translated attributes as choice architecture: aligning ...€¦ · translated attributes as choice...

MANAGEMENT SCIENCEArticles in Advance, pp. 1–15

http://pubsonline.informs.org/journal/mnsc/ ISSN 0025-1909 (print), ISSN 1526-5501 (online)

Translated Attributes as Choice Architecture: Aligning Objectivesand Choices Through Decision SignpostsChristoph Ungemach,a Adrian R. Camilleri,b Eric J. Johnson,c Richard P. Larrick,d Elke U. WebereaTUM School of Management, Technische Universität München, 80333 München, Germany; bRMIT University, Melbourne, Victoria 3000,Australia; cColumbia Business School, Columbia University, New York, New York 10027; dFuqua School of Business, Duke University,Durham, North Carolina 27708; ePrinceton University, Princeton, New Jersey 08544Contact: [email protected] (CU); [email protected] (ARC); [email protected] (EJJ); [email protected] (RPL);[email protected] (EUW)

Received: February 2, 2015Accepted: August 23, 2016Published Online in Articles in Advance:March 23, 2017

https://doi.org/10.1287/mnsc.2016.2703

Copyright: © 2017 INFORMS

Abstract. Every attribute can be expressed in multiple ways. For example, car fuel econ-omy can be expressed as fuel efficiency (“miles per gallon”), fuel cost in dollars, or tons ofgreenhouse gases emitted. Each expression, or “translation,” highlights a different aspectof the same attribute. We describe a new mechanism whereby translated attributes canserve as decision “signposts” because they (1) activate otherwise dormant objectives, suchas proenvironmental values and goals, and (2) direct the person toward the option thatbest achieves the activated objective. Across three experiments, we provide evidence forthe occurrence of such signpost effects as well as the underlying psychological mech-anism. We demonstrate that expressing an attribute such as fuel economy in terms ofmultiple translations can increase preference for the option that is better aligned withobjectives congruent with this attribute (e.g., the more fuel-efficient car for those with pro-environmental attitudes), even when the new information is derivable from other knownattributes. We discuss how using translated attributes appropriately can help align a per-son’s choices with their personal objectives.

History: Accepted by Yuval Rottenstreich, judgment and decision making.Funding: Financial support from the National Science Foundation [Grant SES 0951516] is gratefully

acknowledged. The second author was supported by an Alcoa Foundation Fellowship (2787_2012)received from the American Australian Association.

Supplemental Material: The online appendix is available at https://doi.org/10.1287/mnsc.2016.2703.

Keywords: behavioral decision making • behavioral economics • consumer behavior • choice architecture • theory • government • regulations •applications • marketing • segmentation • translated attributes • proenvironmental behavior

1. IntroductionChoice options typically vary on multiple attributes.For example, a car’s attributes include price, fuel econ-omy, safety rating, and type of navigation system.In turn, each attribute can have multiple translations,that is, different ways the attribute can be expressed.For instance, a car’s fuel economy can be expressedin different ways. Figure 1 shows the U.S. Environ-mental Protection Agency (EPA) fuel economy andenvironment label, which expresses fuel economy interms of miles travelled per gallon of fuel, the esti-mated annual fuel cost (AFC), and a 1-to-10 green-house gas rating (GGR), among others. These differ-ent expressions, which we term “translated attributes,”are closely related to each other and highly corre-lated. Nevertheless, each translation also highlightsdifferent aspects of the same attribute. A fundamentaldecision faced by policy makers is to determine howmuch information should be presented to people tohelp them make better decisions for themselves andfor society. On the one hand, more information could

overwhelm a person. On the other, less informationcould deprive a person of important facts.

Little may be gained by presenting one translatedattribute instead of another because the two trans-lated attributes are often perfectly correlated. Froma strictly rational point of view, translated attributesare completely redundant if the translation equiva-lence is known. However, the theoretical perspectiveof bounded rationality (Simon 1982) and supportingempirical evidence suggest that people tend to con-struct their preferences (Lichtenstein and Slovic 2006;Payne et al. 1988, 1993; Ungemach et al. 2011) usinginformation about choice options at face value (e.g.,the concreteness principle, Slovic 1972; or What YouSee Is All There Is [WYSIATI], Kahneman 2011). Bondet al. (2008) have shown that people often overlookkey personal objectives when making choices becausethey possess a myopic mental representation of thechoice. Assuming that consideration of personal objec-tives is beneficial to effective decision making (Keeney1992), these findings imply that people often end upwith suboptimal decision outcomes that are not fully

1

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

http://pubsonline.informs.org/journal/mnsc/

mailto:[email protected]





https://doi.org/10.1287/mnsc.2016.2703

Ungemach et al.: Translated Attributes as Choice Architecture2 Management Science, Articles in Advance, pp. 1–15, ©2017 INFORMS

Figure 1. (Color online) The Current U.S. Environmental Protection Agency Fuel Economy and Environment Label forGasoline Vehicles

aligned with personal objectives. Bond et al. (2008)have suggested that recognition of important per-sonal objectives may be assisted through guides thatdescribe additional features of products such as con-sumer reports. We propose that translated attributesplay a similar role in improving choice: they allowpeople to recognize important personal objectives thatwould otherwise be overlooked. Therefore, our gen-eral hypothesis is that translated attributes have thepotential to influence preferences because they serveas decision “signposts” helping people first recognizeand then pursue options congruent with their personalobjectives.Previous literature has shown that partitioning a

global attribute into components can increase theweight accorded to that global attribute (Dawes 1979,Fischhoff et al. 1978, Weber et al. 1988). For exam-ple, the global attribute “job security” can be split intocomponent attributes such as “stability of the firm”and “personal job security.” Although related to par-titioning, the notion of translated attributes has threefeatures that distinguish it from this earlier literature.First, translated attributes are monotone transforma-tions of each other and highly correlated, whereas com-ponents of a global attribute described in the attributesplitting literature may not be correlated at all. Sec-ond, translated attributes differ from derived attributesused in advertising, such as a car being described as“masculine,” because translated attributes allmap ontoactual physical properties of the object. Third, andmost importantly, whereas the attribute splitting lit-erature demonstrates the phenomenon that the sumof component attributes attracts more weight than the

global attribute, we identify and test two different pro-cesses by which translated attributes affect decisions.Specifically, we distinguish the established effects ofnoncompensatory heuristics such as counting superi-ority across attributes and using the majority rule (e.g.,Alba and Marmorstein 1987, Russo and Dosher 1983,Zhang et al. 2006), and our own novel explanation:activation of relevant objectives and direction towardachieving them through decision signposts.

1.1. Translated Attributes as Decision SignpostsOn a roadside, a signpost has two important fea-tures: First, it indicates the presence of a destinationthat might be of personal relevance, such as a nearbytown. Second, it points out how to reach the destina-tion by providing the actual direction and distanceto travel. Similarly, translated attributes possess bothfeatures. First, translated attributes can produce acti-vation of personally relevant objectives. This impliesthat a translated attribute will have the most impacton those who have objectives matching those aspectshighlighted by the translated attribute. Second, a deci-sion signpost conveys direction toward achieving thoseobjectives by explicitly describing the degree to whichdifferent options meet the person’s objectives.

People have many objectives that can arise from dif-ferent sources, such as values and goals. Values arestable dispositions that structure and guide specificbeliefs, norms, and attitudes (Feather 1995, Rokeach1973). Goals are motivational constructs directed toachieve a desirable end state (van Osselaer andJaniszewski 2012). Goals direct movement that leads todegrees of achievement allowing progress (and failure)

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

Ungemach et al.: Translated Attributes as Choice ArchitectureManagement Science, Articles in Advance, pp. 1–15, ©2017 INFORMS 3

to be gauged (Huang et al. 2012). Goals, such as loos-ing 5 lbs over the summer, are typically more specificand temporary than values, such as living healthily.Values can direct which goals are considered impor-tant, and prioritized objectives can direct attention tocongruent information, which in turn affects behav-ior (Stern and Dietz 1994). However, Bond et al. (2008)demonstrate that people often focus too narrowly andforget many objectives that they consider to be valu-able, unless aided.We argue that our proposed signpost effect is con-

ceptually distinct from previous work on framingand priming. Traditional “framing effects” occur whenbehavior changes in response to different representa-tions of information that is equivalent in basic struc-ture and final consequences. For example, Levin et al.(1998) distinguish three types of framing effects, eachrelying on different cognitivemechanisms: risky choiceframing, attribute framing, and goal framing. Each ofthese framing categories relies on changes in valance:risky choice framing contrasts options in terms of gainand loss; attribute framing contrasts object character-istics positively and negatively; goal framing contrastspositive consequences of engaging in a behavior withnegative consequences of not engaging in the behavior.Our proposed translated attribute effect is distinct fromthese framing effects because it does not rely on such“valence shifts.” Indeed, we would expect the sameactivation-and-direction effect regardless of positive ornegative valence.

