When not to accentuate the positive: Re-examining valence effects in attribute framing

Organizational Behavior and Human Decision Processes 124 (2014) 95–109

Contents lists available at ScienceDirect

Organizational Behavior and Human Decision Processes journal homepage: www.elsevier.com/locate/obhdp

When not to accentuate the positive: Re-examining valence effects in attribute framing Traci H. Freling a,⇑, Leslie H. Vincent b, David H. Henard c a

University of Texas at Arlington, 217 College of Business Administration, Arlington, TX 76019, USA University of Kentucky, Gatton College of Business & Economics, Lexington, KY 40506, USA c North Carolina State University, Poole College of Management, Raleigh, NC 27695, USA b

a r t i c l e

i n f o

Article history: Received 26 March 2012 Accepted 7 December 2013 Accepted by Harris Sondak Keywords: Message framing Framing effects Construal level theory Meta-analysis

a b s t r a c t While the expanding body of attribute framing literature provides keen insights into individual judgments and evaluations, a lack of theoretical perspective inhibits scholars from more fully extending research foci beyond a relatively straightforward examination of message content. The current research applies construal level theory to attribute framing research. The authors conduct a meta-analysis of 107 published articles and then conceptually expand this knowledge base by synthesizing attribute framing research and construal level concepts. Results suggest that attribute framing is most effective when there is congruence between the construal level evoked in a frame and the evaluator’s psychological distance from the framed event. A follow-up experiment confirms that the congruence between a frame’s construal level and psychological distance—not simply its valence—appears to be driving attribute framing effects. This research proposes to shift the focus in attribute framing research from that of message composition to a more complex relationship between the message and the recipient. Ó 2014 Elsevier Inc. All rights reserved.

Introduction ‘‘4 out of 5 dentists surveyed would recommend sugarless gum. . .’’ This now infamous advertising tagline for Trident gum has been used for nearly 50 years and is a prime example of how marketing managers successfully use framing in persuasive messages. Parallel to its use in practice, research on framing effects and their impact on decision-making continues to proliferate and bears tribute to the interest level in the subject area. In line with Krishnamurthy, Carter, and Blair (2001) we define framing, in general, as presenting individuals with logically equivalent options in semantically different ways. Framing scholars traditionally focus their research on one of three frame types: attribute, risky choice, or goal framing. While each of these focal areas provides insight into various facets of choice, Levin, Schneider, and Gaeth (1998) warn that the three different types of framing should be examined independently to avoid unnecessary complexity and confusion that can result from their idiosyncratic characteristics. In this current research, we therefore focus our attention on the effects of attribute framing, wherein the object of the frame is an attribute of the decision option.

⇑ Corresponding author. Fax: +1 (817) 272 2854. E-mail addresses: [email protected] (T.H. Freling), [email protected] (L.H. Vincent), [email protected] (D.H. Henard). http://dx.doi.org/10.1016/j.obhdp.2013.12.007 0749-5978/Ó 2014 Elsevier Inc. All rights reserved.

Extant empirical works overwhelmingly indicate that people are more receptive to positive (e.g., 4 out of 5 dentists recommend Trident sugarless gum) vs. negative (e.g., Only 1 out of 5 dentists does not recommend Trident sugarless gum) attribute frames. Krishnamurthy et al. (2001) explain that positive framing is more effective because it ‘‘generates more positive associations and thus seems more attractive than negatively framed options’’ (p. 383). Levin et al. (1998) support this contention, stating ‘‘even at the most basic level the valence of a description often has a substantial influence on the processing of that information’’ (p. 184). Given this, one could view knowledge of attribute framing as fait accompli, concluding that attribute framing effects are so straightforward that the results are, statistically speaking, nearly always positive and that any differences in outcomes are simply a matter of degree or a result of study artifacts. While valence effects in attribute framing are ‘‘a reliable phenomenon’’ (Levin, Gaeth, Schreiber, & Lauriola, 2002, p. 413), viewing them as straightforward is problematic. Research regularly reveals that seemingly straightforward relationships are often more complex when viewed from different levels of analysis. Toward this end, researchers have identified various moderators of valence effects including the nature of the product (Khan & Dhar, 2010), personal involvement with the framed issue (Chan & Mukhopadhyay, 2010), and processing motivation and opportunity (Shiv, Edell Britton, & Payne, 2004). Together, these studies suggest

96

T.H. Freling et al. / Organizational Behavior and Human Decision Processes 124 (2014) 95–109

that something more than valence could be driving attribute framing effects. Since the publication of seminal works in this area (Krishnamurthy et al., 2001; Levin et al., 1998), there has been a continued expansion of published research. Thus, the first goal of this manuscript is to conduct a meta-analysis of the research stream to update the empirical base of knowledge on attribute framing. While much of the early work on attribute framing involved exploring the impact of positive vs. negative message attributes (i.e., valence effects), a growing trend toward investigating other issues such as differing frames of reference and temporal contexts has developed. It is our view that this is an impactful and meaningful research evolution. The second goal of this manuscript is therefore to determine if components of construal level theory constitute important structural determinants of framing effects that could possibly encompass both earlier generalizations focusing on valence and more recent work. We believe that a theory-driven, micro-focused examination of the attribute framing literature will yield insights that build upon and extend existing research. Construal level theory (Liberman & Trope, 1998; Trope & Liberman, 2010; Trope, Liberman, & Wakslak, 2007) is a useful and relevant conceptual lens through which to view attribute framing effects. This is due to its treatment of events and issues as differing in terms of construal level and psychological distance, which together can impact resulting evaluations. The dimensions of both construal level and psychological distance map favorably with key variables manipulated in attribute framing research. Furthermore, by incorporating construal level theory into the extant attribute framing literature, we are better able to meaningfully understand the nuances of concomitant effects. While much of the framing literature has focused on message construction, using a construal level perspective to guide our investigation allows us to examine the interaction between the message and the recipient, thus providing a richer understanding of the phenomena. This approach makes the subsequent findings relevant for any individual in an organization who is responsible for crafting persuasive messages in a host of managerial, negotiation, selling, evaluation, or promotional situations. In the following section, we discuss the conceptual foundations that underpin existing attribute framing research. We then conduct a meta-analysis designed to both update the current base of attribute framing research knowledge and expand that knowledge with a fine-grained theoretical perspective using construal-level theory. Building on the meta-analysis, we advance the attribute framing literature by conducting an experiment that investigates the outcome effects emanating from the congruency between the evoked construal level of a message frame and the perceived psychological distance of intended message recipients. The manuscript concludes with a discussion of the results and implications for framing scholars and practitioners responsible for developing persuasive messages.

Conceptual development Attribute framing Kahneman and Tversky (1979) were the first researchers to demonstrate that framing (i.e., different wording of formally identical problems) makes individuals code decision outcomes as gains or losses relative to a reference point. Since that groundbreaking work, empirical research on framing effects has flourished across multiple research domains including cognition, psycholinguistics, perception, social psychology, health psychology, clinical psychology, educational psychology, and marketing (Kühberger,

1998). While the term ‘‘framing’’ includes all of the various ways decision situations are presented that lead decision-makers to construct markedly different representations of such situations (Kühberger, 1995), we focus exclusively on attribute framing, in which a single attribute within a given context is the subject of the framing manipulation (e.g., describing ground beef as ‘‘80% lean’’ or ‘‘20% fat’’). We distinguish attribute framing from two other types of framing identified by Levin et al. (1998): risky choice framing, which describes the outcomes of a potential choice involving options differing in level of risk (e.g., presenting two programs differing in risk level for reducing cholesterol described in terms of either positive or negative outcomes); and goal framing, where the goal of an action or behavior is framed (e.g., stressing either the positive consequences of reducing red meat in one’s diet or the negative consequences of failing to do so). Levin et al. (1998) cogently assert that research exploring these different types of frames is qualitatively different because attribute, goal, and risky choice framing involve different mechanisms and consequences, and vary in terms of the information that is framed, the presumed outcome of the frame, and the manner in which effects of the frame are measured (see Table 1, p. 151). Levin et al. (2002) empirically corroborate these theoretical propositions using a within-subjects framing manipulation in a study conducted across two sessions in which each subject saw both framing conditions and all three types of frames. Among other key insights, Levin et al. (2002) demonstrate significant effects for attribute and risky choice framing, but not goal framing and conduct direct test of dependency suggesting the three types of framing are governed by difference processes that are independent of each other. These results provide further empirical support for the decision to solely concentrate on attribute framing in this meta-analysis. Another contribution of Levin et al. (1998) is their identification of a ‘‘valence-consistent shift’’ that is found in most attribute framing studies, wherein a positive description of attributes leads to more favorable evaluations than a negative frame. A classic demonstration of this valence-consistent shift is provided by Levin and Gaeth (1988), where ground beef was rated as better tasting and less greasy among subjects exposed to a ‘‘75% lean’’ frame compared to those in a ‘‘25% fat’’ frame. In other attribute framing studies, subjects evaluate issues described in terms of ‘‘success’’ or ‘‘survival’’ rates vs. ‘‘failure’’ or ‘‘mortality’’ rates (Davis & Bobko, 1986; Dunegan, 1993; Levin, Schnittjer, & Thee, 1988; Linville, Fischer, & Fischhoff, 1993; Marteau, 1989), or assess gambling contexts that are portrayed in terms of probability of ‘‘winning’’ or ‘‘losing’’ (Levin, Snyder, & Chapman, 1989; Levin et al., 1986). In such studies, the alternative framed in a more positive light is routinely rated more favorably than when described negatively. While this valence-consistent shift has been amply demonstrated in the literature, the body of work on attribute framing has tremendously expanded since Levin et al.’s (1998) article and warrants a new review and synthesis. Interestingly, much of the recent attribute framing research foregoes a valence manipulation and explores the effects of presenting numeric information in different formats (e.g., dollars vs. cents), providing different frames of reference (e.g., self vs. others), or varying the temporal context (e.g., now vs. in the future). While earlier attribute framing research primarily assessed effects in terms of evaluations, many recent studies use alternative criterion variables such as behaviors, behavioral intentions, estimates, and predictions. To explore and better understand these important qualitative differences, we draw upon construal level theory (CLT) to integrate previous attribute framing research findings into a theoretical framework that allows us to make specific predictions about when effects should be stronger across different types and contexts of attribute framing.

