ArticlePDF Available

Status Quo Bias in Decision-Making

Authors:

Abstract and Figures

Most real decisions, unlike those of economics texts, have a status quo alternative--that is, doing nothing or maintaining one's current or previous decision. A series of decision-making experiments shows that individuals disproportionately stick with the status quo. Data on the selections of health plans and retirement programs by faculty members reveal that the status quo bias is substantial in important real decisions. Economics, psychology, and decision theory provide possible explanations for this bias. Applications are discussed ranging from marketing techniques, to industrial organization, to the advance of science. Copyright 1988 by Kluwer Academic Publishers
Content may be subject to copyright.
Journal of Risk and Uncertainty, 1: 7-59 (1988)
0 1988 Kluwer Academic Publishers, Boston
Status Quo Bias in Decision Making
WILLIAM SAMUELSON
Boston University
RICHARD ZECKHAUSER
Harvard University
Key words: decision making, experimental economics, status quo bias, choice model, behavioral
economics, rationality
Abstract
Most real decisions, unlike those of economics texts, have a status quo alternative-that is, doing noth-
ing or maintaining one’s current or previous decision. A series of decision-making experiments shows
that individuals disproportionately stick with the status quo. Data on the selections of health plans and
retirement programs by faculty members reveal that the status quo bias is substantial in important real
decisions. Economics, psychology, and decision theory provide possible explanations for this bias. Ap-
plications are discussed ranging from marketing techniques, to industrial organization, to the advance
of science.
“To do nothing is within the power
of
all men.”
Samuel Johnson
How do individuals make decisions? This question is of crucial interest to
researchers in economics, political science, psychology, sociology, history, and
law. Current economic thinking embraces the concept of rational choice as a pre-
scriptive and descriptive paradigm. That is, economists believe that economic
agents-individuals, managers, government regulators-should (and in large part
do) choose among alternatives in accordance with well-defined preferences.
In the canonical model of decision making under certainty, individuals select
one of a known set of alternative choices with certain outcomes. They are endowed
with preferences satisfying the basic choice axioms-that is, they have a transitive
ranking of these alternatives. Rational choice simply means that they select their
most preferred alternative in this ranking. If we know the decision maker’s rank-
ing, we can predict his or her choice infallibly. For instance, an individual’s choice
should not be affected by removing or adding an irrelevant (i.e., not top-ranked)
alternative. Conversely, when we observe his or her actual choice, we know it was
his or her top-ranked alternative.
8
WILLIAM SAMUELSON AND RICHARD ZECKHAUSER
The theory of rational decision making under uncertainty, first formalized by
Savage (19.54) requires the individual to assign probabilities to the possible out-
comes and to calibrate utilities to value these outcomes. The decision maker
selects the alternative that offers the highest expected utility. A critical feature of
this approach is that transitivity is preserved for the more general category, deci-
sion making under uncertainty. Most of the decisions discussed here involve what
Frank Knight referred to as
risk
(probabilities of the outcomes are well defined) or
uncertainty (only subjective probabilities can be assigned to outcomes). In a num-
ber of instances, the decision maker’s preferences are uncertain.
A fundamental property of the rational choice model, under certainty or uncer-
tainty, is that only preference-relevant features of the alternatives influence the in-
dividual’s decision. Thus, neither the order in which the alternatives are presented
nor any labels they carry should affect the individual’s choice. Of course, in real-
world decision problems the alternatives often come with influential labels. In-
deed, one alternative inevitably carries the label status quo-that is, doing nothing
or maintaining one’s current or previous decision is almost always a possibility.
Faced with new options, decision makers often stick with the status quo altema-
tive, for example, to follow customary company policy, to elect an incumbent to
still another term in office, to purchase the same product brands, or to stay in the
same job. Thus, with respect to the canonical model, a key question is whether the
framing of an alternative-whether it is in the status quo position or not-will
significantly affect the likelihood of its being chosen.’
This article reports the results of a series of decision-making experiments
designed to test for status quo effects. The main finding is that decision makers ex-
hibit a significant status quo bias. Subjects in our experiments adhered to status
quo choices more frequently than would be predicted by the canonical model.
The vehicle for the experiments was a questionnaire consisting of a series of
decision problems, each requiring a choice from among a fixed number of alter-
natives. While controlling for preferences and holding constant the set of choice
alternatives, the experimental design varied the framing of the alternatives. Under
neutralframing, a menu of potential alternatives with no specific labels attached
was presented; all options were on an equal footing, as in the usual depiction of the
canonical model. Under status quo framing, one of the choice alternatives was
placed in the status quo position and the others became alternatives to the status
quo. In some of the experiments, the status quo condition was manipulated by the
experimenters. In the remainder, which involved sequential decisions, the sub-
ject’s initial choice self-selected the status quo option for a subsequent choice.
