- Email: [email protected]
Influencing weight bias: The impact of biased questionnaire anchors on stereotype beliefs and judgments
Obesity Research & Clinical Practice (2015) 9, 448—457
ORIGINAL ARTICLE
Influencing weight bias: The impact of biased questionnaire anchors on stereotype beliefs and judgments R.A. Carels a,∗, J. Rossi a, M. Taylor b, J. Borushok b, A. Kiefner-Burmeister c, N. Cross b, N. Hinman b, J.M. Burmeister b a
East Carolina University, United States Bowling Green State University, United States c University of Findlay, United States b
Received 3 December 2014 ; received in revised form 2 February 2015; accepted 5 February 2015
KEYWORDS Cognitive aspects of survey methodology; Anchor effect; Weight bias; Social consensus; Source credibility
Summary Objectives: In this investigation, biased questionnaire response anchors were designed to indirectly manipulate respondents’ estimates of their peers’ stereotypic beliefs or the estimates of scientific research findings about individuals with obesity. The current study tested the hypothesis that biased response anchors could influence personal beliefs about obesity. Methods: Two-hundred adults participated in the study. A simple manipulation of questionnaire items (i.e., asking respondents to estimate peers’ beliefs or scientific research findings) using biased response scale anchors was designed to subtly relay information about certain personality traits of individuals with obesity. Results: The anchor manipulation significantly influenced participants’ immediate and follow-up weight biased beliefs as well as participants’ evaluation of an obese job applicant’s potential for employment. Conclusion: Social judgments about obese individuals may be susceptible to subtle manipulation of response anchors and may be impacted by the source of comparison information (e.g., peers; scientific research). © 2015 Asian Oceanian Association for the Study of Obesity. Published by Elsevier Ltd. All rights reserved.
∗ Corresponding author at: East Carolina University, Department of Psychology, Greenville, NC 27858-4353, United States. Tel.: +1 2527375070. E-mail address: [email protected] (R.A. Carels).
http://dx.doi.org/10.1016/j.orcp.2015.02.002 1871-403X/© 2015 Asian Oceanian Association for the Study of Obesity. Published by Elsevier Ltd. All rights reserved.
Anchor effects and weight bias
Introduction Weight stigma and discrimination are ubiquitous and produce a number of detrimental individual and societal costs [1]. Individuals with obesity are often characterized unfavorably (e.g., ‘‘lazy,’’ ‘‘weakwilled’’) and negative weight-based stereotypes are widespread [1]. Research examining weight stigma and attitude change suggests that attitudes toward people with obesity may be influenced by direct manipulation of social consensus information (i.e., information on attitudes that are purportedly from peers and the credibility of the source [2]). Specifically, Puhl and colleagues [2] showed that students from Yale changed their explicit beliefs and attitudes regarding individuals with obesity to more closely align with purported social consensus and scientific research information when it is discrepant with their personal beliefs and attitudes [2]. Moreover, their findings imply that source credibility, such as whether information reportedly originates from scientific research or an in-group source (i.e., Ivy league versus community college students), may be a key factor in determining whether presented information influences judgment. Shifts in attitudes are susceptible to both direct manipulation of social consensus information and the credibility of the source through techniques, such as providing direct feedback to participants that one’s views differ from one’s peers or experts [2] as well as indirect manipulations of social consensus information and the credibility of the source through techniques, such as asking respondents to judge the frequency of particular attitudes of peers or experts utilizing biased questionnaire response anchor sets [3—5]. Research on the cognitive aspects of survey methodology suggests that response option sets, skewed in either a positive or negative direction influence reported response frequencies and ratings, and may also influence social judgments. Biased response anchors influence judgments through factors such as cognitive heuristics (i.e., mental shortcuts used to make rapid, automatic judgments [3,6]). Cognitive heuristics can lead to systematic and potentially detrimental biases and fallacies when information is presented in a way that skews a person’s attitudes or beliefs in a harmful direction [3]. Beyond cognitive heuristics, Schwartz’s [4] research reveals that response options also create a frame of reference for responding. In other words, respondents consider the available range of response options to frame their estimation of behavioral frequencies, especially when perceived personal knowledge regarding these events or behaviors is
449 sparse. In the specific domain of weight bias, people may use biased response anchors to frame their estimates of social consensus or scientific research community supported information regarding characteristics of overweight individuals. This framing may lead respondents to adjust their attitudes and beliefs about overweight individuals to more closely align with biased anchor sets, leading to increased stigmatization of individuals with obesity. Similar research examining attitudes and social judgments toward another stigmatized group, African Americans, also provides valuable evidence adding support to the current study’s premise. Wittenbrink and Henly [5] found that participants exposed to a measure suggestive of negative beliefs towards African Americans (the measure featured numerical response anchor sets skewed to imply more negative information about African Americans), subsequently espoused more negative beliefs about African Americans (i.e., rated them as having more negative traits and fewer positive traits). Importantly, this anchor manipulation not only influenced trait beliefs in a negative way, but also produced outcomes suggestive of increased generalizable prejudice. In addition, people exposed to the negatively biased anchor (suggestive of negative beliefs towards African Americans came from an unidentified source) were significantly more confident that an African American defendant in a mock jury case was guilty, when compared to people exposed to a positively biased anchor set. This important study suggests that exposure to biased information via response anchor sets may lead to the development, solidification, or strengthening of prejudicial beliefs. No previous weight bias research has examined the impact of manipulating questionnaire response anchor sets on attitudes toward people with obesity. Additionally, the current study builds upon previous studies which have examined the relationship between extreme response anchors, source information, attitude formation and change, and social judgment. The investigation’s first purpose was to examine whether exposing individuals to biased numerical response sets during an independent manipulation would influence subsequent prevalence ratings of various desirable and undesirable traits of individuals with obesity immediately following active exposure to the biased response anchors and one-week following. It was hypothesized that prior exposure to biased anchor response sets would predict subsequent attitudes about individuals with obesity immediately and one week following response anchor set exposure. Specifically, it was conjectured that people who received a response anchor
450 manipulation that featured numerical responses sets skewed to represent greater prevalence of unfavorable external attitudes towards obese individuals would later rate individuals with obesity in a more undesirable way as compared to individuals who received a response anchor manipulation that featured numerical responses sets skewed to represent greater prevalence of favorable external attitudes towards obese individuals. It was hypothesized that individuals who received the negative response anchor manipulation would later rate individuals with obesity more undesirably than individuals who received the positive response anchor manipulation. Secondly, the study sought to examine potential differences in attitudes towards obese individuals (immediate and follow-up) based on the source of the prevalence information. Two relevant informational sources were manipulated: peer views (i.e., students at their university) and scientific research findings. The peer source manipulation served as a subtle social consensus manipulation, while the scientific researcher manipulation served as a subtle source credibility manipulation. Consistent with findings from Puhl’s [2] research that directly manipulated the impact of social consensus and source credibility on weight bias, it was hypothesized that both indirect manipulation of peer consensus and scientific research information (via biased response anchors) would predict immediate and delayed attitudes towards individuals with obesity, with scientific research providing a somewhat stronger effect. A final purpose of the study was to uncover whether biased numerical response sets could influence subsequent attitudes regarding hiring likelihood of an employee candidate with obesity. Specifically, it was hypothesized that people who received a negative response anchor manipulation would be less likely to hire an individual with obesity than individuals who received a positive response anchor manipulation. Information from this investigation is important and relevant, because any source that conveys biased numerical information regarding stereotypes or stigmatized individuals may subtly create, maintain, and strengthen prejudice toward this group.
Methods Participants Participants (N = 200) were undergraduate students at a mid-sized Midwestern University who earned psychology course research credit in exchange for participation. Sample size was selectedto detect a
R.A. Carels et al. moderate effect size employing ANCOVA. Participants (79.5% female; 84.5% Caucasian) had a mean age of 19.22 (SD = 2.43) and a mean self-reported BMI of 25.0 (SD = 5.28). Of the 200 participants who completed the initial survey, 162 completed a follow-up survey one week later. Participants signed up and completed the entire study online.
