5D Library › 关系断裂生存分析、断联期记忆衰减与复联博弈窗口 · English

第四章 可信的信号

信号博弈

两位前任之间,常常隔着一个疑问:有什么改变了吗?想复合的一方可以这样说,但话不花成本,没有改变的人也可以说同样的话。信号理论问的是:什么样的行动能把想装作一样的两种人分开 (Spence 1973)。答案是:这个行动对假装的一方的成本,必须高于对没有假装的一方的成本。

p07 把信号取为 mm 个月稳定、不施压、对方可以亲眼看到的表现:保住工作、守住承诺、平静地回应。对真正改变了的人,这样的一个月成本是 1;对没有改变的人成本更高,因为要逆着习惯维持。

推导

# Chapter 4. p07: When can a display of change be believed? Two sender
# types: "changed" (H) and "unchanged" (L). The signal is m months of
# steady, pressure-free conduct that the other person can observe
# (work, health, calm replies). A month of it costs H 1 unit and L 3
# units (it is harder to keep up if nothing has changed). Being given
# another chance is worth V = 12 to either type. A separating
# equilibrium needs H willing (m <= V / 1) and L unwilling (m >= V / 3).
# A post that costs nothing to either type separates no one.
V = 12.0

def band(c_h, c_l):
    return V / c_l, V / c_h

if __name__ == "__main__":
    print("   cost a month (H, L)   months that separate   least-cost m")
    for c_h, c_l in ((1, 3), (1, 2), (1, 1.2), (0, 0)):
        if c_l == c_h:
            print("   {:>19}   {:>20}   {:>12}".format(
                "({}, {})".format(c_h, c_l), "none (pooling)", "-"))
            continue
        lo, hi = band(c_h, c_l)
        print("   {:>19}   {:9.1f} to {:6.1f}   {:12.1f}".format(
            "({}, {})".format(c_h, c_l), lo, hi, lo))
    print("   H's net gain at least-cost m (1, 3): {:.1f}".format(V - 4.0))
    print("   L's gain from mimicking at m = 4: {:.1f}".format(V - 3 * 4.0))
#    cost a month (H, L)   months that separate   least-cost m
#                 (1, 3)         4.0 to   12.0            4.0
#                 (1, 2)         6.0 to   12.0            6.0
#               (1, 1.2)        10.0 to   12.0           10.0
#                 (0, 0)         none (pooling)              -
#    H's net gain at least-cost m (1, 3): 8.0
#    L's gain from mimicking at m = 4: 0.0

当成本是 1 与 3、再一次机会对两种人都值 12 时,4.0 至 12.0 个月之间的任何长度都能把两种人分开;成本最低的是 4.0 个月,改变了的人净得 8.0,而没有改变的人模仿它只得 0.0。成本接近时(1 与 1.2),需要 10.0 个月。当信号对任何人都没有成本,例如一条开心的动态,任何长度都分不开:两种人都会发,它不携带任何信息。

对方能推断什么

同一个道理可以写成贝叶斯法则。p08 从「发信号的人真的改变了」的信念 0.3 出发,按观察到的东西更新。

推导

# Chapter 4. p08: What the other person can infer. Prior that the
# sender has really changed: 0.3. Each observation has a probability
# under H and under L; the posterior is updated by Bayes' rule.
#   cheap display (a happy post): H 0.90, L 0.80
#   four months of steady conduct: H 0.80, L 0.15
#   a direct demand for commitment: H 0.20, L 0.60
#   a reply that respects a "no": H 0.90, L 0.40
SIG = {"cheap display": (0.90, 0.80), "four steady months": (0.80, 0.15),
       "direct demand": (0.20, 0.60), "respects a no": (0.90, 0.40)}

def post(prior, seq):
    p = prior
    for s in seq:
        h, l = SIG[s]
        p = p * h / (p * h + (1 - p) * l)
    return p

if __name__ == "__main__":
    print("   evidence seen                              posterior")
    for seq in ((), ("cheap display",), ("cheap display",) * 3,
                ("four steady months",), ("direct demand",),
                ("four steady months", "direct demand"),
                ("four steady months", "respects a no")):
        lab = " + ".join(seq) if seq else "nothing (prior)"
        if len(seq) == 3:
            lab = "cheap display, three times"
        print("   {:<42} {:9.3f}".format(lab, post(0.3, seq)))
#    evidence seen                              posterior
#    nothing (prior)                                0.300
#    cheap display                                  0.325
#    cheap display, three times                     0.379
#    four steady months                             0.696
#    direct demand                                  0.125
#    four steady months + direct demand             0.432
#    four steady months + respects a no             0.837

一次廉价的展示把信念从 0.300 提高到 0.325,三次也只到 0.379:几乎没有作用,因为没有改变的人发得几乎一样多。四个月稳定的表现把它提高到 0.696。一次直接索要承诺作用相反:单独出现时把信念降到 0.125,在四个好月之后也仍会把它拉回 0.432。平静地接受一个「不」,是表中最有力的证据,在四个月稳定表现之后把信念提高到 0.837。一个人能展示的最有说服力的东西,是难以伪装、而且什么也不索取的东西。

假说 · 尚未验证

作者的诠释(尚未验证): 分手之后,前任对另一方「已经改变」的信念,随可观察的、不施压的表现持续时间而上升,在明确索要承诺之后下降。什么会证明它错: 在一项预先登记、随机分配分手后表现描述的情景实验中,稳定表现的持续时间对「是否改变」的判断没有影响,或者索要承诺反而提高这一判断。

小结

展示更好的生活,只有在没有改变的人难以维持时才可信。言语、动态与索取都很廉价,携带的信息很少甚至是负的;几个月稳定的表现与对对方答案的尊重,携带的信息最多。

参考文献

Spence, Michael. 1973. “Job Market Signaling.” The Quarterly Journal of Economics 87 (3): 355 to 374.

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).