Finally, we argue that the effect of signposts is dis-tinct from “bottom-up” priming effects, which findthat a goal (e.g., becoming educated) can be activatedthrough the subliminal presentation of means to com-plete that goal (e.g., study; Shah and Kruglanski 2003).First, when a translated attribute acts as decision sign-post, the objective is directly presented and compre-hended. Second, translated attributes, when acting asdecision signposts, also provide directional informa-tion about how to choose and act, which primes do not.

1.2. Translated Attributes as Choice ArchitectureBecause they act as decision signposts, translated at-tributes represent a type of choice architecture inter-vention. Choice architecture refers to the applicationof behavioral insights to understand the influence thatdifferent ways of presenting information about choiceoptions can have on decisions and behavior (Sunstein2011, Thaler and Sunstein 2008, Johnson et al. 2012).Examples of application domains include personalhealth (e.g., Kling et al. 2012), retirement savings (e.g.,O’Donoghue and Rabin 1999), and the environment(e.g., Camilleri and Larrick 2014, Larrick and Soll 2008).Choice architecture builds on insights about the effectof defaults (Johnson and Goldstein 2003), the numberof alternatives presented (Payne et al. 1993), and the

partitioning of options and attributes (Fox et al. 2005;for reviews, see Johnson et al. 2012 and Camilleri andLarrick 2015). These interventions, sometimes referredto as “nudges,” often operate beneath conscious aware-ness to help improve individual and social welfare(Smith et al. 2013, Thaler and Sunstein 2008). How-ever, this lack of awareness exposes such interventionsto the criticism that they manipulate individuals byrestricting their autonomy to act upon their own prefer-ences (e.g., Hausman and Welsh 2010). As we demon-strate below, the presentation of translated attributescan be an effective and benevolent aspect of choicearchitecture that does not restrict individuals’ auton-omy. Instead, it allows people to select options that aremore consistent with their personal objectives. This isalso another important theoretical distinction betweensignposts and counting effects (such as attribute split-ting effects and majority rule strategies). Whereas sim-ply steering people to an option that is superior onmultiple positive (translated) attributes is potentially a“trick,” guiding people to see what they care about andto act on this preference is helpful to the individual.

In summary, our novel contribution is the idea thatthe presentation of translated attributes can be utilizedselectively as decision signposts to help people matchoptions with their personal objectives. Specifically, wepredict that the effect of shifting preferences throughthe provision of translated attributes will be strongestfor those who have personal objectives congruent withthose highlighted by the translated attributes. Whena translation is distinct enough to activate an over-looked objective, a signpost effect is expected to occurin addition to any simple counting heuristic effects,which are agnostic with regard to the type of trans-lated attributes presented. Finally, we predict this sign-post effect to occur only when the presented attributesprovide directional information, allowing the person toidentify options that are aligned with activated per-sonal objectives.

The rest of this paper is arranged as follows: In Exper-iment 1, we demonstrate how translated attributes canaffect preferences dependent on the congruency be-tween the translated attribute and a person’s values.In Experiment 2, we show that signpost effects can-not be explained by differences in perceived knowledgeabout the relation between attributes provided by onetranslated attribute versus another. In Experiment 3,weshowhowtranslatedattributes act asdecision signpostsby explicitly manipulating the activation and directionfeatures of translated attributes.

2. Experiment 1—The Signpost EffectThe purpose of Experiment 1 was to test the value acti-vationmechanismproposedaspartof thehypothesized

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


signpost effect. We chose to focus on the environmen-tal domain, specifically choices between cars that dif-fer in terms of their fuel economy, because it providesan acknowledged paradox of consumers appearing toselect options that are not in their economic self-interest(National Highway Traffic SafetyAdministration 2010).We asked participants to choose between pairs of carswhere the global attributes were price and fuel econ-omy, which were arranged to trade off against oneanother. In line with our proposed translated attributeframework, we used three highly correlated transla-tions for the car fuel economyattribute: annual fuel cost,gallons (of gas consumed) per 100 miles, and green-house gas rating. We assumed that annual fuel costwas likely to activate financial objectives, that gallonsper 100 miles was likely to activate fuel consumptionobjectives, and that greenhouse gas rating was likely toactivate preexisting proenvironmental values. This lat-ter assumption was guided by the fact that greenhousegases have been identified as themost significant driverof climate change (IPCC2013) and thegreenhouse effect(U.S. Environmental Protection Agency 2014). In addi-tion, a significant proportion of sustainability-relatedproduct claims are based on carbon emissions (Cohenand Vandenbergh 2012), and consumers are familiarwith its usage. Two advantages of focusing on theactivation of proenvironmental values were that we(a) knew that such values varied in strength acrossthe population and (b) environmental attitude couldbe assessedwith awell-validatedmeasure. Participantswerepresented either oneor twoof these translated fueleconomy attributes. Our experimental design kept thetype of translation orthogonal to the number of trans-lated attributes presented, which allowed us to distin-guish between the effect of counting heuristics and theproposed signpost mechanism.Based on past research, we predicted that when cars

were presented with more translated fuel economyattributes, participants would be more likely to pre-fer the more fuel-efficient car (Alba and Marmorstein

Table 1. Set of Nine Choice Problems Used in Experiments 1 and 3

Car A ($) Car B ($)Difference in total cost

Choice problem Price Annual fuel costa Price Annual fuel costa over five years (car A−B) ($)

1 29,999 3,964 33,699 2,775 2,2462 25,799 2,775 29,999 2,220 −1,4253 25,799 3,964 33,699 2,775 −1,9544 25,799 2,775 33,699 2,220 −5,1255 29,999 3,964 33,699 2,220 5,0216 25,799 3,964 29,999 2,220 4,5217 25,799 3,964 33,699 2,220 8218 29,999 2,775 33,699 2,220 −9259 25,799 3,964 29,999 2,775 1,746

Notes. The lower price car is always car A. The ordering of the cars during the experiments was counterbalanced.aThe fuel cost assumes 15,000 miles driven annually for five years, costing $3.70 for gas and no discount rate.

1987, Russo and Dosher 1983, Zhang et al. 2006). Con-gruent with our signpost effect hypothesis, we furtherpredicted that the effect of translated attributes wouldbe moderated, and be stronger for participants whovalued the environment. Specifically, we predicted thatmore proenvironmental participants would be morelikely to choose the fuel-efficient car when the pre-sented translated attribute was associated with theenvironmental impact of fuel economy (i.e., when pre-sented with greenhouse gas rating). This is because theenvironmental attribute was expected to activate pre-existing environmental values and highlight the oth-erwise overlooked proenvironmental objective in thedecision. However, this moderation was not expectedto occur when the presented translated attribute wasnot associated with the environmental impact of fueleconomy (e.g., annual fuel cost).

2.1. Method2.1.1. Participants. Three hundred forty-one Americanparticipants (52% female; mean age � 31.8; SD � 10.1)were recruited through Amazon’s Mechanical Turk(MTurk) and paid a flat fee for their participation.2.1.2. Materials. Choice Design. We designed a set ofnine choice problems (see Table 1). Carswere describedin terms of price and either one or two different trans-lated attributes of fuel economy. In the cases wherethere was one fuel economy attribute, half the partici-pants were presented with “annual fuel cost” and theother half with “greenhouse gas rating.” In the caseswhere there were two fuel economy attributes, half theparticipants were presented with “annual fuel cost”together with “greenhouse gas rating,” and the otherhalf of participants were presented with “annual fuelcost” together with “gallons per 100 miles” (GPM).Annual fuel cost, greenhouse gas rating, and gallonsper 100 miles were highly correlated. Annual fuel costwas calculated assuming 15,000 miles driven annuallyand $3.70 per gallon of gas (similar assumptions under-lie the calculated values presented on the American

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


EPA label). Gallons per 100 miles was described as thefuel efficiency of the car in terms of how many gallonsof fuel were required to travel 100 miles. Greenhousegas rating was described as a 1-to-10 rating comparinga car’s fuel economy and tailpipe carbon dioxide (CO2)emissions to those of all other new cars, where a ratingof 10 was best (i.e., fewest CO2 emissions). The valuesfor the different attribute levels reflected typical rangesencountered in actual car choices in a factorial design.

