5D Library › Computational Attachment Theory and Couple Interaction Dynamics · 中文
Chapter 6 Repair Attempts: The Damper in the Conflict Loop
Never fighting is not the goal
One often overlooked finding of Gottman’s research is that couples who stay together are not all couples who rarely fight. Some have little conflict and tend to avoid disagreement, others argue often and intensely, and both can last; the difference lies not in how much conflict they have but in whether positive elements during conflict are enough to buffer the negative, and whether escalation can be pulled back (Gottman 1994).
Gottman calls a word or action that stops negative emotion from escalating further a repair attempt: “wait, are we getting off the point?”, a self-mocking joke, “you have a point there”, or “I need a break”. He stresses that whether a repair attempt succeeds often depends less on how cleverly it is put than on whether the other accepts it; and acceptance in turn depends on the friendship and goodwill the two have built up outside conflict (Gottman 1999).
Note
“Gentle” is not always better. Overall and McNulty’s review notes that direct, even oppositional, communication during conflict may benefit a relationship in the long run when problems are serious and change is needed, whereas gentle or avoidant approaches help with minor problems but may leave serious ones unresolved (Overall and McNulty 2017). The aim of a repair attempt is to stop runaway escalation, not to make one partner give up a legitimate demand.
Writing repair as damping
In the loop model of Chapter 2, escalation comes from the two partners amplifying each other’s reactions. A repair attempt can be seen as adding a damping term to the loop: when negativity passes a certain level, there is some probability of pulling it down.
Calculation
A conflict is a sequence of turns. Negativity rises on average by 0.30 per turn (each reply a little sharper), with random fluctuation. When , the speaker makes a repair attempt with probability and the listener accepts it with probability ; an accepted repair halves . The conflict “settles for now” when falls below 0.3 and ends “flooded” (one partner shuts down or leaves) when reaches 6.
# Repair attempts as damping. A conflict is a sequence of speaking turns. Negativity n rises on
# average by s = 0.30 per turn (each reply a little sharper), with noise. Whenever n > 1, the speaker
# makes a repair attempt (a pause, a softer restatement, humour, "I see your point") with
# probability p_make; the listener accepts it with probability p_accept, and an accepted repair
# halves n. A conflict ends "settled for now" when n falls below 0.3, or "flooded" when n reaches 6
# (someone shuts down or leaves). Illustrative parameters; only the shape of the result matters
# (Gottman 1999 describes repair attempts and why their acceptance matters).
import random
def episode(p_make, p_accept, rng, s=0.30, cut=0.5, start=1.0, cap=6.0, settle=0.3, max_turns=200):
n, turns = start, 0
while settle <= n < cap and turns < max_turns:
n += s + rng.gauss(0, 0.5)
if n > 1 and rng.random() < p_make and rng.random() < p_accept:
n *= 1 - cut
turns += 1
return turns, n >= cap
rng = random.Random(2)
print("share of conflicts that end flooded, and median length in turns (4,000 simulated conflicts each)")
print("p_make \\ p_accept | 0.2 | 0.5 | 0.8")
for pm in (0.1, 0.3, 0.5, 0.8):
cells = []
for pa in (0.2, 0.5, 0.8):
res = [episode(pm, pa, rng) for _ in range(4000)]
med = sorted(t for t, _ in res)[len(res) // 2]
cells.append(f"{sum(f for _, f in res) / len(res):4.0%} flooded, {med:2d}")
print(f" {pm:.1f} | " + " | ".join(cells))
print("p_make * p_accept is what counts: 0.8 x 0.2 (=0.16) floods far more often than 0.5 x 0.5 (=0.25),")
print("and the change from mostly flooded to mostly settled happens between products of about 0.1 and 0.4")
# share of conflicts that end flooded, and median length in turns (4,000 simulated conflicts each)
# p_make \ p_accept | 0.2 | 0.5 | 0.8
# 0.1 | 89% flooded, 16 | 87% flooded, 18 | 82% flooded, 21
# 0.3 | 86% flooded, 19 | 66% flooded, 27 | 30% flooded, 29
# 0.5 | 78% flooded, 22 | 26% flooded, 27 | 3% flooded, 20
# 0.8 | 61% flooded, 27 | 3% flooded, 19 | 0% flooded, 13
# p_make * p_accept is what counts: 0.8 x 0.2 (=0.16) floods far more often than 0.5 x 0.5 (=0.25),
# and the change from mostly flooded to mostly settled happens between products of about 0.1 and 0.4The model shows two things. First, making a repair and having it accepted multiply: when one partner tries very hard to repair () and the other seldom accepts (), more than six in ten conflicts still end flooded; when both make and accept only moderately (0.5 and 0.5), the flooded share falls to about a quarter. Second, the change from “mostly flooded” to “mostly settled” happens over a fairly narrow band, products of about 0.1 to 0.4: in the model there is a threshold-like relation between damping and the slope of escalation. What decides how a conflict ends is not whether there is conflict, but whether repair can cut off the slope of escalation.
