Frames and Games

Jordi Brandts1 and Christiane Schwieren2

1
Institut d’Anàlisi Econòmica (CSIC)
2
Universitat Pompeu Fabra

1 Introduction
Economists are proud of being able to point to situations in which the fun-
damental forces identified by economic analysis work inexorably. One such
case is the effect that supply reductions have on prices in the international
oil market. One can safely predict that prices will increase and that any at-
tempts by politicians and journalists to prevent the increase by presenting the
situation in a particular light will not work. It is known, however, that things
are not always that simple. Numerous studies have shown that behaviour of-
ten depends on the way in which logically equivalent choice situations and
strategically equivalent situations are described or presented to people. Such
so-called framing effects have been identified in a number of different contexts.
Kuhberger [8] surveys some of the relevant literature.
The existence of framing effects poses important challenges to the scientific
analysis of society. After all, social science is based on the idea that human
behaviour can be captured and understood by simplified representations of
things. If framing effects are pervasive, if every change in the circumstances
surrounding social situations affects people’s decisions substantially, the anal-
ysis of humans’ social behaviour will be an extremely difficult task. At the
same time the question arises why people are affected by the framing of situ-
ations.
In this paper we study experimentally some of the possible limits of framing
effects. We do that in the context of different representations of dilemma
games. Andreoni [1] compares behaviour in a public good (PG) and in a certain
public bad (PB) game. He finds that subjects are more willing to cooperate
in the PG case, even though the two situations are strategically equivalent.
This result invites further investigation, because it is not directly consistent
with some of the recent models of social preferences like those of Fehr and
Schmidt [6], Bolton and Ockenfels [3] and Charness and Rabin [5]. The result
also differs from most other results on framing effects in this kind of games.
First, the strength of the effect is surprising. Two meta-analyses (Levin et al.
176 Jordi Brandts and Christiane Schwieren

[9]; Kuhberger [8]) recently showed that PG/PB frames are not as effective
in producing “framing-effects” as is the classical Asian disease situation of
Tversky & Kahneman [11]. Both meta-analyses see the reason for this in the
specific structure of “game-theoretic” (Kuhberger) or “goal-framing” (Levin)
situations. Both choices are risky, and it is not clear, which one is riskier.
Furthermore, Andreoni’s effect goes in the opposite direction from what has
been found in most studies of PG/PB frames. Usually, the negative (loss, PB)
frame has been found to have a stronger impact on responses than the positive
(gain, PG) frame. For example, in a classical study by Brewer & Kramer [4],
subjects left more of the common resource in the commons frame than in the
PG frame.
Our experiments are motivated by two specific hypotheses about the limits
of framing effects. The first is that there is a kind of — in a loose sense —
continuity in the relation between frames and behaviour. If this were the case,
then very small changes in the way a situation is presented would have minor
effects and only larger differences would lead to larger differences in behaviour.
It would mean that not any small change would matter. The second hypothesis
is that variations in the parameters that govern monetary incentives lead to
similar changes — in direction and magnitude — under different framing
conditions. This hypothesis is motivated by the general notion that perhaps
models should not be expected to accurately predict levels of behaviour but
should have the ability of accommodating observed comparative statics in
the sense of the shifts in behaviour in response to parameter changes. In our
experiments small and large changes in the presentation correspond to what
– intuitively – are minor vs. more extensive changes in the wording used in
the experimental instructions. With respect to the variations in the relevant
parameters we follow the approach of Goeree, Holt and Laury [7], who have
subjects make decisions for different PG situations, which vary in several
dimensions.

2 Basic Experimental Design

A PG game with ten different parameter-combinations and three different
frames is the basis for this research. The parameters used were taken from
Goeree et al. (see Table 1). Their study changed parameters such that the
“external” and “internal” return of contributions differed between the situa-
tions analyzed. The external return is defined as the return the investment has
for others in the group, whereas the internal return is the value of the invest-
ment for oneself. In our experiment, parameters were varied within subjects,
i.e., each subject had to take ten contribution-decisions.
Frames were varied between subjects. First, we compared two frames with
a subtle linguistic difference, as had been used before successfully by Brewer
et al. [4]. One group of participants played the game with a simple PG frame,
the same used by Goeree et al., describing a situation where money could be
 Frames and Games 177

Table 1. Parameter structure of the ten decisions

Decision 1 2 3 4 5 6 7 8 9 10
Group size 4 2 4 4 2 4 2 2 4 2
Internal return 4 4 4 2 4 4 2 4 2 4
External return 2 4 6 2 6 4 6 2 6 12