Environmental Values. Environmental values weremeasured with the New Ecological Paradigm–Revised(NEP-R) scale (Dunlap et al. 2000), which is a standardmeasure of attitudes toward sustainability. Participantsrated 15 statements (e.g., “Humans have the right tomodify the natural environment to suit their needs”)on five-point scales ranging from “strongly disagree”to “strongly agree.” Scores on the NEP-R scale rangefrom 15 to 75. Higher scores indicate stronger proenvi-ronmental values.

2.1.3. Design. We employed a 2 (number of fuel-effi-ciency attributes: 1 vs. 2)× 2 (environmental attribute:present vs. absent) × 9 (choice set) mixed design. Thefirst two factors were between subjects and the thirdfactor was within subjects. The main dependent vari-able was whether or not participants selected the fuel-efficient car.

2.1.4. Procedure. Participants first read an instructionscreen inwhich theywere told to assume they averaged15,000 miles of driving each year and that they wouldkeep the car for five years before giving it away. Par-ticipants were also shown a summary of the price andfuel-efficiency attribute definitions they would later bepresented with. During the choice task, participantssaw two options in a table format with price andfuel economy attributes (see Online Appendix A). Thechoice screens also displayed the attribute definitions.Participants made a total of nine choices between

different pairs of cars. After these choices, participantsrated the extent to which they considered the environ-mental and the cost aspects of the cars, respectively,when making their choices. These responses weremade on five-point scales ranging from “not at all”to “exclusively.” Participants were then asked to judgehow important a number of different car attributes(e.g., price, car type, safety rating, miles per gallon(MPG)) would be to them ifmaking a real car purchase,on five-point scales ranging from “unimportant” to“extremely important.” Finally, participants completeditems relating to demographics, past driving behavior,environmental attitudes (NEP-R scale), and an atten-tion check.

2.2. ResultsTwenty-one participants incorrectly answered the at-tention check question and were removed from the

subsequent analysis, leaving 320 participants. How-ever, as described below, all conclusions remain un-changed if we include these data in the analysis.2.2.1. Activation of Environmental Values. To assesswhether our manipulation was successful in activat-ing environmental values, we checked the effect ofpresenting the greenhouse gas rating attribute onthe self-rating of considering the environmental andcost aspects of the cars in participants’ decisions. Asexpected, participants reported considering the envi-ronmental aspect to a greater extent when the GGRwas present (mean� 2.95, SD� 1.12) than when it wasabsent (mean � 1.81, SD � 1.00, t(312.16) � −9.553, p <0.0001). Conversely, in conditions with the GGR rat-ing present, the cost aspect was considered to a lesserextent (mean� 3.78, SD� 0.85) than in conditions with-out the GGR rating (mean�4.20, SD�0.75), t(312.94)�4.836, p < 0.0001).2.2.2. Preferences. Figure 2 shows the observed pro-portions of fuel-efficient car choices in the four exper-imental groups collapsed across choice sets. On aver-age, participants chose the fuel-efficient option 54% ofthe time (SD � 31%). In line with a counting expla-nation, participants made more fuel-efficient choiceswhen fuel economy was expressed via two translatedattributes (right pair of bars) than one (left pair of bars).

To investigate the signpost effect, we examined therelationship between the observed likelihood of choos-ing the environmental option across participants’ envi-ronmental attitudes as indicated by their NEP-R scores(see Figure 3). As predicted, the probability of choosing

Figure 2. Proportion of Choices for the Fuel-EfficientOption in Experiment 1 as a Function of the Number ofFuel-Efficiency Attributes and Whether an EnvironmentalAttribute Was Presented or Not

Pro

port

ion

choo

sing

the

fuel

-effi

cien

t car

1 2

No. of fuel-efficiency attributes

1.00

0.75

0.50

0.25

0

Environmental attribute:Absent

Present

Note. Error bars are the standard error of the mean (SEM).

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


Figure 3. Probability of Choosing the Fuel-Efficient Car inExperiment 1 Modeled as a Function of EnvironmentalValues, Number of Fuel-Efficiency Attributes, and Whetheran Environmental Attribute Was Presented or Not

the fuel-efficient car increased with increasing NEP-Rscores when the GGR attribute was present (rightpanel). However, this relationship was not observed inthe conditions without the GGR attribute (left panel).To test that this effect was statistically significant,

we compared several nested multilevel logistic regres-sion models with selection of the fuel-efficient car asthe dependent variable. All models contained randomintercepts for participants and choice problems. For themain hypotheses, we included simple fixed effects forthe number of attributes presented, a dummy variableindicating the presence of the GGR attribute, and thetwo-wayGGRpresent×NEP-R interaction to test the sign-post effect. Themodel confirmed that participantsweremore likely to select the fuel-efficient option when fuelefficiency was expressed by two translated attributescompared to one (χ2(1) � 9.16, p < 0.01). In addi-tion, and in line with our prediction, the relationshipbetween environmental attitudes and the probabilityof choosing the fuel-efficient car was moderated bythe presence of the GGR attribute, as indicated by asignificant GGRpresent×NEP-R interaction (χ2(1)� 12.2,p < 0.001).1 Finally, a comparison with a model thatalso included a number of attributes × NEP-R inter-action showed that the relationship between environ-mental attitudes and the probability of choosing thefuel-efficient car was not moderated by the number ofattributes presented (i.e., number of attribute×NEP-Rinteraction, χ2(1)� 0.62, p > 0.05).

2.3. DiscussionThe observations made in Experiment 1 illustrate thatboth the number of attributes and the type of attributes

independently influence people’s preferences. First,presenting more translated fuel economy attributes in-creased the likelihood of choosing the more fuel-effi-cient car, regardless of the type of attributes. Second,and in line with our hypothesized signpost effect, thepresentation of an environmental translated attributeincreased the likelihood of choosing the more fuel-efficient car, regardless of the number of total presentedattributes. Importantly, the increase in fuel-efficientchoices was driven by those with higher proenviron-mental values and cannot be explained by a simplecounting heuristic. Rather, it appears that the pres-ence of the environmental translated attribute acti-vated respondents’ preexisting but latent proenviron-mental value, and subsequently aligned choices withpersonal values.

An alternative explanation for our observations isthat participants may have received different informa-tion when presented with one attribute combinationcompared to another. Therefore, the purpose of Exper-iment 2 was to contrast our activation-and-directionaccount with this potential informational account.

3. Experiment 2—Effects of InformationAccording to our value-activation-and-direction ac-count, translated attributes serve as decision-signposts:They activate a person’s preexisting objectives and thenhelp the person to identify which option best alignswith their activated objective. According to the infor-mational account, a personmight fail to realize that cer-tain attributes are translations of one another and, as aresult of this oversight, have a different knowledge basewith relation to one attribute combination compared toanother. Moreover, a person might gain different infor-mation about the relation between attributes through-out the experiment depending on the experimentalmanipulation they were assigned to. For example, par-ticipants presentedwith a greenhouse gas ratingmightfind it easier to learn the relation between environmen-tal impact and annual fuel cost compared to partici-pants presented with gallons per 100 miles and annualfuel cost.

To rule out this informational account, we replicatedExperiment 1 while manipulating whether knowledgeabout the relation between translated attributes wasmeasured before versus after the choice task. If theinformational account were true, we would expect thatknowledge regarding the relation between translatedattributes would be different before versus after thechoice task, or as a function of which attributes werepresented during the choice task.

3.1. Methods3.1.1. Participants. Seven hundred ninety-nine Ameri-can participants (44% female; mean age� 31; SD� 11.2)were recruited through MTurk and paid a flat fee fortheir participation.

