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

第二章 价格尾数与数字精确性的信号模型

左位数字效应

托马斯与莫维茨的实验显示,买家对价格的判断偏重最左边的数字:一分钱的降价若改变了最左边的数字,就被感觉为比改变后面数字的同额降价大得多 (Thomas and Morwitz 2005)。拉塞特拉、波普与赛德诺在二手车市场估计,买家对里程表首位以后的数字只给出部分注意力,注意力不足参数约为 0.3 (Lacetera et al. 2012)。程序把这一机制写成:感知价格 = 首位数字所代表的整数 + (1−θ)(1-\theta) × 其余部分。

推导

# Chapter 2. p03: The left-digit effect as partial inattention.
# A buyer reads the leading digits fully and the rest with weight
# (1 - theta): for a price with d digits, perceived price
#   P~ = L + (1 - theta) (P - L),  L = P rounded down to its first digit.
# Lacetera, Pope and Sydnor (2012) estimated theta about 0.3 for car
# odometers. We compare just-below and round prices, then ask how much
# conversion a just-below price buys with the WTP model of p02.
from math import log, erf, sqrt

def lead(p):
    s = str(int(p))
    return int(s[0]) * 10 ** (len(s) - 1)

def perceived(p, theta):
    L = lead(p)
    return L + (1 - theta) * (p - L)

def buy_share(price, med=30.0, sd=1.0):
    return 0.5 * (1 - erf((log(price) - log(med)) / sd / sqrt(2)))

if __name__ == "__main__":
    pairs = ((49, 50), (298, 300), (9800, 10000), (29800, 30000))
    for theta in (0.0, 0.3):
        print("   theta = {:.1f}".format(theta))
        for a, b in pairs:
            pa, pb = perceived(a, theta), perceived(b, theta)
            print("     RM {:>6,} feels like {:9.1f}; RM {:>6,} like {:9.1f};"
                  .format(a, pa, b, pb))
            print("       gap {:.1%} of the round price (nominal {:.1%})"
                  .format((pb - pa) / b, (b - a) / b))
    s49 = buy_share(perceived(49, 0.3))
    s50 = buy_share(perceived(50, 0.3))
    print("\n   share buying, theta 0.3: RM 49 {:.4f} vs RM 50 {:.4f}"
          " ({:+.1%})".format(s49, s50, s49 / s50 - 1))
    print("   revenue per visitor: RM 49 {:.2f} vs RM 50 {:.2f}".format(
        49 * s49, 50 * s50))
#    theta = 0.0
#      RM     49 feels like      49.0; RM     50 like      50.0;
#        gap 2.0% of the round price (nominal 2.0%)
#      RM    298 feels like     298.0; RM    300 like     300.0;
#        gap 0.7% of the round price (nominal 0.7%)
#      RM  9,800 feels like    9800.0; RM 10,000 like   10000.0;
#        gap 2.0% of the round price (nominal 2.0%)
#      RM 29,800 feels like   29800.0; RM 30,000 like   30000.0;
#        gap 0.7% of the round price (nominal 0.7%)
#    theta = 0.3
#      RM     49 feels like      46.3; RM     50 like      50.0;
#        gap 7.4% of the round price (nominal 2.0%)
#      RM    298 feels like     268.6; RM    300 like     300.0;
#        gap 10.5% of the round price (nominal 0.7%)
#      RM  9,800 feels like    9560.0; RM 10,000 like   10000.0;
#        gap 4.4% of the round price (nominal 2.0%)
#      RM 29,800 feels like   26860.0; RM 30,000 like   30000.0;
#        gap 10.5% of the round price (nominal 0.7%)
#
#    share buying, theta 0.3: RM 49 0.3322 vs RM 50 0.3047 (+9.0%)
#    revenue per visitor: RM 49 16.28 vs RM 50 15.24

当 θ=0\theta = 0 时,RM 49 与 RM 50 的感知差距就是名义差距 2.0%。当 θ=0.3\theta = 0.3 时,RM 49 被感知为 46.3,感知差距放大到 7.4%;RM 298 被感知为 268.6,差距 10.5%(名义只有 0.7%);RM 9,800 被感知为 9560.0。在第一章的估值模型中,RM 49 的购买比例为 0.3322,RM 50 为 0.3047(+9.0%),每位访客带来的收入分别为 RM 16.28 与 RM 15.24。

整数价格与权威感

左位数字效应解释了前端冲动消费为什么常用 49、298 这样的尾数。反过来,整数价格让人感觉「对」:瓦德瓦与张发现,对于凭感受评价的购买,整数价格提高产品评价,而对于凭计算评价的购买,非整数价格更好 (Wadhwa and Zhang 2015)。因此高客单的牌照或顾问服务写 RM 10,000 这样的整数、低客单的冲动购买写 RM 49 这样的尾数,可以同时成立:两者针对的是不同的评价方式。注意 RM 9,800 并不是整数价格,它仍然是一个「首位以下打折」的价格。