“invested” in the public account or “kept” in the private account. In the PB

frame subjects had to make a choice between “keeping” money in the public
account or “investing” in the private account. Thus, in the PG situations,
subjects were asked to do something good, whereas in the PB situations sub-
jects could avoid doing something “bad”. Our third frame differed in more
aspects from the first two frames, but kept payoffs constant. It was designed
following the PB frame Andreoni has used. Now the difference to a PG is not
just linguistic. The frame describes the situa-tion such that when investing
in the private account, some money is taken from each player in the group,
while investing in the public account doesn’t affect others.
144 students at Universitat Autonoma de Barcelona, from various faculties,
participated voluntarily, for performance-based payment in an experiment on
decision-making. In one session participated between 12 and 20 subjects. In
each session, subjects took ten different decisions, but were exposed to only
one frame. The order of the decisions was kept constant across participants.
At the end of the experiment, one situation of the ten was randomly chosen
and participants were paid according to their decision taken in this situation.
In addition, they received a show-up fee of =
C 3.

3 Results
The main result of our experiment is that differences between frames are
less important than the parameters of the decision task for determining con-
tributions to a common good. However, one frame did have a stronger ef-
fect on contributions. Furthermore, the effect of the frames differed between
parameter-constellations. The first figure shows the average contributions for
each frame and each decision. To allow for a comparison with Goeree et al.’s
original results in the same situations, we added their data to our figures.
The figure shows that there are no big differences between the frames in most
decisions, but that there are differences for all frames between the parameter-
constellations. It also shows that the frames do not always influence contri-
butions in the same direction.
With respect to the percentage of the endowment contributed in each
frame, over all decisions, there is no significant difference for our three frames.
While Goeree et al.’s result is very similar, Andreoni’s data differ largerly from
ours. In a later step we discuss whether this might be caused by the specific pa-
rameters he is using. But, first we take another look at contributions, but now
178 Jordi Brandts and Christiane Schwieren

18
16

mean contribution
14
g
12
10 a
8 pg
6
pb
4
2
0
1 2 3 4 5 6 7 8 9 10
decision

Fig. 1. Average token contributions per frame and decision (g = Goeree’s data, a
= our “andreoni”-frame).

looking only at the percentage of participants contributing zero to the PG.

Doing this, differences between the frames become more pronounced. Specifi-
cally, our “Andreoni” frame differs significantly from our other two frames. It
is nearly always highest with respect to the percentage of subjects contributing
zero, and never lowest. However, in Andreoni’s original study, the percentage
of people contributing zero was even higher than in our version of his frame. A
pos-sible explanation for this is that the parameters Andreoni uses are closest
to our situation four with respect to internal and external return and num-ber
of people in the social dilemma — and our situation four was exactly the one
with the highest percentage of people contributing zero for all our frames. If
we look additionally at the percentage of subjects contributing their full en-
dowment in each decision and each frame, it becomes obvious that our frame
“Andreoni” also has a very large number of subjects con-tributing fully. As
it also has the largest percentage of subjects contributing zero in nearly all
situations, average contribution in this frame looks very similar to the other
frames, but this average stems from different behavior than in the other two
frames.
Finally, one can look at whether there is a difference between treatments
with respect to the percentage of subjects contributing more to the public
account than to their private account. Differences exist between decision sit-
uations, but only for few situations between frames, and again, the “pb” and
“pg” frame are very similar, whereas the “Andreoni” frame differs sometimes
from the other two.
A statistical analysis (U-tests) shows no significant difference between the
frames with respect to overall contributions (over all decisions). When looking
at the percentage of full and zero contributions, differences between frames
become significant, always the “Andreoni”-frame being the one that differs
from the other two frames (significance levels always < .005). The linguis-
tically different PG and PB frame differ with respect to the percentage of
full contributions (p < .005). More people contribute fully in the PG than in
 Frames and Games 179

the PB frame. With respect to zero contributions, both frames are equal. An
ANOVA using “parameter structure” as independent factor and contribution
as the dependent variable reveals that for each frame the type of decision is
an important predictor of the amount contributed (significance level for PB
and PB < .01 and for Andreoni and Overall < .001). Decisions are made in-
dependent for this analysis by subtracting contributions in each decision from
contributions in decision one.