97

T.H. Freling et al. / Organizational Behavior and Human Decision Processes 124 (2014) 95–109 Table 1 Main effect results for attribute framing effects.

Evaluations Estimates Behaviors a b *

Number of samples (k)

Number of observations (N)

Mean correlation (r)a

Weighted correlation (rW)b

Weighted variance (vart)

95% Confidence Interval (CIBS)

80% Credibility Interval (CI)

Unaccounted variance (v2)

Fail-safe sample size (NfsR)

359 70 175

51,665 9785 26,876

.25 .30 .23

.25 .24 .21

.041 .018 .015

.22–.28 .20–.29 .19–.23

.02 to .52 .05–.43 .05–.37

3274.56* 491.55* 618.21*

413,085 16,467 91,196

The ‘‘mean correlation’’ is a simple average among all of the coded effect sizes reported for each relationship and is un-weighted. The ‘‘weighted correlation’’ is the reliability-corrected or sample-size weighted mean correlation to account for sampling error. p 6 .05.

Construal level theory According to CLT, an association forms between an individual’s perceptions of his psychological distance from an event being evaluated and the construal level of the information he receives about that event (Liberman & Trope, 1998). This, in turn, influences that individual’s outcome evaluations. Applying CLT terminology to attribute framing research, an ‘‘event’’ translates as the object or situation being evaluated, ‘‘construal level’’ is the mental representation evoked by the attribute framing information provided, and ‘‘psychological distance’’ equates to any dimension that affects how closely the individual perceives himself to be from the framed event or issue. Although attribute framing studies have not previously drawn upon CLT to explain observed effects, its central constructs—construal level and psychological distance—are readily applicable to this body of work. In line with Trope and Liberman (2010), we conceptualize construal level as the degree of perceived abstractness that an event holds for an individual. Low-level construals are relatively unstructured, contextualized representations that include subordinate and incidental features of the event. In contrast, high-level construals are structured, de-contextualized representations that include only a few superordinate core features of the events (Trope & Liberman, 2003). CLT posits that individuals construct different representations of the same information at varying levels of abstraction (Liberman & Trope, 1998). For example, one might have a career ambition of ‘‘being successful’’ with a personal ambition of ‘‘spending quality time with my family’’ (both abstract, high-level construals) or a career ambition of ‘‘being a productive researcher and a respected teacher’’ with a personal ambition of ‘‘visiting Disneyworld next week with my family’’ (both concrete, low-level construals). How an individual construes an event is affected by that person’s psychological distance from the event—his perceptions of temporal distance (when an event occurs), spatial distance (where it is likely to occur), social distance (to whom it occurs), or hypothetical distance (whether it occurs). The closer an event is to the individual, the more concrete it is perceived; the greater the distance from the event, the more abstract the perception. CLT research demonstrates events that take place farther into the future (Liberman, Sagristano, & Trope, 2002; Liberman & Trope, 1998; Wakslak, Nussbaum, Liberman, & Trope, 2008), that occur in more remote locations (Fujita, Henderson, Eng, Trope, & Liberman, 2006a; Henderson, Fujita, Trope, & Liberman, 2006), that are less likely to occur (Todorov, Goren, & Trope, 2007; Wakslak, Trope, Liberman, & Alony, 2006), and that happen to people less similar to the evaluator (Liviatan, Trope, & Liberman, 2008; Smith & Trope, 2006) are associated with greater psychological distance and thus perceived as more abstract. Beyond these four primary psychological distance dimensions, recent research identifies additional distance variables that could also affect construal level, such as informational distance (i.e., the amount of knowledge or relevant data the consumers possesses about decision options), experiential distance (i.e., the degree to which information is based on direct

experience), affective distance (i.e., the emotional intensity of the decision context), and perspective distance (i.e., one’s commitment level in the decision process) (Fieldler, 2007). Studies exploring the effects of attribute framing vary considerably in terms of how the framing information is presented to subjects to evaluate. For example, Shiv et al. (2004) broadly position the stimulus product in their experiment as simply ‘‘better than’’ or ‘‘worse than’’ the competitor (high-level construal). In contrast, Buda and Zhang (2000) frame the stimulus product in their experiment using relatively more concrete details such as ‘‘85% of customers were satisfied with the product’’ vs. ‘‘15% of customers were dissatisfied with the product’’ (low-level construal). There is also substantial variation in studies exploring the effects of attribute framing, in terms of how psychologically distant the decision event is from the person exposed to the frame. To illustrate, Dunegan (1996) requires student subjects to make business resource allocation decisions (a scenario that is high in hypothetical distance for this sample), while student subjects in Agrawal and Duhachek’s (2010) experiments evaluate ad appeals for anti-drinking messages (a scenario that is low in hypothetical distance for this sample). Thus, both construal level and psychological distance appear to be appropriate conceptual constructs to enhance our understanding of attribute framing effects. Hypotheses Viewing attribute framing research through a CLT lens allows scholars to investigate certain nuances and micro-level dynamics at play in persuasive messaging. That is, the application of a CLT perspective to attribute framing data provides an opportunity to test theoretical factors that could have an impact on relationships of interest. In the current research, meta-analysis becomes an important tool for both testing and expanding theory. Applying CLT to attribute framing, we assert that an individual’s construal level is largely dictated by the relative abstractness of the information presented in the frame—a research design factor that is wholly under the control of the message developer and one that can affect perceptions of the attribute frame. Based on CLT research, we further anticipate that the psychological distance inherent in the decision context or task makes that information more or less persuasive depending on its congruence with how the information in the frame is construed. Specifically, attribute framing effects should be stronger when the construal level and psychological distance in the framing manipulation are congruent. We expect that when the construal level of the frame is aligned with subjects’ perceptions of how distant or proximal the event is, the framing effect should be stronger and vice versa. This prediction is consistent with research by Zhao and Xie (2011) who explore the effectiveness of temporal attribute frames when social distance is also manipulated. Their findings suggest that congruency of construal levels between distal others and distant time results in others’ recommendations having a stronger impact on subjects’ preferences and choice shifts for the distant future than for the near future (when the construal levels between distal

98

T.H. Freling et al. / Organizational Behavior and Human Decision Processes 124 (2014) 95–109

others and proximal time did not match). Other researchers have employed CLT to explore goal framing in recent empirical work. Lee, Keller, and Sternthal (2010) explore the effectiveness of prevention vs. promotion frames at high vs. low levels of construal level, demonstrating that attitudes and performance on subsequent tasks (i.e., solving anagrams) are enhanced when subjects process information construed at a level that fits with their regulatory focus. White, MacDonnell, and Dahl (2011) test a similar ‘‘matching hypothesis,’’ providing evidence that loss (gain) goal frames more effectively influence consumer behavior when paired with low-level, concrete (high-level, abstract) marketing appeals. These newly published studies inform our theorizing about how CLT applies to attribute framing. Based in part on findings by Zhao and Xie (2011), Lee et al. (2010), and White et al. (2011), we predict stronger attribute framing effects for congruent pairings as compared to incongruent pairings (i.e., psychologically close scenarios with high-level construal or psychologically distant scenarios with low-level construal). Consistent with these framing studies and recent CLT research showing that a ‘‘match in messaging’’ results in increased fluency, ease of understanding, greater processing meaning, and stronger consumer responses (Ulkumen & Cheema, 2011), we expect stronger framing effects to occur when individuals in a psychologically close decision situation (e.g., evaluating a new product) receive information framed at a lower-level construal (e.g., 85% satisfied vs. 15% dissatisfied) (Buda & Zhang, 2000) and when individuals in a psychologically distant decision situation (e.g., making real estate investment decisions) are exposed to a frame evoking a higher-level construal (e.g., higher profits vs. lower expenses) (Brockner, Wiesenfeld, & Martin, 1995). We hypothesize this relationship across four facets of psychological distance (i.e., temporal, hypothetical, affective, informational).1

message concerns a more psychologically distant issue, whereas negative information—which leads to a lower level of construal— should be more effective for psychologically closer events. For example, Eyal et al. (2004) demonstrated that, as temporal distance from an event increased, subjects generated more arguments in favor of (and fewer arguments against) a particular action. Similarly, in research conducted by Herzog, Hansen, and Wanke (2007), distant-future actions were construed in terms of their pro aspects while near-future actions were construed in terms of their con aspects. As such, subjects found it easier to generate arguments in support of (against) actions pertaining to the distant (near) future. Likewise, Labroo and Patrick (2009) found that presenting an event as benign increased subjects’ abstract construal and the adoption of abstract, future goals while presenting an event as dangerous caused subjects to focus attention on concrete, proximal concerns and reduce the adoption of abstract, distant goals. In sum, framing an event by highlighting its positive aspects and inducing a higher level of construal is likely to be more persuasive when the individual evaluating the message perceives the event to be psychologically distant. Conversely, negative framing is likely to induce a lower level of construal, which will be more effective when the event is more psychologically close to the individual evaluating the message. This relatively recent development in CLT research is important because if this subordination of negative information does apply to attribute framing, it could provide a compelling explanation for the valence-consistent shift historically documented in attribute framing research, which would be a meaningful development for framing researchers. Based on construal level theory and these recent empirical findings, we offer the following prediction about the impact of valence information on construal level in attribute framing:

Hypothesis 1. Stronger attribute framing effects will occur when the construal level of the information presented is congruent with subjects’ psychological distance from the framed event.

