5D Library › Relationship Rupture Survival Analysis, No-Contact Memory Decay and Reconciliation Game Windows · 中文

Chapter 5. How Long Reunions Last

Two kinds of reunion

Couples who break up and get back together, sometimes more than once, are common, and studies of such on-again, off-again relationships report lower satisfaction and more uncertainty than in relationships that never broke (Dailey et al. 2009). The popular conclusion is that reunions are doomed. A simple mixture model shows why that conclusion is too quick. Suppose a reunited couple is one of two kinds: in some, the cause of the rupture has been dealt with, and the monthly chance of a new rupture is small (0.01); in others it has not, and the chance is large (0.08).

Calculation

# Chapter 5. p09: How long do reunions last? A reunited couple is one
# of two kinds: the cause of the rupture was dealt with (share q;
# monthly hazard of a new rupture 0.01) or it was not (monthly hazard
# 0.08). The survival curve of all reunions is the mixture
#   S(t) = q exp(-0.01 t) + (1 - q) exp(-0.08 t),
# and the hazard among couples still together falls over time because
# the unresolved couples leave first. Months 12, 24 and 36.
from math import exp

H_FIX, H_NOT = 0.01, 0.08

def surv(q, t):
    return q * exp(-H_FIX * t) + (1 - q) * exp(-H_NOT * t)

def hazard(q, t):
    a, b = q * exp(-H_FIX * t), (1 - q) * exp(-H_NOT * t)
    return (a * H_FIX + b * H_NOT) / (a + b)

if __name__ == "__main__":
    print("   share resolved q   together at 12   at 24   at 36"
          "   hazard m1   m36")
    for q in (0.2, 0.4, 0.6):
        print("   {:16.1f}   {:14.3f}   {:5.3f}   {:5.3f}   {:9.4f}"
              "   {:5.4f}".format(q, surv(q, 12), surv(q, 24), surv(q, 36),
                                  hazard(q, 1), hazard(q, 36)))
    q = 0.4
    print("   q = 0.4: share of couples together at 36 months who had")
    a = q * exp(-H_FIX * 36)
    print("   resolved the cause: {:.3f}".format(a / surv(q, 36)))
#    share resolved q   together at 12   at 24   at 36   hazard m1   m36
#                 0.2            0.484   0.275   0.184      0.0652   0.0270
#                 0.4            0.585   0.403   0.313      0.0508   0.0175
#                 0.6            0.685   0.531   0.441      0.0368   0.0136
#    q = 0.4: share of couples together at 36 months who had
#    resolved the cause: 0.892

If one reunion in five has dealt with the cause, 0.484 of couples are still together after a year and 0.184 after three. If two in five have, the figures are 0.585 and 0.313; if three in five, 0.685 and 0.441. The hazard among couples still together falls with time, from 0.0508 in the first month to 0.0175 in month 36 when two in five have dealt with the cause, because the unresolved couples leave first. By the third year, 0.892 of the couples still together are ones that dealt with the cause. The average reunion looks fragile; the reunions that last are mostly of one kind.

Share of reunited couples still together, by the share that dealt with the cause of the rupture (p09).

What earlier cycles say

A couple’s own history is evidence about which kind they are. p10 treats each earlier reunion that ended within a year as an observation and updates the share resolved by Bayes’ rule.

Calculation

# Chapter 5. p10: What earlier break-ups say. Use the two kinds of
# p09 (resolved: monthly re-rupture hazard 0.01; not: 0.08) and a prior
# share 0.4 resolved. If a couple has already broken up and reunited k
# times, each earlier reunion that ended within 12 months is evidence:
# chance of that under "resolved" is 1 - exp(-0.12) and under "not"
# 1 - exp(-0.96). Posterior share resolved, and the chance of lasting
# three more years, after k such cycles.
from math import exp

P_RES, P_NOT = 1 - exp(-0.12), 1 - exp(-0.96)

if __name__ == "__main__":
    print("   earlier quick cycles k   P(resolved)   P(together 36 months)")
    for k in range(5):
        a, b = 0.4 * P_RES ** k, 0.6 * P_NOT ** k
        q = a / (a + b)
        s = q * exp(-0.36) + (1 - q) * exp(-2.88)
        print("   {:22d}   {:11.3f}   {:21.3f}".format(k, q, s))
    print("   likelihood ratio per cycle: {:.2f}".format(P_NOT / P_RES))
#    earlier quick cycles k   P(resolved)   P(together 36 months)
#                         0         0.400                   0.313
#                         1         0.109                   0.126
#                         2         0.022                   0.070
#                         3         0.004                   0.059
#                         4         0.001                   0.057
#    likelihood ratio per cycle: 5.46

Each quick cycle is 5.46 times as likely for an unresolved couple as for a resolved one. After one such cycle the chance that the couple has dealt with the cause falls from 0.400 to 0.109, and the chance of lasting three more years from 0.313 to 0.126; after two cycles they are 0.022 and 0.070, and after four, 0.001 and 0.057. A repeated pattern is strong evidence that the cause has not been addressed. What would change the arithmetic is not a stronger feeling but a new fact: a specific change in the circumstances that caused the earlier ruptures.

Hypothesis · not established

The author’s interpretation (not yet tested): the survival of reunited couples is better described by a mixture of a low-hazard group and a high-hazard group than by a single hazard, and membership of the low-hazard group is predicted by a concrete change in the cause of the earlier rupture. What would show it wrong: in a cohort of reunited couples followed for three years, a single constant hazard fits as well as a two-group mixture, or reported changes in the earlier cause do not predict which couples last.

Summary

Reunions are not doomed on average, but they are a mixture. The longer a reunited couple lasts, the more likely they are of the kind that dealt with the cause; and each quick repeat of the cycle is strong evidence that they are not.

References

Dailey, René M., Abigail Pfiester, Borae Jin, Gary Beck, and Glenn Clark. 2009. “On-Again/off-Again Dating Relationships: How Are They Different from Other Dating Relationships?” Personal Relationships 16 (1): 23 to 47.

Fan, Y. (2026). Relationship Rupture Survival Analysis, No-Contact Memory Decay and Reconciliation Game Windows: Negative-Affect Dissipation Rates, Dual-Process Nostalgia Peaks and Signaling Equilibria [关系断裂生存分析、断联期记忆衰减与复联博弈窗口] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090241 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).