Procedure The study was approved by the University’s Human Subjects Review Board. At baseline, participants completed a demographic information questionnaire and measures of potential covariates (Attitudes Towards Obese Persons Scale, Marlowe—Crowne Social Desirability Scale). Participants were then exposed to an experimental manipulation via a questionnaire modified for each condition (see Biased Response Scales below). Using a 2 (peer consensus versus scientific research condition) × 2 (positive versus negative bias response anchor condition) design, participants were randomized into one of four experimental conditions: (1) peer consensus & positive bias response anchors (N = 50; follow-up N = 41), (2) peer consensus & negative bias response anchors (N = 42; follow-up N = 36), (3) scientific research & positive bias response anchors (N = 53; follow-up N = 42), and (4) scientific research & negative bias response anchors (N = 54; follow-up N = 43). Participants were exposed to the experimental manipulation utilizing modified questions from the Obese Persons Trait Scale [OPTS; 2]. Standard use of the OPTS asks respondents to estimate the percentage prevalence (0—100%) of 10 desirable and 10 undesirable personality traits. However, rather than asking participants to report estimates based on their own beliefs, for the main manipulation in this study, participants were asked to estimate the prevalence of various traits of individuals with obesity reported by either ‘‘other university students’’ or by ‘‘scientific research.’’ In the prompt for each question, participants were to estimate the prevalence of certain attributes of obese individuals for either peers or the scientific research community. For example, in the peer consensus condition, participants were asked, ‘‘According to [name of university] students, what percentage of obese people are lazy?’’ In the scientific research condition, participants were asked, ‘‘According to scientific research, what percentage of obese people are lazy?’’ In the response to each question, participants were exposed to a positive or negative bias response anchor condition. The response anchor
Anchor effects and weight bias Table 1
451
Examples of biases response scales for each condition.
Information source
Example question
Positive response anchors
Negative response anchors
Peer consensus
‘‘According to [name of university] students, what percentage of obese people are GENEROUS?’’
Scientific research
‘‘According to Scientific Research, what percentage of obese people are GENEROUS?’’
Peer consensus
‘‘According to [name of university] students, what percentage of obese people are LAZY?’’
Scientific research
‘‘According to Scientific Research, what percentage of obese people are LAZY?’’
Less than 60% 61—68% 69—76% 77—84% 85—92% 93—100% Less than 60% 61—68% 69—76% 77—84% 85—92% 93—100% 0—10% 11—20% 21—30% 31—40% 41—50% Greater than 50% 0—10% 11—20% 21—30% 31—40% 41—50% Greater than 50%
0—7% 8—15% 16—23% 24—31% 32—39% Greater than 40% 0—7% 8—15% 16—23% 24—31% 32—39% Greater than 40% Less than 50% 51—60% 61—70% 71—80% 81—90% 91—100% Less than 50% 51—60% 61—70% 71—80% 81—90% 91—100%
Note. Response options were modified to range by 5,7,10 and 15 points (e.g. 0—5%, 0—7%, 0—10% and 0—15%) for various questions. The biased response scales consisted of 20 items with ten positive (i.e. humerous, generous, sociable, friendly, outgoing, intelligent, honest, productive, warm personality & organized) and ten negative (i.e. lazy, self-indulgent, undisciplined, gluttonous, unhealthy, sluggish, lack of willpower, unclean, insecure & unattractive) traits modeled after the Obese Persons Trait Scale (OPTS).
manipulation used response option sets that were either positively or negatively skewed. The positive response anchor condition was biased toward representing a high prevalence of desirable personality traits and low prevalence of undesirable personality traits among people with obesity. Conversely, the negative response anchor condition was biased toward representing a high prevalence of undesirable personality traits and low prevalence of desirable personality traits among people with obesity. Within both response conditions, answer choice increments were randomly varied between 5, 7, 10, and 15 percentage points in order to disguise the manipulation. See Table 1 for an example question for 7% and 10% response anchor manipulations. After the experimental manipulation, participants completed the dependent measures (See additional information on Obese Persons Trait Survey and a hiring vignette below). Lastly, participants completed the Obese Persons Trait Survey one week later. Following the completion of all measures, participants were given the opportunity to share comments about the study and were debriefed.
Measures Obese Persons Trait Scale (OPTS [2]) The OPTS was administered following the experimental manipulation and again one week following the study. As indicated earlier, the OPTS consists of 20 items listing stereotypical traits [2]. Ten undesirable common weight stereotypes (e.g., undisciplined) form an undesirable traits subscale and 10 desirable common weight stereotypes (e.g., intelligent) form a desirable trait subscale. Participants were asked to estimate the percentage (0—100%) of obese persons who possess these traits. For desirable traits, Cronbach’s ˛ was .90 postmanipulation and .90 at one-week follow-up. For undesirable traits, Cronbach’s ˛ was .92 postmanipulation and .91 at one-week follow-up Hiring scenario To measure behavioral components of weight bias, participants were asked to engage in a hiring scenario shortly following the experimental manipulation. They were asked to pretend to be a Student Tour Coordinator at their university who is responsible for hiring a new tour guide. Participants were
452 given criteria to help determine who would be a good candidate (e.g., ‘‘educating prospective students about the admission process’’) and were then asked to review the cover letter and casual photo of an applicant. The cover letter was created to suggest that the applicant would be a good candidate (e.g., ‘‘I am an honors student at [name of the university] majoring in Education’’) and the casual photo was of a young, college-aged woman who was obese. Afterwards, participants were asked to rate how likely they would be to hire this applicant on a 9-point Likert scale (ranging from Very Unlikely to Very Likely). Marlowe—Crowne Social Desirability Scale (MC-SDS [7]) The MC-SDS was administered at baseline to control for socially desirable responding. The MC-SDS is a widely used, well-validated measure of social desirability response bias. Participants rated 33 items as True or False (e.g., ‘‘I never resent being asked to return a favor’’). Attitudes Towards Obese Persons (ATOP [8]) The ATOP was administered at baseline to examine for baseline differences in weight bias between conditions. The ATOP consists of 20 items assessing positive (e.g., ‘‘obese people are usually sociable’’) and negative (e.g., ‘‘obese people tend to have family problems’’) attitudes towards people with obesity [8]. Participants responded using a 6point Likert scale ranging from Strongly Disagree to Strongly Agree. Cronbach’s ˛ was .82 for these items.
Data analysis Statistical analysis was conducted such that demographic differences between groups and socially desirable responding were entered as covariates, and conditions (peer consensus versus scientific research condition and positive versus negative bias response anchor conditions) were entered as fixed factors. Thus, all hypotheses were tested using 2 × 2 Analysis of Covariance (ANCOVA). Given the theoretical approach and expected direction of findings, all hypothesis testing was conducted using one-way significance tests.
Results Covariates Demographic information and baseline weight bias (measured using the ATOP) were tested in order
R.A. Carels et al. to ensure that there was equal representation in all four conditions. Baseline weight bias did not differ across conditions. Gender was included as a covariate, because there was a disproportionate number of females in both peer consensus conditions, 2 (3) = 8.02, p < .045. BMI was also included as a covariate because people in the peer consensus negative response anchor condition had significantly lower BMIs than those in the scientific research positive response anchor condition, F(3, 168) = 3.34, p = .02. Finally, social desirability was included as a covariate, as this variable was significantly related to hiring likelihood r(169) = .201, p = .01.
Experimental manipulation: effects of interval spacing on change scores A manipulation check was conducted on positive and negative response anchors. In reaction to the response anchor manipulation, participants receiving the negative response bias anchors estimated that their peers and the scientific community possessed significantly higher estimates of the prevalence of undesirable traits in individuals with obesity and significantly lower estimates of the prevalence of desirable traits of individuals with obesity. Two exceptions were the negative trait ‘‘unclean’’ t(195) = 1.18, p = .241 and the positive trait ‘‘productive’’ t(194) = 1.60, p = .11. Next, we examined whether the size of the anchor interval affected the amount of change in personality trait ratings when responding to each question. Intervals existed such that some traits were rated in 5% (e.g., ‘‘95—100%,’’ ‘‘90—95%,’’’’85—90%,’’ ‘‘80—85%,’’ and ‘‘Less than 80%’’), 7%, 10%, and 15% intervals. There was a significant effect of anchor interval F(3, 610) = 16.581, p < .001 with 2 = .08. Post hoc comparisons using the Bonferroni correction indicated that the 5% interval anchor sets produced greater changes (M = 22.78, SE = .62) from the inferred value than did the 7% (M = 20.05, SE = .63), the 10% (M = 17.94, SE = .69), and the 15% (M = 17.05, SE = .62) intervals. Similarly there were significant mean change score differences between the 7% and 15% intervals. There were no significant differences between the 10% and 15% intervals.
