Why this page exists

Open any book about selling and you will meet the same dozen studies. They appear so often, and so confidently, that it is reasonable to assume somebody checked them. Mostly nobody has. Four or five retellings back there is a real paper, and it often says something narrower and more conditional than the version in circulation.

That produces no debunking. Most of these findings are real. But in nearly every case the striking number survived into popular use and its conditions did not — and the conditions are the practical content. The table below is the whole evidence base of the book, open for anyone to check.

How to read the labels

Holds up — reproduced by other researchers at roughly the reported size.
Conditional — real, but only under stated conditions, or rests on a single study or site.
Contested — failed to replicate, shrank substantially, or the usually quoted claim is not what the paper measured.
Not a study — trade practice, corporate anecdote or unpublished proprietary research. Useful, but not evidence of the same kind.
Synthesis — the entry is itself a meta-analysis or a paper that reconciles two earlier results. It is the yardstick, not a thing being measured.

All thirty sources

Chapter numbers refer to The Fourth Bottle. Full citations are in Appendix C of the book.

#StudyHeadline figureReplication statusCh
1Iyengar & Lepper, JPSP — jam display
2000
60% / 40% stopped; 3% / 30% boughtContested
Contested. Near-zero mean in Scheibehenne 2010 meta-analysis; restored as conditional by Chernev 2015
4
2Scheibehenne, Greifeneder & Todd — meta-analysis, ~50 studies
2010
Mean effect of assortment size ≈ zeroSynthesis4
3Chernev, Böckenholt & Goodman, JCP — 99 observations, N=7,202
2015
Four moderators: set complexity, task difficulty, preference uncertainty, decision goalSynthesis
Reconciles 1 and 2
4, 5, 7
4P&G Head & Shoulders SKU cut 26→15
1997
Sales +10%, reported Washington PostNot a study
Single company, single line. Corporate anecdote, not a study
4
5Fragrance-trade testing guidance
current
Cap of 3–4 fragrances per session; olfactory adaptationNot a study
Industry practice, convergent across sources; not an experiment
4, 5
6Karmarkar, Shiv & Knutson, JMR — price primacy, fMRI
2015
Price-first shifts valuation to “is it worth it”; product-first to “do I like it”. Purchase rates did not differConditional
Single study, student sample, small items
1
7Mason, Lee, Wiley & Ames, JESP — precise first offers
2013
N=1,254. Precise offers ($5,015 vs $5,000) read as better-informed; counterparts conceded moreConditional
Later work finds limits; see 8
2
8Lee, Loschelder, Schweinsberg, Mason & Galinsky, OBHDP
2018
Precise offers deter aggressive counteroffersSynthesis
Extends 7
2
9Yang, Kimes & Sessarego, IJHM — menu price formats
2009
N=201. No currency symbol → +$5.55 per head, ≈8%. Spelled-out prices no different from $Conditional
Single site (CIA café)
3, 9
10Ariely — The Economist subscription demonstration
2008
3 options: 16/0/84. 2 options: 68/32Contested
Two large 2014 replication efforts largely failed with realistic stimuli
6
11Huber, Payne & Puto — asymmetric dominance
1982
Original decoy effectConditional
Holds in abstract two-attribute paradigms only
6
12Prelec & Loewenstein, Marketing Science — pain of paying
1998
Pain driven by payment saliency and timingHolds up
Framework, widely built on
8, 9
13Prelec & Simester, Marketing Letters — “Always leave home without it”
2001
Credit-card bids up to 100% higher than cash, real Celtics ticketsConditional
Directionally replicated; magnitude varies
8
14Shiv, Carmon & Ariely, JMR — marketing placebo
2005
Discounted energy drink → worse puzzle performance than full priceHolds up
Conceptually replicated (Fell & Eisend 2013)
10
15Shampanier, Mazar & Ariely, Marketing Science — zero price
2007
Hershey 1¢ / Lindt 15¢ → 27/73. Hershey free → invertsHolds up
Robust
10
16Jia, Huang, Zhang, Shi & Zhang — final price neglect
2023
Separated discounts beat integrated ones on higher-priced itemsConditional
Recent, not yet replicated
10
17Regan, JESP — unsolicited Coke
1971
Raffle purchase 25% → 58%Holds up
Classic; reciprocity robust
11
18Strohmetz, Rind, Fisher & Lynn, JASP — candy and tipping
2002
Small candy gift raises tipsHolds up
Replicated in tipping literature
11
19Rind & Strohmetz, JASP — helpful message on the check
1999
Mean tip ≈17% → ≈20%Conditional
Single server, single site
11
20Server self-introduction by name
—
Tip rate 15% → 23% (buffet setting)Conditional
Single study; widely cited in tipping work
11
21Buell & Norton, Management Science — labor illusion
2011
Five experiments. Visible effort → preference for longer waits, same resultsConditional
Robust within domain
11
22Peck & Shu, JCR — mere touch
2009
Touch raises perceived ownership and valuation; imagery-of-touch works tooConditional
Four studies; touch valence matters
11
23Goldstein, Cialdini & Griskevicius, JCR — hotel towels
2008
35.1% → 44.1% (p=.05); Exp 2 37.2% → 44.5%Holds up
Five replications, N=2,466, all p>.14. Bayesian synthesis of 7 studies: BF₁₀=36.89, effect real but ≈6 points, not 9
12
24Scheibehenne, Jamil & Wagenmakers, Psych Science — Bayesian synthesis
2016
Reconciles 23Synthesis12
25Ein-Gar, Shiv & Tormala, JCR — blemishing effect
2012
Small negative info raises favourability — only when it follows positives, is minor, and processing is low-effortHolds up
Four studies, lab and field
12
26Worchel, Lee & Adewole, JPSP — cookie jar
1975
2-cookie jar rated more desirable than 10-cookie jar. Price measure not significantContested
Desire yes, willingness-to-pay unproven
12
27Freedman & Fraser, JPSP — foot in the door
1966
76% vs 16.7% agreed to large lawn signConditional
Classic; effect smaller in modern meta-analyses
12
28Langer, Blank & Chanowitz, JPSP — photocopier
1978
5 pages: 60% / 94% / 93%. 20 pages: placebic reason stops workingConditional
Replicated within its conditions
13
29Rackham / Huthwaite — photo-store closing study
1988
Low-value: 72%→76% after closing training. High-value (avg $109): 42%→33%Not a study
Proprietary consulting research, never peer-reviewed, raw data never released
13
30Rackham / Huthwaite — early-closing study
1988
30 high-close calls → 11 sales (37%). 30 low-close calls → 21 sales (70%)Holds up
Same caveat as 29
13