Hypothesis 2. Stronger attribute framing effects will occur when the frame’s valence evokes a construal level that is congruent with the individual’s psychological distance from the framed event.

Because attribute framing research is replete with demonstrations of the valence-consistent shift (Levin et al., 1998), we also hypothesize—from a construal level perspective—how information valence impacts attribute framing effects. Interestingly, CLT scholars are increasingly treating information valence as a manifestation of construal level and suggest that positive information is construed at a higher level than negative information and that it is more valued in decision-making (Eyal, Liberman, Trope, & Walther, 2004). This is theorized to be true because negative information is only important when positive information is present, whereas the importance of positive information does not depend on the existence of negative information. For example, in deciding whether to accept a new job opportunity, an individual might consider the positive factors (e.g., assuming greater responsibility and earning more money) as well as any associated downsides such as learning a new position, having a longer commute, or increased overnight travel. If the job opportunity had no apparent benefits, individuals would likely not inquire about potential drawbacks and simply decline the offer. In contrast, the same individual would likely inquire about the job’s benefits even in the absence of perceived negatives. A handful of recent CLT studies examine the differential effects of positive vs. negative information in individual choice models and demonstrate that negative information is construed at a lower level than positive information in regards to decision-making. This research also suggests that positive information—which evokes a more abstract mindset—should be more persuasive when the 1 While CLT suggests that social, spatial, experiential, and perspective distances are also potential moderating factors, the articles comprising our dataset provided either insufficient information or sample size to include these factors.

Methodology Dataset development We identified extant empirical research studies focusing on attribute framing in multiple ways. First, we collected the articles referenced in seminal review pieces by Krishnamurthy et al. (2001) and Levin et al. (1998). Second, we manually searched relevant academic journals including European Journal of Social Psychology, International Journal of Advertising, Journal of Advertising, Journal of Advertising Research, Journal of Applied Psychology, Journal of Applied Social Psychology, Journal of Behavioral Decision Making, Journal of Business Research, Journal of Consumer Psychology, Journal of Consumer Research, Journal of Experimental Social Psychology, Journal of Marketing, Journal of Marketing Research, Journal of Product & Brand Management, and Organizational Behavior & Human Decision Processes for the years 1980–2012. Reviewing the reference sections in the papers identified in these two initial search efforts uncovered additional studies from articles, books, and dissertations. We then conducted a forward citation search of the seminal papers in this topic area. As final measures to locate as many relevant manuscripts as possible, we conducted keyword searches of appropriate electronic databases and also requested known research on attribute framing via academic list serves (e.g., ELMAR) to identify any additional studies potentially missed in our earlier data collection efforts. After identifying these studies, the appropriateness of each one to our research focus was evaluated. Studies were deemed eligible if the following two conditions were met: (1) the study focused on

T.H. Freling et al. / Organizational Behavior and Human Decision Processes 124 (2014) 95–109

the relationship between attribute framing and respondent outcomes; and (2) either the correlation between the attribute framing effect and the outcome effect or a statistical equivalent was reported (see Glass, McGaw, & Smith, 1981; Janiszewski, Noel, & Sawyer, 2003). We acknowledge that while unpublished manuscripts could provide additional sources of data, the mature nature of the attribute framing literature coupled with the often less rigorous nature of some working and conference papers led us to err on the side of rigor and include only peer reviewed published studies in this research. A less developed literature stream might elicit a different approach.2 Following existing taxonomies in the CLT literature (e.g., Trope et al., 2007), we categorized outcome effects into three distinct categories: evaluations (including attitudes and perceptions regarding the stimuli), estimates (predictions based on the stimuli), and behaviors (including choice and behavioral intentions related to the stimuli). To correctly categorize the frame under investigation as an attribute frame, and to build an appropriate dataset, we used extant literature as a guide (Krishnamurthy et al., 2001; Levin et al., 1998, 2002). We followed accepted meta-analytic procedures (see Eysenck, 1978; Sharpe, 1997) in establishing decision rules for determining which articles would comprise our dataset, paying special attention meta-analytic experts’ warning against ‘‘comparing apples to oranges.’’ This resulted in our exclusion of studies examining the effects of other types of frames (i.e., goal framing and risky choice framing), including investigations involving games of chance and monetary lotteries that differed in terms of associated outcome risk. We also consulted a risky choice framing meta-analysis (Kühberger, 1998) for more specific guidance on inclusion criteria relating to experimental task characteristics (e.g., response mode, unit of analysis, manipulation quality). First, with respect to response modes elicited in attribute framing studies, only studies in which an attribute frame addresses features of one object and elicit a reaction from one subject were included. This decision was based on research suggesting that presenting two options and forcing subjects to make a choice artificially enlarges differences between conditions (Pany & Reckers, 1987; Perner, Gschaider, Kühberger, & Schrofner, 1999). Second, in regard to manipulation quality, we applied a strict definition of framing as a ‘‘semantic manipulation of prospects whereby the exact same situation is simply redescribed’’ (Kühberger, 1998; p. 24). As such, studies featuring ‘‘loose’’ framing manipulations, where other individual factors and contextual features of a situation might result in frames that were not logically equivalent or directly comparable were excluded. Finally, Kühberger (1998) notes that experiments conducted at the group level are heterogeneous and introduce additional social and procedural features that may lead to a frame being construed differently. Given this, only studies requiring individuals to evaluate the stimuli and eliminated any group decision-making scenarios in which subjects interacted with other research participants to arrive at a consensual decision were included. Two of the authors who were blind to the hypotheses independently coded the data. Inter-coder agreement was 95.8% with discrepancies rectified through discussion and reference to the coding scheme and confirmation from a third, independent referee. In all, 107 studies containing 604 framing effects and 88,326 observations were retained for analysis (see Appendices A–C for dataset development specifics).

2 Following Geyskens, Steenkamp, and Kumar (1999), we elected to only include published, peer-reviewed research in this analysis to maximize the empirical rigor of our data population. The rationale for this decision was that unpublished studies have not undergone the same rigorous review process as published studies, and that efforts by other meta-analysts to uncover unpublished work have not yielded much success (cf. Krashnikov & Jayachandran, 2008; Palmatier, Dant, Grewal, & Evans, 2006).

99

Overview of meta-analytic procedures We employed analytic techniques prescribed by Hunter and Schmidt (2004) in data collection and analysis. The effect size coded for the analyses is the point biserial correlation coefficient, which is an appropriate metric for our research in that it provides a means for easy interpretation and meaningful comparison across the effect sizes reported in the attribute framing literature (Hunter & Schmidt, 2004). More importantly, it is the dominant metric found across the population of attribute framing studies, given that the manipulation of the frame is dichotomous in nature (e.g., positive vs. negative, self vs. others, now vs. future) while outcome variables are inherently continuous (i.e., evaluations, estimates, and behaviors). We identified potential outliers within our dataset using the sample-adjusted meta-analytic deviancy (SAMD) statistic (Huffcutt & Arthur, 1995). In identifying outliers, this procedure uses a bootstrapping technique where the overall sample-size weighted or reliability-corrected between attribute framing3 and its respective outcomes is calculated k  1 times to understand if one sample is biasing the analysis. This analysis detected a single observation as an outlier, which was subsequently removed from the dataset. We additionally calculated the estimated correlation (rW) between the attribute frame and outcomes associated with attribute framing. To calculate this overall correlation, each study was weighted by its corresponding sample size. When reported, each was further corrected for systematic variance. The results were then averaged across all studies to ensure that sampling error is accounted for in the estimate of the overall effect of framing. From this, we calculated the average study variance (vart) and an estimate of the heterogeneity (i.e., chi-square statistic) across observed effect sizes within our dataset to ascertain the amount of variance within our observed effects that is explained by sampling error and study artifacts (see Hunter & Schmidt, 2004). To help in the interpretation of the significance of the correlation between attribute framing and behaviors, estimates and evaluations, we computed the 95% bootstrapped confidence interval (CIBS) and the 80% credibility interval (CI) for each framing relationship. Since collective data often violate the distributional assumptions of parametric tests, the use of bootstrapped confidence intervals that are based on a non-parametric distribution is appropriate and provides a more powerful estimate than traditional confidence intervals (Rosenberg, Adams, & Gurevitch, 2000). Finally, the fail-safe sample size (NFS) was calculated to assess the possibility of publication bias or the ‘‘file-drawer’’ problem (Rosenthal, 1979). This information estimates the number of unpublished studies with an effect size of zero that would have to exist to render the observed effects non-significant at the alpha = .05 level (Janiszewski et al., 2003). A larger NFS value conveys greater confidence in the robustness of obtained results. Coding procedures Prior studies employing meta-analysis suggest four potential sources of variation among observed effect sizes within a research domain: research context; measurement method; estimation procedure; and model specification (Assmus, Farley, & Lehmann, 1984; Sultan, Farley, & Lehmann, 1990). Given that our analysis uses correlation as the metric of the observed effects (which are 3 When reported in original studies, we use the reliability-corrected mean (samplesize weighted mean corrected for systematic variance due to variability in the reported reliability of the measure) under the assumption that correlations from larger samples (central limit theorem) and estimated from more reliable data produce a mean correlation closer to the true population mean. When the reliability-corrected mean cannot be estimated due to the absence of reliability data in the original studies, we use the next most rigorous estimate of the population mean, the sample-size weighted mean.

