Motivated Use of Numerical Anchors for Judgments Relevant to the Self

Abstract

The anchoring effect has been replicated so extensively that it is generally thought to be ubiquitous. However, anchoring has primarily been tested in domains in which people are motivated to reach accurate conclusions rather than biased conclusions. Is the anchoring effect robust even when the anchors are threatening? In three studies, participants made a series of probability judgments about their own futures paired with either optimistic anchors (e.g., “Do you think that the chances that your current relationship will last a lifetime are more or less than 95%?”), pessimistic anchors (e.g., “more or less than 10%?”), or no anchors. A fourth study experimentally manipulated motivation to ignore the anchor with financial incentives. Across studies, anchors that implied high probabilities of unwanted events occurring were ineffective. Together, these studies suggest that anchoring has an important boundary condition: personally threatening anchors are ignored as a result of motivated reasoning processes.

Full text

1
Numerical Anchors
Motivated Use of Numerical Anchors for Judgments Relevant to the Self
Samantha Joel1, Stephanie S. Spielmann2, & Geoff MacDonald3
1 University of Utah
2 Wayne State University
3 University of Toronto
Word count: 9993
Keywords: judgment and decision making, anchoring, motivation, self, romantic relationships

2
Numerical Anchors
Abstract
The anchoring effect has been replicated so extensively that it is generally thought to be
ubiquitous. However, anchoring has primarily been tested in domains in which people are
motivated to reach accurate conclusions rather than biased conclusions. Is the anchoring
effect robust even when the anchors are threatening? In three studies, participants made a
series of probability judgments about their own futures paired with either optimistic
anchors (e.g., “Do you think that the chances that your current relationship will last a
lifetime are more or less than 95%?”), pessimistic anchors (e.g., “more or less than 10%?”),
or no anchors. A fourth study experimentally manipulated motivation to ignore the anchor
with financial incentives. Across studies, anchors that implied high probabilities of
unwanted events occurring were ineffective. Together, these studies suggest that
anchoring has an important boundary condition: personally threatening anchors are
ignored as a result of motivated reasoning processes.

3
Numerical Anchors
Motivated Use of Numerical Anchors for Judgments Relevant to the Self
Suppose you were asked to estimate the chances that someone will break your heart
in the next five years, from 0-100%. How would you answer this question? You might draw
from relevant information that you have on hand, such as your own relationship
experiences, in an effort to make a more accurate judgment. You might also use a
potentially relevant anchor point, such as average rates of infidelity or divorce. However,
imagine that before being asked this question, you were tallying the grades for a term
paper, and the class average happened to be 60%. Would this arbitrary number influence
your own perceived likelihood of being left broken-hearted? What if the class average had
been 80%—would that higher number prompt you to make a higher estimate?
Current research on the anchoring phenomenon suggests people should be
influenced by such an anchor. When people need to make judgments about topics for which
they have insufficient information, any available information can be used as an anchor
point. Thus, arbitrary or irrelevant information can be overly influential, a phenomenon
referred to as the anchoring effect (Tversky & Kahneman, 1974). For example, Chapman
and Johnson (1999) asked participants to turn their social security numbers into monetary
figures. Next, participants were asked to indicate the minimum amount for which they
would be willing to sell a particular lottery ticket. The researchers found that participants
used their social security numbers as anchors, such that participants with higher social
security numbers provided higher minimum amounts.
The anchoring effect has been demonstrated in hundreds of studies (see Furnham &
Boo, 2011, for a review). Indeed, this phenomenon has been so systematically replicated
that anchoring has been called “arguably one of the most important truths about human

4
Numerical Anchors
judgment” (Simmons, LeBoeuf, & Nelson, 2010, p. 917), as well as “a truly ubiquitous and
robust phenomenon” (Mussweiler & Strack, 1999, p. 137). However, no study that we know
of has examined anchoring in the context of judgments that participants should personally
prefer (naturally or because of incentives) to be in one direction rather than another. We
propose that the anchoring effect may be less robust or even eliminated when the
perceiver is strongly motivated to reach conclusions that are inconsistent with the anchors.
In particular, we propose that anchoring may have important boundary conditions in
contexts involving meaningful, real-life consequences for the self.
The Anchoring of Unbiased Judgments
The vast majority of anchoring research has examined how anchors impact people’s
judgments in the domain of general knowledge (see Furnham & Boo, 2011, for review), a
domain in which people generally have no reason to be biased toward one judgment or
another. For example, numerous studies have examined how people’s judgments of facts
(e.g., the height of the Brandenburg Gate) are moved in the upward direction by high
anchors (e.g., 150 meters), and moved in the downward direction by low anchors (e.g., 25
meters; Strack & Mussweiler, 1997).
Some research has examined anchoring in the more consequential domain of legal
sentences (e.g., Chapman & Bornstein, 1996; Englich & Musseiler, 2001; Englich,
Mussweiler, & Strack, 2006). However, such research has placed participants in the role of
a neutral third party, motivating accurate—rather than biased—judgments. For example, in
some studies, legal professionals have been presented with hypothetical criminal cases and
asked to decide on the lengths of the defendants’ sentences (Englich & Mussweiler, 2001;
Englich et al., 2006). In such cases, although the domain may be consequential—such that

5
Numerical Anchors
participants may be particularly motivated to reach a fair and just conclusion—there is no
reason for participants to be motivated to reach a particular conclusion. For example, none
of the anchors in these studies were personally threatening such as by deciding a length of
time the participant would themselves be in jail.
Limited research has examined the influence of anchors on participants’ judgments
about their own performance: a domain in which people are potentially biased.
Participants have been asked to predict how many anagram puzzles they would be able to
solve (Cervone & Peake, 1986), or how many sentences they would be able to unscramble
(Switzer & Sniezek, 1991). Participants estimated that they could solve more anagram
puzzles, or unscramble more sentences, after being given high anchors rather than low
anchors. However, in these studies, participants actually engaged in the task at hand,
meaning that participants knew that their estimates would be compared to their
performance. Because people are less likely to make self-serving evaluations about
themselves in cases where the evaluation can be readily compared against objective
standards (e.g., Dunning, Meyerowitz, & Holzberg, 1989; Felson, 1981), these studies
appear to incentivize for accurate judgments more so than particular judgments.
Plous (1989) examined whether anchors can influence people’s judgments about
the probability of a nuclear war. Nuclear war was a topic of great concern at the time of this
research (Mayton, 1986); therefore, this research may be the closest in the existing
literature to a test of our hypothesis that threatening anchors are ineffective. Across six
samples, participants were asked to judge the probability of nuclear war after being given a
low anchor (1%), a high anchor (90%), or no anchor. The researchers found that the
estimates provided in the low-anchor conditions (M = 10.8), the no-anchor conditions (M =

6
Numerical Anchors
19.1), and the high-anchor conditions (M = 25.7) were significantly different from one
another. Thus, this study points to the possibility that threatening anchors may indeed be
effective at swaying people’s probability judgments: anchors may be able to lead to the
perception that nuclear war is more likely. However, although nuclear war appears on its
face to be a threatening outcome, data on participants’ motives were not collected, and
therefore it is unclear just how motivated participants were to conclude that nuclear war
was unlikely.
Some research has experimentally manipulated people’s motivations in the context
of anchoring. However, such research has focused exclusively on accuracy motivation,
whereby participants are provided with monetary incentives for choosing correct answers
and reaching accurate conclusions (e.g., Epley & Gilovich, 2006; Simmons et al., 2010;
Tversky & Kahneman, 1974; Wright & Anderson, 1989). No study that we know of has
manipulated people’s directional motivation; incentivizing participants to reach one
conclusion over another.
Overall, we know very little about how the anchoring phenomenon operates when
people are motivated to reach a certain conclusion, such as a judgment or decision that is
consistent with their own pre-existing goals, beliefs, preferences, and biases. The anchoring
effect has yet to be tested in highly personally relevant domains—contexts in which people
should have naturally strong motivation to reach certain conclusions over others—nor has
the anchoring effect been tested in contexts where participants are incentivized to prefer
certain conclusions.

