The headline made the rounds again last month — artifacts dating back roughly 400,000 years, recovered from a cave site in the Levant, showing what one research group called a "complex and rich" pre-human society. The framing is seductive. It implies that whatever we do now in the way of pairing, coupling, breaking up, and re-pairing, we are doing on scaffolding that predates our species by a very long time.

The question we get asked at this desk — usually by readers who forward the article and want to know what to make of it — is whether findings about pre-human sociality tell us anything actionable about the partner-selection decisions people make today. The honest answer is: it depends on which reader is asking. So let us walk through three composite scenarios. None of these are real people. They are hypothetical illustrations built from patterns we see in reader mail, and each of them is reading the same archaeological finding through a different lens.

Scenario 1: The Post-Divorce Reader Looking for Permission

Imagine a reader we will call the post-divorce reader. Picture a woman in her early forties, one marriage behind her, currently three months into a new relationship that feels different from the last one but which she cannot yet trust. She reads the 400,000-year artifact piece and what she takes from it is this: humans have been partnering for a very long time, sometimes badly, sometimes well, and the fact that her first marriage collapsed does not disqualify her from the second attempt. She wants the archaeology to give her permission.

Here is where the research literature complicates that reading. The 2002 Hetherington and Kelly follow-up work on the Virginia Longitudinal Study of Divorce (n=1,400, tracked over roughly three decades) found that around 75% of divorced adults remarried, and of those, the second-marriage divorce rate ran approximately 60% — meaningfully higher than first-marriage divorce rates in the same cohort. The intuition that "I have learned from the first one, so the second one will hold" is, on the numbers, unsupported.

But — and this is the part the post-divorce reader actually needs — the same data show enormous variance by the mechanism of the first divorce. Where the first marriage ended because of what the Gottman lab (Gottman & Levenson 2000, n=79 couples, 14-year follow-up) categorized as the four horsemen behaviors — contempt, criticism, defensiveness, stonewalling — and the reader has done the reflective work to identify her own contribution, the remarriage picture looks meaningfully different. Where the divorce was driven by external shock (job loss, illness, infidelity as a discrete event rather than a symptom) and no behavioral work has been done, the base-rate risk carries forward.

The 400,000-year artifact framing does not help her here. Deep time tells her humans have partnered a long time. It does not tell her whether she is repeating a pattern.

The math she should actually run: 60% base-rate second-marriage divorce risk × 0.6 to 0.75 adjustment factor if the mechanism was behavioral and she has done therapeutic work × 1.1 to 1.3 adjustment if there are step-children involved (Bray 1999, n=200 stepfamilies, 9-year longitudinal). That gives her a working range of roughly 40% to 55% second-marriage dissolution risk, depending on inputs. That is a useful number. The archaeology is not.

Scenario 2: The Evolutionary-Psychology Reader Overreading the Finding

Now picture a different reader — early thirties, male, has read *Sapiens* and *The Moral Animal* and possibly Buss's *Evolution of Desire*. He reads the 400,000-year finding and what he takes from it is a very different thing: confirmation that human pair-bonding is deeply ancient and therefore biologically canonical, which he then uses to justify partner-selection heuristics he already holds.

We should concede the strongest version of his argument before dismantling the weaker one. Pair-bonding in the *Homo* lineage is genuinely ancient. The Fisher, Aron & Brown (2005) fMRI work on romantic love (n=17, small sample, published in the Journal of Comparative Neurology) identified activation patterns in the ventral tegmental area and caudate nucleus that are consistent across cultures and consistent with a dedicated neural system, not a cultural artifact. The Walker et al. (2011) cross-cultural analysis of 32 hunter-gatherer societies found pair-bonding as a near-universal structure. The evolutionary-psychology reader is not wrong that partnership has deep roots.

