Data Driven Viewpoint: Below is a study with far reaching implications for making choices in politics and all other areas of life. It is present in marketing, advertising, sermons, political speeches, news stories and op-ed pieces. It is either controlled for or abused in surveys. If we are aware of the role of positional bias in our thinking we may be able to inoculate ourselves from it when we see it in play and perhaps make better, more rationally informed decisions. (And yes, anyone who reads this blog will not that Data Driven Viewpoints almost always precede the article that is the subject of my commentary.)
First Is Best
Dana R. Carney1*, Mahzarin R. Banaji2
We experience the world serially rather than simultaneously. A century of research on human and nonhuman animals has suggested that the first experience in a series of two or more is cognitively privileged. We report three experiments designed to test the effect of first position on implicit preference and choiceusing targets that range from individual humans and social groups to consumer goods. Experiment 1 demonstrated an implicit preference to buy goods from the first salesperson encountered and to join teams encountered first, even when the difference in encounter is mere seconds. In Experiment 2 the first of two consumer items presented in quick succession was more likely to be chosen. In Experiment 3 an alternative hypothesis that first position merely accentuates the valence of options was ruled out by demonstrating that first position enhances preference for the first even when it is evaluatively negative in meaning (a criminal). Together, these experiments demonstrate a “first is best” effect and we offer possible interpretations based on evolutionary mechanisms of this “bound” on rational behavior and suggest that automaticity of judgment may be a helpful principle in clarifying previous inconsistencies in the empirical record on the effects of order on preference and choice.
Experiment 1 Top
In Experiment 1 we tested the “first is best” hypothesis using three different pairs of stimuli: two male salespersons, two female salespersons, and two teams. Participants were given each of the items of a pair in sequence and then participants indicated their preference.
Participants and procedure.
One hundred twenty-three participants were presented with three different choice-pairs and then probed for preference after each pair. Participants were introduced to a pair of teams they may wish to join (the “Hadleys” and the “Rodsons”), a pair of male salespersons from whom they may purchase a car (“Jim” and “Jon”), and a pair of female salespersons from whom they may purchase a car (“Lisa” and “Lori”). Four photos of each individual (e.g., 4 different photos of “Lisa”) and 4 different team member names (e.g., 4 different members of the “Hadley” team) were shown for a total of 24 stimuli. Team-member names were balanced for word-length and letter-usage and faces were balanced within pair for attractiveness and emotional expression. Each option within a pair was presented sequentially for 30-seconds and participants were forced to maximally consider both options. Immediately after each choice-pair was presented, participants completed a measure which assessed automatic preference for each option (an Implicit Association Test, or IAT) . Self-reported preference was also measured. Order of choice-pairs was randomized as was which option was presented first. All of this research (Experiments 1–3) was approved by the human subjects review board at Harvard University. Written and signed informed consent was obtained from participants in Experiments 1 and 3 and verbal consent was obtained from the field participants in Experiment 2.
The measure of automatic preference was an evaluative IAT for each choice-pair. In these tests, each category label–in this case, team names and salespersons–were paired with “Better” and “Worse.” For example, the IAT assessed the degree to which Lisa versus Lori (and also Jim vs. Jon and Hadleys vs. Rodsons) were spontaneously associated with positive and negative attributes, with order of first paired block (Lisa + Better; Lori + Worse) counterbalanced. The difference in average response latency to the paired blocks was divided by the pooled SD yielding an index of implicit preference for one person (or team) versus the other (the score is known as a D-score) . In the IATs, the items for the salespersons were 4 photographs of each individual or team. The items for “Better” and “Worse” were roughly synonyms for the words (e.g., wonderful, best; horrible, worst).
Deliberative, self-reported preference was measured with: (a) a 7-point scale labeled “I strongly prefer Lisa to Lori,” to “I strongly prefer Lori to Lisa”; (b) two “feeling thermometers” of Lisa and Lori preference from 0 (cold) –100 (warm); the difference between the two thermometers was z-scored and averaged with the z-scored 7-point rating item (all option-pairs were rated on these self-reported items). Both self-report and implicit measures contained a midpoint of zero indicating no preference. Order of preference-measures (implicit vs. explicit) was counterbalanced.