7
Numerical Anchors
The Anchoring of Biased Judgments
The notion that anchors may be less effective when they contradict a person’s
preferred conclusions is consistent with an attitudinal perspective on the anchoring effect
(Wegener, Petty, Detweiler-Bedell, & Jarvis, 2001). Drawing from attitude change theories
such as the elaboration likelihood model (Petty & Cacioppo, 1986), an anchor can be
thought of as a persuasive message about what judgment one should make. This message
may influence the perceiver through thoughtful, elaborative processes, or through less-
thoughtful, low-effort processes (Blankenship, Wegener, Petty, Detweiler-Bedell, & Macy,
2008).
Within this attitudinal framework, one of the most empirically supported models of
anchoring—the selective accessibility model (Strack & Mussweiler, 1997)—represents an
elaborative route through which anchors can affect judgments. The selective accessibility
model posits that when people are trying to make an accurate estimate, the presence of an
anchor prompts people to consider similarities between the true answer and the anchor.
Anchor-consistent information becomes selectively accessible, leading the perceiver to
generate an estimate closer to the anchor. The attitudinal framework suggests that seeking
confirming information is one elaborative way in which the perceiver may process the
anchor. However, in certain circumstances, the perceiver may seek disconfirming
information instead. For example, Wegener et al. (2001) found that anchors are less
effective when they are implausibly extreme, and argued that this is because implausible
anchors lead people to consider ways in which the true answer is different from the anchor.
People often seek disconfirming information when they are motivated to reject a
persuasive message (Edwards & Smith, 1996). Consistent with this, Wegener et al. (2001)

8
Numerical Anchors
suggested that people may seek disconfirming information in response to anchors when
they are motivated to reject the anchors; i.e., when they “have some reason to ‘disagree’
with the value suggested by an anchor” (p.67). Through such processes, anchors that run
counter to the perceiver’s biases may be rendered ineffective. These ideas have yet to be
empirically tested.
Self-relevant domains offer an ideal context for testing the impact of motivational
biases on anchoring. People tend to be highly motivated to reach certain conclusions when
making judgments that have meaningful, real-life consequences for the self. Specifically,
people make judgments that are important to the self in a motivated fashion such that goal-
consistent information (i.e., information that supports one’s desired outcome) receives
more weight than goal-inconsistent information (e.g., Kunda, 1987; 1990). For example, in
the domain of romantic relationships, people tend to make overly optimistic predictions
about the longevity of their relationships because they are motivated to ignore the negative
aspects of their relationships (T. MacDonald & Ross, 1999). Such domains offer a suitable
test of whether the anchoring effect persists in the face of motivational bias.
The Present Research
The goal of the present research was to test a potential boundary condition to the
anchoring effect. Can arbitrary anchors push people toward particular judgments even
when those judgments contradict their own preferences? Or, might anchors that threaten
preferred conclusions represent an important exception to the otherwise robust anchoring
phenomenon? We predicted that anchors are processed consistently with people’s
personal goals, such that anchors suggesting negative outcomes for the self are relatively
ineffective.

9
Numerical Anchors
In Studies 1 and 2, we examined how arbitrary anchors affect people’s estimates of
the likelihood of positive and negative events occurring in domains high in personal
relevance (e.g., relationship events, life events). We predicted that anchors would be highly
effective when they suggested positive outcomes for the self, but less effective when they
suggested negative outcomes for the self.
In Study 3, we examined the effects of arbitrary anchors in conjunction with a
different motivation. People in romantic relationships tend to defensively derogate
alternative potential partners because they threaten their current relationship (e.g.,
Johnson & Rusbult, 1989). Therefore, we expected that anchors would be effective when
they suggested a negative future with an alternative romantic partner, but less effective
when they suggested a positive future with an alternative romantic partner.
Finally, in Study 4, we experimentally manipulated people’s motivation to reach a
particular conclusion with a financial incentive. Specifically, we told participants that they
would be assigned to one of two arbitrary groups (Copper versus Bronze), and asked
participants to judge the likelihood that we would assign them to the Copper group. Some
participants were told that those in the Bronze group would receive a financial bonus. We
predicted that anchors suggesting a high probability of being assigned to the Copper group
would be ineffective, but only for participants who were informed about the bonus and
thus had motivation to ignore the anchor.
Data and syntax for all four studies can be accessed through the Open Science
Framework: https://osf.io/nsed6/?view_only=3f99dbfdf93c424c8fc9124219bf5ac3.
Complete lists of probability judgment questions, their respective anchors, and
participants’ mean probability estimates can be found in Supplementary Tables 1-5.

10
Numerical Anchors
Standardized probability estimates were used for all analyses; however, the same patterns
of results emerge using unstandardized estimates (See Supplementary Appendix A).
Study 1
In Study 1, we focused on judgments about romantic relationships because people
care deeply about the outcomes of their relationships. Thus. this domain should promote a
desire to reach biased rather than objectively accurate conclusions. Most people hope and
expect that their relationships will succeed (see Fletcher & Kerr, 2010, for review). They
view their relationships in an overly optimistic light, and they make strongly biased
judgments about their romantic futures (e.g., Baker & Emery, 1993; T. MacDonald & Ross,
1999). Participants in romantic relationships were presented with positive and negative
relationship events that could plausibly happen in the future, and asked to estimate the
likelihood that each event would occur. Prior to making their estimates, participants were
randomly assigned to receive either optimistic numerical anchors, pessimistic numerical
anchors, or no anchors. We predicted that people’s motivation to have a successful
relationship would impact the relative effectiveness of the anchors, such that anchors
suggesting an optimistic romantic future (e.g., anchors suggesting that positive events are
likely) would lead to judgments significantly different from those in the control condition,
whereas anchors suggesting a pessimistic future (e.g., anchors suggesting that negative
events are likely) would be ineffective.
Participants
A total of 631 North American participants in relationships were recruited online.1
Sample size was chosen based on a power analysis to achieve 90% power to detect small
effects. Thirty-three participants were excluded for not following instructions, 16 because

11
Numerical Anchors
they were single, and 15 expressed suspicion about the purpose of the study. The final
sample was 567 participants (227 male), with an average age of 30 (range = 18 to 67), and
an average relationship length of 5 years (range = 1 month to 42 years); 296 participants
were dating, 54 were engaged, and 217 were common-law or married. This final sample is
large enough to detect a small effect size (ηp2 = .02, f = .14) at 87% power.
Materials and Procedure
Participants were randomly assigned to one of three experimental conditions:
optimistic, pessimistic, or control. All participants were asked to judge the probability that
each of nine relationship events would occur in their future, from 0-100%. Five of the
relationship events were positive (e.g., “What do you think the chances are that your
current relationship will last a lifetime?”) and four events were negative (e.g., “What do you
think the chances are that your partner will one day fall out of love with you?”).
Pilot testing confirmed that relationship events were considered highly personally
relevant, and thus appropriate for testing the boundary conditions of the anchoring effect.
Pilot participants (N = 196) in romantic relationships rated the personal meaning of the
nine relationship events (1 = not at all personally meaningful, 9 = extremely personally
meaningful), compared to the personal meaning of nine world events that could happen in
the future (e.g., “The polar ice cap fully melts away in the next 50 years”). We chose world
events because they are similar to measures that have been used in past anchoring
research (e.g., Kahneman & Tversky, 1974). A paired-samples t-test indicated that
relationship events, M = 7.30, SD = 1.48, were seen as significantly more personally
meaningful than world events, M = 5.10, SD = 1.61, t(195) = 12.86, p < .001, d = .92.

