Shane Bonetti
Department of Economics
University of St Andrews
St Andrews, Scotland
December 1996
Revised May 1997
Abstract
Several leading experimental economists have independently
proposed that deception should be proscribed on methodological
grounds as an experimental technique. The basis for this
prescription is the assertion that the psychological reaction to
suspected manipulation jeopardises experimental control and
validity, and contaminates the subject pool. According to this view,
honesty is a methodological public good and deception is
equivalent to not contributing. This paper reviews the literature on
the consequences of the use of deception. It is concluded that there
is little evidence to support the argument that deception should be
proscribed. It is argued that there are potential gains from
deception in data validity and experimental control. These gains
are illustrated by examining ultimatum games and public goods
experiments.
JEL Classification: C9
§ I am grateful to two anonymous referees for their comments, and to Alan Lewis, Friedel Bolle
and Peter Lunt for their kind help.
The Orthodoxy on Deception
Several eminent experimental economists have independently proposed a simple
rule to govern the conduct of experiments. Douglas Davis and Charles Holt
For want of a better term, I refer to the proponents of this ban on deception as
The Honest Johns, by adoption of the modal given name of the four experimental
prescriptions. The next section compares the views of The Prohibitionists with
inaccurate. First, to the limited extent that the concern of The Prohibitionists is
with the reputation of experimental economists per se rather than with the effect
Gottlieb, 1981; Smith & Richardson, 1985a, 1985b). Second, it is not the case that
frequency by psychologists (Dunston & Ross, 1986; Gross & Fleming, 1982;
Levenson et al., 1976; Sieber et al., 1995). Of course, that psychologists show no
would eschew deception. They have not, therefore it does not! This is
deception is a public bad. This implies that there may be private benefits from
deception but that experimenters acting egoistically will ignore the costs
Evidence
It is reasonably clear that The Prohibitionists oppose deception on the ground
that it causes the data produced to be invalid. That is, the argument against
Psychologists have been experimenting with human subjects for more than a
scrutiny for at least the past four decades. Important early contributors include
Vinacke (1954), Rosenthal (1963) and Kelman (1967). It seems likely that
and evidence.
There is a school of thought which argues that this is not the case. Davis &
the use of deception. Davis & Holt are unable to offer any argument in support
sound. That is, there are differences between the methodology of valid
Whatever the merits of this reverse analogy on purely logical grounds, this
paper proceeds from the assumption that the psychological responses of human
subjects generally do not change when the subjects are moved from a psychology
laboratory to an economics laboratory. That is, the starting point for this paper is
be of interest to economists.
The Prohibitionist view has two key elements: the direct effect of deception
The essence of The Prohibitionist direct argument is that deceiving subjects taints
likely to expect future deception (Beins, 1993; Christensen, 1977; Krupat &
Garonzik, 1994). However, the mere anticipation of deception does not of itself
cause, or establish the existence of, an alteration of behaviour on the basis of that
experimental subjects in the room suspected that the seizure was a confabulation,
More than anecdote is required to test the effect of deception. Fortunately, the
proposition that anticipated deception alters behaviour has been tested in a wide
range of experimental settings. The most important work has been performed in
three fields: (i) obedience and conformity; (ii) attribution; and (iii) social
dilemmas. The evidence varies according to the experimental setting, but tends
toward the conclusion that anticipated deception does not usually alter
measured behavior.
(i) Obedience experiments test the extent to which individuals obey the
Conformity experiments test the effect of group pressure upon the modification
and distortion of individual opinions, attitudes, perceptions and actions. The
classic studies are those of Asch (1951, 1955). The influence of suspicion on
Asch's line-judgement task. He found that informing subjects that they might be
deceived did not influence conformity. Chipman (1966) drew the same
conclusion.