100

T.H. Freling et al. / Organizational Behavior and Human Decision Processes 124 (2014) 95–109

unaffected by model specification) and that the model’s estimation procedure is invariant, we restricted the examination of moderators to those relating to research context and measurement method as possible explanations for differences in attribute framing effects. Based on this, we identified variables that are theoretically justifiable as potential moderating factors and that could be coded from the extant studies. With respect to the research context of attribute framing studies, we used the construal level theory (CLT) literature as a guide to develop a conceptually-focused coding scheme. We first coded which type of frame each study utilized. Valence frames entail presenting otherwise identical information in a positive vs. negative format, as when Jain, Lindsey, Agrawal, and Maheswaran (2007) describe the advertised toothpaste brand as ‘‘better than’’ the comparison brand or the comparison brand of toothpaste as ‘‘worse than’’ the advertised brand. Other studies, such as Shu and Gneezy (2010), employ temporal frames by varying the time context of the information presented (e.g., measuring redemption intentions for gift certificates expiring soon vs. sometime in the future). We also coded numeric frames like those used by DelVecchio, Krishnan, and Smith (2007), which gauged consumer reactions to ‘‘cents off’’ vs. ‘‘percent off’’ promotional offers for shampoo. Finally, our coding scheme included referent frames which presented identical information from different perspectives. To illustrate, White and Peloza (2009) measure consumer attitudes toward ads for charity that benefit the self vs. others. We coded the construal level of the information presented in the attribute frame for each study (i.e., whether the frame contained predominantly abstract information vs. concrete information) by adapting coding schemes recently employed by other CLT researchers (cf. Bornemann & Homburg, 2011; Magee, Milliken, & Lurie, 2010). Specifically, we carefully reviewed and rated each frame on a 5-point scale, where 1 = very concrete, 2 = somewhat concrete, 3 = both concrete and abstract, 4 = somewhat abstract, and 5 = very abstract. (Intercoder reliabilities for construal level coding were a = .97.) We then had two independent coders evaluate and rate the few frames that initially received a ‘‘3’’ rating as either more abstract than concrete, more concrete than abstract, or equally concrete and abstract. This allowed us to develop a high construal level (i.e., more abstract) and low construal level (i.e., more concrete) for analysis. (Results for this phase of coding was also highly reliable: a = .90.) Additionally, using the information available for each study, we coded four psychological distance variables: temporal distance (i.e., whether the decision scenario required an immediate vs. future response), hypothetical distance (i.e., whether the decision scenario was likely vs. unlikely for subjects), affective distance (i.e., whether or not the decision scenario was emotionally intense), and informational distance (i.e., whether the decision scenario involved an event that was familiar vs. novel to subjects). We treated these research context moderators as theoretical moderators and test specific predictions regarding how their interplay impacts the effectiveness of attribute framing. We included several measurement factors, recording information about each study’s sample composition (i.e., whether the sample contained students or nonstudents, US or international subjects, and women only or males and females). Additionally, we coded details about the experimental task (i.e., whether subjects performed a product-related task or some other decision task). Given that measurement factors are less theoretically interesting and practically important, we treated these characteristics as control variables in the meta-analysis in that they were included in our GLS regression (see Lynch, 1982; Peterson, 2001) and focus discussion around the substantive research context moderators featured in our hypotheses.

Moderator analysis procedures To explore the influence of moderators in explaining the effects of attribute framing on its correlates, a weighted generalized least squares regression (GLS) approach was used (Geyskens et al., 1999; Lipsey & Wilson, 2001). We used the following equation to estimate the impact of our proposed moderators on each framing effect separately: 1

b ¼ ðX 0 R1 XÞ X 0 R1 d where d is the transformed correlation associated with the framing effect coded from the dataset (Raudenbush, Becker, & Kalaian, 1988), X is the matrix of moderators hypothesized to influence these framing effects (and included both research context and our control moderators together), and R is a diagonal vector of the variance assigned to each observation (from the sample size of each study included in our dataset). However, given that our study focuses on three separate outcomes of attribute framing, utilizing this univariate approach has some limitations in that it ignores the potential correlation, or within study effects, that could be present among the three framing outcomes (i.e., evaluations, estimates, and behaviors) (Riley, 2009; Riley, Thompson, & Abrams, 2008). Theory suggests that there is a relationship between subject evaluations and behavior, therefore suggesting the need to account for within study correlation. Research suggests that one can explore the potential bias of within study correlation through a series of sensitivity analyses. According to Riley (2009), ‘‘the most popular approach [to examining the impact of within study correlation] is a sensitivity analysis over an (informed) range of imputed correlations; for simplicity this usually assumes a common within study correlation across studies’’ (pp. 807–808). To help in the selection of the appropriate range of within correlations among evaluations and behaviors to use in our sensitivity analysis (i.e., to make sure that the presence of within correlations did not bias the results), we were guided by Kim and Hunter (1993) which found that the average correlation between attitudes and behavioral intentions is 0.65 (uncorrected weighted mean) and that across the various moderator analyses, this correlation ranged from 0.46 to 0.91. They also found that the average correlation between attitudes and behaviors is 0.47 (uncorrected weighted mean) and across the various moderator analyses, the correlation ranged from 0.26 to 0.86. We use these values to estimate the impact of within study correlation in our dataset. Taking into account the relationship between evaluations and behaviors, we utilized the methodology by Raudenbush et al. (1988) where the intercorrelation that is present among outcomes effects when they are studied together within the same sample is specified. To model the interdependencies present within these data, we first computed the variance–covariance matrix for each sample (Ri) and analyzed these together using a full block diagonal matrix in our analysis on framing effects (R) (Raudenbush et al., 1988). Following the procedures set forth by Raudenbush et al. (1988), we transformed the point biserial correlation to the d statistic and use the following equation to understand the impact of our proposed moderators on attribute framing relationships using GLS estimation:

d ¼ Xb þ e; where b is estimated as: 1

b ¼ ðX 0 R1 XÞ X 0 R1 d; and b* is estimated via the variance–covariance matrix with the following:

101

T.H. Freling et al. / Organizational Behavior and Human Decision Processes 124 (2014) 95–109 1

V b ¼ ðX 0 R1 XÞ : We ran the analysis four times using the following estimations of the evaluation–behavior relationship in our calculation of the variance–covariance matrix for the impact of framing on evaluations and behaviors: r = .25, r = .45, r = .65, and r = .85. Results Main effects Table 1 provides an overview of the association between attribute framing and the three outcome effects (i.e., 107 studies, 604 framing effects, 88,326 individual observations). In addition to reporting estimates of the mean true score correlations, it is also important in meta-analysis to describe variability in the correlations. Accordingly, we report 80% credibility intervals and 95% confidence intervals around the estimated population correlations. While some meta-analyses report only confidence intervals (e.g., Ernst Kossek & Ozeki, 1998) and others report only credibility intervals (e.g., Vinchur, Schippmann, Switzer, & Roth, 1998), it is important to report both because each tells different things about the nature of the correlations (Judge & Piccolo, 2004). Confidence intervals provide an estimate of the variability around the estimated mean correlation. A 95% confidence interval excluding zero indicates that one can be 95% confident that the average true correlation is non-zero (5% of average correlations would lie beyond the upper limit of the distribution). Credibility intervals provide an estimate of the variability of individual correlations across studies. An 80% credibility interval excluding zero indicates that 90% of the individual correlations in the meta-analysis exclude zero (for positive correlations, 10% are zero or less and 10% lie at or beyond the upper bound of the interval). Thus, confidence intervals estimate variability in the mean correlation, whereas credibility intervals estimate variability in the individual correlations across the studies. As shown, the sample size-weighted and reliability-corrected correlation between attribute framing and evaluations outcomes is .25, indicating a medium-sized statistical relationship (Rosenthal & Rosnow, 2008). The 95% bootstrapped confidence interval around the mean correlation ranges from .22 to .28 and the fail-safe sample size (NFS = 413,085) suggests that there is little indication of publication bias. Similarly, the adjusted correlation between attribute framing and estimates outcomes is .24, also indicative of a medium-sized relationship. The 95% bootstrapped

confidence interval around the mean correlation ranges from .20 to .29 and the NFS = 16,467. The adjusted correlation between attribute framing and behaviors outcomes is .21, with a 95% confidence interval of .19–.23 and a fail-safe sample size (NFS = 91,196) that indicates file drawer effect is not an issue for this relationship. The statistical evidence presented in Table 1 suggests that attribute framing has a positive and statistically significant impact on evaluations, estimates, and behaviors outcomes. These results both expand and confirm previously published research findings. Yet the data range, variance, and heterogeneity indicate that an investigation of potential moderating variables is warranted. As such, we next examine factors that might attenuate or mitigate the relationships between attribute framing and the three outcome effects. Moderator effects Table 2 highlights the influence of both theoretical and control variables in moderating each of the three outcome effects. Regression results reveal that the correlations observed in prior research between attribute framing and key outcomes are significantly impacted by several moderators. Furthermore, results from our sensitivity analyses demonstrate that these effects are robust against varying levels of within study correlation among evaluations and behaviors. The univariate GLS regression analysis results are presented next to highlight the impact of the theoretical moderators. Hypothesis testing Results indicate that the construal level of the attribute frame is a statistically significant moderator of its impact on evaluations (ß = .10, p 6 .05), estimates (ß = .19, p 6 .05), and behaviors (ß = .04, p 6 .05). These results also hold true for each of the psychological distance moderators across all three criterion variables, thus providing initial support for the view that psychological distance moderates the relationship between framing and outcome effects. To fully evaluate our hypotheses, which predict stronger framing effects when there is congruence between the attribute frame’s construal level and the subject’s perceived psychological distance from the framed event, we conducted additional analyses. Specifically, we examined the relationship between construal level and psychological distance for the high vs. low level of each dimension across each outcome effect variable. Results appear in Table 3.

