5D Library › 说服文案代数、损失框架不对称性与结账微界面工程 · English

第一章 损失框架的前景理论算术

价值函数与决策权重

卡尼曼与特沃斯基的前景理论把结果相对于参照点分成收益与损失,并用一条在参照点处折弯的价值函数来描述主观价值 (Kahneman and Tversky 1979)。累积前景理论给出的参数估计是:收益 v(x)=x0.88v(x) = x^{0.88},损失 v(x)=−2.25(−x)0.88v(x) = -2.25\,(-x)^{0.88};概率经过权重函数 w(p)=pγ/(pγ+(1−p)γ)1/γw(p) = p^{\gamma}/(p^{\gamma} + (1-p)^{\gamma})^{1/\gamma} 变换,收益 γ=0.61\gamma = 0.61,损失 γ=0.69\gamma = 0.69 (Tversky and Kahneman 1992)。

推导

# Chapter 1. p01: Loss and gain frames under cumulative prospect theory.
# Tversky and Kahneman (1992): v(x) = x^a for gains, -lam (-x)^b for
# losses, a = b = 0.88, lam = 2.25; decision weights
# w(p) = p^g / (p^g + (1 - p)^g)^(1/g), g = 0.61 (gains), 0.69 (losses).
# A frame does not change the outcome; it changes which branch of v is
# used. We compare the subjective weight of the same RM amount presented
# as a gain to secure or as a loss to avoid, sure and uncertain.
A = B = 0.88
LAM = 2.25

def v(x):
    return x ** A if x >= 0 else -LAM * (-x) ** B

def w(p, g):
    return p ** g / (p ** g + (1 - p) ** g) ** (1 / g)

def money_equivalent(u):
    # the sure gain that has the same value as u (inverse of v on gains)
    return u ** (1 / A)

if __name__ == "__main__":
    print("   amount   v(gain)   |v(loss)|   ratio")
    for x in (10, 100, 1000):
        print("   {:6d}  {:8.2f}  {:10.2f}   {:.3f}".format(
            x, v(x), -v(-x), -v(-x) / v(x)))
    print("\n   uncertain outcomes of RM 1,000 with probability p")
    print("   p      w+(p)   w-(p)   gain frame  loss frame  ratio")
    for p in (0.05, 0.10, 0.30, 0.50, 0.90):
        g = w(p, 0.61) * v(1000)
        l = -w(p, 0.69) * v(-1000)
        fmt = "   {:.2f}  {:6.3f}  {:6.3f}  {:10.2f}  {:10.2f}  {:5.3f}"
        print(fmt.format(
            p, w(p, 0.61), w(p, 0.69), g, l, l / g))
    l = -w(0.10, 0.69) * v(-1000)
    print("\n   'avoid a 10% chance of losing RM 1,000' feels like a sure"
          " gain of")
    print("   RM {:.2f}; the expected value is RM 100.00".format(
        money_equivalent(l)))
#    amount   v(gain)   |v(loss)|   ratio
#        10      7.59       17.07   2.250
#       100     57.54      129.47   2.250
#      1000    436.52      982.16   2.250
#
#    uncertain outcomes of RM 1,000 with probability p
#    p      w+(p)   w-(p)   gain frame  loss frame  ratio
#    0.05   0.132   0.111       57.46      109.45  1.905
#    0.10   0.186   0.170       81.32      167.11  2.055
#    0.30   0.318   0.328      138.97      321.73  2.315
#    0.50   0.421   0.454      183.62      445.89  2.428
#    0.90   0.712   0.775      310.68      761.08  2.450
#
#    'avoid a 10% chance of losing RM 1,000' feels like a sure gain of
#    RM 335.85; the expected value is RM 100.00

由于收益与损失的曲率相同,同额的确定损失与确定收益的主观值之比恰为 2.25,与金额无关。对于不确定的结果,两种框架的比值随概率变化:概率为 0.05 时为 1.905,0.10 时为 2.055,0.90 时为 2.450。一个具体的读法是:「避免一成机会损失 RM 1,000」在主观上相当于确定得到 RM 335.85,而它的期望值只有 RM 100.00;小概率被高估,损失又被放大。