Undesirable personality traits Following the experimental manipulation, there were significant main effects for the scientific research versus peer consensus manipulation and for the positive versus negative bias anchor manipulation immediately following the manipulations.
Anchor effects and weight bias
453
Figure 1 Undesirable traits post manipulation and follow-up.
The effects existed such that those individuals in the scientific research conditions made significantly higher undesirable personal attributions of obese individuals than those in the peer consensus conditions. Similarly, participants in the negative response bias anchor conditions had significantly higher undesirable personal judgments of individuals with obesity than those in the positive response anchor conditions immediately following the intervention (Table 2 and Fig. 1). At follow-up, there was a significant main effect such that participants in the negative response anchor conditions made significantly higher undesirable attributions than those in the positive response anchor conditions (Table 2 and Fig. 1). There was no interaction effect immediately or one week following experimental manipulation.
conditions and a significant interaction effect (Table 3 and Fig. 2). Differences existed such that those in the scientific research, negative response bias anchor condition rated individuals with obesity significantly less desirably than individuals in the peer consensus conditions. Furthermore, there was a significant interaction effect which demonstrated that participants in the scientific research negative response anchor condition showed lower attributions of desirable traits for individuals with obesity than participants in any of the other three conditions. Regarding desirable trait attributions at follow-up, there were no significant main or interaction effects between peer consensus and scientific research conditions and between positive and negative response anchors conditions.
Hiring likelihood Desirable personality traits For desirable trait attributions, immediately following the intervention, there was a main effect for the peer consensus versus scientific research
For the hiring scenario, there was a significant interaction effect (Table 4 and Fig. 2), such that those in the scientific research positive response bias anchor condition were more likely to hire
Figure 2 Post-manipulation desirable trait and hiring likelihood.
454
Table 2 Condition
ANCOVA table of undesirable traits post manipulation and at one-week follow-up. Positive post (N)
Positive post M (SD)
Negative post (N)
Scientific research 47
56.49 42 (18.11) Peer consensus 45 51.50 38 (18.77) Marginal means 92 54.05 80 (18.51) Source F(1,165) = 2.68, p = .05 with 2 = .016* Anchor valance F(1,165) = 2.812, p = .04 with 2 = .017* Interaction effects F(1,165) = .033, p = .86 with 2 < .001
Table 3 Condition
Negative post M (SD)
Post marginal means N (M) (SD)
Positive follow up N
62.77 (15.84) 58.60 (21.73) 60.79 (18.86)
89 (61.81) (13.00) 83 (54.75) (20.36) 172 (57.19) (18.91)
39
Positive follow up M (SD)
Negative follow up N
54.09 38 (17.96) 36 53.11 34 (17.15) 75 53.62 72 (17.47) F(1, 140) = 2.091, p = .08 with 2 = .015 F(1, 140) = 2.785, p = .05 with 2 = .020* F(1, 140) = 1.386, p = .12 with 2 = .010
Negative follow up M (SD)
Marginal means post N (M) (SD)
63.64 (13.09) 56.29 (20.59) 60.17 (17.31)
77 (58.80) (16.36) 70 (54.66) (11.40) 147 (56.83) (17.64)
Negative follow up M (SD)
Marginal means post N (M) (SD)
55.71 (9.83) 60.01 (17.65) 57.74 (14.14)
77 (58.90) (11.40) 70 (59.88) (11.40) 147 (59.36) (13.58)
ANCOVA of desirable traits post manipulation and at one-week follow-up. Positive post (N)
Scientific research 47
Positive post M (SD)
Negative post (N)
Post marginal means N (M) (SD)
Positive follow up N
57.40 (12.94) 65.09 (16.85) 61.05 (15.32)
89 (61.81) (13.00) 83 (65.30) (15.54) 172 (63.49) (14.35)
39
Positive follow up M (SD)
Negative follow up N
62.00 38 (12.08) 36 59.75 34 (13.95) 75 60.92 72 (12.92) F(1, 140) = .711, p = .201 with 2 < .01 F(1, 140) = .316, p = .290 with 2 < .01 F(1, 140) = 2.52, p = .058 with 2 = .02
R.A. Carels et al.
65.75 42 (11.86) Peer consensus 45 65.47 38 (14.54) 92 Marginal means 65.61 80 (16.16) F(1,165) = 3.91, p = .03 with 2 = .023* Source Anchor valance F(1,165) = 2.13, p = .08 with 2 = .013 Interaction effects F(1,165) = 3.673, p = .03 with 2 = .022*
Negative post M (SD)
Anchor effects and weight bias Table 4
455
ANOVA prediction of hiring likelihood by condition.
Condition
Positive (N)
Scientific research
47
Positive M (SD)
7.21 (1.63) Peer consensus 45 6.87 (2.06) Marginal means 92 7.04 (1.85) Trait prevalence versus peer consensus Positive versus negative Interaction effect
Negative (N) 41
Negative M (SD)
Marginal means (N)
6.22 88 (1.96) 38 7.05 83 (1.83) 79 6.62 171 (1.93) F(1, 164) = .293, p = .30 with 2 = .002 F(1, 164) = .576, p = .15 with 2 = .003 F(1,164) = 3.704, p = .03 with 2 = .022*
the applicant than participants in the scientific research negative response bias anchor condition. There was no statistically significant difference between the peer consensus positive response bias anchor condition and the peer consensus negative response bias anchor condition.