Table 2 Moderator results for attribute framing. Factor

a

Construal level High vs. Low Psychological distance High vs. Low Temporal distance High vs. Low Probability distance High vs. Low Affective distance High vs. Low Informational distance Control variables Frame type Subject composition Geographic composition Gender composition Experimental task Number of observations

Outcome effect – estimated separately

Evaluations and Behaviors – estimated together modeling within-study variance at different levelsa

Evaluations

r = 0.25

r = 0.45

r = 0.65

r = 0.85

Estimates

Behaviors

.10*

.19*

.04*

0.03*

0.03*

0.04*

0.21*

.04* .44* .14* .24*

.21* .07* .24* .14*

.22* .30* .34* .13*

0.54* 0.30* 0.02* 0.14*

0.53* 0.30* 0.01+ 0.14*

0.50* 0.33* 0.02* 0.16*

0.63* 0.89* 0.04* 0.57*

.29* .06* .45* .35* .62*

.08* .13* .05* – .16*

.20* .15* .03* .94* .23*

0.17* 0.12* 0.29* 0.26* 0.64*

0.18* 0.12* 0.31* 0.30* 0.67*

0.20* 0.12* 0.35* 0.35* 0.72

0.14* 0.33* 0.68* 0.52* 0.91*

359

69

175

To address the potential for interdependence among our dependent variables, we specified the correlation among Evaluations and Behaviors at different levels (i.e., r = .25, r = .45, r = .65, and r = .85) and ran a series of robustness checks to ensure that interdependence was not an alternate explanation of our findings. * p 6 .05. + p 6 .10.

102

T.H. Freling et al. / Organizational Behavior and Human Decision Processes 124 (2014) 95–109

Table 3 Weighted univariate results for theoretical moderators. Sample size

Number of observations

Mean effect size

Weighted variance

8184 6504

66 62

.3024 .2817

.074 .061

2694 4352

28 27

.3293 .2184

.037 .007

11,046 7456

69 47

.2979 .1897

.031 .007

5503 7037

52 58

.2816 .2469

.087 .045

5300 4334

47 25

.2290 .1917

.017 .008

7305 10,459

64 75

.2723 .2142

.061 .037

1944 3460

21 24

.3733 .2253

.040 .015

7305 10,459

52 50

.2064 .1977

.015 .011

8479 7513

71 63

.2660 .2263

.019 .025

3571 2575

28 15

.2829 .1601

.026 .010

8006 575

51 38

.2247 .1894

.020 .013

a

High temporal distance Evaluations* High construal level Low construal level Estimates* High construal level Low construal level Behaviors* High construal level Low construal level High hypothetical distanceb Evaluations* High construal level Low construal level Behaviors* High construal level Low construal level High affective distance Evaluations* High construal level Low construal level Estimates* High construal level Low construal level Behaviors* High construal level Low construal level High informational distance Evaluations* High construal level Low construal level Estimates* High construal level Low construal level Behaviors* High construal level Low construal level *

p 6 .05. For reporting simplicity, each psychological distance reported indicates the ‘‘high’’ observations. Analysis of the corresponding ‘‘low’’ observations reveals perfectly inverse relationships to those demonstrated above. b Insufficient sample sizes for high and low construal levels across Estimates outcomes precluded a statistical examination. a

In support of hypothesis 1, studies featuring framing manipulations with CLT congruency between the psychological distance of the framed issue and the construal level of information described have significantly stronger effects. This is true across all four dimensions of psychological distance coded and analyzed including: temporal distance (evaluations: r = .30 vs. r = .28; estimates: r = .33 vs. r = .22; and, behaviors: r = .30 vs. r = .19); hypothetical distance (evaluations: r = .28 vs. r = .25; and, behaviors: r = .23 vs. r = .19); affective distance (evaluations: r = .27 vs. r = .21; estimates: r = .37 vs. r = .23; and, behaviors: r = .21 vs. r = .20); and, informational distance (evaluations: r = .27 vs. r = .23; estimates: r = .28 vs. r = .16; and, behaviors: r = .22 vs. r = .19. In sum, these results lend support to our hypothesis and suggest the interactive effect between construal level and psychological distance determines the relative effectiveness of attribute frames. To test hypothesis 2, we culled 209 valenced attribute frame observations from our dataset and conducted additional univariate analyses on this data.4 We sought to determine if positive attribute

4

Only the relationship between valenced attribute frames and evaluations (N = 209) were included in these univariate analyses. We also explored the relationships between valenced attribute frames and estimates (N = 15) and behaviors (N = 96). The same pattern of relationships (i.e., all more positive, all p 6 .05) were found, so these results are not reported in the text.

frames (theorized to evoke a higher construal level) are more effective when greater psychological (temporal, hypothetical, affective, and informational) distance characterizes the decision scenario. Results support both our expectations and recent CLT-based theorizing on information valence. Specifically, we found statistically significant stronger correlations (all p 6 .05) between attribute framing and evaluations in studies characterized by higher temporal distance (r = .37 vs. r = .23), higher hypothetical distance (r = .32 vs. r = .25), higher affective distance (r = .31 vs. r = .26), and higher informational distance (r = .30 vs. r = .25). Taken together, these results indicate that the long-established valence-consistent shift found in attribute framing research is likely a reliable phenomenon (Levin et al., 2002) due to the interaction between the message and its recipient instead of an outcome effect solely emanating from the constructed message.

Analysis of control moderators Table 4 details the influence of control moderators on each outcome effect. For all three outcome variables, studies using a valenced decision frame have positive, statistically significant results (p 6 .05). Results for evaluations (ß = .29, p 6 .05), estimates (ß = .08, p 6 .05) and behaviors (ß = .20, p 6 .05) indicate that

103

T.H. Freling et al. / Organizational Behavior and Human Decision Processes 124 (2014) 95–109 Table 4 Weighted univariate results for control moderators. Sample size

Number of observations

Mean effect size

Weighted variance

30,796 20,869 1675 8110 14,631 12,245

209 150 15 55 96 79

.2648 .2163 .2446 .2394 .2111 .2067

.046 .026 .025 .018 .022 .011

42,990 8675 8247 1538 20,809 5971

308 51 59 11 142 33

.2536 .2087 .2318 .2854 .2174 .1806

.040 .032 .022 .003 .006 .014

41,968 9697 8598 1187 18,361 8419

291 68 59 11 120 55

.2566 .2006 .2254 .3371 .2155 .1952

.042 .023 .015 .053 .016 .019

47,265 4400 26,225 555

348 12 172 3

.2620 .0772 .2107 .1355

.040 .003 .017 .004

30,158 21,507 3630 6155 10,110 16,670

176 183 23 47 59 116

.2184 .2852 .2031 .2623 .1929 .2189

.023 .061 .009 .025 .010 .021

a

Valence frame Evaluations* Estimates Behaviors*

Subject compositionb Evaluations* Estimates

*

Behaviors Geographic compositionc Evaluations* Estimates

*

Behaviors* Gender compositiond Evaluations* Behaviors* Experimental taske Evaluations* Estimates* Behaviors*

*

p 6 .05. For each outcome effect, the first line represents studies manipulating valence while the second represents studies that did not. For each outcome effect, the first line represents student subjects while the second represents non-students. c For each outcome effect, the first line represents US subjects while the second represents non-US. d For each outcome effect, the first line represents mixed gender while the second represents single gender; for estimates outcomes, there was no variation in gender composition to allow for comparison. e For each outcome effect, the first line represents product-related while the second represents non product-related. a

b

studies where the attribute frame involves a valence manipulation are significantly different than studies where valence is not manipulated. Results reveal that the correlation between attribute framing and evaluations is statistically stronger in studies manipulating valence (r = .26), as compared to those where there is no valence manipulation (r = .22). Valence frames have a similar effect on estimates (r = .24 vs. r = .24) and behaviors (r = .21 vs. r = .21). Inconsistencies across the observed attribute framing correlations can also be explained by differences in the research design, samples, experimental stimuli, and measures used in studies comprising our dataset. For evaluations, attribute framing effects are significantly different for studies utilizing student vs. non-student subjects (ß = .06, p 6 .05), US respondents vs. non-US respondents (ß = .45, p 6 .05), samples comprised of mixed vs. single gender (ß = .35, p 6 .05), and when subjects performed a product-related evaluation vs. another experimental task (ß = .62, p 6 .05). Correlations are substantially stronger between attribute framing and evaluations in studies with respondents who were students (r = .25 vs. r = .21), American (r = .26 vs. r = .20), and mixed gender (r = .26 vs. r = .08), as well as when the experimental task did not involve a product-related evaluation (r = .22 vs. r = .29). Significant moderators of attribute framing effects on estimates include student vs. non-student respondents (ß = .13, p 6 .05), US vs. non-US subjects (ß = .05, p 6 .05), and product evaluation vs. non-product evaluation (ß = .16, p 6 .05). Substantially stronger correlations between attribute framing and estimates were observed for studies that used non-student (r = .29 vs. r = .23) and

non-US respondents (r = .34 vs. r = .23), and that involved a non product-related experimental task (r = .26 vs. r = .20). For behaviors, attribute framing effects were significantly different for studies utilizing student vs. non-student subjects (ß = .15, p 6 .05), for studies with US respondents vs. non-US respondents (ß = .03, p 6 .05), for studies utilizing samples comprised of mixed vs. single gender (ß = .94, p 6 .05), and for studies in which subjects performed a product-related evaluation vs. another experimental task (ß = .23, p 6 .05). Stronger correlations between attribute framing and behaviors were observed in studies with samples comprised of students (r = .22 vs. r = .18), US subjects (r = .22 vs. r = .20), and mixed genders (r = .21 vs. r = .14), and when the experimental task focused on a non-product-related evaluation (r = .19 vs. r = .22).