Results and Discussion
Repeated measures ANOVA across choice-pairs revealed a main effect of primacy on automatic preference. Automatic preference for first items presented was compared with automatic preference for second items presented (across choice-pairs: Hadleys & Rodsons, Jim & Jon, and Lisa & Lori). Regardless of the actual option, the one presented first compared to the one presented next was significantly more strongly associated with the concept “better” rather than “worse”, F(1, 121) = 20.20, p<.001; effect size r = .38 (Figure 1). There was no difference in self-reported preference for firsts versus seconds, F(1, 121) = .08, p = .78.
Figure 1. The effect of presentation order on automatic preference for Hadleys/Rodsons, Jim/Jon, and Lisa/Lori.
Higher scores indicate a preference for Hadleys, Jim, and Lisa (vs. Rodsons, Jon, and Lori); lower scores indicate a preference for Rodsons, Jon, and Lori (vs. Hadleys, Jim, and Lisa). Error bars are standard errors of the mean.doi:10.1371/journal.pone.0035088.g001
Experiment 2 Top
The IAT is a very particular kind of test of automatic preference, and in Experiment 2 we introduced a different and more realistic measure of preference in the form of a direct choice of a consumer good. When choosing between two pieces of similarly packaged and flavored bubble gum, we either (a) imposed upon participants a time constraint that forced a spontaneous and immediate decision, or (b) allowed a deliberative choice with more time available. Consumer items varied in no way other than serial position; choice of gum was the dependent variable.
Participants and procedure.
Two-hundred seven participants were recruited from a train station in Boston, MA. Adults sitting alone were approached by an experimenter who was blind to the hypothesis. Participants were asked to participate in a study on consumer choice (<1% declined). For remuneration they were offered the small consumer item they chose. Two pieces of similar-looking bubblegum (1 piece of “Bubble Yum” and 1 piece of “Bubblicious;” equal in size and shape) were placed sequentially on a white clipboard. First placement of gum brand and side of clipboard (left vs. right) were counterbalanced. After rapid sequential placement of the two items, participants reached out and grabbed (and kept) their choice.
In the rapid decision task participants were instructed to make their selection fast “within one second or so” whereas in the deliberative decision task they were instructed to select “after you have thought about it.”
Results and Discussion
A 2×2 χ2 tested the effect of order (first vs. second) on choice for each deliberation condition (automatic vs. controlled) separately. Consistent with results from Experiment 1, participants in the rapid decision task chose the chewing gum presented first (62%) significantly more often than the gum presented second (38%) whereas, participants in the deliberative choice condition were equally as likely to choose the gum presented first (51%) or second (49%): Χ2(df = 1, N = 113) = 6.53, p<.02; effect size r = .24 (Figure 2).
Figure 2. The effect of presentation order on actual choice for bubblegum presented first versus second.doi:10.1371/journal.pone.0035088.g002
Experiments 1 and 2 suggest that firsts are preferred when the choice bypasses deliberative thought across several target domains (individuals, groups, consumer goods) using two measures of preference (the IAT and simple choice). In all cases, preferences were obtained in the automatic choice condition and not in the more consciously controllable response. However, both Experiments 1 and 2 leave open a possibility that challenges the “first is best” hypothesis: It is possible that the first item in a pair does not engender greater preference but a more extreme evaluative response to whatever the existing valence of that object. In Experiments 1 and 2, the stimuli were mildly or extremely positive objects and the observed effect may simply reflect an enhancement of existing valence. In other words, had the first item been obviously aversive, the result could be a stronger negative reaction to the first, resulting in a choice of the second. Experiment 3 was conducted to test this alternative explanation of polarization.
Experiment 3 Top
We measured automatic preference for negatively valenced options: two criminals convicted of violent crimes. If “first is best,” the first criminal presented should be seen as more worthy of parole; if polarization is the effect, the first criminal presented should be viewed as less worthy of parole.
