Guilt, Shame, and Prosociality
Oslo, October 2008 Tore Ellingsen 1
Wanted: Positive Theories of
• Question: Why do people give?
• Some answers:
– Out of compassion
– Out of obligation
– To avoid guilt
– To avoid shame
• Objective: Shed light on the roles of guilt
Oslo, October 2008 Tore Ellingsen 2
Vehicle: Dictator Games
Basic Dictator Game experiment:
Two subjects. One subject, the Dictator,
gets a monetary endowment M from the
experimenter, the other subject, the
Recipient, gets nothing. The Dictator
decides how to allocate the endowment
between herself and the recipient.
Oslo, October 2008 Tore Ellingsen 3
• Strictly speaking, economic experiments
with monetary payoffs are not “games” but
“game forms”. Only when the (von
Neumann-Morgenstern) utility function has
been defined, can we identify the game
• The identification of such utility functions is
what concerns me here.
Oslo, October 2008 Tore Ellingsen 4
Dictator Games: Different
• FF: Face-to-face
– Natural, but what is “the situation”?
• SB: Single-blind (separate rooms)
– Typical treatment, fixates the situation
• DB: Double-blind (anonymous choice)
– Reduces experimenter concerns
• TB: “Triple-blind” (recipient unawareness)
– Reduces effect of joint context
Oslo, October 2008 Tore Ellingsen 5
A SEK 120 Dictator Game (DB)
Ellingsen and Johannesson (EHB2008)
0 20 40 60 >60
Oslo, October 2008 Tore Ellingsen 6
Dictator Games: General features
• The SEK 120 game distributed 6 SEK 20
bills. With finer grids, the following pattern
– Peaks at 0 and even splits
– Troughs after 0 and before the even split.
– Few donations above the even split.
Oslo, October 2008 Tore Ellingsen 7
• Let us look for utility functions U that may
“rationalize” the choices that we observe.
• Initially, we consider utility functions that
depend merely on the material outcomes.
• As a convention:
– “sub-utility” functions f are increasing,
– parameters are non-negative.
• Of course, people are heterogeneous.
Oslo, October 2008 Tore Ellingsen 8
What model would fit? (i)
• Could it be simply altruism? Let
where ci is i’s consumption.
• Probably not, because:
– It is unreasonable that the utility maximization
problem would have an interior solution given the
restriction on transfers.
– Indeed, subjects should give nothing, because people
don’t share their money with random strangers
outside the lab.
– Interior final allocations are sensitive to initial
allocations (taking options). Bardsley (EE, 2008) and
List (JPE, 2007).
Oslo, October 2008 Tore Ellingsen 9
What model would fit? (ii)
• Could it be altruism and “narrow bracketing”?
where si is i’s endowment share.
• Probably not.
– Narrow bracketing surely plays a role. The “situation”
is: how to share manna from heaven.
– But only a strange distribution of α (in combination
with strange functions f1 and f2) can rationalize the
Oslo, October 2008 Tore Ellingsen 10
What model would fit? (iii)
• Assume narrow bracketing from now on. Could it
be fairness? Let
where si is i’s share.
• Explains why few give more than half.
• [Can also explain Bardsley (EE, 2008) and List
• But why give exactly 0.5? If s*=0.5, we’d expect
sj<0.5 is f2 is smooth. (More on this later.)
• And what about the troughs just above 0 and
just below 0.5? (More on this later.)
Oslo, October 2008 Tore Ellingsen 11
• Anonymity evidence challenges the fairness
theory but is sparse/weak:
– Hoffman et al (AER, 1996) find that double-blind
design reduces contributions somewhat (although not
statistically sig.; n≈40/treatment); an experimenter
– Koch and Norman (SEJ, 2008) likewise find that
triple-blind design reduces contributions further
(again, not stat. sig.; n=36/treatment); a recipient
Oslo, October 2008 Tore Ellingsen 12
Counterevidence (ii): Exit
• Exit evidence: Dana et al. (OBHDP, 2006).
– $10 Dicator Game,
– Unexpected $9 exit option – (exit implies
receiver unawareness about game).
• Case 1: Standard. Receiver aware unless exit.
• Case 2: Private. Receiver always unaware.
– Models considered above predict: No exit.
– Standard game: 33% exit. (n=61)
– Private game: 4% exit. (n=24)
Oslo, October 2008 Tore Ellingsen 13
Another nail in the coffin
• Broberg, Ellingsen and Johannesson (EL, 2007) go further and elicit
exit reservation prices.
• 119 subjects. Double-blind. Letters format. BDM-elicitation
Oslo, October 2008 Tore Ellingsen 14
• 64% of subjects violate purely outcome-
based preferences (exit res. price < 100).
• Relationship between donation and exit
reservation price is negative. If motivated
solely by altruism or fairness the
relationship should be positive.
Oslo, October 2008 Tore Ellingsen 15
How can we explain exits?
• Exit evidence suggests that we must “somehow
incorporate beliefs into the utility function.”
• Two approaches:
– Caring about the recipient’s expectations (ex ante
beliefs about action).
– Caring about the recipient’s judgment (ex post beliefs
• Dana et al. formally discuss expectations, but
informally also discusses judgment.
Oslo, October 2008 Tore Ellingsen 16
Expectations: Guilt aversion
• For example, let
where sE is the share i thinks that j expects.