In both parts of the experiment, status quo framing was found to have predict-
able and significant effects on subjects’ decision making. Individuals exhibited a
significant status quo bias across a range of decisions. The degree of bias varied
with the strength of the individual’s discernible preference and with the number of
alternatives in the choice set. The stronger was an individual’s preference for a se-
lected alternative, the weaker was the bias. The more options that were included in
the choice set, the stronger was the relative bias for the status quo.
STATUS QUO BIAS IN DECISION MAKING
9
To illustrate our findings, consider an election contest between two candidates
who would be expected to divide the vote evenly if neither were an incumbent (the
neutral setting). (This example should be regarded as a metaphor; we do not claim
that our experimental results actually explain election outcomes.‘) Now suppose
that one of these candidates is the incumbent office holder, a status generally ack-
nowledged as a significant advantage in an election. An extrapolation of our ex-
perimental results indicates that the incumbent office holder (the status quo alter-
native) would claim an election victory by a margin of 59% to 41%. Conversely, a
candidate who would command as few as 39% of the voters in the neutral setting
could still earn a narrow election victory as an incumbent. With multiple can-
didates in a plurality election, the status quo advantage is more dramatic. Con-
sider a race among four candidates, each of whom would win 25% of the vote in the
neutral setting. Here, the incumbent earns 38.5% of the vote, and each challenger
20.5%. In turn, an incumbent candidate who would earn as little as 9% of the vote
in a neutral election can still earn a 25.4% plurality.
The finding that individuals exhibit significant status quo bias in relatively sim-
ple hypothetical decision tasks challenges the presumption (held implicitly by
many economists) that the rational choice model provides a valid descriptive model
for all economic behavior. (In Section 3, we explore possible explanations for
status quo bias that are consistent with rational behavior.) In particular, this find-
ing challenges perfect optimizing models that claim (at least) allegorical signifi-
cance in explaining actual behavior in a complicated imperfect world. Even in
simple experimental settings, perfect models are violated.
In themselves, the experiments do not address the larger question of the impor-
tance of status quo bias in actual private and public decision making. Those who
are skeptical of economic experiments purporting to demonstrate deviations from
rationality contend that actual economic agents, with real resources at stake, will
make it their business to act rationally. For several reasons, however, we believe
that the skeptic’s argument applies only weakly to the status quo findings. First,
the status quo bias is not a mistake-like a calculation error or an error in maxi-
mizing-that once pointed out is easily recognized and corrected. This bias is con-
siderably more subtle. In the debriefing discussions following the experiments,
subjects expressed surprise at the existence of the bias. Most were readily per-
suaded of the aggregate pattern of behavior (and the reasons for it), but seemed un-
aware (and slightly skeptical) that they personaly would fall prey to this bias.
Furthermore, even if the bias is recognized, there appear to be no obvious ways to
avoid it beyond calling on the decision maker to weigh all options evenhandedly.
Second, we would argue that the controlled experiments’ hypothetical decision
tasks provide fewer reasons for the expression of status quo bias than do real-
world decisions. Many, if not most, subjects did not consciously perceive the dif-
ferences in framing across decision problems in the experiment. When they did
recognize the framing, they stated that it should not make much of a difference. By
contrast, one would expect the status quo characteristic to have a much greater im-
pact on actual decision making. Despite a desire to weigh all options evenhand-
10
WILLIAM SAMUELSON AND RICHARD ZECKHAUSER
edly, a decision maker in the real world may have a considerable commitment to,
or psychological investment in, the status quo option. The individual may retain
the status quo out of convenience, habit or inertia, policy (company or govern-
ment) or custom, because of fear or innate conservatism, or through simple
rationalization. His or her past choice may have become known to others and, un-
like the subject in a compressed-time laboratory setting, he or she may have lived
with the status quo choice for some time. Moreover, many real-world decisions are
made by a person acting as part of an organization or group, which may exert ad-
ditional pressures for status quo choices. Finally, in our experiments, an alterna-
tive to the status quo was always explicitly identified. In day-to-day decision mak-
ing, by contrast, a decision maker may not even recognize the potential for
a choice. When, as is often the case in the real world, the first decision is to recog-
nize that there is a decision, such a recognition may not occur, and the status
quo is then even more likely to prevail. In sum, many of the forces that would
encourage status quo choices in the real world are not reproduced in a laboratory
setting.3
Critics might complain, however, that our laboratory decisions were unrep-
resentative. To this charge we have no definitive answer. However, in Section 2, we
report on two field studies involving the actual choices of employees of Harvard
University in choosing health coverage and of faculty members nationwide on the
division between TIAA (bonds) and CREF (stocks) for their retirement in-
vestments. Both studies discovered significant status quo bias. We leave to future
research the task of identifying the characteristics of decisions that make a strong
status quo bias likely.