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


3.1.2. Design. The design was a simplified version ofExperiment 1. Specially, we used a 2 (translated at-tribute: GPM vs. GGR)× 2 (knowledgemeasure: beforevs. after choice task) between-subjects design. For eachcar, participants were presented with price, AFC, andone of the translated attribute variables (i.e., GPM orGGR). To limit opportunities to learn about the correla-tionbetween the translatedattributes,participantswerepresented with a single choice scenario in this experi-ment (Problem 2 from Experiment 1; see Table 1).3.1.3. Materials. We measured knowledge of the rela-tion between AFC andGGR in twoways. First, we useda general “relatedness” measure that asked partici-pants how related the two metrics were on a 10-pointscale from “not at all related” to “completely related.”On this same page and using the same scale, we alsoasked how related AFC and engine longevity were,and how related GGR and crash safety rating were (seeOnline Appendix B). These two additional questionsmeasured comparisons with unrelated variables andwere expected to produce very low relatedness scores,thereby allowing us to evaluate the validity of our relat-edness scale. Second, we used directional questionsthat presented the participant with a list of AFC val-ues ($3,800, $2,900, $2,150, $1,650) in random order, forwhich matching scores had to be identified within alist of GGR values ranging from 1 to 10 (the answerswere 2, 4, 6, and 8, respectively). This question allowedus to determine whether participants correctly iden-tified the (negative) direction of the relation betweenAFC and GGR.3.1.4. Procedure. Participants for whom knowledgewas to be measured prior to choice were presentedwith the current U.S. fuel economy label (see Fig-ure 1), together with definitions of AFC and GGR,and answered the three relatedness questions. On thenext page, participants answered the four directionalquestions. They then completed the choice task, whichwas followed by two attention check questions. Finally,we asked participants to complete the NEP-R scaleand demographic questions. Participants for whomknowledge was to be measured after choice had asimilar experience; however, the choice task was pre-sented first.

3.2. ResultsThirty-eight participants incorrectly answered bothattention check questions and were removed from thesubsequent analysis, leaving 761 participants. How-ever, as described below, the conclusions remain thesame even when these excluded data are included inthe analysis.3.2.1. Knowledge. We first contrasted the three relat-edness questions to lend support to the use of related-ness as a valid measure. The mean relatedness score

between AFC and GGR was 6.4 out of 10 (SD� 2.4).The mean relatedness score between AFC and enginelongevity was 4.2 out of 10 (SD � 2.6). The mean relat-edness score between GGR and crash safety rating was2.1 out of 10 (SD � 1.9). The mean relatedness scorebetween AFC and GGR was significantly higher thanthe related scores between the other two comparisons(both p’s < 0.001), suggesting that the relatedness mea-sure was valid.

We next examined whether participants’ responseto the relatedness between AFC and GGR varied asa function of experimental group. An inspection ofthe means, which ranged between 6.2 and 6.6, sug-gested little variance between groups. An analysis ofvariance with translated attribute (GPM vs. GGR) andknowledge measure timing (before vs. after choicetask) entered as independent variables and relatednessentered as the dependent variable revealed no effectof translated attribute (F(1, 760) � 1.54, p � 0.21), noeffect of knowledge measure timing (F(1, 760) � 0.51,p � 0.48), and no interaction between these two vari-ables (F(1, 760) � 0.03, p � 0.84).2 Moreover, the distri-bution of responses to the relatedness question was notsignificantly different between those who answeredthe knowledge questions before versus after the choicetask (two-sample Kolmogorov–Smirnov test, DMax �

0.04, p � 0.94). Analyses using the data collected fromthe four directional questions produced similarly nullresults (see the electronic companion). In summary,there was no evidence for a difference in knowledgeregarding the relation between AFC and GGR as afunction of which label was presented or whether theknowledge questions were asked before or after thechoice task.3.2.2. Choice. As can be seen in Figure 4, the pro-portion of participants choosing the fuel-efficient carwas higher when the environmental attribute GGRwas present compared to when it was absent (i.e.,when GPM was presented). Moreover, this differenceappears to hold regardless of when the knowledgemeasure was administered. To statistically confirmthis interpretation, we ran a logistical regression withtranslated attribute (GPM vs. GGR) and knowledgemeasure timing (before vs. after choice task) enteredas independent variables, and choice as the dependentvariable. The analysis revealed a main effect of trans-lated attribute (χ2(1, N � 761) � 48.32, p < 0.001), noeffect of knowledge measure timing (χ2(1, N � 761) �1.36, p � 0.24), and no interaction between these twovariables (χ2(1, N � 761) � 1.41, p � 0.24).3 In otherwords, participants were significantly more likely toprefer the efficient car when presented with AFC andGGR compared to AFC and GPM, irrespective of thetiming of the knowledge question.

To demonstrate a signpost effect for Experiment 2,we ran a second logistic regression, this time also

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


Figure 4. Proportion of Choices for the Fuel-EfficientOption in Experiment 2 as a Function of the Timing of theKnowledge Measure and Whether an EnvironmentalAttribute Was Absent or Present

Pro

port

ion

choo

sing

the

fuel

-effi

cien

t car

0

1

Before After

Knowledge question timing

Environmental attributeAbsentPresent

adding themean-centeredNEP-R scores as an indepen-dent variable, which was also crossed with the otherindependent variables. Again, the analysis revealed aneffect of translated attribute (χ2(1, N � 761) � 48.14,p < 0.0001), no effect of knowledge measure timing(χ2(1, N � 761) � 2.31, p � 0.13), and no interactionbetween these two variables (χ2(1, N � 761) � 1.55, p �

0.21). Moreover, there was a significant effect of NEP-Rscores (χ2(1, N � 761)� 4.52, p � 0.03). Importantly, andaspredicted, therewasa significant interactionbetweentranslated attribute andNEP-R scores (χ2(1, N � 761) �15.01, p � 0.0002),4 indicating that participantswere sig-nificantly more likely to prefer the efficient car whenpresented with AFC and GGR compared to AFC andGPM, and this was especially true for those who had arelatively high NEP-R score. This pattern is illustratedin Figure 5.Finally, there was no correlation between choice and

relatedness (Spearman’s ρ � 0.01, p � 0.73), which sug-gests that choices were not influenced by the knowl-edge that participants possessed regarding the relationbetween AFC and GGR.

3.3. DiscussionThe observations made in Experiment 2 replicated thesignpost effect found in Experiment 1: Participantstended to have a stronger preference for the fuel-effi-cient car when presented with AFC and GGR com-pared to AFC and GPM, especially for those withhigher proenvironmental attitudes. Importantly, thiseffect cannot be explained by a knowledge account be-cause understanding of the relationship between AFC

Figure 5. Probability of Choosing the Fuel-Efficient Car inExperiment 2 as a Function of Environmental Values andWhether an Environmental Attribute Was Present or Not

and GGR did not differ before versus after participat-ing in the choice task, and regardless of which trans-lated attribute was presented. In addition, we observedthe signpost effect within a single choice trial, rulingout a learnt knowledge account. Moreover, the ten-dency to choose the efficient car was not determined bythe participants’ perceived relationship between AFCand GGR. Therefore, these observations lend furthersupport to our value-activation-and-direction accountof the data: individuals possess a collection of knowl-edge and values, and different translated attributescan activate and direct otherwise dormant objectives,which only then impact upon choices.

4. Experiment 3—Boundary Conditionsfor Signposts

In our conceptual framework, the signpost effect oper-ates by the following two processes: (1) a translatedattribute activates some latent objective of the person(e.g., a value, goal, etc.), and (2) the translated attributedirects the people to the option that is most congru-ent with the activated objective. This conceptualizationpredicts several conditions under which a translatedattribute should not cause a signpost effect. First, theeffect should not occur if the objective associated withthe translated attributes is already activated. Recallthat we expect that presenting the greenhouse gasrating attribute will cause participants with preexist-ing proenvironmental values to weight environmentalinformation more heavily during choice. However, wewould not expect this if, for some reason, the partic-ipants were already thinking about how much theyvalue the environment.

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


The second process would not operate if the optionmost congruent with the activated objective werealready known. This could be possible if participantswere explicitly taught the relationship between at-tributes and, more importantly, the direction of therelationship such that they could independently makethe translation. For example, participants who haveproenvironmental objectives and understand that ahigh annual fuel cost implies high greenhouse gasemissions could use the first attribute to infer thesecond and thus act on their objectives in spite ofthe incongruence between the presented translatedattribute (i.e., annual fuel cost) and their activatedobjectives (i.e., proenvironmental behavior).