累积前景理论的价值函数:损失一侧的斜率是收益一侧的 2.25 倍(p01)。

从估值倍数到转化率

估值的放大并不直接等于转化率的放大。程序假设访客的支付意愿服从对数正态分布(中位数 RM 30,对数标准差 1.0),某种框架把每个人的感知价值乘以 kk。

推导

# Chapter 1. p02: From a valuation multiplier to a conversion rate.
# Willingness to pay W is lognormal across visitors (median RM 30,
# log-s.d. 1.0). A frame that multiplies perceived value by k raises the
# share buying at price P from P(W > P) to P(kW > P). k = 2.25 is the
# theoretical ceiling of p01 (the whole value read as a loss avoided);
# meta-analyses of message framing find average effects close to zero,
# so k near 1.00 to 1.10 is the realistic range to plan with.
from math import log, erf, sqrt

MED, SD = 30.0, 1.0

def buy_share(price, k=1.0):
    z = (log(price / k) - log(MED)) / SD
    return 0.5 * (1 - erf(z / sqrt(2)))

if __name__ == "__main__":
    print("   price    k=1.00   k=1.05   k=1.10   k=1.25   k=2.25")
    for price in (49, 98, 298):
        row = [buy_share(price, k) for k in (1.0, 1.05, 1.10, 1.25, 2.25)]
        print("   RM{:4d}  ".format(price) + "  ".join(
            "{:7.4f}".format(x) for x in row))
    base = buy_share(49)
    print("\n   relative lift at RM 49:")
    for k in (1.05, 1.10, 1.25, 2.25):
        lift = buy_share(49, k) / base - 1
        print("     k = {:.2f}: {:+.1%}".format(k, lift))
#    price    k=1.00   k=1.05   k=1.10   k=1.25   k=2.25
#    RM  49   0.3118   0.3293   0.3463   0.3946   0.6256
#    RM  98   0.1183   0.1282   0.1382   0.1684   0.3546
#    RM 298   0.0108   0.0123   0.0139   0.0191   0.0688
#
#    relative lift at RM 49:
#      k = 1.05: +5.6%
#      k = 1.10: +11.1%
#      k = 1.25: +26.5%
#      k = 2.25: +100.6%

在 RM 49 的价格上,k=1.05k = 1.05 使购买比例从 0.3118 升到 0.3293(+5.6%),k=1.10k = 1.10 为 +11.1%,k=1.25k = 1.25 为 +26.5%;即使取理论上限 k=2.25k = 2.25(把全部价值都读成「避免的损失」),也只是 +100.6%。

证据

框架效应的实证大小。 对健康信息的元分析发现,收益框架与损失框架在说服效果上的平均差异接近于零 (O’Keefe and Jensen 2007)。价值函数的 2.25 描述的是对结果本身的估值,而一句文案只是改变描述,读者未必把它当作真实的得失。规划时应当用 1.05 到 1.10 这样的倍数,而不是 2.25。

注

「损失」必须是真实的。 「不解锁就会错过某月的止损窗口」这类文案,暗示购买能避免某种现实损失;本书系列的立场是不主张、也不检验任何测算应验,因此这样的暗示就是一个无法兑现的事实陈述,在许多法域属于误导性陈述 (Laws of Malaysia 1999)。真实可用的「损失」是对报告本身的:优惠截止、名额、价格变化。第五章把两者的账算在一起。

小结

损失框架的 2.25 倍是价值函数在参照点两侧的斜率比;不确定结果下比值在 1.905 到 2.450 之间。换算成转化率,现实的框架效应只值几个到十几个百分点,理论上限也不过翻倍。

参考文献

Kahneman, Daniel, and Amos Tversky. 1979. “Prospect Theory: An Analysis of Decision under Risk.” Econometrica 47 (2): 263–91.
Laws of Malaysia. 1999. Consumer Protection Act 1999 (Act 599). Part II: Misleading and deceptive conduct, false representation and unfair practice.
O’Keefe, Daniel J., and Jakob D. Jensen. 2007. “The Relative Persuasiveness of Gain-Framed and Loss-Framed Messages for Encouraging Disease Prevention Behaviors: A Meta-Analytic Review.” Journal of Health Communication 12 (7): 623–44.
Tversky, Amos, and Daniel Kahneman. 1992. “Advances in Prospect Theory: Cumulative Representation of Uncertainty.” Journal of Risk and Uncertainty 5 (4): 297–323.

Fan, Y. (2026). Persuasive Copywriting Algebra, Loss-Framing Asymmetry and Checkout Micro-Architecture: Prospect-Theory Valuation Ratios, Numerical Specificity Credibility and Cognitive-Fluency Conversion Bounds [说服文案代数、损失框架不对称性与结账微界面工程] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090205 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).