What does Star Ratings, Self-Selection and Bayesian Ranking conclude, in numbers, about “Five Stars Do Not Sell Best: the Spiegel Curve at Source and a Small-Sample Reading”?
Answer: Chapter 2 returns to the 2017 report of the Spiegel Research Center: the report states the peak in three ways, whose intersection with its source article is 4.2 to 4.5 stars; it prints no curve values or confidence intervals and does not test the proposed psychological mechanism; and an illustrative catalogue shows that neither small-sample noise nor moderate boosting explains the whole inverted U.
Scope of the numbers: Chapter 1 explains the J-shaped star distribution with two filters, a purchasing bias and an under-reporting bias; with illustrative parameters the purchasing bias lifts the mean by +0.532 stars. Marked boxes separate published evidence, runnable Python calculations (21 programs), classical sources and the author's hypotheses.
Data table
The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 2017, 4.2, 4.5
peak statements (stars), Spiegel 2017 report and its source paper p9 'typically peaks' 4.0 to 4.7 width 0.7 p9 'less likely' (vs) 4.2 to 4.7 width 0.5 p18 summary 4.2 to 4.5 width 0.3 IJA 2017 abstract 4.2 to 4.5 width 0.3 common to all four: 4.2 to 4.5 data behind the claims (as described in the report) CPG: 2 retailers, 22 categories, 111460 SKUs, 5066 SKUs/category gift retailer: 8611 views/product, 1.99 views/user prompted study: 122000 reviews, 9.0 per product, verified 0.533 effects as printed, restated as multipliers five reviews vs none: times 3.70 lower-priced times 2.90, higher-priced times 4.80, ratio 1.66 verified minus anonymous mean rating: +0.45 stars badge: times 1.15 purchase likelihood
Show the full program
# Chapter 2. p21: what the Spiegel report actually printed, put in one ledger.
# Source: Spiegel Research Center (Northwestern University, Medill) with
# PowerReviews, "How Online Reviews Influence Sales", 2017. Every constant
# below is copied from that report; the rest is arithmetic on them.
# The report publishes no purchase-probability values for the rating curve,
# only charts, so no curve height is entered here.
import math
# three peak statements printed in the same report
RANGES = (
("p9 'typically peaks'", 4.0, 4.7),
("p9 'less likely' (vs)", 4.2, 4.7),
("p18 summary", 4.2, 4.5),
)
PAPER = ("IJA 2017 abstract", 4.2, 4.5) # Maslowska, Malthouse, Bernritter
# data described in the report
CPG_RETAILERS, CPG_CATEGORIES, CPG_SKUS = 2, 22, 111460
GIFT_VIEWS, GIFT_PRODUCTS = 15.5 * 1e6, 1800
GIFT_USERS = 7.8 * 1e6
ANON_REVIEWS, VERIFIED_REVIEWS, PRODUCTS_3 = 57000, 65000, 13500
LIFT_FIVE = 2.70 # "270% greater" with five reviews vs none
LIFT_LOW, LIFT_HIGH = 1.90, 3.80 # lower- vs higher-priced, gift retailer
MEAN_VERIFIED, MEAN_ANON = 4.34, 3.89
BADGE = 0.15 # verified-buyer badge, purchase likelihood
lo = max(r[1] for r in RANGES + (PAPER,))
hi = min(r[2] for r in RANGES + (PAPER,))
print("peak statements (stars), Spiegel 2017 report and its source paper")
for name, a, b in RANGES + (PAPER,):
print(" %-24s %.1f to %.1f width %.1f" % (name, a, b, b - a))
print(" common to all four: %.1f to %.1f" % (lo, hi))
print("data behind the claims (as described in the report)")
print(" CPG: %d retailers, %d categories, %d SKUs, %.0f SKUs/category"
% (CPG_RETAILERS, CPG_CATEGORIES, CPG_SKUS,
CPG_SKUS / CPG_CATEGORIES))
print(" gift retailer: %.0f views/product, %.2f views/user"
% (GIFT_VIEWS / GIFT_PRODUCTS, GIFT_VIEWS / GIFT_USERS))
tot = ANON_REVIEWS + VERIFIED_REVIEWS
print(" prompted study: %d reviews, %.1f per product, verified %.3f"
% (tot, tot / PRODUCTS_3, VERIFIED_REVIEWS / tot))
print("effects as printed, restated as multipliers")
print(" five reviews vs none: times %.2f" % (1.0 + LIFT_FIVE))
print(" lower-priced times %.2f, higher-priced times %.2f, ratio %.2f"
% (1.0 + LIFT_LOW, 1.0 + LIFT_HIGH, (1 + LIFT_HIGH) / (1 + LIFT_LOW)))
print(" verified minus anonymous mean rating: %+.2f stars"
% (MEAN_VERIFIED - MEAN_ANON))
print(" badge: times %.2f purchase likelihood" % (1.0 + BADGE))Single source and citation
Fan, Y. (2026). J-Shaped Self-Selection in Star Ratings, Bayesian-Average Ranking and Rating Inflation from the Nine-Rank System to Online Marketplaces (Version 1.0.0), Chapter 2: Five Stars Do Not Sell Best: the Spiegel Curve at Source and a Small-Sample Reading. 5D Library. DOI pending
DOI pending · monograph page goes live after publication · 5DL-B-212 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026212,
author = {Fan, Yafu},
title = {J-Shaped Self-Selection in Star Ratings, Bayesian-Average Ranking and Rating Inflation from the Nine-Rank System to Online Marketplaces},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}Other chapter answers in this book
- Chapter 1 · Where the J Comes From: a U-Shaped Urge to Post and Two Self-Selection Biases
- Chapter 3 · The Wilson Lower Bound: Why Three Five-Star Reviews Rank Below Two Hundred at 4.6
- Chapter 4 · The Dirichlet-Prior Bayesian Average: Pseudo-Reviews Borrowed from the Shop
- Chapter 5 · Sorting Is Soliciting: the Shelf Feedback Loop, the Cost of Exploring and the First Five Reviews
- Chapter 6 · Rating Inflation: Norm Chasing, Information Decay and the Friction of Asking
- Chapter 7 · Classic Texts East and West: Liu Yi on the Nine-Rank System and Galton's "Vox Populi"
- Chapter 8 · Closing the Loop: Shop Sorting, the Timing of Review Requests and a Falsification Plan