In Experiment 3, we tested the extent to which bothof these mechanisms underlie the signpost effect usingthe car choice paradigm introduced in Experiments 1and 2. We activated participants’ environmental values(or not) prior to choice by having them fill out an envi-ronmental value questionnaire before (vs. after) choos-ing between the cars. In addition, we educated theparticipants (or not) about the direction of the relation-ship between the annual fuel cost and environmentalimpact prior to the choice phase through a tutorial.

We predicted that in situations where participantswere not reminded of their environmental valuesor not educated about the direction of the relation-ship between annual fuel cost and the environmentalimpact, we would replicate the original signpost effect:participants’ likelihood of choosing the fuel-efficientcar would be aligned with their environmental val-ues when presented with the greenhouse gas ratingattribute but not when presented with the annual fuelcost attribute. In contrast, we predicted that when par-ticipants were reminded of their environmental val-ues or educated about the direction of the relation-ship between annual fuel cost and the environmentalimpact prior to choice, then the signpost effect wouldnot be observed: participants’ likelihood of choosingthe fuel-efficient car would be aligned with their envi-ronmental values regardless of which attribute waspresented.

Finally, the measurement of environmental attitudesbefore and after the presentation of the translatedattributes as part of the design also allowed us to test analternative explanation for the signpost effect observedin Experiments 1 and 2, namely, that the translatedattribute created new environmental values that theparticipant originally did not possess. If our manip-ulation of presenting environmental attribute transla-tions did indeed create new values, then we wouldexpect the distribution of environmental values to dif-fer between experimental groups and expect higherenvironmental attitude scores after the decision tasksinvolving the greenhouse ratings. However, if environ-mental attribute translations serve only as signposts for

existing values, as predicted by our own account, wewould expect the distribution of environmental atti-tudes to be similar across experimental groups andstages of the experiment.

4.1. Methods4.1.1. Participants. Six hundred six American partic-ipants (58% female; mean age � 31.6; SD � 10.8) wererecruited through MTurk and paid a flat fee for theirparticipation.

4.1.2. Design. We employed a 2 (translation: annualfuel cost vs. greenhouse gas rating)×2 (tutorial: presentvs. absent)× 2 (order of NEP-R scale: before choice vs.after choice) mixed design. The main dependent vari-able was whether or not the participant selected thefuel-efficient car.

4.1.3. Materials and Procedure. Participants were ran-domly assigned to one of the eight conditions andpresented with the same nine car choices used inExperiment 1. For each choice pair, participants viewedthe price of the car and one fuel-efficiency translation(see Online Appendix C). Half of the participants werepresented with the annual fuel cost, and the other halfwith the greenhouse gas rating. Half of the participantscompleted the NEP-R scale before the choice task, andthe other half completed it after the choice task. Inaddition, half of the participants watched a brief ani-mation explicitly describing the direction of the rela-tionship between the annual fuel cost and the green-house gas emissions before the choice task and after theNEP-R scale (see Online Appendix D). To complete thetutorial, participants had to correctly answer a questionregarding the positive relationship between the twovariables. After the nine choices, all participants had toanswer two manipulation check questions to evaluatewhether they had understood the relationship betweenannual fuel cost and the greenhouse gas emissions ofthe car. The first question asked participants to ratehow strong the relationship between the annual fuelcost and the greenhouse gas emissions of a car is on afive-point scale (“unrelated” to “completely related”).The second question asked participants to select thecar with the fewest greenhouse gas emissions out ofa set of four cars described only by their annual fuelcost. Finally, participants provided basic demographicinformation. As there were no attention check ques-tions, the data from all 606 participants were used forthe subsequent analysis.

4.2. Results4.2.1. Manipulation Check. To assess whether our tu-torial manipulation was successful in explaining therelationship between annual fuel cost and the green-house gas emissions of a car, we analyzed the perceivedrelationship strength between these attributes across

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


the tutorial conditions. As expected, participants whowent through the tutorial reported a stronger rela-tionship between a car’s annual fuel cost and itsgreenhouse gas emissions (mean � 3.94, SD � 0.86)than participants who did not go through the tutorial(mean � 3.49, SD � 0.84, t(601.6) � −6.47, p < 0.001).Participants who went through the tutorial were alsomore likely to correctly identify the car with the lowestemissions (85%) than participants who did not (78%,χ2(1, N � 604)� 10.52, p < 0.01).We also tested whether the tutorial affected par-

ticipant’s environmental values by comparing NEP-Rscores between conditions. We found no significantdifferences between NEP-R scores of participants whowent through the tutorial (mean � 53.31, SD � 9.73)and participants who did not (mean� 53.92, SD� 9.59,t(603.73) � 0.77, p � 0.44). Similarly, there was no dif-ference between the NEP-R scores of participants whoanswered the NEP-R questions before (mean � 53.64,SD� 9.41) and after the choice task (mean� 53.60, SD�

9.91, t(602.37) � 0.05, p � 0.963). Finally, there was nodifference between the NEP-R scores of participantswho had seen the greenhouse gas rating (mean� 54.03,SD � 9.61) and participants who had seen the annualfuel cost (mean � 53.21, SD � 9.70, t(603.99) � −1.04,p � 0.3). Thus, there was no evidence for the creation ofenvironmental values as a result of our experimentalmanipulations.

4.2.2. Preferences. To investigate whether the sign-post effect was observed across the different experi-mental conditions, we examined again the observedlikelihood of choosing the environmental option acrossparticipants’ environmental attitudes. This relation-ship is visualized in Figure 6, where we see a signposteffect replicated only in the lower left panel: a strongpositive relationship between NEP-R scores and theproportion of fuel-efficient car choices was observedfor participants presented with the greenhouse gas rat-ing (gray line). In contrast, for participants presentedwith the same information as annual fuel cost (blackline), we observed no (or a slightly negative) relation-ship between the NEP-R scores and the proportion offuel-efficient choices.Comparing the lower left panel with the other pan-

els, we see that this signpost effect was not observedin the remaining panels. Participants made choicesrelated to their environmental preferences regardlessof whether GGR was present or absent when (1) thetutorial explained the relationship between annual fuelcost and the greenhouse gas emissions (right panels,Figure 6) or (2) the NEP-R scale was presented beforechoice (top panels, Figure 6). When participants werereminded of their environmental values or how touse fuel cost as a proxy for environmental outcomes,they did not need a signpost; the signpost effect was

Figure 6. Percentage of Fuel-Efficient Car Choices inExperiment 3 as a Function of Environmental Values(Centered NEP-R Scores), Type of Fuel-Efficiency Attribute,Whether the NEP-R Scale Was Presented First or Last, andWhether a Tutorial Was Presented or Not

Rt

R

observed only when there was no tutorial and theNEP-R scale was presented after choice.

This observed effect of our manipulations and theirinteractions with the relationship between environ-mental values and choice was confirmed using mul-tilevel logistic regression models with the selection ofthe fuel-efficient car as the dependent variable, randomintercepts for participants and choice problems andfixed effects for the type of attribute presented (AFC vs.GGR), the presence of the tutorial (present vs. absent),the order of theNEP-R scale (first vs. last), and centeredNEP-R scores. To examine the effects of the manipula-tions on the interaction between type of attribute andNEP-R scores (the signpost effect), we also includedthe higher order interactions between the independentvariables.

In linewith theproposedmechanismsunderlying thesignpost effect, the model indicated that the effect ofNEP-R scores on the probability of choosing the envi-ronmental option depended on the type of attributepresented and was moderated by the presence of thetutorial and the order of the NEP-R question: the four-way interaction (tutorial ×NEP-R order×GGRpresent ×NEP-R score) was significant (χ2(11)� 32.5, p < 0.001).