The range of explanations for the existence of status quo bias (Section 3 presents
an extensive discussion) suggests that this phenomenon will be far more pervasive
in actual decision making than the experimental results alone would suggest. The
status quo bias is best viewed as a deeply rooted decision-making practice stem-
ming partly from a mental illusion and partly from psychological inclination.
Some examples of status quo effects in practice should be instructive.
A
small
town in
Germany.
Some years ago, the West German government under-
took a strip-mining project that by law required the relocation of a small town
underlain by the lignite being mined. At its own expense, the government of-
fered to relocate the town in a similar valley nearby. Government specialists
suggested scores of town planning options, but the townspeople selected a plan
extraordinarily like the serpentine layout of the old town-a layout that had
evolved over centuries without (conscious) rhyme or reason.4
Decision making by habit.
For 26 years, a colleague of ours chose the same lunch
every working day: a ham and cheese sandwich on rye at a local diner. On
March 3, 1968 (a Thursday), he ordered a chicken salad sandwich on whole
wheat; since then he has eaten chicken salad for lunch every working day.
Brand allegiance.
In 1980, the Schlitz Brewing Company launched a series of live
beer taste tests on network television (during half times of National Football
STATUS QUO BIAS IN DECISION MAKING
11
League games) in an effort to regain its reputation as a premium beer. (It had
fallen from second to fourth place in market share.) A panel of 100 confirmed
Budweiser drinkers (each had signed an affidavit that he drank at least two six-
packs of Bud a week) were served Budweiser and Schlitz in unmarked con-
tainers and asked which they preferred. Schlitz’s advertising gamble paid off.
On live television, between 45 percent and 55 percent of confirmed Budweiser
drinkers said they preferred Schlitz. Similar results were obtained when con-
firmed Miller drinkers participated in the test.5
The decisions made in these examples display a strong affinity for the status
quo. Offered a score of plans, citizens duplicated the layout of their town. The
lunchtime diner’s relationship with his chosen sandwich has outlasted several
marriages. Taste notwithstanding, beer drinkers are loyal to their chosen brands.
In each case, status quo bias appears to be operating. The historical layout of the
town, owing little or nothing to city planning, is likely to be highly inefficient for
twentieth-century life. Nonetheless, the old plan is preferred to presumably superi-
or alternatives, even when the cost of switching is negligible. Conceivably, any
layout would have been retained simply by virtue of a centuries-long history. If so,
this is a violation of the canonical model of decision making.
Similarly our lunchtime companion appears to be a creature of habit, which
may rule out any meaningful exploration of his genuine preferences. How does
one explain the one-time switch in his consumption decision? Did he abandon
ham and cheese deliberately or on a whim? Or was ham unavailable that day, forc-
ing him to accept an alternative choice, which he then discovered he preferred?
Beer drinkers are not the only consumer segment loyal to its chosen brands. The
greatest marketing error in recent decades-the substitution of “new” for “old’
Coca Cola-stemmed from a failure to recognize status quo bias.6 In blind taste tests,
consumers (including loyal Coke drinkers) were found to prefer the sweeter taste of
new Coke over old by a large margin. But the company did not think about informed
consumer preferences-that is, their reactions when fully aware of the brands they
were tasting. Coke drinkers’loyalty to the status quo (Coke Classic currently outsells
new Coke by three to one) far outweighed the taste distinctions recorded in blind
taste tests. In short, so far as marketing was concerned, blind taste tests, despite their
objectivity (or, more aptly, because of it), proved to be irrelevant.
We have attempted to test the strength of status quo effects experimentally and to
speculate on their significance. The paper is organized as follows: Section 1 con-
tains a discussion and analysis of the controlled experiments. Section 2 examines
status quo bias in two field studies. One study examines the choice of health in-
surance plans by Harvard employees. The other examines the division of retire-
ment contributions between TIAA and CREF funds of faculty throughout the na-
tion. To examine status quo bias in each case, we compare the choices of new
enrollees as opposed to those who have already made choices. Section 3 draws on
economics and psychology to provide explanations for the status quo bias. Section
4 considers a range of applications.
12
WILLIAM SAMUELSON AND RICHARD ZECKHAUSER
1. Experimental tests
Controlled experiments were conducted using a questionnaire consisting of a
series of decision questions. Each question begins with a brief description of a
decision facing an individual, a manager, or a government policymaker, followed
by a set of mutually exclusive alternative actions or policies from which to choose.