Discussion Previous research has shown that providing people with direct, explicitly positive peer consensus information or scientific research biased information about people with obesity can increase the positivity of their short-term social attitudes towards people with obesity [2,9]. While important, these studies possess fairly strong manipulations of peer attitudes or scientific research. The current study suggests that attitudes toward individuals with obesity are susceptible to indirect, subtle manipulations of peer consensus and scientific research information by manipulating questionnaire response anchors. The current research study suggests that attitudes may be influenced when response option numerical values are skewed in a positive or negative direction on a simple questionnaire. In this investigation, regardless of whether participants estimated the attitudes of their peers or estimated the outcomes of scientific research, participants exposed to the positive response bias anchor manipulation later reported that individuals with obesity possessed fewer undesirable personality traits (e.g., lazy) immediately and one week following the experimental manipulation, relative to participants exposed to the negative response anchor manipulation. Additionally, in the scientific research condition, participants exposed to the positive response bias anchor manipulation later reported that individuals with obesity possessed more desirable personality traits (e.g., industrious)
Marginal means M (SD) 6.75 (1.85) 6.95 (1.95) 6.85 (1.89)
immediately following the experimental manipulation. Finally, after the experimental manipulation, participants exposed to the scientific research, positive response bias anchor manipulation also expressed greater likelihood that they would hire an applicant with excess weight for employment relative to participants exposed to the scientific research, negative response bias anchor manipulation. These findings are consistent with several independent lines of research. Past research on the cognitive aspects of survey methods indicates that people may be indirectly influenced by biased information presumed to come from either the majority of their peers or from perceived high credibility sources, likely because they automatically employ the anchoring heuristic to make quick judgments under conditions of uncertainty [4,10]. In the current study, biased anchor sets influenced participant estimates of their peers’ attitudes or their estimates of the outcomes of scientific research. More importantly, when later asked to report their own attitudes toward individuals with obesity, their reporting were predicted by the experimental manipulation. For example, when the experimental manipulation asked participants to provide an estimate of the percentage of individuals with obesity that are lazy according to scientific research using a negative response bias anchor set, they were more likely to indicate that people with obesity are lazy on a subsequent measure of weight bias, as compared to participants provided with a positive bias anchor set who were asked to provide the same estimate. Interestingly, one week following the experimental manipulation, the manipulation’s influence on judgments of undesirable traits remained. Much like Wittenbrink and Henley [5] study, the current study expanded on the assessment of attitudes to include an assessment of generalizable prejudice (i.e., likelihood of hiring). Among those in the scientific research condition, participants who received the negatively
456 biased anchor sets reported that they would be less likely to hire a job applicant with obesity than participants who received the positive response bias anchor sets. The disadvantages in hiring, wages, promotions, and job termination resulting from weight stigma are becoming increasingly well documented [1]. As such, the results from this investigation are hardly surprising, with the exception that the preferences in hiring were influenced by a subtle, indirect manipulation whereby respondents were asked to make judgments regarding scientific research. In other words, by simply asking participants to estimate scientific research findings regarding personality traits of individuals with obesity utilizing a measure with biased response anchors, subsequent judgments on the suitability of employment were influenced. This influence likely occurred outside of respondents’ awareness and is consistent with research on priming and the malleability of automatic stereotypes and prejudice [11]. Nevertheless, it is concerning, especially considering that individuals with obesity are not protected under workplace discrimination laws. Research indicates that extreme response anchors have been shown to exert a stronger influence on decisions under conditions that feature relatively higher ambiguity, lower familiarity, lower individual knowledge about a subject, and a more trustworthy source of information [12—15]. In the current investigation, respondents had little knowledge or expertise regarding the true estimates of specific personality traits that people with obesity possess. Also, it is probable that participants had a limited knowledge about what other students at their university might believe regarding people with obesity. These are conditions that would make a respondent susceptible to anchoring effects. However, it is notable, that while attitudes were changed under both the scientific research and peer consensus conditions, the scientific research condition had a significantly greater influence on desirable trait estimation and hiring likelihood than the peer consensus condition. It is not entirely clear why hiring decisions and desirable trait estimations are more responsive to engaging in an estimation of scientific research-based prevalence of personality traits in obese individuals using biased response anchors as compared to engaging in estimation of peer consensus on the prevalence of personality traits in obese individuals using biased response anchors. Perhaps peer influence is sufficient to influence attitude change toward a target regarding undesirable personality attributes, but
R.A. Carels et al. hiring decisions or attitude change toward a target regarding desirable personality attributes requires greater source credibility. Although the methodology used to produce attitude change in this study was very different that the methodology used in Puhl et al’s study [2], the findings are consistent with Puhl and colleagues [2] research, which demonstrated a modestly stronger effect for a scientific research manipulation as compared to a peer consensus manipulation on the outcome of improving short-term social judgments regarding the traits of individuals with obesity. The between subjects design of this investigation leaves the findings vulnerable to the interpretation that between group differences somehow accounted for the findings. However, differences between conditions were included as covariates in all analyses and several post hoc analyses were consistent with the argument that the experimental anchor manipulation produced the intended effects. First, the size of the anchor interval manipulation (5% versus 7% versus 10% versus 15%) was associated with the magnitude of change for participant estimates of the percentage of obese people whom they believe possess certain traits. In other words, when participants were asked to provide estimates of their peers’ attitudes or estimates of the findings from scientific research regarding the personality traits of individuals with obesity using more extreme anchor sets (those with a small range of very high or low numerical values), they tended to respond more extremely in the expected direction on subsequent attitude assessments of identical personality attributes. Therefore, the strength of the anchor manipulation had a direct impact on subsequent ratings. Similarly, and perhaps no less importantly, the response anchor manipulation itself resulted in respondents choosing differing estimates of the percentage of obese people possessing different personality traits according to peers or the outcomes of scientific research in nearly all cases with only two exceptions (the personality traits ‘‘unclean’’ and ‘‘productive’’.) While it is reasonable to assume that the response anchor manipulations modified attitudes, it cannot be ruled-out that feedback influenced respondents’ assumptions about the acceptability of expressing or not expressing negative beliefs. Therefore, rather than change attitudes, the manipulation may have motivated some participants to more readily express their negative attitudes and inhibited other participants from expressing their negative attitudes. Also, while evidence for the existence of changes in weight bias and its potential impact on hiring was detected,
Anchor effects and weight bias
457
the measures utilized in this investigation were not measures of actual behavioral displays of this bias. Thus, application of the current findings to the ‘‘real’’ discrimination and bias that people with obesity may experience will be an important focus for future research. Finally, the primarily Caucasian, female university sample limits generalizability of the current study’s findings to samples of differing gender, race, age, and education. Response scales are not simply measurement devices that respondents blithely complete without consideration. Rather, respondents are likely to assume that the response alternatives reflect ‘‘real’’ knowledge or the beliefs of others. When the questions are ambiguous or the answers to the questions are not readily accessible, response alternatives may serve as a frame of reference in estimating and evaluating behavior [10]. As such, social judgments may be quite susceptible to subtle manipulation. As a caution to researchers, it is probably safe to assume that people’s attitudes are influenced by the very measures that they take. If questions are extreme, respondents may assume more extreme attitudes than they originally possessed. Additionally, people may make biased judgments about individuals who are overweight or obese due to exposure to biased visual and numerical information in their environment. Again, the exposure does not have to be extreme or direct, research suggests that providing individuals with subtle, indirect cues regarding purported scientific research findings or peers’ views may be enough to influence recipient attitudes. As such, exposure to negative media depictions of personal characteristics of individuals with obesity may lead people to express more negative beliefs and attitudes towards overweight individuals [16]. These subtle influences may have lasting effects on an individual’s attitudes and behavior and the impact of such manipulations on weight bias and prejudice is an important area of future inquiry.
Conflict of interests The authors declare not conflict of interests.
References [1] Puhl RM, Heuer CA. The stigma of obesity: a review and update. Obesity 2009;17(5):1—24. [2] Puhl RM, Schwartz M, Brownell KD. Impact of perceived consensus on stereotypes about obese people: a new approach for reducing bias. Health Psychol 2005;24(5):517—25. [3] Tversky A, Kahneman D. Judgment under uncertainty: heuristics and biases. Science 1974;125:1124—31. [4] Schwarz N. Self-report: how the questions shape the answers. Am Psychol 1999;54(2):93—105. [5] Wittenbrink B, Henley JR. Creating social reality: informational social influence and the content of stereotypic beliefs. Personal Soc Psychol Bull 1996;22:598—610. [6] Belsky G, Gilovich T. Why people make big money mistakes. New York: Simon and Schuster; 1999. [7] Crowne DP, Marlowe D. A new scale of social desirability independent of psychopathology. J Consult Psychol 1960;24:349—54. [8] Allison DB, Basile VC, Yuker HE. The measurement of attitudes toward and beliefs about obese persons. Int J Eat Disord 1991;10(5):599—607. [9] Stangor C, Sechrist GB, Jost JT. Changing beliefs by providing consensus information. Personal Soc Psychol Bull 2001;27:486—96. [10] Furnham A, Boob HC. A literature review of the anchoring effect. J Socioecon 2011;40:35—42. [11] Blair IV. The malleability of automatic stereotypes and prejudice. Personal Soc Psychol Bull 2002;6(3):242—61. [12] Van Excel N, Brouwer W, van der Berg B, Koopmanschap M. With a little help from an anchor: discussion and evidence for anchoring effects in contingent valuation. J Socioecon 2006;35(836):853. [13] Mussweiler T, Strack F. Considering the impossible: explaining the effects of implausible anchors. Soc Cognit 2001;19(145):160. [14] Mussweiler T, Strack F. Numeric judgments under uncertainty: the role of knowledge in anchoring. J Exp Soc Psychol 2000;36:495—518. [15] Smith AR, Windschitl PD, Bruchmann K. Knowledge matters: anchoring effects are moderated by knowledge level. Eur J Soc Psychol 2013;43:97—108. [16] Pearl RL, Puhl RM, Brownell KD. Positive media portrayals of obese persons: impact on attitudes and image preferences. Health Psychol 2012;31(6):821—9.