精确数字的可信度

文案中写「误差小于 12 秒」,比写「极准」更可信吗?实验研究发现,精确的数字更容易被当作事实,并改变读者的锚定 (Janiszewski and Uy 2008)。程序用一个简单的贝叶斯模型说明这种可信度从何而来:说话者可能诚实,也可能夸大;诚实者给出精确数字的概率为 qhq_h,夸大者为 qpq_p。

推导

# Chapter 2. p04: When does a precise number earn trust?
# A claim comes from an honest source (prior pi) or a puffing one.
# Honest sources state precise figures with probability qh, puffers with
# qp (precision is costlier to fake: a precise figure can be checked).
# Seeing a precise figure, P(honest | precise) = pi qh / (pi qh +
# (1 - pi) qp). If a checked false figure is caught with probability c,
# puffers who fear checking lower qp; the trust gain then grows.
def posterior(pi, qh, qp):
    return pi * qh / (pi * qh + (1 - pi) * qp)

if __name__ == "__main__":
    pi = 0.5
    print("   prior 0.50   qh    qp    P(honest|precise)  P(honest|vague)")
    for qh, qp in ((0.6, 0.6), (0.6, 0.4), (0.7, 0.3), (0.8, 0.1)):
        print("              {:.1f}   {:.1f}   {:17.3f}  {:15.3f}".format(
            qh, qp, posterior(pi, qh, qp), posterior(pi, 1 - qh, 1 - qp)))
    print("\n   puffers' precision falls as checking risk c rises:"
          " qp = 0.6 (1 - c)")
    for c in (0.0, 0.25, 0.5, 0.75):
        qp = 0.6 * (1 - c)
        print("     c = {:.2f}: qp {:.2f}, P(honest|precise) {:.3f}".format(
            c, qp, posterior(pi, 0.6, qp)))
    print("\n   with no checking (c = 0) precision carries no information;")
    print("   its value comes entirely from the chance of being checked.")
#    prior 0.50   qh    qp    P(honest|precise)  P(honest|vague)
#               0.6   0.6               0.500            0.500
#               0.6   0.4               0.600            0.400
#               0.7   0.3               0.700            0.300
#               0.8   0.1               0.889            0.182
#
#    puffers' precision falls as checking risk c rises: qp = 0.6 (1 - c)
#      c = 0.00: qp 0.60, P(honest|precise) 0.500
#      c = 0.25: qp 0.45, P(honest|precise) 0.571
#      c = 0.50: qp 0.30, P(honest|precise) 0.667
#      c = 0.75: qp 0.15, P(honest|precise) 0.800
#
#    with no checking (c = 0) precision carries no information;
#    its value comes entirely from the chance of being checked.

若两类说话者给出精确数字的概率相同(都是 0.6),看到精确数字后的后验与先验相同,都是 0.500:精确本身不带任何信息。只有当夸大者因为害怕被核查而较少给出精确数字时,精确才有信息量:若被核查的概率为 0.50,后验升到 0.667;若诚实者与夸大者分别为 0.8 与 0.1,后验为 0.889。所以,精确数字的说服力完全来自它可以被核查,这同时意味着:写下的精确数字必须是能经得起核查的真实数字,否则它恰恰是最容易被揭穿的那一句。

假说 · 尚未验证

作者的假说(尚未验证): 在测算报告的销售页上,把「极准」换成可以核查的精确陈述(例如程序的计算误差),对付费转化的提升主要出现在已经读过免费部分的访客身上。什么会证明它错: 若按访客是否读完免费部分分层后,两层的提升没有显著差异,这一假说就不成立。

小结

尾数定价的作用可以用注意力不足参数计算:θ=0.3\theta = 0.3 时 RM 49 与 RM 50 的感知差距是名义差距的三倍多。整数价格适合凭感受评价的高客单购买。精确数字的可信度来自可核查性,没有核查就没有信息。

参考文献

Janiszewski, Chris, and Dan Uy. 2008. “Precision of the Anchor Influences the Amount of Adjustment.” Psychological Science 19 (2): 121–27.
Lacetera, Nicola, Devin G. Pope, and Justin R. Sydnor. 2012. “Heuristic Thinking and Limited Attention in the Car Market.” American Economic Review 102 (5): 2206–36.
Thomas, Manoj, and Vicki Morwitz. 2005. “Penny Wise and Pound Foolish: The Left-Digit Effect in Price Cognition.” Journal of Consumer Research 32 (1): 54–64.
Wadhwa, Monica, and Kuangjie Zhang. 2015. “This Number Just Feels Right: The Impact of Roundedness of Price Numbers on Product Evaluations.” Journal of Consumer Research 41 (5): 1172–85.

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).