The subject plays the role of the decision maker and is asked to indicate his pre-
ferred choice among the alternatives. In many of the decisions, one alternative oc-
cupies the status quo position. In Part One of the questionnaire, the wording of the
decision problem frames one of the alternatives as the status quo. That is, the
status quo labeling is exogenously given. In Part Two, subjects face a sequential
decision task. In an initial decision, each subject chooses from a set of alternatives.
This choice becomes the self-selected status quo point for a subsequent decision.
1.1 Test design
To test for status quo effects, Part One’s experimental design used two versions of
the decision questions. In the neutral version, the subject faces a new decision and
must choose from several alternatives, all on an equal footing. In the status quo
version, one alternative occupies the position of the status quo. Question 2 of Part
One illustrates the experimental design: the neutral version is shown first,
followed by the status quo version.
2. You are a serious reader of the financial pages but until recently have had few
funds to invest. That is when you inherited a large sum of money from your great
uncle. You are considering different portfolios. Your choices are:
__ a) Invest in moderate-risk Co. - b) Invest in high-risk Co. B. Over
A. Over a year’s time, the stock a year’s time, the stock has a .4
has .5 chance of increasing 30% chance of doubling in value, a .3
in value, a .2 chance of being chance of being unchanged, and a
unchanged, and a .3 chance of .3 chance of declining 40% in
declining 20% in value. value.
__ c) Invest in treasury bills. Over __ d) Invest in municipal bonds.
a year’s time, these will yield a Over a year’s time, they will
nearly certain return of 9%. yield a tax-free return of 6%.
2’. You are a serious reader of the financial pages but until recently have had
few funds to invest. That is when you inherited a portfolio of cash and securities
from your great uncle. A significant portion of this portfolio is invested in
moderate-risk Company A. You are deliberating whether to leave the portfolio
intact or to change it by investing in other securities. (The tax and broker com-
STATUS QUO BIAS IN DECISION h4AKING
13
mission consequences of any change are insignificant.) Your choices are
(check one):
__ a) Retain the investment in moderate-risk Company A. Over a year’s time,
the stock has a .5 chance of increasing 30% in value, a .2 chance of being
unchanged, and a .3 chance of declining 20% in value.
__ b) Invest in high-risk Company B. Over a year’s time, the stock has a .4
chance of doubling in value, a .3 chance of being unchanged, and a .3
chance of declining 40% in value.
- c) Invest in treasury bills. Over a year’s time, they will yield a nearly cer-
tain return of 9%.
- d) Invest in municipal bonds. Over a year’s time, these will yield a tax-free
rate of return of 6%.
The entire questionnaire is shown in the Appendix.
In the neutral (NEUT) version of the question, the four choices are presented as
new alternatives, whereas the status quo (SQ) version portrays the first alternative
as the status quo: retain the investment in moderate-risk Company A. In all, five
different versions of this decision problem were tested: one neutral version and
four SQ versions, each assigning a different option to the SQ position. Across the
five versions of the question, a particular option occupied three possible positions:
as a neutral alternative (one case), as the SQ option (one case), or as an alternative
to the status quo (ASQ) option (three cases).
Testing for status quo effects proceeded according to a straightforward experi-
mental design. Each subject was presented with a single version of each of the Part
One questions. (No subject answered the same question or different versions of the
same question twice.) Different versions of each question were tested across the
aggregate sample of subjects. In addition, the number of available alternatives in
the decision problems was varied between two and four in an effort to test whether
a numbers effect influenced the degree of status quo bias.
Thus, in addition to the four-alternative version shown earlier, a decision prob-
lem was also presented in 2 two-alternative versions: one pairing options a and b,
the other pairing options c and d. Each such question was portrayed in a neutral
version and in two status-quo versions. In all, there were six separate two-
alternative versions for each question. Each question was also tested using a set of
three alternatives; this required four versions: one for the neutral case and three
SQ versions. Thus, the total number of versions tested across all conditions was fif-
teen (6 two-alternative versions, 4 three-alternative versions, and 5 four-alterna-
tive versions).