However, there are two apparently conflicting results. Stang (1976) had 65
found less conformity by suspicious subjects, but suggested that this was a
a set of stimulus items and a set of response items. Subjects who had been
conditioning.
precisely the same conclusion. After a survey of conformity and attitude change
research he concludes that "the experimental results seem to defy description by
and characteristics to another person. Wiener & Erker (1986) conducted a 2x2
variable is varied over all possible values of another independent variable. They
divided their sample of 64 undergraduates into two groups. Half of the sample
were informed that they might be misinformed. The two experimental conditions
(iii) Social dilemma experiments are more familiar to economists. Allen (1983)
had a 2x2 factorial design with two levels of deception (deception present vs
deception absent) and two levels of forewarning (presence of deception
experimental condition. With regard to the deception factor, it was found that
reasonable concern". A decade later, West & Gunn (1978) drew the same
conclusion:
"Even a casual reading of the research literature fails to confirm the
... expectation that the proportion of subjects classified as
suspicious of experimental procedures has evidenced a significant
increase in recent years".
Key support for this view comes from Stricker's (1967) review of 16 studies
which report the existence and extent of suspicion of deception among subjects
four percent, and the range was from zero to 23 percent. Similar results are
behave little different from "naive" subjects (Fillenbaum, 1966). Even this small
proportion of undeceived subjects may overstate any "tainting" effect. Brock &
Becker (1966) found that behaviour in a subsequent experiment was affected only
for subjects who had performed similar tasks in an earlier deceptive experiment
(Brock & Becker, 1966; Gruder et al., 1977; Mills, 1976; Stricker et al., 1969; Walsh,
1976).
Why Deceive?
This brief review of the available evidence reveals that the concerns which The
exaggerated. Data supportive of The Prohibitionist view are rare, and the balance
of the evidence is certainly to the contrary. Deceiving experimental subjects
taint the behaviour of experimental subjects. However, this does not establish the
need for deception, or the gains which deception might bring. One way to
The experimental work of Binmore, Shaked & Sutton (1985) is a good starting
place. They examine a two-stage ultimatum game. Their hypotheses are that
human agents' utility functions do not include fairness or the welfare of others,
and that human players choose Nash equilibria rather than focal points. There is
is to be found in a less than obvious place. Binmore and his colleagues do engage
the game until the first stage has been completed. However, more illuminating
for our purposes is the instructions which Binmore et al. offer their subjects: :
"How do we want you to play? YOU WILL BE DOING US A
FAVOUR IF YOU SIMPLY SET OUT TO MAXIMIZE YOUR
WINNINGS". (The emphasis and capitals appear in the original)
This is, of course, quite extraordinary advice to offer subjects, as Thaler (1988)
has remarked. What exactly is wrong with the Binmore et al. experimental
design, and with this advice to their subjects? There is powerful evidence that
accordance with them. This means that useful experimentation on responses over
which subjects can exercise voluntary control cannot be done if the investigator
tells the subject the theoretical hypothesis beforehand. It means in addition that if
you tell subjects how you would like them to behave, then on the whole that is
how they can be expected to behave (Crano & Brewer, 1973; Flay & Hamid, 1977;
humans respond to threats in bargaining, whether they free ride, care about
fairness, take sunk costs into account or commit the gambler's fallacy, there is a
general rule: beware of telling subjects how they should behave or how they are
usually invalidate the reliability of experimental results (Turner, 1981; Roese &
Jamieson, 1993), unless the experimental hypothesis involves and requires a test
of the consequence of such instructions. That is, such cues are usually only
The inference might be drawn that this canon of experimental design implies
only that: (i) subjects should not usually be told how to behave; (ii) subjects
should not usually be told the experimental hypothesis, for fear that they will
will wonder for themselves what it is the experimenter is after, what hypothesis
the experimenter is testing. From the subject's point of view, the "typically
the study's real nature" (Stricker et al., 1969), or more prosaically "to penetrate
the experimenter's inscrutability" (Orne, 1969, citing Riecken, 1962, 31). There is
therefore an argument for filling this gap in the subjects' minds by announcing a
purpose of the experiment which is not the true purpose. This "deception ... is an
the subject which would interfere with his research" (Orne & Holland, 1968), or
"to keep ... subjects naive about the purpose of the experiment so that they can
respond ... spontaneously" (Kelman, 1967). If the deception is plausible, then
there is little danger that the experimental subjects will figure out the true
The typical experimental subject will always attend to cues, information and
the attention of the subjects can be effectively distracted, thus ensuring that the
behavior which is measured is more natural and spontaneous, and less affected
and contrived.