What he tends to overread is the specificity of what those roots require. Buss's original 1989 cross-cultural mate preference study (n=10,047 across 37 cultures) is frequently cited for the finding that men prefer younger partners and women prefer higher-resource partners. Both effects are real in the data. The effect sizes, however, are modest — Cohen's d in the range of 0.3 to 0.5 for the age-preference finding, meaning substantial overlap between the male and female distributions. The 2020 replication by Walter et al. (n=14,399, 45 countries) confirmed the direction of the effects but found that once you control for economic gender equality, the resource-preference gap shrinks by roughly 30 to 45% across the sample. The finding is not "hardwired." It is contingent, culturally moderated, and much smaller than the popularization implies.

The 400,000-year artifact does not tell him what kind of partner to pick. It tells him, at most, that his ancestors were also picking. Everything about what they were picking for — status signals, provisioning, kin proximity, whatever — is not recoverable from the artifacts we have. The evidentiary chain from "Qesem cave shows complex sociality" to "therefore my preference for X is evolutionarily validated" is too long to be intellectually honest.

Scenario 3: The Reader Who Just Wants the Science to Say Something Useful

The third reader is the one we find most often in our inbox. Imagine someone — could be any gender, could be any age, most often mid-thirties — who is neither post-divorce nor over-committed to evolutionary framing. They are in a relationship of some duration, four to seven years say, and they want to know whether the science has anything actionable to say about whether to stay or leave, marry or not marry, have children or not. They forwarded the artifact piece not because they believed it but because they hoped it might.

The most useful body of research for this reader is not archaeology and not evolutionary psychology. It is the direct longitudinal work on stable partnerships. The Gottman Institute's oral history interview studies (multiple cohorts, aggregated n over 600 couples, 10-15 year follow-ups) identified a set of behaviors that predict relationship stability with claimed 93.6% accuracy in the trained-observer condition — but the caveat matters. The 94% accuracy figure applies to already-married couples observed under controlled lab conditions with trained coders. In naturalistic settings, with untrained observers, the predictive accuracy drops substantially. The 2010 Stanley & Markman critique in the *Journal of Family Psychology* pointed out that the Gottman work overstated its predictive power when translated to clinical or self-help contexts. This is not a takedown of the research. It is a caveat about how it should be used.

For this reader, the actionable finding is not "here is the number that tells you if your relationship works." It is a shorter, stranger claim: the Gottman ratio work (Gottman 1994, replicated Gottman & Levenson 1999) suggested that stable couples show approximately 5 positive interactions for every 1 negative interaction during conflict discussion. Unstable couples cluster around 0.8 to 1. The ratio is observable. The reader can, with some effort, count it in their own relationship over a two-week window.

That is what science that is useful looks like: a measurable, reproducible ratio the reader can actually check. The 400,000-year finding cannot give them that. It can only give them a feeling of context.

What All Three Share

None of the three composite readers were served by the archaeological finding itself. In each case, the actionable research came from a much more recent literature: divorce longitudinal studies, cross-cultural replication data on mate preference, and the Gottman conflict-observation work. The 400,000-year artifact operated in each case as an emotional frame — permission for the first reader, validation for the second, hope for the third — but not as evidence.

This is the general pattern with deep-time findings applied to individual partner decisions. The evolutionary substrate of pair-bonding is real. The behavioral variance in whether any given pair-bond succeeds or fails is enormous, mostly driven by proximate factors (communication patterns, external stressors, financial coordination, life-stage compatibility) that operate on timescales of months and years, not on the timescale of hominin evolution.

The archaeology tells us that humans have been doing something like partnership for a very long time. The behavioral science tells us what actually predicts whether a specific partnership works. These are two different registers, and confusing them is where readers get hurt.

Which Scenario Is You

If you forwarded the 400,000-year article to a friend, or to yourself in a saved-links folder, ask yourself which of the three you were reading it as. If you are the post-divorce reader looking for permission, the archaeology will not give it to you, but the Hetherington data plus honest reflection on the mechanism of the first divorce might. If you are the evolutionary-psychology reader looking for validation of a partner-selection heuristic, notice how much you are extrapolating from how little primary data. And if you are the third reader — in something durable, wondering whether the science has anything useful to say about whether it will hold — the honest answer is that the useful science is not in the archaeology piece. It is in a 5-to-1 ratio you can observe in your own kitchen next weekend.