Five things worth knowing before you quote any of this

The standing caveats

The most-cited evidence in professional selling is the least checkable

Huthwaite’s 35,000-call corpus underpins most modern B2B sales training. It was proprietary consulting research, never peer-reviewed, and the raw data was never released. The book cites it, because it is the largest body of evidence there is, and says this plainly rather than treating it as settled science. The source even disagrees with itself on how long the study ran: the preface says twelve years, chapter one says ten.

A famous result about price order found no difference in what people bought

The price-primacy work (Karmarkar, Shiv & Knutson, 2015) found that showing the price first shifts the brain toward “is it worth it” and the product first toward “do I like it”. Purchase rates did not differ. Any advice built on it has to say so.

The decoy effect is a laboratory phenomenon first

Asymmetric dominance is real and reliable in abstract two-attribute choices. Two large replication efforts in 2014 largely failed to reproduce it with realistic stimuli. The book presents it as a layout principle, not a lever you can pull.

The towel sign is real, and it is smaller than you were told

Famous number: a 9-point lift. Five replications with 2,466 guests found nothing. A Bayesian synthesis of all seven studies (Scheibehenne, Jamil & Wagenmakers, 2016) put the Bayes factor at 36.89 and the effect at roughly 6 points. Both the “it works” and the “it failed” camps were reading one half of the evidence.

Scarcity raises desire. The study usually cited never showed it raises price

In the cookie-jar experiment (Worchel, Lee & Adewole, 1975) the two-cookie jar was rated more desirable than the ten-cookie jar. The willingness-to-pay measure was not significant. That distinction matters a great deal if you are about to cut your stock to raise your margin.

Questions this page gets asked

Did the jam study replicate?

Partly. Iyengar and Lepper found in 2000 that a six-flavour display outsold a twenty-four-flavour one roughly ten to one. A 2010 meta-analysis of about fifty studies by Scheibehenne, Greifeneder and Todd put the average effect near zero. A 2015 analysis of 99 observations by Chernev, Bockenholt and Goodman found the effect is real but depends on four moderators: set complexity, task difficulty, preference uncertainty and decision goal.

Is the hotel towel social proof effect real?

Yes, and it is smaller than the famous figure. The original 2008 study reported a rise in towel reuse from 35.1 to 44.1 per cent. Five later replications with 2,466 guests found nothing significant, every p-value above .14. A 2016 Bayesian synthesis of all seven studies concluded the effect exists and is about six percentage points, not nine.

Is SPIN Selling based on real data?

It is based on the largest dataset in professional selling and the least verifiable one. Huthwaite analysed around 35,000 sales calls, but the research was proprietary, was never peer-reviewed and the raw data was never published. The headline finding, that closing training lowers success on high-value sales from 42 to 33 per cent, cannot be independently checked.

Does the decoy effect work on real products?

Much less reliably than on paper. It is robust in abstract two-attribute laboratory choices. Two substantial replication efforts in 2014 largely failed to reproduce it using realistic stimuli, including the subscription example that made it famous. Treat it as a way to lay out a menu rather than a lever that moves revenue.

Which findings in the book have the strongest evidence?

Reciprocity, the zero-price effect and the pain-of-paying framework come through replication best. The labour-illusion work holds within its domain across five experiments. These are also, not coincidentally, the findings that keep working after the buyer understands what you are doing.

Why include findings you say are weak?

Three findings in the first volume have weak evidence and are labelled as such wherever they appear. They are in the book because they are so widely taught that leaving them out would be the greater disservice: a reader who meets them in a training course deserves to know what the evidence for them actually is.

Where these thirty studies turn into something you can use

The table says what the evidence is. The book says what to do about it: the situation each finding applies to, the words to use, and the conditions under which it stops working.

Buy The Fourth Bottle — $19PDF and EPUB · 219 pages · instant download · a full chapter is free to read first