The analysis of free riding and public goods problems is one of the most difficult
public goods (Andreoni, 1988; Bohm, 1972; Isaac et al., 1985; Isaac & Walker,
1988; Isaac et al., 1994; Ledyard, 1995; Marwell & Ames, 1982; Molander, 1992;
Schneider & Pommerhene, 1981; Weimann, 1994). The standard public goods
repeated game in which they are endowed with tokens which they must invest in
either private or public investment. The private rate of return (by convention set
to unity) exceeds the public rate of return (called the marginal per capita return
or MPCR). However, public investment yields a benefit for all subjects so that the
(1) N i= Ei – xi + MPCR Σ xj
j=1
where:
N = payoff
Ei = endowment
xi = public investment by subject i
MPCR = marginal per capita return
Σxj = sum of individual public investments
N = number of subjects
This equation means that each subject's payoff is equal to his or her endowment
(Ei), less the amount contributed to the public good by that subject (xi), plus the
return to the subject from the public good investment of all subjects (MPCR
Σ xj). In general, there is a public good problem if:
(2) (1/N) < MPCR < 1
where N is the group size and MPCR is the marginal per capita return. If (2)
holds, the Nash equilibrium is xi = 0 ∀ i. There are many ways in which
deceptive experimental design can cast light upon the puzzles which public
goods experiments reveal. Indeed, deception is not unknown in the public goods
and free rider literature (E.g. Bohm, 1972; Schneider & Pommerehne, 1981;
Andreoni, 1988, page 295, esp. note 7; Weimann, 1994). Four examples illustrate
choices may depend on the actions of other players. It may be that any given
player is more likely to be cooperative if the other players are very cooperative,
and more likely to free ride if the other players are very selfish. However, as
Weimann (1994, page 187) remarks, in standard public goods experiments this
others do not cooperate, subjects react in a very natural way: because they do not
like to be exploited they also stop cooperation" (Weimann, 1994, page 198). The
discovery of this evidence supporting what Weimann calls "exploitation
increases with group size. That is, free riding will be more severe the larger is the
social group (Sandler, 1992). The laboratory and field evidence usually
demonstrates no such group size effect (Isaac & Walker, 1988; Isaac et al., 1994;
Lipford, 1995; Olson & Caddell, 1994). Clearly, it is not the actual size of the
group which is the potentially important variable, but the perceived size of the
group. Far greater experimental data can be derived for any given experimental
cost by using deceptive procedures in which subjects are told the size of group of
which they are a member, and receive fictitious information regarding the
contributions of the other members of the group. In particular, very large group
experiments are possible using this technique when they would be prohibitively
expensive if deception were prohibited. Isaac et al. (1994) attempt to solve this
cost problem by using course-related credit points as the payoff in their
experiment. A major problem with this method is that it necessarily limits the
hypothesis testing in at least two ways. First, a wider range of the group sizes
can be studied for any given number of experimental subjects. Second, the
behaviour. Satisficing means that players are "happy enough" to end up with
more than they started with, which requires a minimum profitable coalition i.e.
MPCR x N* >1 where N* is the expected number of players in the coalition. If the
probability that any randomly chosen person will cooperate is α then the
profitable coalition is MPCR . N* > 1 which implies MPCR . αN > 1. For given α
and MPCR, the larger is N the greater is the chance that this condition will be
satisfied.