12
Numerical Anchors
For participants in the optimistic condition, all questions were preceded by
optimistic anchors: numbers that suggested a high probability of positive events occurring
(e.g., “Do you think the chances that your current relationship will last a lifetime are more
or less than 95%?”), and a low probability of negative events occurring (e.g., “Do you think
the chances that your partner will one day leave you broken-hearted are more or less than
5%?). Participants responded yes or no, then provided a percentage estimate. Participants
in the pessimistic condition received pessimistic anchors: numbers suggesting a low
probability of positive events occurring, and a high probability of negative events
occurring. Participants in the control condition received no anchors. Consistent with past
anchoring research (Strack & Mussweiler, 1997), we selected the anchors by taking the 15th
and 85th percentiles of baseline estimates provided by a separate pilot sample.2
Furthermore, some participants were told that the anchors were “randomly generated.” As
in past anchoring research (e.g., Mussweiler & Strack, 1999; Simmons et al., 2010), we
included this to ensure that participants did not assume that the anchors were meaningful
(Grice, 1975).
Estimates for each of the nine events were first standardized across the conditions.
Scores for the five positive events were aggregated such that higher mean estimates
represent more optimistic judgments about one’s romantic future. The four negative event
scores were aggregated such that lower mean estimates represent more optimistic
judgments about one’s romantic future.
Results and Discussion
Recall that pessimistic anchors are motivationally inconsistent, particularly when
paired with negative events (suggesting a high probability of negative relationship

13
Numerical Anchors
outcomes). We conducted a between-participants multivariate ANOVA with anchoring
condition (optimistic, pessimistic, and control) as the predictor, and with probability
estimates for positive events and negative events entered as separate dependent variables.
A Wilks’ Lambda multivariate test indicated that anchoring condition significantly
influenced participants’ probability estimates, F(4,1124) = 8.67, p < .001, ηp2 = .03. We next
examined the effects of experimental condition on probability estimates of positive and
negative events separately. Confidence intervals were calculated using estimated marginal
means, and the p-values were calculated using pairwise comparisons with a Sidak
correction. See Figure 1 for raw probability estimates for positive and negative relational
events.
There was a main effect of experimental condition on probability estimates for
positive relational events, F(2,563) = 15.34, p < .001, ηp2 = .05. Participants who received
optimistic anchors, M = .24, CI95%[.13, .34], SD = .65, made significantly more optimistic
predictions than those in the no-anchor control condition, M = .02, CI95%[-.09, .12], SD = .73,
p = .01, and those in the pessimistic anchors condition, M = -.19, CI95%[-.29, 0.08], SD = .79, p
< .001. Participants who received pessimistic anchors made significantly more pessimistic
predictions than participants who received no anchors, p = .02.
There was also a significant main effect of experimental condition on probability
estimates for negative relational events, F(2,563) = 14.02, p < .001, ηp2 = .05. Optimistic
anchors, M = -.29, CI95%[-.40, -.17], SD = .71, resulted in significantly more optimistic
predictions compared to pessimistic anchors, M = .14, CI95%[.03, .25], SD = .88, p < .001, and
compared to no anchors, M = .002, CI95%[-.11, .11], SD = .75, p = .001. However, pessimistic
anchors did not result in significantly more pessimistic predictions compared to no

14
Numerical Anchors
anchors, p = .22. Effects were not moderated by whether or not participants believed that
the anchors were meaningful: the same pattern of results emerged for participants who
were explicitly told that the anchors were randomly generated.1
These results support our hypotheses. Anchors were generally effective at
influencing people’s estimates about the future of their current romantic relationships.
However, particularly threatening anchors – anchors that suggested a high probability of
negative relationship events occurring – did not affect probability judgments. This null
effect emerged despite a sample size of nearly 200 participants per condition, with 87%
power to detect even a small effect size.
Study 2
The purpose of Study 2 was to test the generalizability of our results. First, we
examined how anchors affect probability estimates regarding personal, non-romantic life
events (e.g., getting fired), expecting to again find that personally threatening anchors
would be ineffective. That is, the relationship effects observed in Study 1 are not specific to
the relational domain, but extend to any domain associated with strong motivational bias.
Second, to ensure our results could not be attributed to an inability to replicate anchoring
effects in more standard domains, we included a condition involving estimates of future
world events. Third, we aimed to demonstrate that the relationship effects of Study 1 are
specific to one’s own relationship (due to motivational bias), and do not generalize to any
romantic relationship. When a person makes judgments about a relationship that they are
not particularly invested in (e.g., about a disliked other’s relationship), standard anchoring
effects should emerge.

15
Numerical Anchors
Overall, the study was a 3 x 3 experimental design. Participants were randomly
assigned to make judgments about the probability of one of three types of events: non-
relational personal events, world events, or a disliked other’s relationship events. These
events were paired with either optimistic anchors, pessimistic anchors, or no anchors
(control). We expected to replicate Study 1 with personal events but replicate standard
anchoring effects with world and disliked other events. These results would suggest that
selective, motivated use of numerical anchors extends to any domain that is relevant and
important to the self, and is not unique to romantic relationships.
Participants
A total of 1,932 North American participants in romantic relationships completed
the study online. Ninety-nine participants were excluded for not following the instructions,
61 participants had already completed one of our previous anchoring studies, 15
participants were excluded because they were single, and 9 expressed suspicions about the
purpose of the study. The final sample was 1,748 participants (959 male), with an average
age of 29 (range = 18 to 79) and an average relationship length of 4.5 years (range = 1
month to 48 years); 677 participants were dating, 99 were engaged, and 971 were
common-law or married (one participant did not respond). This sample is large enough to
detect a small effect size (ηp2 = .02, f = .14) at 99.9% power.
Procedure and Materials
Participants were randomly assigned to be presented with one of three types of
events: non-relational life events, world events, or disliked other’s relationship events.
Non-Relational Life Events. We included five positive life events (e.g., “What do
you think the chances are that you will live past the age of 80?”) and four negative life

16
Numerical Anchors
events (e.g., “What do you think the chances are that you will one day get fired from your
job?”), adapted from Weinstein (1980). Pilot results confirmed that these life events were
perceived as significantly more personally meaningful than either the world events or the
disliked other’s relationship events.3
World Events. We included five positive world events (e.g., “Malaria is eradicated
by 2050”), and four negative world events (e.g., “The polar ice cap fully melts away in the
next 50 years”) that could conceivably happen in the future. These are the same events
used for piloting in Study 1. We constructed these events to mirror the types of questions
used in classic anchoring studies: global judgments that are of less personal relevance to
the decision maker.
Disliked Other’s Relationship Events. Participants were told to think about a
person they dislike who is currently in a romantic relationship. Participants wrote down
the name of this disliked person as well as the name of the disliked person’s partner. They
were then presented with the same nine relationship events from Study 1, but phrased to
be about the disliked person’s relationship (e.g., “What do you think the chances are that
the quality of Jane’s relationship will improve over time?”).
As in Study 1, each event was paired with either an optimistic anchor, a pessimistic
anchor, or no anchor (control condition). Anchors were chosen based on pilot sample
ratings.4 Participants were told that these anchors were randomly generated and that they
contained no useful information. All participants were asked to estimate the probability of
each of nine event occurring, from 0-100%. Within each judgment topic, estimates for each
of the nine events were first standardized across the anchoring conditions. Standardized
scores for the five positive events and the four negative events were each aggregated. Each

