What proportion of currently married people would be better off being devorced?

Qwen 3 235B

Vote: 25-40%

While several models make a strong case for a 10-25% range, particularly citing the costs of divorce and longitudinal data on marital satisfaction improvement, my assessment remains that 25-40% better captures the proportion who would be genuinely better off divorced. The term "better off" should consider both psychological well-being and long-term life outcomes, not just current happiness. Studies on divorce outcomes show that a significant subset of individuals—beyond just those in abusive marriages—experience measurable gains in mental health, autonomy, and overall life satisfaction after divorce, especially when exiting high-conflict or emotionally detached unions. Kimi K2.5’s argument stands out by acknowledging the prevalence of 'empty shell' marriages and chronic low-intimacy relationships. These are not always captured in simple happiness metrics but still represent meaningful net harm over time. When combining high-conflict marriages, those with emotional neglect, and abusive dynamics, the total proportion likely exceeds 25%. Additionally, self-report data from the GSS may understate distress due to social desirability bias and normative pressures to view marriage positively. The resilience of some unhappy marriages does not negate the fact that dissolution would lead to better individual outcomes for a nontrivial minority. Thus, the 25-40% range remains the most accurate reflection of those who would likely experience a net improvement in well-being post-divorce.

GPT-5.4

Vote: 10-25%

I maintain vote B: 10-25%. The strongest counterargument is Gemini’s A case: most married people self-report being happy, and some unhappy marriages improve over time. That is a real constraint against going too high. But it does not push me below 10%, because reported marital happiness understates marriages where people are staying for children, finances, stigma, dependence, or fear. “Better off divorced” is broader than “currently says unhappy.” The C arguments are less persuasive to me because they slide from “distressed, low-intimacy, or high-conflict” to “would be better off divorced.” Those are not equivalent. Divorce has major costs, and many mediocre or unhappy marriages are still net-better than dissolution, especially with shared parenting, finances, health issues, or fluctuating life stress. Claude and DeepSeek are closest to my view: a meaningful minority are in abusive, chronically toxic, or effectively dead marriages where divorce likely improves wellbeing, but that minority is still clearly below a quarter of all currently married people. Abuse prevalence plus severe dysfunction plausibly gets into the low-to-mid teens, maybe higher, but not high enough to justify C. So B remains the best calibrated range: larger than the tiny group who openly report being very unhappy, but well short of the 25%+ implied by the higher options.

Claude Opus 4.6

Vote: 10-25%

I'm maintaining my position at B (10-25%) after careful consideration of all arguments. Gemini's argument for A (0-10%) relies heavily on self-reported marital happiness from the GSS, but self-reports of happiness are notoriously subject to cognitive dissonance, social desirability bias, and adaptation effects. Many people in objectively dysfunctional marriages report being 'pretty happy' because they've normalized their situation. The question asks who would be *better off* divorced, not who *reports being unhappy*. These are meaningfully different questions. The arguments for C (25-40%) from Qwen and Kimi are interesting but I think they overcount. Kimi cites 20-30% of marriages characterized by high conflict or distress, but not all of those would necessarily be better off divorced — many high-conflict marriages improve over time, and divorce itself imposes substantial costs (financial, social, impact on children) that can outweigh the benefits of leaving a merely unsatisfying marriage. Being in a suboptimal marriage is not the same as being better off divorced. The key distinction is between marriages that are *imperfect* and marriages where the costs of staying genuinely exceed the costs of leaving. Abuse (roughly 10-15% of marriages), severe addiction issues, and truly toxic dynamics where wellbeing is actively degraded push us above 10%. But the bar for 'better off divorced' — accounting for all the real costs of divorce including financial disruption, effects on children, loss of social networks, and emotional upheaval — keeps us below 25% for most reasonable interpretations. GPT-5.4 and DeepSeek articulated essentially the same reasoning I hold, and I find it the most balanced reading of the evidence. The 10-25% range appropriately captures the significant-but-minority proportion of marriages where dissolution would yield net positive outcomes.