– Literature: Geanakoplos, Pearce, Stacchetti (GEB,
1989), Charness and Dufwenberg (Ema, 2006), and
Battigalli and Dufwenberg (JET,forthcoming).
• Note: If the reason for guilt is concern for
recipient’s disappointment, this is an other-
regarding preference (“let-down aversion”).
Oslo, October 2008 Tore Ellingsen 17
Evidence for guilt aversion
• Dictator donations correlate positively with
– Dufwenberg and Gneezy (GEB, 2000)
• Back-transfers in trust games are
positively correlated with reported sE.
– Charness and Dufwenberg (Ema, 2006)
– Bacharach, Guerra, Zizzo, (T&D, 2007)
• Alternative interpretation: (False)
Oslo, October 2008 Tore Ellingsen 18
Guilt aversion or false consensus?
• Ellingsen et al (2008)
elicit recipient beliefs
them to dictators. 80
• 84 pairs. 60
• The figure shows a
bubble-plot of the 20
relationship between 0
the recipient’s belief -20
-20 0 20 40 60 80 100 120 140
and the dictator’s Belief (SEK)
Oslo, October 2008 Tore Ellingsen 19
Judgment: Shame and pride
“Nature, when she formed man for society,
endowed him with an original desire to please,
and an original aversion to offend his brethren.
She taught him to feel pleasure in their
favourable, and pain in their unfavourable
regard. She rendered their approbation most
flattering and most agreeable to him for its own
sake; and their disapprobation most mortifying
and most offensive.”
Adam Smith (1790, Pt III, Sec I, Para 13)
Oslo, October 2008 Tore Ellingsen 20
A social esteem model
• For example, let
where φB is i’s belief about j’s belief about φi;
see Andreoni and Bernheim (wp,2008).
– Related literature:
• Glazer and Konrad (AER, 1996), Prendergast and Stole
(EER, 2001), Bénabou and Tirole (AER, 2006), Ellingsen and
Johannesson (2007; AER, 2008), Tadelis (2008).
• Assume continuous distribution of φi.
• Compute perfect Bayesian equilibria and apply
D1, a signaling game refinement.
Oslo, October 2008 Tore Ellingsen 21
• The model can explain…
– Absence of si>0.5 (as before – yet not trivial)
– Prevalence of si=0.5 (new)
– Troughs just above 0 and just below 0.5 (new)
– Anonymity evidence (new)
– Exit evidence (new)
• …i.e., all puzzles so far!
• And it has additional implications.
Oslo, October 2008 Tore Ellingsen 22
• A USD 20 non-anonymous Dictator Game.
• Suppose that with probability p the donation is x
(small) regardless of dictator’s choice; this in
known to both.
• Pure fairness larger donations.
• Social esteem concern donate x.
• Two “conditions” (x=0, x=1). Each dictator sees
one, and makes choices for p=0, p=1/4, p=1/2,
p=3/4. (n=30/treatment and role).
Oslo, October 2008 Tore Ellingsen 23
Oslo, October 2008 Tore Ellingsen 24
• Richer psychological model may sometimes be
– Cues matter
• Haley and Fessler (EHB, 2005)
– Communication matters
• Prior argumentation: Mohlin and Johannesson (JEBO, 2008).
• Anticipated feedback: Ellingsen and Johannesson (EHB,
– People shield themselves from information
• Dana, Weber, Kuang (ET, 2007).
Oslo, October 2008 Tore Ellingsen 25
Fairness is conditional
• Dictator role allocation procedure matters.
– E.g., Hoffman et al (GEB, 1994)
• Prior recipient behavior matters.
– E.g., Falk and Kosfeld (AER, 2006).
• Thus, both φ and f3(φ) could depend on
Oslo, October 2008 Tore Ellingsen 26
• Cues that one is being watched induce larger
– Haley and Fessler (EHB, 2005)
– Bateson and Roberts (BL, 2006)
– Burnham (HN, 2007)
• Cues of norms matter
– Framing (labels, language) matters. E.g., Hoffman et
al (AER, 1996)
– Guessing, watching; Krupka and Weber (wp, 2007)
– Facing opponent guesses; Ellingsen et al (wp, 2008)
Oslo, October 2008 Tore Ellingsen 27
• Loewenstein and Small (RGP, 2007)
suggest that social behavior is driven by
interplay of instincts and cognition.
• Neuroscientific evidence:
– Personal moral dilemmas (trolley problems)
activate other brain regions than impersonal
dilemmas; Greene et al (Science, 2001)
– Lesions to VMPFC (“social emotions center”)
induce utilitarian decisions; Koenigs et al
Oslo, October 2008 Tore Ellingsen 28
Dual process and giving
• Little work, but pilots suggest that
cognitive load releases selfish instincts in
• In a (mini-) Dictator Game, dictators
become more selfish when outcome for
recipient is uncertain (and refuse to learn
– Dana, Weber, Kuang (ET, 2007)
• Alt B: 5,5 Alt A: 6,1 Alt A’: 6,5
Oslo, October 2008 Tore Ellingsen 29
• When recipient can make moral
arguments in writing before dictator
decides, donations go up by 70%+; Mohlin
and Johannesson (JEBO, 2008).
• When recipient can send a written
message after the donation, donations go
up by 40%+; Ellingsen and Johannesson
Oslo, October 2008 Tore Ellingsen 30
Ellingsen/Johannesson, Evolution and Human Behavior 2008
25 No Feedback
0 20 40 60 >60
Oslo, October 2008 Tore Ellingsen 31