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


Next, we broke down the higher order interaction toexamine the ability of theGGRattribute to align choiceswith personal values by running separate models forthe different attribute groups. When the same modelwas run only for the group of participants who saw theAFC attribute, the three-way interaction indicated thatthe effect of NEP-R scores on choosing the efficient cardepended on the presence of the tutorial and the orderof the NEP-R score (tutorial × NEP-R order × NEP-Rscore interaction; χ2(4)� 14.0, p < 0.01).However,whenthe same model was run only for the group of partici-pants who saw the GGR attribute, there was no signifi-cant three-way interaction: χ2(4)� 6.32, p > 0.05. This isin line with our assumption that the effect of the NEP-Rscore on choice was facilitated by the presence of thesignpost itself (GGR attribute) and observed regardlessof the presence of the tutorial (χ2(1) � 0.02, p � 0.9) orthe order of the NEP-R score (χ2(1) � 0.16, p � 0.69) ortheir interaction (χ2(4)� 6.32, p � 0.18).

4.3. DiscussionIn Experiment 3, we replicated the signpost effect intro-duced in Experiments 1 and 2. In addition, we showedthat the signpost effect was only observed under con-ditions in line with our proposed mechanism. Specifi-cally, we demonstrated thatwhen participantswere nottaught the direction of the relationship between annualfuel cost and the environmental impact (i.e., no tuto-rial) and their environmental values were not activated(i.e., NEP-R scale after choice), the environmental GGRattribute acted as a signpost providing both activationand direction. As a result, we replicated the originalsignpost effect: participants’ likelihood of choosing thefuel-efficient car was aligned with their environmentalvalues when presented with the GGR attribute but notwhen presented with the AFC attribute.In contrast, when participants were taught the direc-

tion of the relationship between annual fuel cost andthe environmental impact prior to choice (i.e., via thetutorial) or reminded of their environmental values(i.e., by answering the NEP-R scale before choice), acti-vation and direction were already provided withoutthe aid of the signpost. As a result, we observed choicesalignedwith personal environmental values regardlessof the attribute presented (i.e., GGR or AFC). Thus,we showed that activation of congruent values andcommunication of information regarding the relation-ship between the attribute and the objective could beachieved either via reflection on dormant values and atutorial or via the presentation of a translated attributethat could act as a signpost.

In addition, our finding of similar NEP-R score dis-tributions across experimental groups demonstratedthat the observed signpost effects cannot be explainedthrough the creation of environmental values as a resultof our experimental manipulations. Furthermore, the

signpost effect was observed despite an equal numberof attributes presented to all participants, which rulesout the use of a counting heuristic as an explanation inthese observations.

5. General DiscussionIn this paper, we examined the impact of presentingtranslated attributes on choices. Translated attributesrefer to information that is, in principle, already avail-able by a simple transformation via a change in scale ofother, known information. However, given that peopletend to construct preferences on the fly based on avail-able information (e.g., Lichtenstein and Slovic 2006;Payne et al. 1988, 1993; Ungemach et al. 2011), andoften without considering all the objectives they careabout (Bond et al. 2008), we hypothesized and thenobserved that choices were indeed influenced by thepresentation of different translated attributes that pro-vided opportunities to consider and pursue importantbut initially neglected objectives.

5.1. Summary of ExperimentsIn three experiments, we have presented evidencehighlighting how translated attributes can affect thedecisions that people make. In Experiment 1, wedemonstrated how translated attributes could serve asdecision signposts. A signpost is a translated attributethat helps people choose in line with their personalobjectives. We found that value activation through thepresentation of translated attributes was moderated bycongruent preexisting proenvironmental values: Par-ticipants’ likelihood of selecting the fuel-efficient carincreased with proenvironmental value strength, butonly when an environmental translated attribute waspresented. Importantly, we showed that the effects oftranslated attributes on choice were caused by boththe number of attributes presented and the type ofattributes presented. More important for our contribu-tion, the signpost effect in this experiment was drivenby the presence of an environmental attribute, whichhighlights the activation feature of translated attributes.

In Experiment 2, we replicated the signpost effect,this time showing that the signpost effect could notbe explained by differences regarding the understand-ing and learning of the relationship between differ-ent translated attributes. Finally, in Experiment 3, wedemonstrated that the signpost effect was producedby two separate mechanisms by explicitly turning onand off the activation and direction features of translatedattributes. We also ruled out the possibility that pro-environmental values were created through the simplepresentation of environmental attributes.

5.2. Theoretical ImplicationsIn this paper we have introduced the concept of a“translated attribute” and repeatedly demonstrated

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


how the presentation of translated attributes can in-fluence choice by serving as a decision “signpost.”A decision signpost has two important features that aretheoretically important. First, a decision signpost canhelp people make a decision by reminding them of apersonally held objective. This is the activation featureof translated attributes. In our experiments, objectivesthat translated attributes could signpost included per-sonal values and goals. The activation effect of sign-posts is related to the finding that subliminally pre-sented primes about goal achievement means (e.g.,“study”) can activate the associated goal (e.g., “learn-ing”; Shah and Kruglanski 2003). Our work shows thatsuch bottom-up activation is also possible via explicitlystated product attributes.Moreover, our observation of an interaction between

translated attributes and personal attitudes impliesthat themechanism only affects the preferences of indi-viduals who hold relevant objectives, which is clearlydistinct from the effects of a pure counting heuristic.Just as primes can activate different associations in dif-ferent people (e.g., Wheeler and Berger 2007), so toocan attributes activate different objectives. Thus, trans-lated attributes have an ability to help overcome short-comings in the generation process (Bond et al. 2008) byaiding the recognition of existing objectives.

Second, a decision signpost can help point out theoption that is better aligned with this activated ob-jective. This is the directional feature of translated at-tributes. For example, a person attempting to act upona proenvironmental value but presented only withannual fuel cost information may fail to appreciatewhich option is most value congruent. The translatedattribute “greenhouse gas rating” could eliminate suchconfusion. This point marks another clear and vitaldistinction between primes and signposts: The for-mer is limited to merely nonconsciously activating anobjective, whereas the latter additionally and explic-itly reveals the information necessary to achieve theobjective.We have proposed that translated attributes operate

by reminding decision makers of goals they wish toachieve and directing them to how to achieve them.The activation of different sets of knowledge is sim-ilar to many other constructs in judgment and deci-sion making research. For example, Levin et al. (1998)have discussed framing in three contexts—risk, choice,and attributes—by building on a large literature injudgment and decision making (e.g., Kahneman andTversky 1984, Tversky and Kahneman 1981). Levinet al. (1998) argue that judgments involving funda-mentally identical quantitative outcomes yield differ-ent behaviors depending on whether the outcomes aredescribed as gains or losses. This work on attributeframing, however, focuses narrowly on valence. Theidea of framing has received a broader treatment in

other literatures, such as political psychology, whichdescribes how competing issue frames in politicsemphasize different facts, values, and relationships forcomplex problems (see Nelson 2011). For example,automobile speed limits could be framed as a “publichealth” issue (lower speed would save lives and reducehealth care costs) or as a “free choice” issue (speedlimits impede free choice). Framing research in judg-ment and decisionmaking focuses on holding constantthe key quantitative information across frames; fram-ing research in political psychology, on the other hand,recognizes that frames often change beliefs and prefer-ences because they add new information (e.g., on casu-alties caused by accidents) and omit other informa-tion (e.g., added hours spent driving). We believe thatthe psychology of translated attributes falls in an areabetween many of the traditional framing effects stud-ied in judgment and decision making, which often relyon changes in valence for identical outcomes (Levinet al. 1998), and the more expansive notion of fram-ing in political psychology, which often assumes theintroduction of novel facts. For example, the fact thatcar fuel economy affects driving costs and the envi-ronment is not surprising to most consumers. How-ever, the salience of those dimensions when evaluatingalternatives and their precise magnitude can both beneglected during choice.