Available online at www.sciencedirect.com
ScienceDirect
ORIGINAL ARTICLE
Influencing weight bias: The impact of biased questionnaire anchors on stereotype beliefs and judgments R.A. Carels a,∗, J. Rossi a, M. Taylor b, J. Borushok b, A. Kiefner-Burmeister c, N. Cross b, N. Hinman b, J.M. Burmeister b a
East Carolina University, United States Bowling Green State University, United States c University of Findlay, United States b
Received 3 December 2014 ; received in revised form 2 February 2015; accepted 5 February 2015
KEYWORDS Cognitive aspects of survey methodology; Anchor effect; Weight bias; Social consensus; Source credibility
Summary Objectives: In this investigation, biased questionnaire response anchors were designed to indirectly manipulate respondents’ estimates of their peers’ stereotypic beliefs or the estimates of scientific research findings about individuals with obesity. The current study tested the hypothesis that biased response anchors could influence personal beliefs about obesity. Methods: Two-hundred adults participated in the study. A simple manipulation of questionnaire items (i.e., asking respondents to estimate peers’ beliefs or scientific research findings) using biased response scale anchors was designed to subtly relay information about certain personality traits of individuals with obesity. Results: The anchor manipulation significantly influenced participants’ immediate and follow-up weight biased beliefs as well as participants’ evaluation of an obese job applicant’s potential for employment. Conclusion: Social judgments about obese individuals may be susceptible to subtle manipulation of response anchors and may be impacted by the source of comparison information (e.g., peers; scientific research). © 2015 Asian Oceanian Association for the Study of Obesity. Published by Elsevier Ltd. All rights reserved.
∗ Corresponding author at: East Carolina University, Department of Psychology, Greenville, NC 27858-4353, United States. Tel.: +1 2527375070. E-mail address: [email protected] (R.A. Carels).
http://dx.doi.org/10.1016/j.orcp.2015.02.002 1871-403X/© 2015 Asian Oceanian Association for the Study of Obesity. Published by Elsevier Ltd. All rights reserved.
Anchor effects and weight bias
Introduction Weight stigma and discrimination are ubiquitous and produce a number of detrimental individual and societal costs [1]. Individuals with obesity are often characterized unfavorably (e.g., ‘‘lazy,’’ ‘‘weakwilled’’) and negative weight-based stereotypes are widespread [1]. Research examining weight stigma and attitude change suggests that attitudes toward people with obesity may be influenced by direct manipulation of social consensus information (i.e., information on attitudes that are purportedly from peers and the credibility of the source [2]). Specifically, Puhl and colleagues [2] showed that students from Yale changed their explicit beliefs and attitudes regarding individuals with obesity to more closely align with purported social consensus and scientific research information when it is discrepant with their personal beliefs and attitudes [2]. Moreover, their findings imply that source credibility, such as whether information reportedly originates from scientific research or an in-group source (i.e., Ivy league versus community college students), may be a key factor in determining whether presented information influences judgment. Shifts in attitudes are susceptible to both direct manipulation of social consensus information and the credibility of the source through techniques, such as providing direct feedback to participants that one’s views differ from one’s peers or experts [2] as well as indirect manipulations of social consensus information and the credibility of the source through techniques, such as asking respondents to judge the frequency of particular attitudes of peers or experts utilizing biased questionnaire response anchor sets [3—5]. Research on the cognitive aspects of survey methodology suggests that response option sets, skewed in either a positive or negative direction influence reported response frequencies and ratings, and may also influence social judgments. Biased response anchors influence judgments through factors such as cognitive heuristics (i.e., mental shortcuts used to make rapid, automatic judgments [3,6]). Cognitive heuristics can lead to systematic and potentially detrimental biases and fallacies when information is presented in a way that skews a person’s attitudes or beliefs in a harmful direction [3]. Beyond cognitive heuristics, Schwartz’s [4] research reveals that response options also create a frame of reference for responding. In other words, respondents consider the available range of response options to frame their estimation of behavioral frequencies, especially when perceived personal knowledge regarding these events or behaviors is
449 sparse. In the specific domain of weight bias, people may use biased response anchors to frame their estimates of social consensus or scientific research community supported information regarding characteristics of overweight individuals. This framing may lead respondents to adjust their attitudes and beliefs about overweight individuals to more closely align with biased anchor sets, leading to increased stigmatization of individuals with obesity. Similar research examining attitudes and social judgments toward another stigmatized group, African Americans, also provides valuable evidence adding support to the current study’s premise. Wittenbrink and Henly [5] found that participants exposed to a measure suggestive of negative beliefs towards African Americans (the measure featured numerical response anchor sets skewed to imply more negative information about African Americans), subsequently espoused more negative beliefs about African Americans (i.e., rated them as having more negative traits and fewer positive traits). Importantly, this anchor manipulation not only influenced trait beliefs in a negative way, but also produced outcomes suggestive of increased generalizable prejudice. In addition, people exposed to the negatively biased anchor (suggestive of negative beliefs towards African Americans came from an unidentified source) were significantly more confident that an African American defendant in a mock jury case was guilty, when compared to people exposed to a positively biased anchor set. This important study suggests that exposure to biased information via response anchor sets may lead to the development, solidification, or strengthening of prejudicial beliefs. No previous weight bias research has examined the impact of manipulating questionnaire response anchor sets on attitudes toward people with obesity. Additionally, the current study builds upon previous studies which have examined the relationship between extreme response anchors, source information, attitude formation and change, and social judgment. The investigation’s first purpose was to examine whether exposing individuals to biased numerical response sets during an independent manipulation would influence subsequent prevalence ratings of various desirable and undesirable traits of individuals with obesity immediately following active exposure to the biased response anchors and one-week following. It was hypothesized that prior exposure to biased anchor response sets would predict subsequent attitudes about individuals with obesity immediately and one week following response anchor set exposure. Specifically, it was conjectured that people who received a response anchor
450 manipulation that featured numerical responses sets skewed to represent greater prevalence of unfavorable external attitudes towards obese individuals would later rate individuals with obesity in a more undesirable way as compared to individuals who received a response anchor manipulation that featured numerical responses sets skewed to represent greater prevalence of favorable external attitudes towards obese individuals. It was hypothesized that individuals who received the negative response anchor manipulation would later rate individuals with obesity more undesirably than individuals who received the positive response anchor manipulation. Secondly, the study sought to examine potential differences in attitudes towards obese individuals (immediate and follow-up) based on the source of the prevalence information. Two relevant informational sources were manipulated: peer views (i.e., students at their university) and scientific research findings. The peer source manipulation served as a subtle social consensus manipulation, while the scientific researcher manipulation served as a subtle source credibility manipulation. Consistent with findings from Puhl’s [2] research that directly manipulated the impact of social consensus and source credibility on weight bias, it was hypothesized that both indirect manipulation of peer consensus and scientific research information (via biased response anchors) would predict immediate and delayed attitudes towards individuals with obesity, with scientific research providing a somewhat stronger effect. A final purpose of the study was to uncover whether biased numerical response sets could influence subsequent attitudes regarding hiring likelihood of an employee candidate with obesity. Specifically, it was hypothesized that people who received a negative response anchor manipulation would be less likely to hire an individual with obesity than individuals who received a positive response anchor manipulation. Information from this investigation is important and relevant, because any source that conveys biased numerical information regarding stereotypes or stigmatized individuals may subtly create, maintain, and strengthen prejudice toward this group.
Methods Participants Participants (N = 200) were undergraduate students at a mid-sized Midwestern University who earned psychology course research credit in exchange for participation. Sample size was selectedto detect a
R.A. Carels et al. moderate effect size employing ANCOVA. Participants (79.5% female; 84.5% Caucasian) had a mean age of 19.22 (SD = 2.43) and a mean self-reported BMI of 25.0 (SD = 5.28). Of the 200 participants who completed the initial survey, 162 completed a follow-up survey one week later. Participants signed up and completed the entire study online.