17
Numerical Anchors
participant thus had two standardized scores representing their perceived probability of
positive versus negative events occurring, relative to other participants who estimated the
same type of event (life events, world events, or disliked others’ relationship events).
Results and Discussion
Recall that pessimistic anchors are motive-inconsistent, particularly when paired
with negative life events (suggesting a high probability of negative outcomes for the self).
We conducted a 3 (anchor type: optimistic, pessimistic, control) X 3 (judgment topic: life
events, world events, disliked relationship) between-participants MANOVA with positive
and negative estimates as the dependent variables. A Wilks’ Lambda multivariate test
indicated that there were was a main effect of anchor type, F(4,3458) = 92.95, p < .001, ηp2
= .10. Due to the standardization of the estimates within each judgment topic, there was no
main effect of judgment topic, F(4,3458) = .02, p = .999, ηp2 < .001. Finally, there was a
significant interaction between anchor type and judgment topic, F(8,3458) = 6.27, p < .001,
ηp2 = .01.
Examining positive events and negative events separately revealed that there were
main effects of anchor type on estimates about both positive events, F(2,1730) = 123.02, p
< .001, ηp2 = .13, and negative events, F(2,1730) = 127.52, p < .001, ηp2 = .13. However,
these main effects were qualified by interactions between anchor type and judgment topic
for both positive events, F(4,1730) = 2.96, p = .02, ηp2 = .007, and negative events,
F(4,1730) = 9.85, p < .001, ηp2 = .02. We next examined the simple main effects for each
judgment topic as in Study 1. See Figure 2 for raw probability estimates of positive versus
negative life events, world events, and disliked other’s relationship events.

18
Numerical Anchors
Non-Relational Life Events. When participants were asked to judge the probability
of positive self-relevant, non-relational events, all anchors influenced participants’
judgments. Optimistic anchors, M = .28, CI95%[.19, .37], SD = .52, resulted in more optimistic
predictions than pessimistic anchors, M = -.31, CI95%[-.40, -.22], SD = .54, p < .001, or no
anchors M = .02, CI95%[-.07, .11], SD = .55, p < .001. Pessimistic anchors also resulted in
more pessimistic predictions than no anchors, p < .001. However, when people were asked
to judge the probability of negative events happening in their lives, not all anchors were
effective. Optimistic anchors, M = -.26%, CI95%[-.35, -.17], SD = .59, resulted in more
optimistic estimates than pessimistic anchors, M = .08, CI95%[-.02, .17], SD = .58, p < .001, or
no anchors, M = .19, CI95%[.09. .28], SD = .67, p < .001. However, those who received
pessimistic anchors made estimates that were not significantly different from those in the
control condition, and were in fact in the opposite direction from the anchors, p = .28.
These effects replicate those in Study 1: whereas anchors could be used to make people feel
like positive events were less likely to happen to them, anchors could not be used to make
people feel like negative events were more likely to happen to them.
World Events. As predicted, all anchors were effective in the context of world
events. For positive world events, optimistic anchors, M = .36, CI95%[.27, .45], SD = .55,
resulted in more optimistic predictions than pessimistic anchors, M = -.37, CI95%[-.40, -.22],
SD = .50, p < .001, or no anchors, M = .02, CI95%[-.07, .11], SD = .55, p < .001. Further,
pessimistic anchors resulted in significantly more pessimistic predictions compared to no
anchors, p < .001. Similarly, when participants were asked to judge the probability of
negative world events, optimistic anchors, M = -.47, CI95%[-.57, -.38], SD = .57, resulted in
more optimistic judgments than pessimistic anchors, M = .38, CI95%[.29, .47], SD = .57, p <

19
Numerical Anchors
.001, or no anchors, M = .09, CI95%[-.003, .18], SD = .60, p < .001. Again, pessimistic anchors
resulted in more pessimistic judgments compared to no anchors, p < .001. Thus, the
ineffectiveness of anchors in personally relevant judgments cannot be attributed to an
inability on our part to replicate standard anchoring effects.
Disliked Other’s Relationship. When participants judged the probability of
positive relationship events happening to a disliked other, those who received optimistic
anchors, M = .19, CI95%[.10, .28], SD = .76, made more optimistic judgments than those in
the control condition, but not significantly so, M = .06, CI95%[-.03, .15], SD = .77, p = .14.
Pessimistic anchors, M = -.24, CI95%[-.33, -.14], SD = .82, resulted in more pessimistic
judgments than optimistic anchors, p < .001, or no anchors, p < .001. All anchors were
effective at influencing people’s judgments about the likelihood of negative relational
events happening to disliked others. Optimistic anchors, M = -.31, CI95%[-.41, -.22], SD = .77,
resulted in more optimistic estimates than pessimistic anchors, M = .29, CI95%[.20, .38], SD =
.77, p < .001, or no anchors, M = .01, CI95%[-.08, .10], SD = .74, p < .001. Pessimistic anchors
also resulted in more pessimistic estimates than no anchors, p < .001. Thus, as predicted,
anchors made people feel that positive events were less likely to happen to a disliked other,
that negative events were more likely, or that negative events were less likely. Surprisingly,
anchors were not completely effective at making people feel that positive events were more
likely to happen in the romantic relationship of a disliked other.
With one exception, all of Study 2’s hypotheses were confirmed. Whereas most
anchors were effective at influencing people’s judgments, the most personally threatening
anchors – those suggesting a high probability of negative events occurring in one’s future –
were ineffective. This null effect is particularly striking given the very large sample size,

20
Numerical Anchors
and the previously-established ubiquity of the anchoring effect. Even anchors suggesting a
higher probability of negative world events were effective, suggesting that people will
accept a high likelihood of negative world outcomes more readily than they will accept a
high likelihood of negative personal outcomes.
The only other case in which anchors were ineffective (and the only finding running
counter to our hypotheses) was that in which anchors suggested that positive events were
more likely to happen to the relationship of a disliked other. It is possible that disliked
others’ relationships are threatening in a manner we did not anticipate. People are biased
toward believing that their own relationships are better than other people’s relationships
(perceived superiority effect; e.g., Rusbult, Van Lange, Wildschut, Yovetich, & Verette,
2000), and all participants in the current sample were in romantic relationships of their
own. The notion of positive events occurring in the relationships of disliked others may
thus provide a threatening upward comparison, in turn stimulating motivated processing
of anchors.
Study 3
Whereas Studies 1 and 2 demonstrated that motivation for a positive personal
outcome can lead negative anchors to be ineffective, in Study 3 we examined whether
motivation for a negative personal outcome can lead positive anchors to be ineffective.
When might individuals desire negative personal outcomes? People in long-term romantic
relationships tend to be motivated to believe that their romantic alternatives are
undesirable, as this belief supports the conclusion that their current romantic partner is
the best possible partner for them. Attractive alternatives, which present a threat to the
current relationship, tend to be devalued or derogated (e.g., Johnson & Rusbult, 1989;

21
Numerical Anchors
Meyer, Berkman, Karremans, & Lieberman, 2011). Therefore, we predicted that in the case
of romantic alternatives, positive anchors would be threatening and thus ineffective.
Study 3 had a similar design to Study 1 except that we asked participants currently
in relationships to imagine that they were no longer with their current partner.
Participants estimated the probability of positive versus negative romantic relationship
events happening to them with new partners (e.g., “What do you think the chances are that
you would find a new partner who is at least as rewarding as your current partner?” “What
do you think the chances are that you would ultimately wind up without a romantic
partner?”). We predicted that anchors suggesting that negative events would occur in this
alternative hypothetical relationship—suggesting low quality of alternatives—would be
effective at swaying people’s probability estimates. In contrast, we predicted that anchors
suggesting that positive events would occur would be less effective, because upward
estimates would suggest a high quality of alternatives and therefore threatening the
current relationship.
Participants
A total of 146 North American participants were recruited online (56 male). Three
participants were excluded for not following the instructions, 7 because they were single, 8
because they had participated in one of our previous studies, and 4 because they expressed
suspicion about the purpose of the study. The final sample was 124 participants (45 male),
with an average age of 29.93 (range = 18 to 59), and an average relationship length of 5
years (range = 1 month to 35 years); 70 participants were dating, 9 were engaged, and 45
were common-law or married.