Alternatively, it could be argued that the choicearchitects’ selection of attributes might leak informa-tion beyond the literal content indicating the archi-tect’s own preferences or implicit recommendation tothe decision maker (e.g., McKenzie and Nelson 2003,Sher andMacKenzie 2006). For example, assuming thatleakage can come from any source, including policymakers (e.g., McKenzie et al. 2006), the inclusion ofgreenhouse gas rating on national fuel economy labelscould be interpreted as evidence that policy makerswant consumers to choose more environmental cars.This is a plausible account, but it would require addi-tional assumptions to make clear predictions consis-tent with the signpost effect. For example, includingGPM or GGR on a label would leak similar infor-mation about the choice architect’s preferences. How-ever, in Experiment 2 only GGR changed behaviorin a value-consistent way (Figure 4). Moreover, with-out further assumptions, information leakage does notexplain the observed interaction between the trans-lated attribute and individual differences in latent val-ues. Future research can try to disentangle whether asimple activation-and-direction account is sufficient toexplain this behavior or whether an information leak-age account is necessary. For example, one strategy forteasing them apart would be to vary the descriptionof the process by which the presented attributes aregenerated, such as whether the attributes were chosen

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


by a designer or randomly drawn from a large pool ofattributes varying in importance.We argue that translated attributes represent a new

type of choice architecture intervention. Some exam-ples of choice architecture can be described as psy-chological “tricks” by capitalizing on common cogni-tive biases. For example, in Experiment 1, we showedthat provision of translated attributes could shift pref-erences in a predicable direction by simply addingmore translations. Such an outcome capitalizes on peo-ple’s tendency to rely on counting heuristics, that is,favoring options that appear to have more superiorcharacteristics. However, across all of our experiments,we also showed that provision of translated attributescould shift preferences in a direction contingent on theperson’s unique objectives. Instead of steering peopleinto choices irrespective of their personal objectives,translated attributes that work as signposts guide peo-ple in the direction they want to go. Therefore, trans-lated attributes could potentially point different peo-ple in different directions. For example, in our exper-iments, the presentation of the greenhouse gas ratinghighlighted the environmental aspect of fuel efficiency,which mainly influenced those with more proenvi-ronmental attitudes. Similarly, individual differencesin concern for future financial costs (e.g., Lynch et al.2010) might predict preferences for fuel-efficient carsmost strongly when future cost attributes are presentand weakly when they are absent. Thus, instead ofrestricting an individuals’ autonomy and ability toact upon her own preferences—a criticism sometimesused against choice architecture (e.g., Hausman andWelsh 2010)—signposts epitomize a new class of choicearchitecture that is concerned with the alignment ofbehavior with existing objectives. While much of thepopular discussion of choice architecture surroundsnudging, this discussion neglects the primary role ofchoice architecture: helping peoplemake better choicesfor themselves. Signposts represent a tool consistentwith this purpose.

5.3. Practical ImplicationsApart from the psychological insights gained by iden-tifying two separate mechanisms of translated attri-butes—activation and direction—there are also impor-tant practical implications for the design of potentialchoice interventions using translated attributes as ap-propriate choice architecture. Clearly, the possibility ofactivating preexisting objectives that are aligned withsocietal goals makes signposts an interesting choicearchitecture tool for policy makers. For example, theU.S. EPA fuel economy and environment label for newcars provides seven translated attributes for fuel econ-omy, including miles per gallon, annual fuel cost, anda greenhouse gas rating, all of which are highly corre-lated with each other but highlight different objectives

related to fuel economy (Figure 1). Many other productlabels include multiple translated attributes. For exam-ple, food labels describe nutrients in terms of amountper serving (55 mg of sodium) as well as a percentage(2%) of recommended daily values. Home appliances,such as air conditioning (AC) units, state estimatedyearly electricity use in kilowatt hours as well as esti-mated yearly operating cost in dollars. Indeed, we havealso replicated our results in other domains that didnot involve cars, including choices between AC unitsand food options. For example, in one study involv-ing choices between AC units, participants made deci-sions that were much more aligned with their personalenvironmental values when we replaced the techni-cal labels “Seasonal Energy Efficiency Rating” and“BTU/Wattage” with “Environmental Rating.” There-fore, merely changing the label of a metric while keep-ing the actual attribute value constant appears to besufficient to produce a signpost effect.

An initial concern when contemplating the value oftranslated attributes is the potential liability of over-loading people with information (e.g., Jacoby 1984, Leeand Lee 2004). Ourwork shows that the presentation oftranslated attributes actually facilitates decision mak-ing by better aligning objectives and choices. Nonethe-less, there remains a question of coverage: How manytranslations to provide while not overloading peoplewith information? We propose that the optimal designwould provide the minimum information needed tocover most important key objectives that vary withinthe target population. We have not explicitly studied alabel with as many attributes as are present on the cur-rent EPA label and, therefore, cannot say whether all ofthe information on the label is useful in decision mak-ing. However, one possibility is that the presentation ofa large number of different translated attributes causesan individual to selectively seek out the attributes thatseem most relevant to addressing his or her own per-sonal objectives. If that were the case, then we wouldexpect the presentation of many translated attributesto have no negative consequences on decision mak-ing beyond the additional effort required to identifypersonally relevant attributes. Nevertheless, the selec-tion of successful signposts requires careful considera-tion of the specific population segment being targeted,the objectives that are important to this segment, andmatching attributes that have the potential to activateand direct those objectives.More generally, translated attributes could be used

as signposts in many other contexts, including onlinecustomization. For example, the website http://www.fueleconomy.gov allows a person to compare the fueleconomy of different vehicles. The information is pre-sented under different tabs, which the user is freeto click on. One tab of information includes fuel-efficiency attributes such as gallons of fuel per 100

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

http://www.fueleconomy.gov

http://www.fueleconomy.gov


miles,MPG, and annual fuel cost. A second tab of infor-mation includes greenhouse gas emissions per mileand annual barrels of gasoline consumed. Although allof these pieces of information are translated attributes,the website user has some control over what informa-tion is presented. We can easily imagine a more cus-tomizable interface that enables the presentation of tai-lored information to specific users or user segments.Obviously, the effectiveness of translated attributes is

conditional on an appropriate presentation of the infor-mation. Potential benefits of translated attributesmightnot be realized in designs that impede the detection ofthe critical attributes because of the amount or formatof the presented information. An important prerequi-site for successful application of a signpost strategy isthe identification of specific attributes that are mosteffective at reminding people of the important objec-tives they care about.

Finally, signpost effects are not necessarily limitedto attribute translation. Other design features or pre-sentation formats might be used to highlight relevantaspects and guide choice in a similar way. For exam-ple, research on food labeling has shown the benefit ofusing simple color schemes to indicate better (green)and worse (red) levels of an attribute, such as saturatedfat, to help people quickly assess which food optionsare healthy and which are not (Hawley et al. 2013).

5.4. ConclusionIn summary, we have introduced the new conceptof translated attributes as decision signposts—relatedattributes derived from a global dimension by sim-ple transformations that can overcome shortcomingsin the generation of choice objectives and align pref-erences with existing objectives. The effect of decisionsignposts on choice is robust in part because two sep-arate psychological mechanisms support it indepen-dently: activation of inherent objectives such as valuesand goals, and provision of directional information forthe individual to act upon. The utilization of translatedattributes as decision signposts offers a new type ofchoice architecture and will be useful for both policymakers and managers alike as they strive to effectivelycommunicate and guide informed choices.

AcknowledgmentsThe first two authors contributed equally to this project, andthe order of the last three is alphabetical. The authors thankthe department editor and the reviewers for their helpful sug-gestions and comments. This paper also benefited from help-ful discussions with the members of the Center for Researchon Environmental Decisions at Columbia University.

Endnotes1Similar results were observed when also including the data of par-ticipants who failed the attention check. Participants were morelikely to select the fuel-efficient option when fuel efficiency was

expressed by two translated attributes compared to one (χ2(1)� 8.78,p < 0.005), and the relationship between environmental attitudesand the probability of choosing the fuel-efficient car was moderatedby the presence of the GGR attribute, as indicated by a significantGGRpresent ×NEP-R interaction (χ2(1)� 15.68, p < 0.001).2The same analysis, this time including all of the data, also revealedno effect of translated attribute (F(1, 798) � 1.11, p � 0.29), no effectof knowledge measure (F(1, 798)� 0.58, p � 0.44), and no interactionbetween these two variables (F(1, 798)� 0.00, p � 0.99).3The same analysis, this time including all of the data, also revealed amain effect of translated attribute (χ2(1, N � 799)� 52.01, p < 0.001),no effect of knowledge measure (χ2(1, N � 799)� 1.97, p � 0.16), andno interaction between these two variables (χ2(1, N � 799) � 1.33,p � 0.25).4The same analysis, this time including all of the data, also revealed asignificant interaction between translated attribute andNEP-R scores(χ2(1, N � 799)� 14.25, p � 0.0002).

ReferencesAlba JW, Marmorstein H (1987) The effects of frequency knowledge

on consumer decision making. J. Consumer Res. 14(1):14–25.Bond SD, Carlson KA, Keeney RL (2008) Generating objectives: Can

decision makers articulate what they want? Management Sci.54(1):56–70.