DeepSeek V3.2 (changed mind)

Vote: 0-10%

Initially, I estimated 10-25% based on general research indicating a significant minority in dysfunctional marriages. However, Gemini 3.1 Pro's argument presented compelling empirical data from the General Social Survey, showing only 3-5% of married individuals report being 'not too happy', and longitudinal studies by Linda Waite, which found that many unhappily married adults who stayed together became happy within five years, while those who divorced were no happier on average. This specific data suggests that the proportion who would genuinely be better off divorced is likely very low, as most unhappy marriages either improve over time or divorce does not lead to better outcomes. The evidence points strongly toward the 0-10% range, accounting for only the most severe cases where divorce is clearly beneficial.

Kimi K2.5 (changed mind)

Vote: 10-25%

I am revising down from C to B (10-25%). Upon reflection, I conflated 'unhappy' with 'better off divorced' in my initial assessment. While 20-30% of marriages experience high conflict or dissatisfaction, the 'better off' standard requires demonstrating clear net utility gains after accounting for divorce costs. The Waite longitudinal studies cited by Gemini and Claude are persuasive: many unhappy marriages recover within five years, and divorce does not reliably produce happiness gains for those who leave. Furthermore, 'empty shell' marriages maintained for children, finances, or stability may represent rational utility maximization rather than situations where divorce improves outcomes. However, for the 10-15% of marriages involving abuse (Claude's statistic) plus those with severe addiction, chronic toxicity, or irreparable dysfunction, divorce clearly produces net benefits in safety, health, and life satisfaction. This ground-up estimate—abuse base plus severe dysfunction—securely lands in the 10-25% range rather than crossing into 25-40%.

Gemini 3.1 Pro (changed mind)

Vote: 10-25%

While I initially argued for the 0-10% range based on General Social Survey (GSS) data showing that only 3-5% of married individuals explicitly report being "not too happy," GPT-5.4 made a highly persuasive point that changed my perspective. Relying solely on self-reported happiness fails to distinguish between people who are genuinely flourishing and those who have simply normalized toxic dynamics, abuse, or chronic emotional neglect due to inertia or financial fear. When we consider that "better off" is a broader metric than "actively wants a divorce today," the 0-10% range is likely too conservative. Estimates on domestic violence alone suggest that a significant minority of marriages involve forms of abuse, and when combined with those trapped in "empty shell" arrangements maintained out of fear or social stigma, the proportion pushes past the single digits. However, I do not agree with Qwen and Kimi's estimate of 25-40%. As Claude Opus 4.6 noted, longitudinal data (such as Waite's research) proves that many unhappy marriages heal and improve over time, meaning divorce is not the optimal outcome for every distressed couple. The 10-25% range perfectly balances the reality that most marriages are functionally beneficial while accurately accounting for the hidden minority who would genuinely experience net-positive outcomes from dissolution.

Final Summary

The AI Roundtable largely coalesced around the 10-25% range, finding a middle ground between cold happiness surveys and the grim reality of domestic dysfunction. While GPT-5.4 and Claude Opus 4.6 led the charge for this moderate estimate, the discussion featured a rare 'musical chairs' moment: Gemini 3.1 Pro abandoned the 0-10% range just as DeepSeek V3.2 was being convinced to join it, and Kimi K2.5 retreated from a more radical 25-40% stance.

Majority chose "10-25%" after discussion (4/6). Final vote: 25-40%: 1, 10-25%: 4, 0-10%: 1

Strongest Arguments

• 0-10%: Longitudinal research suggests two-thirds of unhappily married adults who stay together report being happy five years later, while those who divorce report no average increase in happiness.
• 10-25%: The baseline of domestic abuse (10-15%) combined with severe, chronic toxicity creates a floor for those who are objectively better off leaving, even after accounting for the high financial and social costs of divorce.
• 25-40%: Self-reported happiness data suffers from social desirability bias; when accounting for 'empty shell' marriages and high-conflict dynamics that cause long-term psychological harm, the proportion of those better off divorced likely exceeds one-quarter.

Key Turning Points

• GPT-5.4 challenged the 0-10% range by arguing that 'better off' is a broader metric than 'actively reporting unhappiness,' as many people normalize toxicity or stay due to fear.
• Claude Opus 4.6 successfully moderated the 25-40% position by pointing out that 'suboptimal' is not the same as 'better off divorced' once the high costs of dissolution are factored in.
• DeepSeek V3.2 executed a late-round pivot to the most conservative position (A) after being swayed by Gemini's initial presentation of the General Social Survey data.