Participants and procedure.
Two criminals' photographs, from the Florida Department of Corrections website (www.dc.state.fl.us), were used. Photos depicted 29 year-old males known to have committed the same violent crimes. Criminals were wearing identical correctional facility outfits; photos were pre-tested to be equally attractive and both expressing neutral facial expressions. The two photos were selected from a larger pool of 17 photos taken from the same website. Two coders (D. R. C. and B. M.) coded all photos for (1) facial expression from −3 (extremely negative) to +3 (extremely positive), and (2) attractiveness from −3 (extremely unattractive) to +3 (extremely attractive). The two photos for which the two coders were in 100% agreement were selected for use.
Thirty-one participants learned that evidence suggests people can make accurate judgments of others after limited exposure to them and were asked to evaluate two criminals and to determine who should “stay in jail” versus “be released on parole.” Participants were shown two criminals, one after the other (order of photos was counterbalanced across participants). Immediately upon seeing the faces of the criminals (named Jim and Jon), a measure of automatic preference assessed participants' speed of associating Jim and Jon with “Worse for Parole” and “Better for Parole” using the same procedure as in Experiment 1. Self-reported preference was also measured.
Results and Discussion
Even when the stimuli were evaluatively negative (imprisoned criminals convicted of violent crimes) and the judgment called for was evaluatively negative (who should remain in prison?), participants automatically associated the first criminal with being more worthy of parole (rather than prison) compared to the second criminal. Regardless of which photo was presented first, it was the one presented first who was judged to be more worthy of parole, F(1, 29) = 4.31, p<.05; effect size r = .36 (Figure 3). Replicating Experiments 1 and 2, no effects were observed on self-reported preference. On the self-reported measure of preference the means were consistent with those of the implicit results: When Jim was seen first he was preferred for parole over Jon (M = .25) and when Jon was shown first he was preferred over Jim (M = −.21). However, this effect was not statistically significant:F(1, 29) = 1.88, p>.18.
Figure 3. The effect of presentation order on automatic preference for Jim the criminal versus Jon the criminal.
Higher scores indicate a preference for Jim (vs. Jon); lower scores indicate a preference for Jon (vs. Jim). Error bars are standard errors of the mean.doi:10.1371/journal.pone.0035088.g003
General Discussion Top
In three experiments, with tests using options from five different social and consumer-item categories (female salespeople, male salespeople, teams, pieces of gum, and criminals), we obtained a consistent result that on a deep, automatic level of human cognition, firsts are consistently preferred and chosen. Despite the research demonstrating how firsts influence many aspects of human and animal cognition, the six published reports on adult humans testing the effect of order on preference have reported somewhat mixed results , , –.
Mantonakis et al. attempted to explain the mixed results by advancing the idea that expertise moderates the effect of primacy on preference. They argued that expert knowledge and deliberation about options trump locally available heuristics such as order. While their account does explain their own and some others' findings, it does not explain why Bruine de Bruin's work , , Li and Epley, and Nisbett and Ross  sometimes found no effect of order and/or other times recency effects. The expertise account alone also doesn't explain the results from the three experiments presented here.
We tested whether firsts would show direct effects on preference and choice using measures of automatic and deliberative cognition and in all three experiments our results suggest it is on measures of automatic cognition that firsts may deliver their impact more consistently. In retrospect, that is probably why existing research, mainly with non-humans, showed an early and automatic tendency to prefer firsts; later work with adult humans which sometimes produced mixed results may have simply been a result of mixed methods, both automatic and controlled, being utilized. Although such a conclusion from the review of existing research may have been possible, it is experiments that use both forms of measuring preferences that can be most convincing. Consistent with our results and the theoretical argument advanced by Mantonakis et al. , we propose that judgments that are relatively devoid of conscious awareness will consistently reveal an effect in which firsts are considered best because firsts are privileged for several reasons that heuristic processes may rely on. We note here that accounts of satisficing are consistent with our notion of an automatic preference for firsts–however we refer to automatic preferences resting below the reaches of introspective access and fully developed only after maximizing – in other words, only after both options have been fully presented and considered.