Camilleri AR, Larrick RP (2014) Metric and scale design as choicearchitecture tools. J. Public Policy Marketing 33(1):108–125.

Camilleri AR, Larrick RP (2015) Choice architecture. Scott R, KosslynS, eds. Emerging Trends in the Social and Behavioral Sciences (JohnWiley & Sons, Hoboken, NJ), 1–15.

Cohen MA, Vandenbergh MP (2012) The potential role of carbonlabeling in a green economy. Energy Econom. 34:S53–S63.

Dawes RM (1979) The robust beauty of improper linear models indecision making. Amer. Psychologist 34(7):571–582.

Dunlap RE, Van Liere KD, Mertig AG, Jones RE (2000) New trends inmeasuring environmental attitudes: Measuring endorsement ofthe new ecological paradigm: A revised NEP scale. J. Soc. Issues56(3):425–442.

Feather NT (1995) Values, valences, and choice: The influences ofvalues on the perceived attractiveness and choice of alternatives.J. Personality Soc. Psych. 68(6):1135–1151.

Fischhoff B, Slovic P, Lichtenstein S (1978) Fault trees: Sensitiv-ity of estimated failure probabilities to problem representa-tion. J. Experiment. Psych.: Human Perception and Performance 4(2):330–344.

Fox CR, Ratner RK, Lieb DS (2005) How subjective grouping ofoptions influences choice and allocation: Diversification biasand the phenomenon of partition dependence. J. Experiment.Psych.: General 134(4):538–551.

Hawley KL, Roberto CA, Bragg MA, Liu PJ, Schwartz MB, BrownellKD (2013) The science on front-of-package food labels. PublicHealth Nutrition 16(3):430–439.

Hausman DM, Welch B (2010) Debate: To nudge or not to nudge.J. Political Philosophy 18(1):123–136.

Huang SC, Zhang Y, Broniarczyk SM (2012) So near and yet so far:The mental representation of goal progress. J. Personality Soc.Psych. 103(2):225–241.

IPCC (Intergovernmental Panel on Climate Change (2013) ClimateChange 2013: The Physical Science Basis. Contribution of WorkingGroup I to the Fifth Assessment Report of the IntergovernmentalPanel on Climate Change, Stocker TF, Qin C, Plattner G-K, Tig-nor M, Allen SK, Boschung J, Nauels A, Xia Y, Bex V, MidgleyPM, eds. (Cambridge University Press, Cambridge, UK), doi:10.1017/CBO9781107415324.

Jacoby J (1984) Perspectives on information overload. J. Consumer Res.10(4):432–435.

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.


Johnson EJ, Goldstein DG (2003) Do defaults save lives? Science302(5649):1338–1339.

Johnson EJ, Shu SB, Dellaert BG, Fox CR, Goldstein DG, Häubl G,Larrick RP, et al. (2012) Beyond nudges: Tools of a choice archi-tecture. Marketing Lett. 23(2):487–504.

Kahneman D (2011) Thinking, Fast and Slow (Farrar, Straus & Giroux,New York).

Kahneman D, Tversky A (1984) Choices, values and frames. Amer.Psychologist 39(1):241–350.

Keeney RL (1992) Value-Focused Thinking: A Path to Creative Decision-Making (Harvard University Press, Cambridge, MA).

Kling JR, Mullainathan S, Shafir E, Vermeulen L, Wrobel M (2012)Comparison friction: Experimental evidence from Medicaredrug plans. Quart. J. Econom. 127(1):199–235.

Larrick RP, Soll JB (2008) The MPG illusion. Science 320(5883):1593–1594.

Lee BK, Lee WN (2004) The effect of information overload on con-sumer choice quality in an on-line environment.Psych. Marketing21(3):159–183.

Levin IP, Schneider SL, Gaeth GJ (1998) All frames are not createdequal: A typology and critical analysis of framing effects. Organ.Behav. Human Decision Processes 76(2):149–188.

Lichtenstein S, Slovic P (2006) The Construction of Preference (Cam-bridge University Press, Cambridge, UK).

Lynch JG, Netemeyer RG, Spiller SA, Zammit A (2010). A general-izable scale of propensity to plan: The long and the short ofplanning for time and for money. J. Consumer Res. 37(1):108–128.

McKenzie CR, Nelson JD (2003) What a speaker’s choice of framereveals: Reference points, frame selection, and framing effects.Psychonomic Bull. Rev. 10(3):596–602.

McKenzie CR, Liersch MJ, Finkelstein SR (2006) Recommendationsimplicit in policy defaults. Psych. Sci. 17(5):414–420.

National Highway Traffic Safety Administration (2010) Light-dutyvehicle greenhouse gas emission standards and corporate aver-age fuel economy standards; final rule. Federal Register 75(88):25323–25728.

Nelson TE (2011) Issue framing. Shapiro RY, Jacobs LR, eds. TheOxford Handbook of American Public Opinion and the Media (OxfordUniversity Press, New York), 189–203.

O’Donoghue T, Rabin M (1999) Doing it now or later. Amer. Econom.Rev. 89(1):103–124.

Payne JW, Bettman JR, Johnson EJ (1988) Adaptive strategy selectionin decision making. J. Experiment. Psych.: Learn., Memory, Cogni-tion 14(3):534–552.

Payne JW, Bettman JR, Johnson EJ (1993) The Adaptive Decision Maker(Cambridge University Press, Cambridge, UK).

RokeachM (1973) The Nature of Human Values (Free Press, New York).Russo EJ, Dosher BA (1983) Strategies for multiattribute binary

choice. J. Experiment. Psych.: Learn., Memory, Cognition 9(4):676–696.

Shah JY, Kruglanski AW (2003) When opportunity knocks: Bottom-up priming of goals by means and its effects on self-regulation.J. Personality Soc. Psych. 84(6):1109–1122.

Sher S, McKenzie CR (2006) Information leakage from logicallyequivalent frames. Cognition 101(3):467–494.

Simon HA (1982) Models of Bounded Rationality (MIT Press, Cam-bridge, MA).

Slovic P (1972) Psychological study of human judgment: Implicationsfor investment decision-making. J. Finance 27(4):779–799.

Smith NC, Goldstein DG, Johnson EJ (2013) Choice without aware-ness: Ethical and policy implications of defaults. J. Public PolicyMarketing 32(2):159–172.

Stern PC, Dietz T (1994) The value basis of environmental concern.J. Soc. Issues 50(3):65–84.

Sunstein CR (2011) Empirically informed regulation. University ofChicago Law Rev. 78(4):1349–1429.

Thaler RH, Sunstein CR (2008) Nudge: Improving Decisions AboutHealth, Wealth, and Happiness (Yale University Press, NewHaven, CT).

Tversky A, Kahneman D (1981) The framing of decisions and thepsychology of choice. Science 211(4481):453–458.

Ungemach C, Stewart N, Reimers S (2011) How incidental valuesfrom the environment affect decisions about money, risk, anddelay. Psych. Sci. 22(2):253–260.

U.S. Environmental Protection Agency (2014) Climate Change Indi-cators in the United States, 2014, 3rd. ed. (EPA, Washington,DC), EPA 430-R-14-004. https://www.epa.gov/climatechange/indicators.

van Osselaer SM, Janiszewski C (2012) A goal-based model of prod-uct evaluation and choice. J. Consumer Res. 39(2):260–292.

Weber M, Eisenführ F, von Winterfeldt D (1988) The effects of split-ting attributes onweights inmultiattribute utility measurement.Management Sci. 34(4):431–445.

Wheeler SC, Berger J (2007) When the same prime leads to differenteffects. J. Consumer Res. 34(3):357–368.

Zhang J, Hsee CK, Xiao Z (2006) The majority rule in individualdecision Making. Organ. Behav. Human Decision Processes 99(1):102–111.

Dow

nloa

ded

from

info

rms.

org

by [

128.

112.

72.1

71]

on 1

9 M

ay 2

017,

at 1

2:21

. Fo

r pe

rson

al u

se o

nly,

all

righ

ts r

eser

ved.

https://www.epa.gov/climatechange/indicators

https://www.epa.gov/climatechange/indicators

translated attributes as choice architecture: aligning ...€¦ · translated attributes as choice...

Documents