22
Numerical Anchors
The effect size of the main effect of anchoring type on probability estimates ranged
from ηp2 = .05 to ηp2 = .13 in Studies 1 and 2. The present sample size was large enough to
detect these effect sizes with between 68.9% power (ηp2 = .05) and 98.9% power (ηp2 =
.13).
Procedure and Materials
Participants were again randomly assigned to one of three conditions: optimistic,
pessimistic, and control. Participants were then asked to make seven probability judgments
about their romantic alternatives. Prior to each probability judgment question, participants
in the optimistic condition received optimistic anchors (e.g., “If you started dating someone
new, do you think that the chances that things would progress into a meaningful romantic
relationship are more or less than 84%?”), whereas participants in the pessimistic
condition received pessimistic anchors (e.g., 9%). To make these numbers look randomly
generated, low anchors were randomly selected for each question between 5% and 30%,
and high anchors were randomly selected for each question between 70% and 95% (these
randomly selected anchors were the same for all participants within a condition). Each
estimate was standardized. Positive events were aggregated such that higher estimates
represent more optimistic judgments, and negative events were aggregated such that
higher estimates represent more pessimistic judgments about one’s romantic future
without one’s current partner. As in Study 2, all participants who received anchors were
told that the anchors were randomly generated and that they contained no useful
information.

23
Numerical Anchors
Results and Discussion
Recall that optimistic anchors are motive-inconsistent in this study, as they suggest
a positive relationship future with someone other than one’s current partner. We
conducted a between-subjects MANOVA with anchoring condition (optimistic, pessimistic,
and control) predicting probability estimates for positive and negative events. A Wilks’
Lambda multivariate test indicated that anchoring condition had a significant effect on
participants’ probability estimates, F(4,240) = 6.05, p < .001, ηp2 = .09. We next examined
the effects of experimental condition on probability estimates of positive and negative
events separately. See Figure 3 for raw probability estimates.
There was a significant main effect of experimental condition on probability
estimates for positive alternative relational events, F(2,121) = 6.79, p = .002, ηp2 = .10.
Participants who received optimistic anchors, M = .29, CI95%[.04, .53], SD = .80, did not
make significantly more optimistic predictions than those in the control condition, M = .03,
CI95%[-.20, .27], SD = .75, p = .37. Participants who received pessimistic anchors, M = -.36,
CI95%[-.62, -.11], SD = .80, made marginally more pessimistic predictions than those in the
control condition, p = .07. The optimistic and pessimistic conditions were significantly
different from one another, p < .001.
There was also a significant main effect of experimental condition on probability
estimates for negative relational events, F(2,121) = 5.90, p = .004, ηp2 = .09. Participants
who received optimistic anchors, M = -.21, CI95%[-.42, -.002], SD = .64, did not make
significantly different predictions from participants in the control condition, M = -.07,
CI95%[-.27, .13], SD = .11, p = .70. However, participants who received pessimistic anchors,
M = .30, CI95%[.08, .51], SD = .70, made significantly more pessimistic predictions than those

24
Numerical Anchors
in the control condition, p = .05. Participants in the optimistic and pessimistic conditions
made predictions that were significantly different from one another, p = .003.
These results further support the conclusion that arbitrary anchors can be
ineffective when they threaten the self. People in relationships are threatened by the notion
that they have high-quality alternatives to their current romantic partner (e.g., Johnson &
Rusbult, 1989; Meyer et al., 2011). In the present study, participants who received
pessimistic anchors—suggesting that a romantic future with someone other than their
current partner would be unsatisfying—provided more pessimistic estimates about their
alternative romantic prospects relative to participants in the no-anchor control. However,
participants who received optimistic anchors—suggesting that they would be able to find
relationship happiness without their current partner—did not provide significantly more
optimistic estimates about their alternative prospects relative to the control participants.
Study 4
The purpose of Study 4 was to demonstrate that the boundary condition observed in
Studies 1-3 is specifically due to motivated reasoning processes. We sought to rule out two
alternative, non-motivational explanations for the effect: differential knowledge, and
differential plausibility of the anchors. In Studies 1-3, motivational bias was examined by
presenting people with judgment-anchor pairs that were consistent versus inconsistent
with their naturally preferred conclusions. However, these judgments and anchors may
have differed in other important ways that may have been responsible for their differential
effectiveness. One such potential difference is the amount of knowledge people have about
each judgment. People may already possess considerable knowledge consistent with their
preferred conclusions (e.g., positive qualities of their romantic relationships). When

25
Numerical Anchors
presented with a conclusion-consistent judgment-anchor pair (e.g., high probability of
relationship success), participants may readily call this information to mind, enhancing the
effectiveness of the anchor. In contrast, participants may possess less knowledge consistent
with threatening conclusions, such that they have less confirmatory information available
when presented with a conclusion-inconsistent judgment-anchor pair (e.g., high probability
of heartbreak). This possibility provides an alternative, non-motivational explanation why
anchors consistent with preferred conclusions are more effective than conclusion-
inconsistent anchors. A second, related issue is that conclusion-inconsistent anchors may
seem less plausible to the participant than conclusion-consistent anchors. As discussed,
highly implausible anchors are less effective than plausible anchors even in non-
motivational contexts (Wegener et al., 2001).
To better tease apart the impact of motivational bias in Study 4, we experimentally
manipulated motivational bias while holding the judgments and anchors constant. The
study was a 2 (motivational bias: present or absent) x 3 (anchor: high, low, or none)
experimental design. Mechanical Turk workers were first asked to write a paragraph of
text. Next, participants were told that based on their writing style, they would be assigned
to one of two groups: Copper or Bronze. Some participants were told that those assigned to
the Bronze group would receive a 50-cent bonus (motivated condition), whereas other
participants were not informed of the bonus (objective condition). Participants were then
asked to judge their own probability of being assigned to the Copper group. To those in the
motivated condition, being assigned to the Copper group represents a failure to obtain a
financial bonus, whereas group assignment should be relatively meaningless to those in the
objective condition. Participants were presented with either optimistic anchors (20%),