Procedure The study was approved by the University’s Human Subjects Review Board. At baseline, participants completed a demographic information questionnaire and measures of potential covariates (Attitudes Towards Obese Persons Scale, Marlowe—Crowne Social Desirability Scale). Participants were then exposed to an experimental manipulation via a questionnaire modified for each condition (see Biased Response Scales below). Using a 2 (peer consensus versus scientific research condition) × 2 (positive versus negative bias response anchor condition) design, participants were randomized into one of four experimental conditions: (1) peer consensus & positive bias response anchors (N = 50; follow-up N = 41), (2) peer consensus & negative bias response anchors (N = 42; follow-up N = 36), (3) scientific research & positive bias response anchors (N = 53; follow-up N = 42), and (4) scientific research & negative bias response anchors (N = 54; follow-up N = 43). Participants were exposed to the experimental manipulation utilizing modified questions from the Obese Persons Trait Scale [OPTS; 2]. Standard use of the OPTS asks respondents to estimate the percentage prevalence (0—100%) of 10 desirable and 10 undesirable personality traits. However, rather than asking participants to report estimates based on their own beliefs, for the main manipulation in this study, participants were asked to estimate the prevalence of various traits of individuals with obesity reported by either ‘‘other university students’’ or by ‘‘scientific research.’’ In the prompt for each question, participants were to estimate the prevalence of certain attributes of obese individuals for either peers or the scientific research community. For example, in the peer consensus condition, participants were asked, ‘‘According to [name of university] students, what percentage of obese people are lazy?’’ In the scientific research condition, participants were asked, ‘‘According to scientific research, what percentage of obese people are lazy?’’ In the response to each question, participants were exposed to a positive or negative bias response anchor condition. The response anchor
Anchor effects and weight bias Table 1
451
Examples of biases response scales for each condition.
Information source
Example question
Positive response anchors
Negative response anchors
Peer consensus
‘‘According to [name of university] students, what percentage of obese people are GENEROUS?’’
Scientific research
‘‘According to Scientific Research, what percentage of obese people are GENEROUS?’’
Peer consensus
‘‘According to [name of university] students, what percentage of obese people are LAZY?’’
Scientific research
‘‘According to Scientific Research, what percentage of obese people are LAZY?’’
Less than 60% 61—68% 69—76% 77—84% 85—92% 93—100% Less than 60% 61—68% 69—76% 77—84% 85—92% 93—100% 0—10% 11—20% 21—30% 31—40% 41—50% Greater than 50% 0—10% 11—20% 21—30% 31—40% 41—50% Greater than 50%
0—7% 8—15% 16—23% 24—31% 32—39% Greater than 40% 0—7% 8—15% 16—23% 24—31% 32—39% Greater than 40% Less than 50% 51—60% 61—70% 71—80% 81—90% 91—100% Less than 50% 51—60% 61—70% 71—80% 81—90% 91—100%
Note. Response options were modified to range by 5,7,10 and 15 points (e.g. 0—5%, 0—7%, 0—10% and 0—15%) for various questions. The biased response scales consisted of 20 items with ten positive (i.e. humerous, generous, sociable, friendly, outgoing, intelligent, honest, productive, warm personality & organized) and ten negative (i.e. lazy, self-indulgent, undisciplined, gluttonous, unhealthy, sluggish, lack of willpower, unclean, insecure & unattractive) traits modeled after the Obese Persons Trait Scale (OPTS).
manipulation used response option sets that were either positively or negatively skewed. The positive response anchor condition was biased toward representing a high prevalence of desirable personality traits and low prevalence of undesirable personality traits among people with obesity. Conversely, the negative response anchor condition was biased toward representing a high prevalence of undesirable personality traits and low prevalence of desirable personality traits among people with obesity. Within both response conditions, answer choice increments were randomly varied between 5, 7, 10, and 15 percentage points in order to disguise the manipulation. See Table 1 for an example question for 7% and 10% response anchor manipulations. After the experimental manipulation, participants completed the dependent measures (See additional information on Obese Persons Trait Survey and a hiring vignette below). Lastly, participants completed the Obese Persons Trait Survey one week later. Following the completion of all measures, participants were given the opportunity to share comments about the study and were debriefed.
Measures Obese Persons Trait Scale (OPTS [2]) The OPTS was administered following the experimental manipulation and again one week following the study. As indicated earlier, the OPTS consists of 20 items listing stereotypical traits [2]. Ten undesirable common weight stereotypes (e.g., undisciplined) form an undesirable traits subscale and 10 desirable common weight stereotypes (e.g., intelligent) form a desirable trait subscale. Participants were asked to estimate the percentage (0—100%) of obese persons who possess these traits. For desirable traits, Cronbach’s ˛ was .90 postmanipulation and .90 at one-week follow-up. For undesirable traits, Cronbach’s ˛ was .92 postmanipulation and .91 at one-week follow-up Hiring scenario To measure behavioral components of weight bias, participants were asked to engage in a hiring scenario shortly following the experimental manipulation. They were asked to pretend to be a Student Tour Coordinator at their university who is responsible for hiring a new tour guide. Participants were
452 given criteria to help determine who would be a good candidate (e.g., ‘‘educating prospective students about the admission process’’) and were then asked to review the cover letter and casual photo of an applicant. The cover letter was created to suggest that the applicant would be a good candidate (e.g., ‘‘I am an honors student at [name of the university] majoring in Education’’) and the casual photo was of a young, college-aged woman who was obese. Afterwards, participants were asked to rate how likely they would be to hire this applicant on a 9-point Likert scale (ranging from Very Unlikely to Very Likely). Marlowe—Crowne Social Desirability Scale (MC-SDS [7]) The MC-SDS was administered at baseline to control for socially desirable responding. The MC-SDS is a widely used, well-validated measure of social desirability response bias. Participants rated 33 items as True or False (e.g., ‘‘I never resent being asked to return a favor’’). Attitudes Towards Obese Persons (ATOP [8]) The ATOP was administered at baseline to examine for baseline differences in weight bias between conditions. The ATOP consists of 20 items assessing positive (e.g., ‘‘obese people are usually sociable’’) and negative (e.g., ‘‘obese people tend to have family problems’’) attitudes towards people with obesity [8]. Participants responded using a 6point Likert scale ranging from Strongly Disagree to Strongly Agree. Cronbach’s ˛ was .82 for these items.
Data analysis Statistical analysis was conducted such that demographic differences between groups and socially desirable responding were entered as covariates, and conditions (peer consensus versus scientific research condition and positive versus negative bias response anchor conditions) were entered as fixed factors. Thus, all hypotheses were tested using 2 × 2 Analysis of Covariance (ANCOVA). Given the theoretical approach and expected direction of findings, all hypothesis testing was conducted using one-way significance tests.
Results Covariates Demographic information and baseline weight bias (measured using the ATOP) were tested in order
R.A. Carels et al. to ensure that there was equal representation in all four conditions. Baseline weight bias did not differ across conditions. Gender was included as a covariate, because there was a disproportionate number of females in both peer consensus conditions, 2 (3) = 8.02, p < .045. BMI was also included as a covariate because people in the peer consensus negative response anchor condition had significantly lower BMIs than those in the scientific research positive response anchor condition, F(3, 168) = 3.34, p = .02. Finally, social desirability was included as a covariate, as this variable was significantly related to hiring likelihood r(169) = .201, p = .01.
Experimental manipulation: effects of interval spacing on change scores A manipulation check was conducted on positive and negative response anchors. In reaction to the response anchor manipulation, participants receiving the negative response bias anchors estimated that their peers and the scientific community possessed significantly higher estimates of the prevalence of undesirable traits in individuals with obesity and significantly lower estimates of the prevalence of desirable traits of individuals with obesity. Two exceptions were the negative trait ‘‘unclean’’ t(195) = 1.18, p = .241 and the positive trait ‘‘productive’’ t(194) = 1.60, p = .11. Next, we examined whether the size of the anchor interval affected the amount of change in personality trait ratings when responding to each question. Intervals existed such that some traits were rated in 5% (e.g., ‘‘95—100%,’’ ‘‘90—95%,’’’’85—90%,’’ ‘‘80—85%,’’ and ‘‘Less than 80%’’), 7%, 10%, and 15% intervals. There was a significant effect of anchor interval F(3, 610) = 16.581, p < .001 with 2 = .08. Post hoc comparisons using the Bonferroni correction indicated that the 5% interval anchor sets produced greater changes (M = 22.78, SE = .62) from the inferred value than did the 7% (M = 20.05, SE = .63), the 10% (M = 17.94, SE = .69), and the 15% (M = 17.05, SE = .62) intervals. Similarly there were significant mean change score differences between the 7% and 15% intervals. There were no significant differences between the 10% and 15% intervals.