Follow-up experiment While meta-analysis is extremely useful for quantitatively synthesizing an immense literature base, the technique is limited to including only variables that can be meaningfully and consistently coded from existing studies. Further, because the relationships examined are correlational, causal interpretations should be made with caution. These methodological limitations prohibit us from making more definitive conclusions about the impact of valencebased attribute frames on consumer evaluations and the construal level mechanisms that we propose drive these effects. To rectify this, we conduct a follow-up experiment to supplement findings

104

T.H. Freling et al. / Organizational Behavior and Human Decision Processes 124 (2014) 95–109

emanating from the meta-analysis. Specifically, we explore how frame valence interacts with social distance—one of the original dimensions of psychological distance that could not be coded in the preceding meta-analysis. Social distance is defined as an individual’s perception of whom an event references (e.g., one’s self or various others) and the degree of similarity (i.e., ‘‘close’’ or ‘‘distant’’) between that individual and any others relevant to the event. The further away the event is perceived from the individual subject, the higher the perceived social distance. According to construal level theory, individuals naturally regard certain people as being closer to them than other people (Trope et al., 2007). By definition, the ‘‘self’’ is the most socially proximal entity, ‘‘similar others’’ are more socially close than ‘‘dissimilar others’’, and ‘‘in-group’’ members are more socially close than ‘‘out-group’’ members (Bar-Anan, Liberman, & Trope, 2006). Thus, an event that is framed as directly involving the subject would be construed as having a low-level of social distance, one involving friends or family would be at a higher level, and an event framed as involving strangers at an even higher perceived level of social distance. Consistent with hypothesis 2, we expect stronger framing effects to occur when individuals in a socially close decision situation receive information at a lower-level construal and when individuals in a socially distant decision situation are presented information at a higher-level construal. Again, we conceptualize information valence as a manifestation of construal level where positive information evokes a higher level of construal and negative information leads to a lower construal level. In line with our meta-analysis results and recent CLT research exploring valence (Lee et al., 2010; Ulkumen & Cheema, 2011; White et al., 2011), we expect stronger framing effects for congruent pairings (e.g., a positively valenced frame for a socially distant scenario), as compared to incongruent pairings (e.g., a positively valenced frame for a socially close scenario). Method A 2 (frame valence: positive vs. negative)  2 (social distance: high vs. low) between-subjects factorial design was used to test our hypothesis. One hundred undergraduate students from a large southwestern university (marketing majors and graduating seniors) participated in the study for course credit. Participants were told that the purpose of the study was to understand how individuals similar to them make decisions. Following Liviatan et al. (2008), a decision scenario in which subjects were asked to evaluate a target individual for inclusion in a group project for class was designed. Social distance (high vs. low) was manipulated by varying the target person’s similarity to the subject on many key dimensions, including major, classification, and the two most recent semesters of coursework completed at this university. To manipulate frame valence, we also stated that one student in the group had worked with the target person on another group project. Consistent with Buda and Zhang (2000), we included information from this one student that ‘‘four out of six group members rated the target person positively’’ vs. ‘‘two out of six group members rated the target person negatively.’’ To develop the stimulus materials, we extensively pretested to understand students’ criteria in choosing group members for project work as well as to assess students’ knowledge of information presented in the scenarios and their perceived relevance of and distance from the decision context. Importantly, pretest results suggested that the manipulation of social distance was not associated with significant differences in other source characteristics such as perceptions of the target person’s trustworthiness, credibility, or attractiveness (Fs < 1). Based on pretest results, four versions of a decision scenario were developed (see Appendix D) that varied

frame valence and social distance. Following exposure to their respective decision scenarios, subjects rated the likelihood that they would invite the target person to join the group using four 7-point semantic differential items (anchored by ‘‘very likely’’. . . ‘‘not at all likely’’, ‘‘very probable’’. . . ‘‘not at all probable’’, ‘‘very possible’’. . . ‘‘not at all possible’’, and ‘‘very certain’’. . . ‘‘not at all certain’’). Consistent with prior research assessing intentions (e.g., Bennett & Harrell, 1975; Dover & Olson, 1977; MacKenzie, 1986; Marks & Kamins, 1988; Smith & Swinyard, 1983), an average of the scale items was used to form a composite behavioral intention measure. We included manipulation checks for social distance, valence, and construal style. For the social distance manipulation check, participants indicated how similar and close to themselves they perceived the target person to be using a 7-point response scale ranging from ‘‘not at all’’ to ‘‘very much’’. These two social-distance items were reverse-scored for analysis so that greater perceived similarity corresponded to lower social distance. As an additional social distance manipulation check, we also had participants complete the Inclusion of Others in Self Scale (IOSS; Aron, Aron, & Smollan, 1992), which measures interpersonal closeness. We followed established procedures (see Block & Keller, 1995; Maheswaran & Meyers-Levy, 1990) to assess the effectiveness of our frame valence manipulation. Specifically, we asked subjects to indicate the extent to which the information they read portrayed the target person positively and whether the information they read presented the target person’s past group project performance in a negative light (both with 7-point scales ranging from ‘‘not at all’’ to ‘‘completely’’). We also included the three confounding-check measures suggested by Block and Keller (1995) by asking subjects to rate the information presented on 7-point scales anchored by ‘‘very credible’’. . .. ‘‘not at all credible’’, ‘‘easy to comprehend’’. . . ‘‘difficult to comprehend’’, ‘‘contained a lot of information’’. . . ‘‘contained little information’’. Finally, we assessed each participant’s construal style, which we expected to correspond with frame valence. Subjects completed the Behavioral Identification Form (BIF; see Alter, Zemla, & Oppenheimer, 2010; Vallacher & Wegner, 1989), which indicates an individual’s preferences for either abstract of concrete description of thirteen everyday behaviors. For example, subjects indicated their relative preference for the description of eating as ‘‘chewing and swallowing’’ (concrete, low-level construal) or ‘‘getting nutrition’’ (abstract, high-level construal). To eliminate any presentation order bias, we alternated which label (Description A or Description B) referred to the concrete and abstract descriptions. Participants indicated their relative preference for the two descriptions on a 7-point scale (anchored by ‘‘strongly prefer Description A’’ and ‘‘strongly prefer Description B’’). We then averaged the resulting preference ratings to generate a score reflecting each participant’s construal style. Results Manipulation checks To ensure that we correctly induced the intended levels of valence and social distance, we first checked the study’s manipulations. Subjects’ responses to the questions relating to their status as a college student verified that all were enrolled full-time and had taken the marketing classes in the socially close scenario (but not the non-marketing classes in the socially distant scenario). The two measures assessing participants’ perceived similarity and closeness to the target person both revealed statistically significant main effects for social distance. Respondents’ ratings were lower when social distance was high for both similarity (Msocially close = 4.23 and Msocially distant = 3.27; F(1, 98) = 25.29, p 6 .001) and for closeness (Msocially and Msocially close = 4.32 distant = 3.21; F(1, 98) = 18.00, p 6 .001). As expected, participants who evaluated