To conserve space, the Appendix presents only the four-alternative version of
each question in the neutral and (one) status quo case. The other versions were
constructed by fixing the appropriate number of alternatives and permuting the
option occupying the SQ position. In the neutral version, the alternatives were
listed in the 2
X
2 format shown in the Appendix, and the order of alternatives was
14
WILLIAM SAMUELSON AND RICHARD ZECKHAUSER
permuted to control for possible order effects. In the SQ versions, the status quo
alternative was always listed first (as option a); the order of the other alternatives
was permuted.7
The subjects in the experiments were students in economics classes at Boston
University School of Management and at the Kennedy School of Government at
Harvard University. In all, 486 students participated. More than three-quarters
were first-year MBA students; the others were senior undergraduate business ma-
jors at BU and students in the public policy and public administration programs
at Harvard. In all cases, the questionnaire was administered in class, and students
were given 20 to 25 minutes to complete it. This was sufficient, but by no means
ample, time to finish the task. Over 96% of the subjects completed all the entries on
the questionnaire; 98% left no more than the last question incomplete. Finally, the
experimental design relied exclusively on the questionnaire format; no monetary
payments were made to any of the subjects in any of the experiments.
1.2 Results
Tables la-lc summarize subject responses to the decision questions for the two-,
three-, and four-alternative versions, respectively. The tables record the percentage
response rate for each choice alternative in each of three positions: the neutral,
status quo, and non-status quo cases. The accompanying fraction records the
number of subjects selecting the alternative from among the total number of sub-
jects responding. For instance, in Question 2 (neutral condition), the moderate-
riskcompanywas chosenoverthe high-riskcompany by 15 of25 subjects (Table la).
The simplest way to look for a status quo bias in subjects’ decisions is to scan the
percentage response rates across conditions for a given choice alternative in a given
decision problem. Tables la-lc reveal an obvious and strong prevailing pattern:
for the large majority of alternatives, the percentage response rate is highest when
the alternative is in the SQ position, lower in the NEUT position, and lowest in the
ASQ position. In Table la, 16 of 24 cases fit precisely this pattern; in Table lb, 13 of
18 cases; in Table lc, 17 of 24 cases. This pattern of relative response rates holds
firm despite marked differences in the absolute levels of response rates across dif-
ferent choice alternatives within and across decision questions. For example, in
Table la, the bid of $115,000 outpolls by a large margin the competing bid of
$125,000, and its dominance is greatest when it occupies the status quo position. At
the opposite end of the spectrum, in Table lc, the color choices tan and white are
much less popular than silver and red. Nonetheless, tan and white are chosen
much more often when they occupy the status quo position. In short, the decline in
response rate moving from SQ to NEUT to ASQ is remarkably consistent across
decision tasks.
An approximate chi-square test was carried out to test for differences between the
SQ and ASQ response rates. The null hypothesis was that the response fractions in
the two cases were generated from the same binomial distribution; the alternative
STATUS QUO BIAS IN DECISION MAKING
15
Table la.
Pairs
Decision Questions
Alternatives
Number and Percent
Chi-square
Status Quo Neutral Non-Status Quo Significance
#l
#I
#2
#2
#3
#3
#4
#4
#5
#5
#6
#6
60-40
50-50
30-70
70-30
Mod. Risk
High Risk
Treasury
Mod. Risk
$120 K
$125 K
$115 K
$125 K
E. Coast
W. Coast1
W.
Coast2
Midwest
Sparse 1500
Dense 1000
Dense 2000
Sparse 1500
Silver
Red
Tan
White
11/18 = .61
13117 = .76
18/29 = .62
13/21 = .62
21143 = .63
27148 = .56
32/63 = .51
19139 = .63
15/20 = .75
7/19 = .37
36138 = .95
7125 = .28
16/20 = .80
30/38 = .I9
16/20 = 230
7/22 = .32
13120 = .65
15122 = .68
19/38 = .50
16125 = .64
14121 = .67
24164 = .38
10125 = .40
11/23 = .48
11124 = .46
13124 = .54
16134 = .41
18134 = .53
15/25 = .60
lo/25 = .40
18134 = .53
16134 = .47
15122 = .68
7122 = .32
19122 = .86
3122 = .14
23131 = .74
8131 = .26
Ill22 = .77
5122 = .23
20132 = .62
12132 = .38
15145 = .33
30145 = .67
14/20 = .70
6120 = .30
14125 = .56
11125 = .44
4117 = .24
II18 = .39
8121 = .38
11129 = .38
21148 = .44
16143 = .37
11/30 = .37
31163 = .49
12119 = .63
5120 = .25
18125 = .I2
2138 = .OS
8f38 = .21
4f20 = .20
15122 = .68
4120 = .20
7122 = .32
7120 = .35
9125 = .36
19/38 = .50
40164 = .62
7121 = .33
12123 = .52
15125 = .60
(.025)
(JO)
(JO)
(.40)
(.50)
W)
(.OOl)
(.40)
(.05)
(.25)
w
(N.-w
hypothesis was that the underlying binomial probability was greater for SQ than
ASQ. Thep values for this test are listed in the last column of the table. The null
hypothesis of indistinguishable SQ and ASQ response rates is rejected at the 10%
significance level in 31 of 54 cases.