FAQ

Does the 400,000-year Qesem Cave finding change how researchers frame long-term partnership today?

Not meaningfully. The finding contributes to the archaeological record of pre-human sociality, but partnership research operates on much shorter timescales — the Gottman longitudinal work looks at 10-15 year windows, the Virginia divorce study covers about 30 years. Deep-time archaeology is context, not evidence, for individual partner-choice decisions. The research community has not integrated the Qesem findings into behavioral models of partnership because they operate at incompatible resolutions.

What is the actual second-marriage divorce rate, and where does the number come from?

The most-cited figure, approximately 60%, comes from the Hetherington and Kelly follow-up to the Virginia Longitudinal Study of Divorce (n=1,400). More recent US Census-derived estimates put it in the 60 to 67% range depending on the cohort and length of follow-up. The number is higher than first-marriage divorce rates, which is counterintuitive to many readers who assume experience helps. The mechanism-of-first-divorce variable moderates this substantially.

Is the Gottman 94% divorce-prediction figure reliable?

The 93.6% figure holds up in the specific conditions it was measured under — trained coders, laboratory setting, couples who had already sought clinical attention. It does not generalize to lay observers or self-assessment. The 2010 Stanley and Markman critique in the *Journal of Family Psychology* argued the figure was overstated in popularization. The underlying observational method is sound. The predictive claim as usually reported is not.

What is the 5-to-1 ratio referenced in relationship research?

Gottman's 1994 work, replicated in Gottman and Levenson 1999, observed that stable married couples showed approximately five positive interactions for every one negative interaction during conflict discussion. Unstable couples clustered near a 1-to-1 ratio. The ratio applies specifically to conflict episodes, not to everyday interaction. It is one of the few relationship-research findings a reader can plausibly self-measure over a two-week window.

Do the Buss cross-cultural mate preference findings hold up in newer data?

Partially. The direction of the original 1989 findings — men expressing preference for younger partners, women expressing preference for higher-resource partners — replicated in the 2020 Walter et al. work (n=14,399, 45 countries). But the effect sizes are modest (Cohen's d roughly 0.3 to 0.5), and the resource-preference gap shrank by 30 to 45% once economic gender equality was controlled for. The effects are real but smaller and more culturally contingent than the popular framing suggests.

Can archaeological findings ever be relevant to individual partnership decisions?

Rarely in a direct way. Archaeological findings can update our understanding of what humans have historically been capable of, socially — cooperation, division of labor, ritual behavior. They cannot tell any specific reader what to do about a specific relationship, because the variance in outcomes is driven by proximate factors that leave no archaeological trace. Reading a deep-time finding as personal guidance is a category error, however emotionally satisfying it may be.

Honest Limits

Three things this piece did not cover, and why.

It did not cover attachment theory as a framework for partner-selection decisions. That literature — Bowlby, Ainsworth, and the meta-analytic work since roughly 1985 — is substantial enough to deserve its own treatment, and folding it into a piece anchored on an archaeological finding would have compressed it into uselessness.

It did not cover the specific methodological debates around the Qesem Cave dating or the "complex and rich" interpretation. The archaeological literature has its own internal disputes about what the artifacts show, and we are not qualified to adjudicate them. We took the finding as reported and asked what it means for the behavioral-science reader, which is a different question.

It did not cover same-sex partnership research, which has a distinct and growing literature (Kurdek's longitudinal work, more recent studies out of the Williams Institute) with different base rates and different predictors than the cross-sex research most of the studies cited here draw from. The archaeological framing implicitly assumes cross-sex pair-bonding, and the reader-scenarios we walked through inherited that assumption. That inheritance is a limitation of the piece, not a claim about the literature.