T.H. Freling et al. / Organizational Behavior and Human Decision Processes 124 (2014) 95–109

the socially distant decision scenario scored significantly lower (M = 2.98) on the IOSS than those exposed to the socially close decision scenario (M = 5.12; t(99) = 11.78, p 6 .001). The two items assessing the frame valence manipulation both revealed a statistically significant main effect for frame valence. Subjects rated the extent to which the message positively portrayed the target person higher in the positive frame valence condition (M = 5.93) than in the negative frame valence condition (M = 4.80; F(1, 98) = 36.41, p 6 .001). They also rated the extent to which the message presented the target person’s past group project performance in a negative light to be lower in the positive frame valence condition (M = 4.13) than in the negative frame valence condition (M = 5.67; F(1, 98) = 24.52, p 6 .001). Importantly, there were no confounds with any of the manipulations. With respect to construal level, subjects who were exposed to the positive frame (high-level construal) preferred the abstract descriptions of the BIF behaviors more strongly than did participants who saw the negative frame (low-level construal) (Mpositive frame = 5.04 and Mnegative frame = 3.11; t(99) = 13.29, p 6 .001). Consistent with expectations, exposure to positive information appeared to lead subjects to prefer the descriptions consistent with an abstract construal mindset, while exposure to negative information led participants to prefer descriptions consistent with a concrete construal mindset. To rule out any additional unintended confounds, subjects were asked to indicate how credible, easy to comprehend, and informative the information presented was. Separate ANOVAs on these variables revealed no statistically significant treatment effects (Fs < 1), suggesting that the research treatments were not confounded with any of these variables. In summation, the manipulation and confound checks suggest that the intended factors were successfully manipulated and that our constructs accurately capture the appropriate underlying dynamics. Results A 2  2 ANOVA revealed nonsignificant main effects of social distance and frame valence (Fs < 1). However, consistent with both our expectations and hypothesis 2, the interaction between social distance and frame valence is significant (F(1, 98) = 4.37, p 6 .05). Stronger framing effects were observed when there was congruence between construal level and social distance. Table 5 displays the pattern of means for subjects’ behavioral intentions and highlights support for our expectations. We analyzed the mean intentions for each frame construal level-social distance pairing in a series of planned contrasts. As expected, the two congruent message conditions are associated with the strongest framing effects. When a positive frame is presented with a socially distant referent (M = 5.76), intentions are significantly more favorable than when subjects see a positive frame with a socially close referent (M = 3.45; t(38) = 11.45, p 6 .001) or a negative frame for a socially distant referent (M = 3.64; t(38) = 10.11, p 6 .001). Similarly, subjects express significantly more favorable intentions when a negative frame is presented with a socially close referent (M = 5.68), as compared to when subjects are exposed to a positive frame with a socially close referent (M = 3.45; t(38) = 9.23, p 6 .001)

Table 5 Behavioral intentions toward the new group member.

Positive frame Mean Standard deviation Negative frame Mean Standard deviation

Low social distance

High social distance

N = 25 3.45 (1.13) N = 25 5.68 (.97)

N = 25 5.76 (1.02) N = 25 3.64 (1.01)

105

or a negative frame with a socially distant referent (M = 3.64; t(38) = 8.67, p 6 .001). Of particular interest, a planned contrast of the two congruent message conditions (i.e., positive frame-socially distant referent vs. negative frame-socially close referent) reveals no statistically significant differences (t(38) = 0.25, p > .10), suggesting that negatively framed messages can be as effective as positively framed messages if the issue being evaluated is perceived as being particularly likely to affect the message recipient. Discussion We proposed that congruence between a frame’s construal level (evoked through valence) and the evaluator’s psychological distance (via social distance) from the framed event would determine the effectiveness of that frame. Our experimental results substantiate this expectation by showing that outcome effects are enhanced when there is congruence between these two variables. Both positively framed events presented in a socially distant scenario and negatively framed events presented in a socially close scenario outperformed incongruent construal-social distance pairings. Importantly, the experimental results reveal that there is no statistical difference between the two congruent attribute framing conditions. Despite historical support for a valence-consistent shift whereby positive frames consistently outperform negative frames, our findings indicate that it is the congruency between the message frame and the message recipient that drives results—and not simply a positive vs. negative message frame. This is a noteworthy finding because it suggests that, as long as the construal level and psychological distance are congruent, both positive and negative messages can be equally effective. These results suggest that message framers can choose to invoke a particular level of construal among audience members with their use of positive or negative information in persuasive messaging, depending on the psychological distance that is likely to characterize the issue or event for message recipients. In contrast to the classic Trident sugarless gum example referenced earlier, consider a recent ad for the Clearblue Easy pregnancy test kit that uses the relatively negative tagline: ‘‘1 in 4 women misread a traditional pregnancy test.’’ In this case, the message is negatively framed to draw attention to the proportion of individuals who might mistakenly interpret test results (rather than the 75% of individuals who understand the results). This negative positioning is consistent with a construal level theory view and highlights the importance of congruence between construal level and social distance. Here, the presentation of negatively framed information is likely to induce a lower level of construal and should thus resonate with potential target consumers who are psychologically close to this issue (e.g., females who are sexually active where pregnancy is a possibility). For scholars and managers interested in maximizing the effectiveness of attribute frames, these experimental results imply that when delivering information about a negative event, the greatest persuasive impact should occur when concrete details are presented and the event is portrayed as close to the message recipient. Conversely, positive information should be presented in an abstract manner that accentuates the perceived distance between the event and the message recipient. General discussion Overview With some recent exceptions, past research examining attribute frames has largely been concerned with the characteristics of the message and its impact on consumer outcomes. Our work extends

106

T.H. Freling et al. / Organizational Behavior and Human Decision Processes 124 (2014) 95–109

the current knowledge of framing effects by applying a theoretical lens to the attribute framing literature. We find strong support for our assertion that the impact of attribute framing may not be as straightforward as it is often portrayed. By including a CLT perspective in our investigation, we empirically demonstrate that it is often the interaction between the message and the recipient that is driving evaluations, estimates, and behaviors. These findings are relevant to a variety of organizational communications designed to influence behavioral decisions (e.g., leadership, negotiation, promotional). The meta-analytic results for CLT-based theoretical moderators provide interesting revelations for any individual tasked with constructing persuasive messages relating to frame valence, construal level, and psychological distance. Importantly, construal level is one aspect of attribute framing that is largely under the control of the frame designer. This suggests that the specific language in a message can be purposefully finessed to strengthen its persuasive impact. By simply considering how target audience members are likely to regard an issue or event (in terms of their relative psychological distance from it), one can then construct a message at an appropriate level of construal to better achieve a desired result. CLT suggests the psychological distance of the decision problem affects decision-making, such that people are predisposed to use high-level construals when thinking of distant events and low-level construals when thinking of proximal events (Bar-Anan et al., 2006; Liberman, Trope, & Wakslak, 2007). Consistent with this premise, our overall results for the psychological distance variables (see Table 2) provide an initial indication that these factors do play a role in individuals’ perceptions of attribute frames. Examining univariate contrasts (see Table 3) provides an even richer assessment of the proposed relationships. These key findings echo other CLT research demonstrating stronger responses when the psychological distance of an object is compatible with its level of abstraction (e.g., Amit, Algom, & Trope, 2009; Lee et al., 2010). High temporal distance characterizes situations that are expected to happen in the future. Our results indicate that when an event occurs farther in the future, a frame construed in a more abstract manner (i.e., high-level construal) leads to consistently stronger framing effects. This finding provides converging evidence that distant future events should be represented in an abstract, structured manner while messages about relatively near future events are more effective when represented in a more concrete, contextualized manner (Liberman et al., 2002; Trope & Liberman, 2000; Trope et al., 2007). Research by Chandran and Menon (2004) substantiates this relationship between temporal distance and construal level by demonstrating a higher perceived likelihood of being affected by Epstein-Barr and cell phone radiation among subjects exposed to frames detailing the specific number of people affected each day (vs. each year) and presenting concrete information about the symptoms of these health hazards. High hypothetical distance characterizes situations in which the subject is unlikely to find oneself. When events are portrayed as unlikely for the subjects, a similar positive relationship between construal level and distance is observed. That is, consistent with prior research (Todorov et al., 2007), attribute framing effects are stronger when subjects see an abstractly construed frame of an event that is unlikely to affect them. Kastenmüller et al. (2010) corroborate this assertion, documenting more favorable evaluations of a terrorism safety policy—a political decision task that represents an unlikely issue for student subjects to ponder—when information about the policy is presented in broad, abstract terms. High affective distance characterizes situations that are relatively less emotional and less intense for subjects. In concert with previous CLT research on emotional distance (Fujita, Trope, Liberman, & Levin-Sagi, 2006b; Labroo & Patrick, 2009), our results indicate that, for events characterized by high affective distance,

frames with a high construal level are most effective. Consistent with this assertion, subjects in Agrawal and Duhachek’s (2010) research estimate higher incidences of undergraduate binge drinking after viewing emotionally-charged anti-drinking messages (low in affective distance) that concretely framed the consequences of binge drinking (low construal level). High informational distance characterizes situations or issues that are likely to be perceived as novel or unique (Fieldler, 2007). We observed stronger framing effects for messages that use abstract language to describe unfamiliar events. Thus, when developing an attribute frame for an event that target audience members will perceive as novel, the response to most decision tasks should be stronger if the information is construed at a higher level. Consistent with this dynamic, Chan and Mukhopadhyay (2010) observe higher behavioral intentions to attend a theatrical performance among student subjects when information about this relatively unfamiliar concept (high informational distance) is broadly framed (high construal level). Although a lack of sufficient extant data meant that social distance could not be incorporated as a theoretical moderator in the meta-analysis, we directly examined this psychological distance variable. In the preceding experiment, we formally manipulated subjects’ perception of similarity to oneself in valence-based frames. Results supported predictions that congruence among construal level and frame valence would lead to stronger framing effects. Subjects expressed more positive evaluations under two conditions: (1) when a positive frame was used in conjunction with a socially distant referent; and, (2) when a negative frame was used in combination with a socially close referent. This finding further demonstrates that both positive and negative frames can be equally effective in eliciting evaluations and highlights circumstances when the frequently mentioned valence-consistent shift might not hold. Furthermore, the results of this experiment have important implications for the optimal way to craft persuasive messages. For psychologically close issues that are likely to resonate with or concern message recipients, negative frames should be more effective than positive frames in influencing evaluations.