Tables la-lc demonstrate the presence of (statistically significant) status quo
bias across decision tasks and across alternatives within decision tasks. Pooling
the data in these tables provides a summary measure of the overall degree of bias.
Toward this end, we consider the simple model described by the equation pair:
SQ=a+bNEUT
and ASQ = c + dNEUT
(1)
where NEUT denotes the percentage of responses for a given alternative under
neutral framing, SQ is the percentage when it occupies the status quo position, and
ASQ the percentage when it is an alternative to the status quo.8 If status quo bias is
present, it follows that SQ > NEUT > ASQ for any given choice alternative. The
16
WILLIAM SAMUELSON AND RICHARD ZECKHAUSER
Table lb. Triples
Decision Questions
Number and Percent
Chi-square
#l
#2
#3
#4
#5
#6
Alternatives Status Quo
60-40 22153 = .42
50-50 10/20 = .50
30-70 6117 = .35
Mod. Risk 9124 = .38
High Risk 9/18 = .50
Treasury 16/48 = .33
$120 K 6120 = .30
$125 K 5125 = .25
$115 K 29135 = .83
E. Coast 20125 = .80
W. Coast1 7121 = .33
W. Coast2 17125 = .68
Sparse 1500 10123 = .43
Dense 1000 4/18 = .22
Dense 2000 lo/23 = .43
Silver 15130 = .50
Red 7118 = .39
Tan 5/28 = .18
Neutral
7120 = .35
8/20 = 40
5120 = .25
7120 = .35
5120 = .25
8/20 = .40
4123 = .17
6123 = .26
13123 = .57
13/30 = .43
5130 = .17
12130 = .40
7117 = .41
3/17 = .18
7117 = .41
16130 = .53
10/30 = .33
4130 = .I2
Non-Status Quo
7137 = .19
21170 = .30
15173 = .20
29166 = .44
15172 = .21
13142 = .31
8160 = .13
5155 = .09
25145 = .56
14/46 = .30
4150 = .08
9146 = .20
14141 = .34
5146 = .ll
21/41 = .51
25146 = .54
19/58 = .33
5/48 = .lO
Significance
(.025)
cw
c-w
O\J.N
co11
WA.1
(.W
UO)
cw
(.OOl)
cw
(.OOl)
(.50)
(.25)
WA.)
WA)
(.70)
(.40)
model formulation posits that an alternative’s response rate in the SQ and ASQ
situations depends positively on the rate in the neutral setting; thus, we expect the
coefficients b and d to be positive. Without further assumptions, the signs of a and
c cannot be predicted. Implicit in the equations is the assumption that SQ and
ASQ depend on2y on NEUT (the alternative’s owlz response rate) and not on the
configuration of responses across all other alternatives. (Of course, such a distinc-
tion is relevant only when there are three or more alternatives.) Even with this sim-
plification, we allow the relationships shown earlier to vary according to the num-
ber of alternatives present in the decision task. For instance, the formulations
SQ = a2 + b2NEUT
and ASQ = c2 + d2NEUT
(2)
denote the particular linear relations for two-alternative decision tasks. The three-
and four-alternative cases are described by analogous equations with appro-
priately numbered coefficients.
The key to estimating the equations is to recognize the adding-up constraints
associated with them. To illustrate, consider a choice between two alternatives, op-
tions 1 and 2, having response rates NEUTl and NEUT2, respectively, under neut-
ral framing. Then, when option 1 occupies the status quo position, it is natural to
insist that the predicted values SQl and ASQ2 satisfy SQl + ASQ2 = 1 for all
STATUS QUO BIAS IN DECISION MAKING
Table Ic.