It may well be that a preference for firsts has its origins in an evolutionary adaptation favoring firsts. Research on filial imprinting in non-human animals suggests that attachment to the first may have evolved as an adaptive mechanism to help organisms rapidly discriminate between those entities that are safe versus dangerous . Research on humans' innate preparedness to prefer their own (mother, family, social group), which is also the one encountered first, may account for the potential adaptive utility of a preference for primacy. A primacy effect on preference may have derived from positivity attributed to first experiences not leading to harm. Automatic measures of preference are able to tap into these more primitive human systems which is why the primacy effect can be consistently harvested using such measures.
It may also be that a primacy effect on preference evolved from observations of phenomena like “pecking order” (we thank Joshua Greene for this suggestion). The animal that has alpha status gets to eat first, the person of highest privilege in a family or group is served first (kings, fathers, guests). Perhaps a preference for the first conferred an advantage leading a preference for first to have been a quality favored in natural selection.
In contrast, when controlled processing is possible, other influences can (as they rationally should) override the automatic reliance on the first. This is both the discovery of the present experiments and it also may explain the sometimes mixed outcomes of previous research.
As a pointer to the mechanism by which firsts may have their power, we note that research on comparative judgments shows that whatever is set as the “gold standard” against which others are compared is itself strongly preferred , . Although order effects have not been shown to account for direction of comparison findings , the basic thrust of this and related work on similarity judgments ,  suggests that firsts may automatically be viewed as the gold standard against which others are to be compared.
Future research will involve further tests of the generality of this phenomenon in different samples and species because such tests are crucial to understanding the nature of this surprising and seemingly important effect. If some of the above speculations are correct, this effect should emerge in infants and young children, as well as in other primates and animals–regardless of whether deliberation was “allowed” or not. Likewise, tests are needed to unlink this primacy effect from other primacy producing results. For example, disambiguating this result from primacy effects in memory can be conducted by having later items in a series repeated more often (to enhance recency) and test whether even under such conditions a preference for first is retained (we thank Jonathan Schooler for this suggestion). Two of our tests (Experiments 1 and 3) were optimized to yield recency effects because when forced to maximally consider all options people tend to choose the most recent item; thus, Experiments 1 and 3 served as extremely conservative tests of our hypothesis (we thank an anonymous reviewer for pointing this out). However, additional tests utilizing paradigms more traditionally optimized to reveal recency (e.g., time delays between first and second options as is the case in persuasion research) should examine whether recency effects are an emergent property of only deliberative cognition–perhaps an indirect measure (such as those used in the three experiments presented here) would demonstrate primacy while self-report measures continue to reveal recency. In the meantime, decision theorists would do well to build in tests of “first is best” in research that will teach us about the impact, boundary conditions and mechanisms of the automatic preference for firsts observed here.
Thanks to Moshe Bar, Brian Lucas, Beth Mela (B. M.), Sarah Piccioli, Kevin Chan, Becca Dyer, Christopher Dial, and Katie Lancaster and members of the Bar and Banaji labs for assistance with and feedback about this research. Thanks to Dave Hamilton, Reid Hastie, Jim Sherman, and Danny Kahneman for comments.
Author Contributions Top
Conceived and designed the experiments: DRC MRB. Performed the experiments: DRC MRB. Analyzed the data: DRC MRB. Contributed reagents/materials/analysis tools: DRC MRB. Wrote the paper: DRC MRB.
- (1979) Information processing models of cognition. Annu Rev Psychol 30: 363–396.FIND THIS ARTICLE ONLINE
- (2002) Heuristics and biases: The psychology of intuitive judgment. New York: Cambridge University Press.
- (1982) Judgment under uncertainty: Heuristics and biases. New York: Cambridge University Press.
- (1980) Human inference: Strategies and shortcomings of social judgment. New Jersey: Prentice-Hall.