26
Numerical Anchors
pessimistic anchors (80%), or no anchors (control). We expected that participants would
selectively ignore the pessimistic anchors—suggesting that their likelihood of failing to
achieve the bonus was high—only when they were aware of the bonus.
Participants
A total of 619 North American participants completed the study via Mechanical
Turk; 28 participants were excluded for not following the instructions, and five
participants expressed suspicions about the purpose of the study. The final sample was 586
participants (199 male), with an average age of 34 (range = 18 to 75). This final sample is
large enough to detect a small effect size (ηp2 = .02, f = .14) at 75% power.
Procedure and Materials
Participants were first asked to write about the factors they think people consider
when deciding whether to invest in a new relationship (this topic was chosen for other
research purposes).
Motivational Bias Manipulation. Upon completing the writing task, all participants
received this message: “Thank you for completing this writing task! Different people have
different writing styles. We will feed your responses into an algorithm to determine your
writing style. You will be sorted into either the Copper group or the Bronze group.”
Participants randomly assigned to the motivated condition saw this additional information:
“Participants sorted into the Bronze group will receive a 50 cent bonus on this HIT. Those
sorted into the Copper group will not receive a bonus.” Those assigned to the objective
condition were not told about the bonus.
No-Bonus Estimate. All participants were next asked to make a single estimate:
“What do you think the chances are that you will be sorted into the Copper group? Please

27
Numerical Anchors
provide a numerical estimate from 0-100%.” The question was preceded by either an
optimistic anchor (“Do you think your chances of being sorted into the Copper group are
more or less than 20%?”), a pessimistic anchor (80%?) or no anchor (control condition).
Estimates were standardized across conditions.
Manipulation check. Participants were asked, “How much do you hope to be sorted
into the Bronze group?” (1 = not at all (indifferent), 7 = Very much).
Results and Discussion
Recall that pessimistic anchors are motive-inconsistent, particularly within the
motivated condition where they suggest a high probability of missing a financial
opportunity. We first conducted an independent t-test comparing responses to our
manipulation check question in the motivated versus objective conditions. The
motivational bias manipulation was effective: participants in the motivated condition (M =
5.41, SD = 1.98) were significantly more motivated to be assigned to the Bronze group
compared to those in the objective condition (M = 3.00, SD = 1.83), t(292) = 10.84, p < .001.
We next conducted a 3 (anchor type: optimistic, pessimistic, control) X 2
(motivational bias: motivated, objective) between-participants ANOVA with the no-bonus
estimate as the dependent variable. There was a main effect of anchor type, F(2,579) =
5.93, p = .003, ηp2 = .02, but no main effect of the motivational bias manipulation, F(1,579) =
.05, p = .83, ηp2 < .001. There was a marginal interaction between anchor type and
motivational bias, F(2,579) = 2.81, p = .06, ηp2 = .01.
We next examined simple main effects in the motivated condition versus the
objective condition. Confidence intervals were calculated using estimated marginal means,
and conditions were compared using pairwise comparisons.5 For participants in the

28
Numerical Anchors
motivated condition, optimistic anchors, M = -.15, CI95%[-.35, .05], SD = 1.16, resulted in
marginally more optimistic estimates compared to no anchors, M = .13, CI95%[-.08, .33], SD
= .94, p = .06. Pessimistic anchors, M = -.03, CI95%[-.22, .16], SD = 1.22, resulted in anchors
that were, if anything, also more optimistic compared to no anchors, although the
difference was not significant, p = .27. Estimates provided in the optimistic versus
pessimistic conditions were not significantly different from each other, p = .38.
For participants in the objective condition, optimistic anchors and pessimistic
anchors influenced their judgments to roughly the same extent. Participants in the no-
anchor control condition, M = -.002, CI95%[-.20, .20], SD = .76, provided estimates that were
marginally higher compared to those in the optimistic condition, M = -.26, CI95%[-.46, -.07],
SD = .91, p = .07, and marginally lower compared to those in the pessimistic condition, M =
.27, CI95%[.06, .47], SD = .89, p = .06. Estimates provided in the optimistic versus pessimistic
conditions were significantly different from each other, p < .001. See Figure 4 for raw
probability estimates of being assigned to the Copper (no-bonus) group.
These results suggest that people selectively ignore numerical anchors specifically
due to motivated reasoning processes. Anchors suggesting a high likelihood of being
assigned to the Copper group were ineffective, but only for participants who knew that
being assigned to the Copper group meant missing out on a financial bonus opportunity.
Because the judgments and anchors were held constant between the motivated versus
objective conditions, this study rules out non-motivational alternative explanations such as
differential knowledge about the judgments or plausibility of the anchors. It seems most
likely that participants in the motivated condition engaged in cognitive processing that

29
Numerical Anchors
undermined the effectiveness of the anchor (e.g., disconfirmatory search; c.f. Wegener et al.
2001) specifically because they disliked the conclusion that the anchor implied.
Mini Meta-Analyses
We conducted mini meta-analyses to compare effect sizes across Studies 1-4 (Goh,
Hall, & Rosenthal, 2016). We separately examined effects of low versus high anchors, on
desirable versus undesirable events, in self-relevant versus non-self-relevant domains, for
a total of eight meta-analyses. Here, low anchors are motive-inconsistent when paired with
desirable events (low probability of desirable events occurring), whereas high anchors are
motive-inconsistent when paired with undesirable events (high probability of undesirable
events occurring).
All examined effect sizes were planned comparison tests between an anchoring
condition and a no-anchor control condition. Cohen’s d values were calculated for each
effect by subtracting the control condition mean from the target anchoring condition mean,
then dividing by the pooled standard deviation for those two conditions. We conducted
eight fixed effect meta-analyses using the metagen function in the “meta” package in r.
These meta-analyses weighted Cohen’s ds by sample size across studies and produced an
overall weighted Cohen’s d value and associated p-value for each meta-analysis. See Tables
1 and 2 for meta-analyses comparing effect sizes in self-relevant contexts and non-self-
relevant contexts, respectively. Seven out of eight of the meta-analyses revealed a
moderate overall effect size that was significantly different from zero, Cohen’s ds = |.36 to
.63|, all ps < .001. As expected, the one exception was the meta-analysis examining the
effects of high anchors on undesirable self-relevant events, Cohen’s d = -.002, p = .98. These

30
Numerical Anchors
results confirm that, across studies, anchors were ineffective only when they suggested a
high probability of an undesired outcome for the self.
General Discussion
These experiments are the first to demonstrate that a classic judgment and decision-
making phenomenon—the anchoring effect—has an important boundary condition. Across
four studies, we found that personally threatening anchors were ineffective at swaying
people’s estimates. In Study 1, we found that romantically attached people were not
swayed by anchors suggesting that their current romantic relationships would fail. In Study
2, we found that these effects extend beyond the romantic domain, as people were not
swayed by anchors suggesting that negative life events would happen to them. In Study 3,
we found that motivation for negative outcomes can also bias the processing of anchors:
consistent with research suggesting that people are motivated to derogate romantic
alternatives (e.g., Johnson & Rusbult, 1989; Meyer et al., 2011), people were not swayed by
anchors suggesting that they would be able to attract romantic partners who are superior
to their current partners. Finally, in Study 4, we experimentally manipulated motivation to
ignore an anchor with financial incentives. People were not swayed by anchors suggesting
that they would be assigned to a “no financial bonus” group, but only if they were aware of
the bonus.
The present work suggests that anchoring is influenced by motivated reasoning
processes, such that anchors that suggest undesired outcomes for the self are ineffective.
These findings highlight a potential shortcoming in anchoring research in that past work
appears not to have investigated judgments for which people are strongly motivated to
reach one conclusion over another. Although it is perhaps disheartening that the current