Quads
Number and Percent
Decision Questions
Alternatives Status Quo Neutral Non-Status Quo
17
Chi-square
Significance
#l
#2
#3
#4
#5
#6
60-40
SO-50
30-70
70-30
Mod. Risk
High Risk
Treasury
Municipal
$120 K
$125
K
$115 K
$130 K
E. Coast
W. Coast1
W. Coast2
Midwest
Sparse 1500
Dense 1000
Dense 2000
Sparse 1500
Silver
Red
Tan
White
7119 = .37
12137 = .32
13124 = .54
25148 = .52
7118 = .39
8129 = 28
13145 = .29
9119 = .41
20162 = .32
13/50 = .26
41154 = .76
3128
= .11
13120 = 65
3125 = .12
19/29 = .66
9160
= .15
12119 = .63
4124 = .17
10129 = .34
6120 = .30
32/42 = .I6
24145 = .53
S/38 = .13
15154 = .28
6128 = .21
6f28
= .21
11/28 = .39
5/28
= .18
9128 = .32
5128 = .18
5/28
= .18
9128 = .32
5/31 = .I6
6131 = .19
18/31 = .58
2131 = .06
24146 = .52
l/46 = .02
18/46 = .39
3146 = .07
9122 = .41
2122 = 39
6122 = .21
5122 = .23
12123 = .52
5123 = .22
2123 = .09
4/23 = .17
7/109 = .06
22191 =.24
29/104
=
.28
13/80 =.16
21193
=
.29
17/82
=
.21
11166
=
.17
19192
=
.21
281122
=
.23
20/142
=
.14
631140
=
.45
6/166
=
.04
33/114
=
.29
9/109
= .08
42/105
=
.40
7174
=
.09
25173
=
.34
2168
=
.03
21163
=
.33
12/72
=
.17
681137
=
.50
20/134
=
.I5
3/141
=
.02
1 l/125 = .09
(.OOl)
C35)
(.W
(.OOl)
(.40)
C50)
(.15)
w4
W)
(.05)
(.OOl)
C.10)
(.005)
(.W
W)
(.30)
(.025)
CO2)
(.95)
(W
(.OOS)
(.OOl)
(.05)
(.OOl)
NEUTl and NEUT2 such that NEUTl + NEUT2 = 1. But this requirement is
satisfied if and only if
b2
= d2 and a2 + c2 +
d2
= 1. This is shown by simply ad-
ding the equations and making a substitution to obtain
SQl + ASQ2 = (a2 + c2 +
d2)
+
(b2
- d2)NEUTl
(3)
Since the left-hand side must sum to unity, so too must the right (for any value of
NEUTl), implying the coefficient restrictions listed earlier. The analogous restric-
tions for the three- and four-alternative cases are
b3
=
d3, a3
+ 2~3 +
d3
= 1 and
b4 = d4, a4 + 3~4 + d4 =
1
(4)
Besides the intercept restrictions, the important constraint is that the equations for
SQ and ASQ have equal slopes?
We used the pooled data in Tables la- lc to estimate the coefficients in the linear
model subject to the coefficient restrictions noted earlier. Under the working
hypothesis that variations in SQ and ASQ (unaccounted for by NEUT) were ran-
18
WILLIAM SAMUELSON AND RICHARD ZECKHAUSER
dom, we used ordinary least squares regression to estimate the model. The regres-
sion was run using all the data (that is, using the observations associated with all
66 decision choices in Tables la-lc). Intercept and slope dummies were included
in the regression to account for coefficient differences in the two-, three-, and four-
alternative cases. Table 2a summarizes the results of this regression, listing the
coefficient estimates and associated
t
values andp values. The weightedR-squared
for the system is .72. A glance at the table shows that none of the dummy variables
is statistically significant (indeed, the
p
values are not even close to the 10% level,
let alone the 5% level), leading to the strong conclusion that there is no systematic
difference in the relationship depending on the number of alternatives. Table 2b
shows the resulting coefficient estimates after dropping all dummies. Note that the
intercept and slope coefficients in Tables 2a and 2b are identical to the second
decimal place, showing how small is the effect of the dummies. (An F-test fails to
reject the hypothesis that these dummies are jointly zero.) Observe also that the ASQ
intercept is insignificantly different from zero, while the sum of the SQ intercept
and slope coefficients is insignificantly different from one. (Given the coefficient
restrictions, one follows from the other.) The restriction c = 0 ensures that ASQ is
nonnegative (at NEUT = 0). In mm, the restriction a + b = 1 ensures that SQ is no
greater than unity (at NEUT = 1). In short, the estimated equations satisfy these
commonsense restrictions (though the restrictions were not imposed directly).
From these rough-and-ready regressions, we conclude that the equations
Table 2a.
Regression Statistics
Dependent Variable: SQ
Parameter
Estimate
Standard
Error
Approx
7’ Ratio Prob
T
Intercept .18 ,033 5.51 .OOOl
Neut .833 ,058 14.22 .OOOl
Int Dum3 -.024 ,030 -.78 ,436
Int Dum4 -.006 .032
-.19 ,850
Slope Dum3 ,018 ,048
.39 .700
Slope Dum4 ,054 ,054 1.00 .320
Dependent Variable: ASQ
Parameter Standard
Estimate Error
T
Ratio
Approx
Prob
T
Intercept -.015 ,033 -.46
.649
Neut
,833
.058 14.22 .OOOl
Int Dum3 ,019 ,021 .90 ,312
Int Dum4 .012 ,022 .55 ,583
Slope Dum3 ,018 ,048
.39 .700
Slope Dum4 ,054 ,054
1.00 ,320
STATUS QUO BIAS IN DECISION MAKING
19
Table 2b.
