the aligned sequences are stacked. This is a sensible device for combining historically independent sequences, and it is why the regularity is visible at all. But it has a consequence that is easy to miss: once sites are aligned on the transition, the transition is at zero by construction, and any signal that moves within a few centuries of it will appear to move at it. The coordinate system cannot distinguish a lead of two hundred years from a coincidence. For example, compare Figure 1 in Bocquet-Appel (2008) to Figure 2 in Bocquet-Appel and Naji (2006). The placement of zero (0), the adoption of agriculture on the horizontal axis, suggest in Bocquet-Appel (2008) that the transition was preceded by a decline in fertility while in Bocquet-Appel and Naji (2006) zero shifts slightly to the right suggesting that the transition was preceded by increasing fertility.

Re-aligning the same material on absolute chronology, within regions where the local front is independently dated, turns a binary question into a three-way one: does the juvenility signal lead the front, coincide with it, or lag it? Each answer belongs to a different model. A lag is farming causing fertility increase. A coincidence is the standard reading. A lead is this model, and only this model predicts it.

This does not dispose of the concurrency objection — it sharpens it into a test. If the signal genuinely coincides once the alignment artefact is removed, the model is in trouble.

6. Predictions

6.1.

The juvenility signal appears in sedentary pre-agricultural populations, ahead of the agricultural front. This is the model's distinguishing prediction. Step 6 makes the release a consequence of residential permanence, not of cultivation, so any pre-agricultural population that achieved sedentism should show it. The Jōmon sequence is the primary case: sedentary, pottery-using, with cemeteries, and pre-agricultural for millennia (Tomohito et al., 2008). The Natufian (Munro et al., 2021) is the second.

6.2.

The signal scales with residential permanence, not with subsistence category. A dose-response relation across degrees of sedentism is a stronger prediction than a binary contrast, and it is harder to produce by accident.

6.3.

Effective population size (Ne) declines through the terminal Paleolithic while the older/younger adult ratio rises. This joint pattern is the test against the climate rival. A climate-driven mechanism that depresses population through mortality produces declining Ne and falling dead-OY. The fertility mechanism in this model produces declining Ne and rising dead-OY. Neither series alone discriminates; together they do. Genetic Ne estimates and skeletal series come from independent evidence, which is what makes the conjunction informative rather than circular.

6.4.

Populations with dependable wild resources sit on the bridge longer. The Northwest Coast is the limiting case, where dependability was high enough that the bridge became a destination. Elsewhere the duration of the sedentary-forager phase before cultivation should scale with resource dependability, and the transition should be fastest where wild resources were seasonal or unreliable. This is the prediction most vulnerable to post-hoc classification of "dependability", so the resource ranking has to be fixed on ecological grounds before the durations are compared.

6.5.

Total fertility falls through the Upper Paleolithic by roughly the reciprocal of the survivorship gain. From Section 4.2, a factor near 0.42 across the Neandertal-to-EUP interval, and an upper bound if the early value is taphonomically depressed. I do not know of a direct archaeological proxy for forager TFR that could test this at the required resolution; it is stated because the model commits to it, not because I can currently check it.

7. Research Design for the Pre-Agricultural Test

7.1. The Selection Problem, Stated First

The juvenility index requires a cemetery. Cemeteries are themselves a marker of residential permanence. The mobile-forager baseline against which a sedentary pre-agricultural signal would be measured is therefore the one state the method structurally cannot observe. Any design that ignores this will recover the sedentism effect as an artefact of which sites produce measurable assemblages.

The resolution is to abandon the binary contrast and use a gradient. Among sites that do yield assemblages, residential permanence still varies considerably, and if the model is right the signal should vary with it.

7.2. Steps

7.2.1.

Rank candidate pre-agricultural sites on residential permanence, blind to their demographic data. Potential ranking criteria are architectural investment (post-built versus expedient structures, floor preparation, rebuilding episodes), storage features and their capacity, commensal species presence (house mouse, sparrow, rat), and seasonality indicators from faunal cementum and shellfish sclerochronology. The ranking must be fixed and recorded before any index is computed, by someone who has not seen the age-at-death data. This is the single most important procedural safeguard in the design.

7.2.2.

Compute both indices on the same assemblages. 15P5 where juvenile preservation permits, dead-OY(15/30) everywhere. Report the juvenile recovery fraction alongside 15P5 so the reader can see where the index is resting on thin material.

7.2.3.

Test for dose-response, not for a group difference. The prediction is monotone increase in 15P5 with permanence rank. Report the rank correlation with its exact permutation distribution rather than a parametric test, given the sample sizes involved.

7.2.4.

Apportion fertility and mortality using the two indices jointly. With 15P5 responding mainly to fertility and dead-OY to both, the pair constrains the two unknowns on each site. Where they disagree, the disagreement is itself informative: it localizes whether a site's signal is a fertility release or a mortality change.

8. Limitations

The model is currently a feasibility argument, not a fitted demographic reconstruction. Four limits matter:

8.1.

The scale of the demographic unit is unspecified. Band fluidity means that local failure is usually merger, so the overshoot has to be evaluated at a regional scale I have not fixed.

8.2.

The survivorship evidence supports a step, not a trend. Section 4.5 is explicit that the archaic end of the series is probably taphonomic. If the Upper Paleolithic rise turns out to be a single-episode change rather than a slow accumulation, Step 1 needs restating in those terms, though the rest of the model survives.

8.3.

The identification in Section 3 is conditional on Step 2. It is the right condition to impose when testing this model, but it means the survivorship result is not independent evidence for density dependence — the two stand or fall together, and that has to be said plainly rather than left for a referee to notice.

8.4.

Recognition, in Step 5, is doing real work and is not operationalized. The model requires populations to identify intensification as the answer to their problem, and I have no measure of that beyond the presence of a prior knowledge and use of the resource in question.