- (1959) Theories of decision-making in economics. Am Econ Rev 49: 253–283. FIND THIS ARTICLE ONLINE
- (1980) Towards a positive theory of consumer choice. J Econ Behav Org 1: 39–60. FIND THIS ARTICLE ONLINE
- (1885) Memory. A Contribution to Experimental Psychology. New York: Teachers College, Columbia University.
- (1954) Primacy versus recency in persuasion as a function of the timing of arguments and measures. J Abnorm Soc Psychol 69: 381–391. FIND THIS ARTICLE ONLINE
- (1959) Recency and primacy in persuasion as a function of the timing of speeches and measurements. J Abnorm Soc Psychol 59: 1–9. FIND THIS ARTICLE ONLINE
- (2005) Primacy and recency effects in extinction and latent inhibition: A selective review with implications for models of learning. Behav Processes 69: 223–235. FIND THIS ARTICLE ONLINE
- (1990) On the importance of being first: A primacy effect in filial imprinting. Anim Behav 40: 472–483. FIND THIS ARTICLE ONLINE
- (1957) Imprinting: The interaction of learned and innate behavior. The critical periods. J Comp Physiol Psychol 50: 6. FIND THIS ARTICLE ONLINE
- (1992) Imprinting and the development of face recognition: From chick to man. Curr Dir Psychol Sci 1: 52–55. FIND THIS ARTICLE ONLINE
- (2002) Implicit partisanship: Taking sides for no reason. J Pers Soc Psychol 83: 367–379. FIND THIS ARTICLE ONLINE
- (1946) Forming impressions of personality. J Abnorm Soc Psychol 41: 258–290. FIND THIS ARTICLE ONLINE
- (1968) Pattern of performance and ability attribution: An unexpected primacy effect. J Pers Soc Psychol 10: 317–340. FIND THIS ARTICLE ONLINE
- (1987) An evaluation of a cognitive theory of response order effects in survey measurement. Public Opin Q 51: 201–219. FIND THIS ARTICLE ONLINE
- (1957) Experimental attempts to minimize the impact of first impressions. In Hovland CI, editor. The order of presentation in persuasion. New Haven, CT: Yale University Press.
- (1928) Modes of emphasis in public speaking. J Appl Psychol 12: 611–620. FIND THIS ARTICLE ONLINE
- (1936) Experimental studies of change in attitude: II. A study of the effect of printed argument on changes in attitude. J Abnorm Soc Psychol 30: 522–532. FIND THIS ARTICLE ONLINE
- (1925) The psychology of belief: IV. The law of primacy in persuasion. J Abnorm Soc Psychol 20: 183–191. FIND THIS ARTICLE ONLINE
- (2003) Understanding and using the Implicit Association Test: I. An improved scoring algorithm. J Pers Soc Psychol 85: 197–216. FIND THIS ARTICLE ONLINE
- (2005) Save the last dance for me: Unwanted serial position effects in jury evaluations. Acta Psychologica 118: 245–260. FIND THIS ARTICLE ONLINE
- (2006) Save the last dance for me II: Unwanted serial position effects in figure skating judgments. Acta Psychol 123: 299–311. FIND THIS ARTICLE ONLINE
- (2009) When the best appears to be saved for last: Serial position effects on choice. J Behav Decis Making 22: 378–389. FIND THIS ARTICLE ONLINE
- (2009) Order In Choice: Effects of Serial Position on Preferences. Psychol Sci 20: 1309–1312. FIND THIS ARTICLE ONLINE
- (1996) Comparative judgments as a function of the direction of comparison versus word order. Public Opin Q 66: 400–409. FIND THIS ARTICLE ONLINE
- (1995) Asking comparative questions: The impact of the direction of comparison. Public Opin Q 59: 347–372. FIND THIS ARTICLE ONLINE
- (1977) Features of similarity. Psychol Rev 84: 327–353. FIND THIS ARTICLE ONLINE
- (1978) Studies of similarity. In Rosch E, Lloys B, editors. Cognition and Categorization. New Jersey: Erlbaum. pp. 81–98.