Flooding and taking a break
Gottman describes a state he calls “flooding”: in intense conflict, heart rate and the stress response rise sharply, and it becomes hard to take in what the other says and easier to become defensive, counter-attack or stonewall. In his clinical writing he advises that once one partner feels flooded, the couple should agree to stop for at least about 20 minutes, and spend the break doing something calming rather than rehearsing the argument in their heads (Gottman 1999).
Calculation
During conflict, heart rate has risen from a resting 72 to 110 beats per minute. Once the break begins it relaxes toward rest with a time constant of 5 minutes, but what one does during the break adds its own push: a calming activity adds little, rehearsing the argument a lot.
# The physiology of a time-out. During a heated conflict heart rate has risen from a resting 72 to
# 110 beats per minute. A break begins at t = 0. Heart rate relaxes toward rest with time constant
# tau = 5 minutes, but whatever the person does during the break adds a drive u (beats per minute,
# per minute): rehearsing grievances keeps it up, slow breathing or a walk keeps it low.
# dHR/dt = -(HR - 72)/tau + u plateau = 72 + u*tau
# Gottman's clinical advice is a break of at least 20 minutes spent on something soothing; the model
# shows why the activity can matter as much as the length. All numbers are illustrative.
import math
REST, PEAK, TAU = 72.0, 110.0, 5.0
def hr(t, u):
plateau = REST + u * TAU
return plateau + (PEAK - plateau) * math.exp(-t / TAU)
def minutes_to(target, u):
plateau = REST + u * TAU
if plateau >= target: return None
return TAU * math.log((PEAK - plateau) / (target - plateau))
acts = (("soothing activity", 0.4), ("neutral (reading, chores)", 1.2), ("rehearsing the argument", 2.6))
print("heart rate (bpm) at the end of a break of ...")
print("during the break | 5 min | 10 min | 20 min | 30 min | time to within 10 bpm of rest")
for name, u in acts:
t10 = minutes_to(REST + 10, u)
print(f"{name:28s} | " + " | ".join(f"{hr(t, u):6.0f}" for t in (5, 10, 20, 30))
+ f" | {f'{t10:.0f} min' if t10 else 'not reached'}")
# heart rate (bpm) at the end of a break of ...
# during the break | 5 min | 10 min | 20 min | 30 min | time to within 10 bpm of rest
# soothing activity | 87 | 79 | 75 | 74 | 8 min
# neutral (reading, chores) | 90 | 82 | 79 | 78 | 10 min
# rehearsing the argument | 94 | 88 | 85 | 85 | not reachedIn this model, a calming activity brings heart rate to within 10 beats of rest in about 8 minutes; rehearsing the argument keeps it about 13 beats above rest even after 30 minutes, and it does not come back. So the content of a break may matter as much as its length. The model’s time constant is illustrative, and real recovery times vary greatly between people and situations; “at least 20 minutes” is Gottman’s clinical advice, and the author has found no randomised trial testing that particular length.
Note
A break is not the silent treatment. A useful break has three features: it is named as a break (“I need twenty minutes to calm down, then let’s talk again”), a time to come back is agreed, and the person really does come back. A “break” with no agreement and no return looks to the other like withdrawal and may trigger exactly the loop of Chapter 2. In relationships with violence or control, any arrangement for breaks must put safety first, with professional help where needed.
Hypothesis · not established
Acceptance matters more than attempts (not yet established). In couples’ conflict diaries, the duration of escalation correlates more strongly with “the share of repairs accepted” than with “the number of repair attempts”; and the share accepted correlates with the frequency of everyday positive interaction outside conflict. Testable prediction and refutation: controlling for the number of conflicts, if escalation duration correlates no less with the number of attempts than with the share accepted, the hypothesis is weakened.
Evidence for professional help
This book is not couple therapy, but it is worth knowing that couple therapy has a substantial evidence base. Meta-analysis shows that couples who receive couple therapy improve more on average than couples who do not (Shadish and Baldwin 2003); a large randomised trial with seriously and chronically distressed married couples compared two behavioural couple therapies and found clear improvement with both (Christensen et al. 2004). If conflict has made you both unhappy for a long time, help from a qualified couple therapist is more likely to be useful than any book.
Chapter summary
Stable couples are not couples who never fight, but couples who can pull conflict back when it escalates. Repair attempts can be seen as damping in the loop: making a repair and having it accepted multiply, so if either is near zero the damping fails; in the model there is a threshold-like relation between damping and the slope of escalation. A break when flooded may help, and what one does during it may matter as much as how long it lasts. Gentle is not always better: the aim of repair is to stop runaway escalation, not to give up a legitimate demand. The next chapter reads this modern research beside classical Chinese texts.
References
Fan, Y. (2026). Computational Attachment Theory and Couple Interaction Dynamics: From Anxious-Avoidant Coupled Oscillators and Gottman's Nonlinear Difference Equations to Updating Early Relational Priors [依恋理论的计算模型与伴侣互动动力学] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23754715 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).