Undesirable personality traits Following the experimental manipulation, there were significant main effects for the scientific research versus peer consensus manipulation and for the positive versus negative bias anchor manipulation immediately following the manipulations.
Anchor effects and weight bias
453
Figure 1 Undesirable traits post manipulation and follow-up.
The effects existed such that those individuals in the scientific research conditions made significantly higher undesirable personal attributions of obese individuals than those in the peer consensus conditions. Similarly, participants in the negative response bias anchor conditions had significantly higher undesirable personal judgments of individuals with obesity than those in the positive response anchor conditions immediately following the intervention (Table 2 and Fig. 1). At follow-up, there was a significant main effect such that participants in the negative response anchor conditions made significantly higher undesirable attributions than those in the positive response anchor conditions (Table 2 and Fig. 1). There was no interaction effect immediately or one week following experimental manipulation.
conditions and a significant interaction effect (Table 3 and Fig. 2). Differences existed such that those in the scientific research, negative response bias anchor condition rated individuals with obesity significantly less desirably than individuals in the peer consensus conditions. Furthermore, there was a significant interaction effect which demonstrated that participants in the scientific research negative response anchor condition showed lower attributions of desirable traits for individuals with obesity than participants in any of the other three conditions. Regarding desirable trait attributions at follow-up, there were no significant main or interaction effects between peer consensus and scientific research conditions and between positive and negative response anchors conditions.
Hiring likelihood Desirable personality traits For desirable trait attributions, immediately following the intervention, there was a main effect for the peer consensus versus scientific research
For the hiring scenario, there was a significant interaction effect (Table 4 and Fig. 2), such that those in the scientific research positive response bias anchor condition were more likely to hire
Figure 2 Post-manipulation desirable trait and hiring likelihood.
454
Table 2 Condition
ANCOVA table of undesirable traits post manipulation and at one-week follow-up. Positive post (N)
Positive post M (SD)
Negative post (N)
Scientific research 47
56.49 42 (18.11) Peer consensus 45 51.50 38 (18.77) Marginal means 92 54.05 80 (18.51) Source F(1,165) = 2.68, p = .05 with 2 = .016* Anchor valance F(1,165) = 2.812, p = .04 with 2 = .017* Interaction effects F(1,165) = .033, p = .86 with 2 < .001
Table 3 Condition
Negative post M (SD)
Post marginal means N (M) (SD)
Positive follow up N
62.77 (15.84) 58.60 (21.73) 60.79 (18.86)
89 (61.81) (13.00) 83 (54.75) (20.36) 172 (57.19) (18.91)
39
Positive follow up M (SD)
Negative follow up N
54.09 38 (17.96) 36 53.11 34 (17.15) 75 53.62 72 (17.47) F(1, 140) = 2.091, p = .08 with 2 = .015 F(1, 140) = 2.785, p = .05 with 2 = .020* F(1, 140) = 1.386, p = .12 with 2 = .010
Negative follow up M (SD)
Marginal means post N (M) (SD)
63.64 (13.09) 56.29 (20.59) 60.17 (17.31)
77 (58.80) (16.36) 70 (54.66) (11.40) 147 (56.83) (17.64)
Negative follow up M (SD)
Marginal means post N (M) (SD)
55.71 (9.83) 60.01 (17.65) 57.74 (14.14)
77 (58.90) (11.40) 70 (59.88) (11.40) 147 (59.36) (13.58)
ANCOVA of desirable traits post manipulation and at one-week follow-up. Positive post (N)
Scientific research 47
Positive post M (SD)
Negative post (N)
Post marginal means N (M) (SD)
Positive follow up N
57.40 (12.94) 65.09 (16.85) 61.05 (15.32)
89 (61.81) (13.00) 83 (65.30) (15.54) 172 (63.49) (14.35)
39
Positive follow up M (SD)
Negative follow up N
62.00 38 (12.08) 36 59.75 34 (13.95) 75 60.92 72 (12.92) F(1, 140) = .711, p = .201 with 2 < .01 F(1, 140) = .316, p = .290 with 2 < .01 F(1, 140) = 2.52, p = .058 with 2 = .02
R.A. Carels et al.
65.75 42 (11.86) Peer consensus 45 65.47 38 (14.54) 92 Marginal means 65.61 80 (16.16) F(1,165) = 3.91, p = .03 with 2 = .023* Source Anchor valance F(1,165) = 2.13, p = .08 with 2 = .013 Interaction effects F(1,165) = 3.673, p = .03 with 2 = .022*
Negative post M (SD)
Anchor effects and weight bias Table 4
455
ANOVA prediction of hiring likelihood by condition.
Condition
Positive (N)
Scientific research
47
Positive M (SD)
7.21 (1.63) Peer consensus 45 6.87 (2.06) Marginal means 92 7.04 (1.85) Trait prevalence versus peer consensus Positive versus negative Interaction effect
Negative (N) 41
Negative M (SD)
Marginal means (N)
6.22 88 (1.96) 38 7.05 83 (1.83) 79 6.62 171 (1.93) F(1, 164) = .293, p = .30 with 2 = .002 F(1, 164) = .576, p = .15 with 2 = .003 F(1,164) = 3.704, p = .03 with 2 = .022*
the applicant than participants in the scientific research negative response bias anchor condition. There was no statistically significant difference between the peer consensus positive response bias anchor condition and the peer consensus negative response bias anchor condition.