31
Numerical Anchors
data suggest that the success of a personal enemy’s romantic relationship is more
threatening to people than the prospect of global warming, the data nevertheless suggest
that judgment and decision making researchers may need to carefully consider the
generalizability of their findings to personally-relevant domains such as romantic
relationships (Joel, MacDonald, & Plaks, 2013).
The present research tested the effect of anchors on judgments about both positive
and negative events. Negative events have been shown to have a much stronger impact
than positive events across a variety of domains (e.g., close relationship, major life events;
see Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001 for review). Relatedly, a key tenet of
prospect theory is that losses loom larger than gains, such that people are more motivated
to avoid losses than acquire gains (Kahneman & Tversky, 1979). Consistent with both of
these literatures, anchors were ineffective in the present research only when they
suggested a high likelihood of undesired events occurring. Anchors suggesting a low
likelihood of desired events occurring were effective despite being inconsistent with
people’s motivations. This pattern of results speaks to the overall robustness of the
anchoring effect: numerical anchors appear to be ineffective only when strong motivational
biases are present.
One important future direction would be to examine the mechanism for the present
effects. How are people selectively ignoring threatening anchors? As discussed, the present
finding are consistent with an attitudinal approach to anchoring (Blankenship et al., 2008;
Wegener et al., 2001), which posits that perceivers with sufficient motivation and
resources “elaborate” on anchors as with other kinds of persuasive messages. Although
elaborative processes often entail generating arguments in support of the anchors

32
Numerical Anchors
(confirmatory search strategies; Mussweiler & Strack, 1997), certain circumstances may
compel people to generate counterarguments, lessening the effectiveness of the anchors
(Wegener et al., 2001). Future research should test whether people engage in
disconfirmatory search strategies in response to personally threatening anchors. For
example, if ignoring anchors is an elaborative process that occurs through seeking out
disconfirming evidence, the present effects may disappear under conditions of high
cognitive load.
Implications and Conclusions
These findings shed light on how people make judgments and decisions about their
personal lives. Specifically, even highly self-relevant judgments and decisions may be
affected by seemingly irrelevant sources of influence, as long as that information is not too
inconsistent with one’s preferred conclusions. Moreover, these effects were obtained even
when participants were told that the anchors were meaningless and contained no useful
information. Future research should examine how anchoring effects might extend beyond
likelihood estimates to personal decisions. For example, salient numerical anchors—such
as the age at which friends become pregnant, or the relationship length at which an
acquaintance got engaged—may influence personal decisions on these issues.
Simultaneously, the ineffectiveness of threatening anchors may shield decision-makers
from discouraging base rates. For example, these results may help to explain why, in one
study, engaged individuals correctly estimated the national divorce rate at a median of
50%, yet estimated the likelihood that they would personally divorce at a median of 0%
(Baker & Emery, 1993).

33
Numerical Anchors
Overall, these findings demonstrate the utility of testing judgment and decision
making (JDM) phenomena in non-traditional JDM domains (Joel et al., 2013). This research
suggests that anchoring can influence even the most psychologically meaningful areas of
our lives. Conversely, these findings demonstrate that this classic heuristic has a key
boundary condition: highly threatening anchors are ineffective. Not only is this heuristic
not as pervasive as previously thought, but it is less reliable precisely in the judgment and
decision-making domains that may have the greatest emotional impact for people.

34
Numerical Anchors
Footnotes
1. Study 1 was composed of two samples. Participants in the first sample (N = 150)
were told nothing about how the anchors were generated. Those in the second
sample (N = 417) were told that the anchors were randomly generated. A Wilks’
Lambda multivariate test indicated that there was neither a significant main effect
of sample, F(2,559) = .60, p = .60, ηp2 = .002, nor an interaction between condition
and sample, F(4,1118) = .84, p = .84, ηp2 = .001. We have therefore merged these
samples. Results hold when hypotheses are tested separately in each sample.
2. A pilot sample of 43 romantically attached participants estimated the odds that each
positive and negative relationship event from Study 1 would happen to them in the
future. These pilot responses provided baseline estimates of each relational event.
Optimistic anchors were created by taking the 85th percentiles of baseline estimates
for positive events, and the 15th percentiles of baseline estimates for negative
events. In contrast, pessimistic anchors were created by taking the 15th percentiles
for positive events and the 85th percentiles for negative events.
3. A pilot sample of participants (N = 196) in romantic relationships rated the personal
meaning of each of the 21 events (1 = not at all personally meaningful, 9 = extremely
personally meaningful). Non-relational life events, disliked other’s relationship
events, and world events were all counterbalanced. A repeated measures ANOVA
indicated that the personal meaning of events depended on event type, F(2,408) =
225.45, p < .001, ηp2 = .53. LSD pairwise comparisons revealed that non-relational
life events, M = 6.36, SD = 1.41, were significantly more meaningful than world
events, M = 4.90, SD = 1.63, p < .001. The world events, in turn, were significantly

35
Numerical Anchors
more meaningful than the disliked other’s relationship events, M = 3.32, SD = 1.76, p
< .001. The disliked other’s relationship events were also significantly less
meaningful than the non-relational life events, p < .001.
4. As in Study 1, the events used in all conditions were first piloted to assess baseline
estimates. Out of 150 participants in relationships, 50 estimated the likelihood of
non-relational life events, 50 estimated the likelihood of world events, and 50
estimated the likelihood of relationship events happening to a romantically attached
person who they did not like. We then took the 15th and 85th percentiles of pilot
ratings to use as the anchors, as in Study 1.
5. The overall impact of the anchors on people’s estimates was considerably weaker in
Study 4 (ηp2 = .02) compared to our previous studies (e.g., ηp2 = .10 in Study 2). It
may be that in judging the likelihood of being assigned to one of two groups,
participants were already anchored at 50%. Regardless, given the weaker power,
and given that our predictions for each comparison were confirmatory rather than
exploratory, we did not impose a Sidak correction on the simple effects tests in this
study.

36
Numerical Anchors
Acknowledgements
This work was supported by doctoral fellowships from the Social Sciences and Humanities
Research Council (SSHRC) awarded to Samantha Joel and Stephanie S. Spielmann, and a
SSHRC Insight Grant awarded to Geoff MacDonald.

37
Numerical Anchors
References
Baker, L. A., & Emery, R. E. (1993). When every relationship is above average: Perceptions
and expectations of divorce at the time of marriage. Law and Human Behavior, 17,
439-450.
Baumeister, R. F., Bratslavsky, E., Finkenauer, C., & Vohs, K. D. (2001). Bad is stronger than
good. Review of General Psychology, 5, 323-370.
Blankenship, K. L., Wegener, D. T., Petty, R. E., Detweiler-Bedell, B., & Macy, C. L. (2008).
Elaboration and consequences of anchored estimates: An attitudinal perspective on
numerical anchoring. Journal of Experimental Social Psychology, 44, 1465-1476.
Cervone, D., & Peake, P. K. (1986). Anchoring, efficacy, and action: The influence of
judgmental heuristics on self-efficacy judgments and behavior. Journal of Personality
and Social Psychology, 50, 492-501.
Chapman, G. B., & Bornstein, B. H. (1996). The more you ask for, the more you get:
Anchoring in personal injury verdicts. Applied Cognitive Psychology, 10, 519-540.
Chapman, G. B., & Johnson, E. J. (1999). Anchoring, activation, and the construction of
values. Organizational Behavior and Human Decision Processes, 79, 115-153.
Dunning, D., Meyerowitz, J. A., & Holzberg, A. D. (1989). Ambiguity and self-evaluation: The
role of idiosyncratic trait definitions in self-serving assessments of ability. Journal of
Personality and Social Psychology, 6, 1082-1090.
Edwards, K., & Smith, E. (1996). A disconfirmation bias in the evaluation of arguments.
Journal of Personality and Social Psychology, 71, 5-24.
Epley, N., & Gilovich, T. (2006). The anchoring-and-adjustment heuristic: Why the
adjustments are insufficient. Psychological Science, 17, 311-318.