Regression Statistics
Dependent Variable: SQ
Parameter Standard Approx
Estimate Error T Ratio Prob
T
Intercept ,177 ,022 1.96 .OOOl
Neut ,830 ,038 21.71 .Oool
Intercept
Neut
Dependent Variable: ASQ
Parameter Standard
Approx
Estimate Error
T
Ratio Prob
T
-.0065 .020 -.32 .I525
,830 .038 21.71 .OOOl
SQ = .17 + .83NEUT
ASQ = .83NEUT
(4)
provide the best summary measures of the extent of status quo bias in the experi-
mental decision tasks. These equations imply that both the absolute and relative
response rate advantages enjoyed by the status quo option, SQ - NEUT and
(SQ - NEUT)/NEUT, diminish as NEUT increases. It is the relatively unpopular
alternatives, not the popular ones, that receive the largest response-rate edge from
occupying the status quo position.
These equations apply regardless of the number of alternatives. This suggests
that the relative bias should be expected to be larger the greater the number of
alternatives. (For instance, in the election example discussed earlier, the incum-
bent’s advantage was computed to be greater with four candidates than with
two.)
1.3 Other status quo effects
The final two questions in Part One took a slightly different aim at status quo ef-
fects. In Question 7 (see Appendix), subjects were presented a continuum of possi-
ble options. As water commissioner, the subject had to choose among numerous
possible water allocations between town residents and farmers during a water
shortage. Here, the status quo was introduced by noting the water distribution
chosen by the previous commissioner during an earlier drought. (The decision de-
scription also provided substantial quantitative information about town and
agricultural demands for water.) Each subject received one of three versions of the
question. These were identical except for the status quo water allocation to the
20
WILLIAM SAMUELSON AND RICHARD ZECKHAUSER
town, which was either 100,000,200,000, or 300,000 acre-feet. We sought to isolate
the impact of status quo anchoring (relative to the influence of other sources of in-
formation) by comparing response results across the three versions. Our working
hypothesis was that, other things equal, the greater the status quo allocation to the
town, the greater would be the actual allocation.
Table 3, which lists the distribution of responses by version, strongly bears out
this hypothesis. Starting from a 100,000 acre-feet SQ allocation and proceeding to
larger ones, each subsequent distribution of responses stochastically dominates
(i.e., can be formed by rightward shifts in) its predecessor. A chi-square test
strongly rejects the hypothesis that the responses across versions are drawn from
the same multinomial distribution. A simple way to gauge the impact of the SQ is
to compare the mean allocations across the versions. These are 153,000, 183,000,
and 200,000 in order of ascending SQ allocations. The influence of the SQ alloca-
tion is obvious. Note, however, that subject decisions are only partially anchored
to the status quo point; that is, they are moved by other factors as well. Thus, a
200,000 (i.e., 300,000 - 100,000) difference in the SQ allocation implies roughly a
50,000 acre-foot impact on the chosen allocation.
Question 8 measures the value consequences of status quo bias. As chief of a
consulting firm, subjects were asked to report their willingness to pay to relocate
their office quarters from an older to a newer (more conveniently located) build-
ing. In a second version, all information was the same except that the company’s
present quarters were in the newer building and the proposed move was to the
older building. In either case, the description stated that as an inducement the
company’s moving costs and other expenses would be paid by the landlord-to-be.
Compensating values were expressed as a percentage of the current rental rate
(which was left unspecified). Letx denote the percentage rent increase the subject
would be just willing to pay for a move from old to new;y denotes the required rent
Table
3. Water Allocations
a) 100,000 a-f
(60 subjects)
Status Quo Allocation
b) 200,000 a-f
(67 subjects)
c) 300,000 a-f
(61 subjects)
Town Allocation
chosen by subjects
Percentage of Responses
50,000
3 1 2
100,000 21 4 5
150,000 52 30 21
200,000 17 48 46
250,000 5 12 20
300,000 2 5 6
Total 100 100 100
STATUS QUO BIAS IN DECISION MAKING
21
percentage reduction for a move from new to old. If the subjects show no bias in
evaluating the move, these values should be the same when expressed relative to
the same base: y = x/(1 + x). That is, for bias-free subjects, y should be nearly
equal to (but slightly less than) X. On the other hand, if status quo bias is signifi-
cant, one would expecty >