Discussion Previous research has shown that providing people with direct, explicitly positive peer consensus information or scientific research biased information about people with obesity can increase the positivity of their short-term social attitudes towards people with obesity [2,9]. While important, these studies possess fairly strong manipulations of peer attitudes or scientific research. The current study suggests that attitudes toward individuals with obesity are susceptible to indirect, subtle manipulations of peer consensus and scientific research information by manipulating questionnaire response anchors. The current research study suggests that attitudes may be influenced when response option numerical values are skewed in a positive or negative direction on a simple questionnaire. In this investigation, regardless of whether participants estimated the attitudes of their peers or estimated the outcomes of scientific research, participants exposed to the positive response bias anchor manipulation later reported that individuals with obesity possessed fewer undesirable personality traits (e.g., lazy) immediately and one week following the experimental manipulation, relative to participants exposed to the negative response anchor manipulation. Additionally, in the scientific research condition, participants exposed to the positive response bias anchor manipulation later reported that individuals with obesity possessed more desirable personality traits (e.g., industrious)
Marginal means M (SD) 6.75 (1.85) 6.95 (1.95) 6.85 (1.89)
immediately following the experimental manipulation. Finally, after the experimental manipulation, participants exposed to the scientific research, positive response bias anchor manipulation also expressed greater likelihood that they would hire an applicant with excess weight for employment relative to participants exposed to the scientific research, negative response bias anchor manipulation. These findings are consistent with several independent lines of research. Past research on the cognitive aspects of survey methods indicates that people may be indirectly influenced by biased information presumed to come from either the majority of their peers or from perceived high credibility sources, likely because they automatically employ the anchoring heuristic to make quick judgments under conditions of uncertainty [4,10]. In the current study, biased anchor sets influenced participant estimates of their peers’ attitudes or their estimates of the outcomes of scientific research. More importantly, when later asked to report their own attitudes toward individuals with obesity, their reporting were predicted by the experimental manipulation. For example, when the experimental manipulation asked participants to provide an estimate of the percentage of individuals with obesity that are lazy according to scientific research using a negative response bias anchor set, they were more likely to indicate that people with obesity are lazy on a subsequent measure of weight bias, as compared to participants provided with a positive bias anchor set who were asked to provide the same estimate. Interestingly, one week following the experimental manipulation, the manipulation’s influence on judgments of undesirable traits remained. Much like Wittenbrink and Henley [5] study, the current study expanded on the assessment of attitudes to include an assessment of generalizable prejudice (i.e., likelihood of hiring). Among those in the scientific research condition, participants who received the negatively
456 biased anchor sets reported that they would be less likely to hire a job applicant with obesity than participants who received the positive response bias anchor sets. The disadvantages in hiring, wages, promotions, and job termination resulting from weight stigma are becoming increasingly well documented [1]. As such, the results from this investigation are hardly surprising, with the exception that the preferences in hiring were influenced by a subtle, indirect manipulation whereby respondents were asked to make judgments regarding scientific research. In other words, by simply asking participants to estimate scientific research findings regarding personality traits of individuals with obesity utilizing a measure with biased response anchors, subsequent judgments on the suitability of employment were influenced. This influence likely occurred outside of respondents’ awareness and is consistent with research on priming and the malleability of automatic stereotypes and prejudice [11]. Nevertheless, it is concerning, especially considering that individuals with obesity are not protected under workplace discrimination laws. Research indicates that extreme response anchors have been shown to exert a stronger influence on decisions under conditions that feature relatively higher ambiguity, lower familiarity, lower individual knowledge about a subject, and a more trustworthy source of information [12—15]. In the current investigation, respondents had little knowledge or expertise regarding the true estimates of specific personality traits that people with obesity possess. Also, it is probable that participants had a limited knowledge about what other students at their university might believe regarding people with obesity. These are conditions that would make a respondent susceptible to anchoring effects. However, it is notable, that while attitudes were changed under both the scientific research and peer consensus conditions, the scientific research condition had a significantly greater influence on desirable trait estimation and hiring likelihood than the peer consensus condition. It is not entirely clear why hiring decisions and desirable trait estimations are more responsive to engaging in an estimation of scientific research-based prevalence of personality traits in obese individuals using biased response anchors as compared to engaging in estimation of peer consensus on the prevalence of personality traits in obese individuals using biased response anchors. Perhaps peer influence is sufficient to influence attitude change toward a target regarding undesirable personality attributes, but
R.A. Carels et al. hiring decisions or attitude change toward a target regarding desirable personality attributes requires greater source credibility. Although the methodology used to produce attitude change in this study was very different that the methodology used in Puhl et al’s study [2], the findings are consistent with Puhl and colleagues [2] research, which demonstrated a modestly stronger effect for a scientific research manipulation as compared to a peer consensus manipulation on the outcome of improving short-term social judgments regarding the traits of individuals with obesity. The between subjects design of this investigation leaves the findings vulnerable to the interpretation that between group differences somehow accounted for the findings. However, differences between conditions were included as covariates in all analyses and several post hoc analyses were consistent with the argument that the experimental anchor manipulation produced the intended effects. First, the size of the anchor interval manipulation (5% versus 7% versus 10% versus 15%) was associated with the magnitude of change for participant estimates of the percentage of obese people whom they believe possess certain traits. In other words, when participants were asked to provide estimates of their peers’ attitudes or estimates of the findings from scientific research regarding the personality traits of individuals with obesity using more extreme anchor sets (those with a small range of very high or low numerical values), they tended to respond more extremely in the expected direction on subsequent attitude assessments of identical personality attributes. Therefore, the strength of the anchor manipulation had a direct impact on subsequent ratings. Similarly, and perhaps no less importantly, the response anchor manipulation itself resulted in respondents choosing differing estimates of the percentage of obese people possessing different personality traits according to peers or the outcomes of scientific research in nearly all cases with only two exceptions (the personality traits ‘‘unclean’’ and ‘‘productive’’.) While it is reasonable to assume that the response anchor manipulations modified attitudes, it cannot be ruled-out that feedback influenced respondents’ assumptions about the acceptability of expressing or not expressing negative beliefs. Therefore, rather than change attitudes, the manipulation may have motivated some participants to more readily express their negative attitudes and inhibited other participants from expressing their negative attitudes. Also, while evidence for the existence of changes in weight bias and its potential impact on hiring was detected,
Anchor effects and weight bias
457
the measures utilized in this investigation were not measures of actual behavioral displays of this bias. Thus, application of the current findings to the ‘‘real’’ discrimination and bias that people with obesity may experience will be an important focus for future research. Finally, the primarily Caucasian, female university sample limits generalizability of the current study’s findings to samples of differing gender, race, age, and education. Response scales are not simply measurement devices that respondents blithely complete without consideration. Rather, respondents are likely to assume that the response alternatives reflect ‘‘real’’ knowledge or the beliefs of others. When the questions are ambiguous or the answers to the questions are not readily accessible, response alternatives may serve as a frame of reference in estimating and evaluating behavior [10]. As such, social judgments may be quite susceptible to subtle manipulation. As a caution to researchers, it is probably safe to assume that people’s attitudes are influenced by the very measures that they take. If questions are extreme, respondents may assume more extreme attitudes than they originally possessed. Additionally, people may make biased judgments about individuals who are overweight or obese due to exposure to biased visual and numerical information in their environment. Again, the exposure does not have to be extreme or direct, research suggests that providing individuals with subtle, indirect cues regarding purported scientific research findings or peers’ views may be enough to influence recipient attitudes. As such, exposure to negative media depictions of personal characteristics of individuals with obesity may lead people to express more negative beliefs and attitudes towards overweight individuals [16]. These subtle influences may have lasting effects on an individual’s attitudes and behavior and the impact of such manipulations on weight bias and prejudice is an important area of future inquiry.
Conflict of interests The authors declare not conflict of interests.
References [1] Puhl RM, Heuer CA. The stigma of obesity: a review and update. Obesity 2009;17(5):1—24. [2] Puhl RM, Schwartz M, Brownell KD. Impact of perceived consensus on stereotypes about obese people: a new approach for reducing bias. Health Psychol 2005;24(5):517—25. [3] Tversky A, Kahneman D. Judgment under uncertainty: heuristics and biases. Science 1974;125:1124—31. [4] Schwarz N. Self-report: how the questions shape the answers. Am Psychol 1999;54(2):93—105. [5] Wittenbrink B, Henley JR. Creating social reality: informational social influence and the content of stereotypic beliefs. Personal Soc Psychol Bull 1996;22:598—610. [6] Belsky G, Gilovich T. Why people make big money mistakes. New York: Simon and Schuster; 1999. [7] Crowne DP, Marlowe D. A new scale of social desirability independent of psychopathology. J Consult Psychol 1960;24:349—54. [8] Allison DB, Basile VC, Yuker HE. The measurement of attitudes toward and beliefs about obese persons. Int J Eat Disord 1991;10(5):599—607. [9] Stangor C, Sechrist GB, Jost JT. Changing beliefs by providing consensus information. Personal Soc Psychol Bull 2001;27:486—96. [10] Furnham A, Boob HC. A literature review of the anchoring effect. J Socioecon 2011;40:35—42. [11] Blair IV. The malleability of automatic stereotypes and prejudice. Personal Soc Psychol Bull 2002;6(3):242—61. [12] Van Excel N, Brouwer W, van der Berg B, Koopmanschap M. With a little help from an anchor: discussion and evidence for anchoring effects in contingent valuation. J Socioecon 2006;35(836):853. [13] Mussweiler T, Strack F. Considering the impossible: explaining the effects of implausible anchors. Soc Cognit 2001;19(145):160. [14] Mussweiler T, Strack F. Numeric judgments under uncertainty: the role of knowledge in anchoring. J Exp Soc Psychol 2000;36:495—518. [15] Smith AR, Windschitl PD, Bruchmann K. Knowledge matters: anchoring effects are moderated by knowledge level. Eur J Soc Psychol 2013;43:97—108. [16] Pearl RL, Puhl RM, Brownell KD. Positive media portrayals of obese persons: impact on attitudes and image preferences. Health Psychol 2012;31(6):821—9.
Available online at www.sciencedirect.com
ScienceDirect