38
Numerical Anchors
Englich, B., & Mussweiler, T. (2001). Sentencing under uncertainty: Anchoring effects in the
courtroom. Journal of Applied Social Psychology, 31, 1535-1551.
Englich, B., Mussweiler, T., & Strack, F. (2006). Playing dice with criminal sentences: The
influence of irrelevant anchors on experts’ judicial decision making. Personality and
Social Psychology Bulletin, 32, 188-200.
Felson, R. B. (1981). Ambiguity and bias in the self-concept. Social Psychology Quarterly, 44,
64-69.
Fletcher, G. J. O., & Kerr, P. S. G. (2010). Through the eyes of love: Reality and illusion in
intimate relationships. Psychological Bulletin, 136, 627-658.
Furnham, A., & Boo, H. C. (2011). A literature review of the anchoring effect. The Journal of
Socio-Economics, 40, 35-42.
Goh, J. X., Hall, J. A., & Rosenthal, R. (2016). Mini Meta-analysis of your own studies: Some
arguments on why and a primer on how. Social and Personality Psychology Compass,
10, 535-539.
Grice, H. P. (1975). Logic and conversation. In P. Cole & J. L. Morgan (Eds.), Syntax and
semantics: Speech acts (pp. 41-58). New York, NY: Academic Press.
Joel, S., MacDonald, G., & Plaks, J. E. (2013). Romantic relationships conceptualized as a
judgment and decision-making domain. Current Directions in Psychological Science,
22, 461-465.
Johnson, D. J., & Rusbult, C. E. (1989). Resisting temptation: Devaluation of alternative
partners as a means of maintaining commitment in close relationships. Journal of
Personality and Social Psychology, 57, 967-980.

39
Numerical Anchors
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk.
Econometrica, 47, 263-291.
Kunda, Z. (1987). Motivated inference: Self-serving generation and evaluation of causal
theories. Journal of Personality and Social Psychology, 53, 636-647.
Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108, 480-498.
MacDonald, T. K., & Ross, M. (1999). Assessing the accuracy of predictions about dating
relationships: How and why do lovers’ predictions differ from those made by
observers? Personality and Social Psychology Bulletin, 25, 1417-1429.
Mayton, D. M. (1986). Spontaneous concern about nuclear war by college students. Basic
and Applied Social Psychology, 7, 185-193.
Meyer, M., Berkman, E. T., Karremans, J. C., & Lieberman, M. D. (2011). Incidental regulation
of attraction: The neural basis of the derogation of attractive alternatives in
romantic relationships. Cognition and Emotion, 25, 490-505.
Mussweiler, T., & Strack, F. (1999). Hypothesis-consistent testing and semantic priming in
the anchoring paradigm: A selective accessibility model. Journal of Experimental
Social Psychology, 35, 136-164.
Petty, R. E., & Cacioppo, J. T. (1986). The Elaboration Likelihood Model of persuasion. In L.
Berkowitz (Ed.), Advances in experimental social psychology (Vol. 19, pp. 123-205).
New York: Academic Press.
Plous, S. (1989). Thinking the unthinkable: The effects of anchoring on likelihood estimates
of nuclear war. Journal of Applied Social Psychology, 19, 67-91.

40
Numerical Anchors
Rusbult, C. E., Van Lange, P. A. M., Wildschut, T., Yovetich, N. A., & Verette, J. (2000).
Perceived superiority in close relationships: Why it exists and persists. Journal of
Personality and Social Psychology, 79, 521-545.
Simmons, J. P., LeBoeuf, R. A., & Nelson, L. D. (2010). The effect of accuracy motivation on
anchoring and adjustment: Do people adjust from provided anchors? Journal of
Personality and Social Psychology, 99, 917-932.
Strack, F., & Mussweiler, T. (1997). Explaining the enigmatic anchoring effect: Mechanisms
of selective accessibility. Journal of Personality and Social Psychology, 73, 437-446.
Switzer, F. S., & Sniezek, J. A. (1991). Judgment processes in motivation: Anchoring and
adjustment effects on judgment and behavior. Organizational Behavior and Human
Decision Processes, 49, 208-229.
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases.
Science, 185, 1124-1131.
Wegener, D. T., Petty, R. E., Detweiler-Bedell, B. T., & Jarvis, W.B. G. (2001). Implications of
attitude change theories for numerical anchoring: Anchor plausibility and the limits
of anchor effectiveness. Journal of Experimental Social Psychology, 37, 62-69.
Weinstein, N. D. (1980). Unrealistic optimism about future life events. Journal of Personality
and Social Psychology, 39, 806-820.
Wright, W. F., & Anderson, U. (1989). Effects of situation familiarity and financial incentives
on use of the anchoring and adjustment heuristic for probability assessment.
Organizational Behavior and Human Decision Processes, 44, 68-82.

41
Numerical Anchors
Table 1: Comparing Effect Sizes in Self-Relevant Conditions Across Studies
Desirable Events
Undesirable Events
Judgment Topic
High Anchors
(Motive-Consistent)
Low Anchors
(Motive-Inconsistent)
High Anchors
(Motive-Inconsistent)
Low Anchors
(Motive-Consistent)
Cohen’s d
n
Cohen’s d
n
Cohen’s d
n
Cohen’s d
n
Study 1
Own Relationship
.32
174,
201
-.17
191,
201
.17
191,
201
-.40
174,
201
Study 2
Life Events
.49
195,
188
-.61
189,
188
-.18
189,
188
-.71
195,
188
Study 3
Alternative
Relationship
.52
38, 45
-.21
41, 45
.32
41, 45
-.51
38, 45
Study 4
Financial Bonus
-.14
105, 93
-.27
96, 93
Weighted Cohen’s d
.42
[.28, .55]
-.36
[-.50, -.23]
-.002
[-.12, .12]
-.50
[-.62, -.37]
Weighted p-value
<.001
< .001
.98
< .001
* Sample sizes (n) are reported in the format: anchor condition n, control condition n.

42
Numerical Anchors
Table 2: Comparing Effect Sizes in Non-Self-Relevant Conditions Across Studies
Desirable Events
Undesirable Events
Judgment Topic
High Anchors
(Motive-Consistent)
Low Anchors
(Motive-Inconsistent)
High Anchors
(Motive-Inconsistent)
Low Anchors
(Motive-Consistent)
Cohen’s d
n
Cohen’s d
n
Cohen’s d
n
Cohen’s d
n
Study 2
World Events
.62
192,
193
-.74
195,
193
.49
195,
193
-.97
192,
193
Disliked Relationship
.17
192,
197
-.38
198,
197
.37
198,
197
-.42
192,
197
Study 4
Financial Bonus
.32
93, 99
-.31
99, 99
Weighted d
.39
[.25, .53]
-.55
[-.70, -.41]
.41
[.28, .53]
-.63
[-.77, -.49]
Weighted p-value
<.001
< .001
< .001
< .001
* Sample sizes (n) are reported in the format: anchor condition n, control condition n.

43
Numerical Anchors
Figure 1. Probability Estimates for Positive and Negative Relationship Events (Study 1)
*95% confidence intervals were calculated using estimated marginal means.

44
Numerical Anchors
Figure 2. Probability Estimates for Positive and Negative Events (Study 2)
Positive Life
Events
Negative Life
Events
Positive World
Events
Negative
World Events
Positive
Disliked Other
Events
Negative
Disliked Other
Events

45
Numerical Anchors
Figure 3. Probability Estimates for Alternative Relationship Events (Study 3)

46
Numerical Anchors
Figure 4. Probability Estimates for Motivated versus Objective Judgments (Study 4)
Motivated Objective