4 Discussion
Overall, we find only a weak effect of frames, which is in line with our hy-
potheses and with results from the meta-analyses cited above. We hardly find
significant differences between the two only linguistically different frames, but
we do find significant differences between the “more different” frame (“an-
dreoni”) and the two basic PG and PB frames. This confirms our hypothesis
that stronger framing manipulations have stronger effects.
Our results further indicate that the direction of the framing effects seems
to be influenced by the parameters chosen. This explains partially why An-
dreoni [1] contrary to most other studies on framing in PG/PB games finds
lower contributions in his PB frame than in his PG frame. The parameter-
constellation of our experiment which ist most similar to Andreoni’s parame-
ters results in the highest percentage of zero-contributions of all our constel-
lations, and is close to the percentage of zero-contributions Andreoni reports.
Another important aspect about the “andreoni” frame used in our experi-
ment is that it leads to the largest variance in results. This frame has both
the highest percentage of zero-contributions and the highest percentage of full
contributions in most decision-situations. Unfortunately, Andreoni does not
report on percentage of full contributions in his experiment.
Boettcher [2]concludes from his review of the existing literature on framing
that “Relatively minor differences in experimental design appear to exagger-
ate or minimize the impact of prospect framing” (p. 355). Andreoni’s study
compared to our experiments is a nice example - if the “right” parameters are
chosen, larger framing effects can be found, whereas the “wrong” parameters
lead to no or very small framing effects. Furthermore, the effect of frames can
go in both directions as the results with our “andreoni”-frame show: It leads
both to more zero and more full contributions.
Our results are far from being conclusive. They confirm, that “goal frames”
are more complicated than simple Asian-disease problems (Levin et al., [9]),
because more than one aspect of the message can be manipulated, and because
it is not obvious which option is the riskier one. Furthermore, there is room
for differences in emotional intensity induced by different terminologies used.
Andreoni’s frame might enhance emotional intensity as opposed to the two
other frames we use, as taking money from someone probably is emotionally
more involving than just making different contributions.
180 Jordi Brandts and Christiane Schwieren

One important problem of our design that could provide an alternative

explanation for the results found is that each subject went through all ten
decisions. This might have some demand-characteristics, inducing subjects to
think carefully about the decision. Research by McElroy and Seta [10] has
shown that subjects are far less susceptible to framing manipulations when
they are prone to or asked to think analytically about their decision.
Our review of the literature showed that there exist some first attempts
to characterize situations in which framing effects occur in PG/PB frames.
However, there is no conclusive evidence yet and systematic research is lacking
on what aspects of the frame and the parameters determine whether framing
has an effect on subjects or not in this kind of situations. Our research wants
to be a first step in this direction. Apart from providing some tentative results
showing what factors might influence the effectiveness of framing manipula-
tions, if underlines the necessity of further, more systematic research in this
direction.

References
1. Andreoni, J (1995) Warm-Glow Versus Cold-Prickle: The Effects of Positive
and Negative Framing on Cooperation in Experiments. The Quarterly Journal
of Economics 110:1-21
2. Boettcher III WA (2004) The Prospects for Prospect Theory: An Empirical
Evaluation of International Relations Applications of Framing and Loss Aver-
sion. Political Psychology 25:331-362
3. Bolton GE, Ockenfels A (2000) ERC: A Theory of Equity, Reciprocity, and
Competition. American Economic Review 90:166-193
4. Brewer MB, Kramer RM (1986) Choice behavior in social dilemmes: Effects
of social identity, group size, and decision framing. Journal of Personality and
Social Psychology 50:543-549
5. Charness G, Rabin M (2003) Understanding Social Preferences with Simple
Tests. General Economics and Teaching 0303002, Economics Working Paper
Archive at WUSTL.
6. Fehr E, Schmidt K (1999) A Theory of Fairness, Competition and Cooperation.
Quarterly Journal of Economics 114:817-868
7. Goeree JK, Holt C, Laury S (2002) Private Costs and Public Benefits: Unrav-
eling the Effects of Altruism and Noisy Behavior. Journal of Public Economics
83: 257-278
8. Kuehberger A (1998) The Influence of Framing on Risky Decisions: A Meta-
analysis. Organizational Behavior and Human Decision Processes 75:23-55
9. Levin IP, Schneider S, Gaeth GJ (1998) All Frames Are Not Created Equal:
A Typology and Critical Analysis of Framing Effects. Organizational Behavior
and Human Decision Processes 76:149-188
10. McElroy T, Seta JJ (2003) Framing effects: An analytic-holistic perspective.
Journal of Experimental Social Psychology 39:610-617