Ledyard (1995) and Andreoni (1989) hypothesise that investing in the public
good may in part be a consequence of a "warm inner glow", that is that such
payoffs. It is usual in public good experiments to avoid the use of the term
"public", possibly because of the fear that economics student subjects will make
the association "public good" and therefore infer that free riding is the
contribution mechanism adopt the terms "group" and "private" to denote the
available investments. However, these terms too may carry implications which
testing of the hypothesis that "warm inner glow" cooperation is the consequence
the payoffs. For instance, behaviour could be compared using the standard
"group/private" terminology" and some neutral terminology like "A/B". These
letters of the alphabet are probably not the best choices and surely some test of
an ordering hypothesis would be required. Letters from later in the alphabet and
more neutral from a student subject point of view would probably be preferable.
(iv) Pulsing
repetition. However, the decay is not monotonic (E.g. Andreoni, 1988, Tables 1
and 3, pages 296, 299; Isaac & Walker, 1988, Figures II and IV, pages 190, 194;
Isaac et al., 1994, Figures 2-4, pages 11-12). There appear sporadic attempts by
(Isaac et al, 1985, page 65). Such "pulsing behaviour could be interpreted as an
(Isaac et al., 1985, 66). It is therefore necessary, if the causes and consequences of
pulsing are to be understood, to "control" (that is, simulate) the behaviour of N–1
experiments.
or "contaminate a subject pool". It does not mean that "we have lost control". Nor
does it "taint" experiments or cause the data they produce to be invalid. Indeed,
there is good reason to think that the selective use of deception can enhance
and dangerous (Aitkenhead & Dordoy, 1983; Oetting, 1978; Rosenthal, 1976;
Swingle, 1968, page 31; Trice, 1986; West & Gunn, 1978). Thus Kelman (1967), one
of the more widely cited sceptics regarding the use of experimental deception,
admits that:
"There are good reasons for using deception in many experiments.
There are many significant problems that probably cannot be
investigated without the use of deception. ... I have not forsworn
the use of deception under any and all circumstances".
Two exceptions to this general consensus may be mentioned.1 First, there are
because it may cause serious harm to subjects, or because subjects cannot give a
Brandt, 1978; Flanagan, 1973; Jung, 1971; Kroger & Wood, 1986; Mixon, 1977a,
1977b; Noble, 1983; Oliansky, 1991; Rubin, 1973, 1985; Shipley, 1977; Zimbardo,
(Eisner, 1977; Forward et al., 1976; Greenwood, 1983; Hendrick, 1977; Krupat,
1977; Mitchell et al., 1977; Petranek, 1985, 1994; cf. Meeus & Raaijmakers, 1985).
There has been comparatively little interest in role playing experiments among
economists, by comparison to their colleagues in management and international
1 In addition to these two, there are the fundamentalist Christian non-empiricist anti-
psychologists (E.g. Foster & Ledbetter, 1987) who assert that the deception inherent in
psychological research is sinful, and that in any case knowledge gained by empirical methods is
invalid because it does not derive from the bible. These would be strange bedfellows indeed for
economists.
relations departments. However, there is a similarity between the expressed
distracting the subjects from the fiction of the experience, subjects will behave in
a way which is natural. This view echoes that of many experimental economists
who believe that if only we could persuade subjects to attend closely to their
other words, there seems to be a view among some experimental economists that
a "steep objective function" is the key feature in the design of an experiment. The
discussion of ultimatum games and data validity above should make clear that
the typical experimental subject will always attend to cues, information and
hunches about features of the design other than the payoffs. Competition or the
deception. Indeed, that some deceptive experimental results have been published
in the economics literature would seem to imply that editors are sensibly
ignoring The Prohibitionist view, or that it has not yet fully taken hold.
Rather than providing glib answers to difficult methodological and design
follow the Roth Rule: "It is important, I think, to avoid establishing rigid