Contributions The dual objectives of the current research were to synthesize and analyze the empirical findings on attribute framing in an effort to take inventory of existing knowledge and also to apply a construal level theory (CLT) perspective to attribute framing in an effort to better understand the dynamics at play. This manuscript first contributes to the literature by making current the empirical body of knowledge on attribute framing and quantitatively summarizing over three decades of research—much of it published in Organizational Behavior & Human Decision Processes (see Appendix A). The application of construal level theory provides evidence that the attribute framing relationships once described as fairly straightforward are actually quite intricate. Therefore, another contribution of the current research is the application of CLT’s core concepts of construal level and psychological distance to extant attribute framing research. Results indicate that the long recognized valence-consistent shift document in attribute framing research is likely more than simply a function of positive vs. negative message positioning, but likely acts via the impact of the event information on an individual’s perceived construal level and psychological distance from the event. By manipulating the mental representation in a framed event, one is able to impact subjects’ evaluations. In fact, while the meta-analysis clearly demonstrates that attribute framing effects are significantly stronger when messages contain positive frames, a closer analysis of studies manipulating

T.H. Freling et al. / Organizational Behavior and Human Decision Processes 124 (2014) 95–109

frame valence suggests the valence-consistent shift described in earlier summary articles is somewhat conditional. Our univariate results suggest that positively valenced (higher construal level) attribute frames are more effective when the psychological distance of the decision scenario is also higher. This result is consistent with recent research in the CLT tradition (Eyal et al., 2004; Herzog et al., 2007; Labroo & Patrick, 2009), which intimate that information valence may be a manifestation of construal level and share a similar relationship with psychological distance. Our results add to this developing literature stream and suggest that the impact of positive attribute framing is more pronounced when subjects are temporally, hypothetically, emotionally, or informationally distant from the decision scenario. Our experiment demonstrating a similar relationship between social distance and frame valence further substantiates the appropriateness of reinterpreting attribute framing effects using a CLT framework. Perhaps the most noteworthy contribution of this research is the finding that the congruence between the evoked construal level of a framing event and subject’ perceived psychological distance from that event appears to influence attribute framing effects. This finding guides scholars and managers as to how to best frame an event regardless of how far in the future, how likely, how emotionally intense, or how familiar a scenario is for the individual interpreting the event. In sum, this current research (1) integrates and analyzes the body of empirical attribute framing knowledge, (2) extends the domain of construal level theory, and (3) uncovers a finer-grained explanation for the valence-consistent shift effect that is so abundantly observed in attribute framing research.

107

By far, most attribute framing studies examine the impact of the frame on evaluations and behaviors; however, some interesting effects occur when the outcome variables of interest involve estimates. While the focus of this analysis was on theoretical moderators and not control variables, it is interesting that in studies which require subjects to make a prediction or estimate, attribute framing is more effective among non-US, non-student subjects. This finding is at odds with attribute framing studies that feature evaluations or behaviors as the outcome variable. Future research would benefit from additional inquiries relating to the effect of attribute framing on individual predictions. Finally, while this manuscript appropriately focused solely on the impact of attribute framing on outcomes, future research should apply CLT and explore the role of congruence among the message being framed and the recipient for goal framing, which frames the relationship between behaviors and goal attainment (Krishnamurthy et al., 2001). While qualitatively different from attribute framing, goal framing is also widely used in persuasive messaging and the vast literature on goal framing could also benefit from a quantitative synthesis and unifying theoretical framework. Acknowledgments The authors thank Ryan Freling, Adwait Khare, Ritesh Saini, and Zhiyong Yang for constructive comments on previous versions of this article. They also thank Xiao-Ping Chen and four anonymous referrers for their insightful comments during the review process.

Limitations and future research Appendix A. Supplementary material While this manuscript expands the attribute framing knowledge base, some limitations should be noted. Any quantitative synthesis is constrained by the nature and scope of the original studies on which it is based and this shortcoming should be borne in mind when findings presented here are interpreted. First, not all published studies on attribute framing reported correlations or sufficient data to calculate a usable effect size; therefore, some empirical studies exploring the effects of attribute framing could not be incorporated into this analysis. Second, the cross-sectional nature of some of the original studies restricts our ability to make confident causal inferences. Although time-series data would be most desirable for these purposes, they are largely unavailable in the original studies and therefore a reliance on cross-sectional data for making causal inferences naturally exists in the attribute framing literature. Third, our analyses were constrained to examining moderating factors that could be coded from the extant literature. While the moderating factors studied here provide scholars and practitioners with useful information, the inability of these codeable moderators to fully account for the variance in the performance outcome correlations indicates that additional measurement and/or contextual factors need to be modeled and reported in future studies on attribute framing. Several avenues of future research emanate from our work. As mentioned previously, this research demonstrates that the effect of framing on outcomes might not be as straightforward as originally thought. We provide a good starting point for exploring the interplay between an attribute frame’s construal level and psychological distance, showing that the effectiveness of attribute framing is contingent upon the relationship between the message and the intended recipient. In this paper, we examined construal level and each psychological distance variable independently of others. Future research should examine the interactions of attribute framing with multiple psychological distance mechanisms simultaneously and explore the differential effects across the key outcomes examined.

Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.obhdp.2013.12. 007. References Agrawal, N., & Duhachek, A. (2010). Emotional compatibility and the effectiveness of antidrinking messages: A defensive processing perspective on shame and guilt. Journal of Marketing Research, 47, 263–273. Alter, A. L., Zemla, J. C., & Oppenheimer, D. M. (2010). Missing the trees for the forest: A construal level account of the illusion of explanatory depth. Journal of Personality and Social Psychology, 99, 436–451. Amit, E., Algom, D., & Trope, Y. (2009). Distance dependent processing of pictures and words. Journal of Experimental Psychology: General, 138(3), 400–415. Aron, A., Aron, E. N., & Smollan, D. (1992). Inclusion of other in the self scale and the structure of interpersonal closeness. Journal of Personality and Social Psychology, 63, 596–612. Assmus, G., Farley, J. U., & Lehmann, D. R. (1984). How advertising affects sales: Meta-analysis of econometric results. Journal of Marketing Research, 21, 65–74. Bar-Anan, Y., Liberman, N., & Trope, Y. (2006). The association between psychological distance and construal level: Evidence from an implicit association test. Journal of Experimental Psychology: General, 135, 609–622. Bennett, P. D., & Harrell, G. D. (1975). The role of confidence in understanding and predicting buyers’ attitudes and purchase intentions. Journal of Consumer Research, 2, 110–117. Block, L. G., & Keller, P. A. (1995). When to accentuate the negative: The effects of perceived efficacy and message framing on intentions to perform a healthrelated behavior. Journal of Marketing Research, 32, 192–203. Bornemann, T., & Homburg, C. (2011). Psychological distance and the dual role of price. Journal of Consumer Research, 38, 490–504. Brockner, J., Wiesenfeld, B. M., & Martin, C. L. (1995). Decision frame, procedural justice, and survivors’ reactions to job layoffs. Organizational Behavior and Human Decision Processes, 63, 59–68. Buda, R., & Zhang, Y. (2000). Consumer product evaluation: The interactive effect of message framing, presentation order, and source credibility. Journal of Product and Brand Management, 9, 229–242. Chan, E., & Mukhopadhyay, A. (2010). When choosing makes a good thing better: Temporal variations in the valuation of hedonic consumption. Journal of Marketing Research, 47, 497–507. Chandran, S., & Menon, G. (2004). When a day means more than a year: Effects of temporal framing on judgments of health risk. Journal of Consumer Research, 31, 375–389.

108

T.H. Freling et al. / Organizational Behavior and Human Decision Processes 124 (2014) 95–109

Davis, M. A., & Bobko, P. (1986). Contextual effects on escalation processes in public sector decision making. Organizational Behavior and Human Decision Processes, 37, 121–138. DelVecchio, D. H., Krishnan, S., & Smith, D. C. (2007). Cents or percent? The effects of promotion framing on price expectations and choice. Journal of Marketing, 71, 158–170. Dover, P. A., & Olson, J. C. (1977). Dynamic changes in expectancy value attitude model as a function of multiple exposures to product information. In B. A. Greenberg & D. N. Bellenger (Eds.), Contemporary marketing thought (pp. 455–459). Chicago: American Marketing Association. Dunegan, K. J. (1993). Framing, cognitive modes, and image theory: Toward an understanding of a glass half full. Journal of Applied Psychology, 78, 491–503. Dunegan, K. J. (1996). Fines, frames, and images: Examining formulation effects on punishment decisions. Organizational Behavior and Human Decision Processes, 68, 58–67. Ernst Kossek, E., & Ozeki, C. (1998). Work–family conflict, policies, and the job–life satisfaction relationship: A review and directions for organizational behavior– human resources research. Journal of Applied Psychology, 83, 139–149. Eyal, T., Liberman, N., Trope, Y., & Walther, E. (2004). The pros and cons of temporally near and distance action. Journal of Personality and Social Psychology, 86, 781–795. Eysenck, H. J. (1978). An exercise in meta-silliness. American Psychologist, 33, 517. Fieldler, K. (2007). Construal level theory as an integrative framework for behavioral decision-making research and consumer psychology. Journal of Consumer Psychology, 17, 101–106. Fujita, K., Henderson, M. D., Eng, J., Trope, Y., & Liberman, N. (2006a). Spatial distance and mental construal of social events. Psychological Science, 17, 278–282. Fujita, K., Trope, Y., Liberman, N., & Levin-Sagi, M. (2006b). Construal levels and selfcontrol. Journal of Personality and Social Psychology, 90, 351–367. Geyskens, I., Steenkamp, J. E. M., & Kumar, N. (1999). A meta-analysis of satisfaction in marketing channel relationships. Journal of Marketing Research, 36, 223–238. Glass, G. V., McGaw, B., & Smith, M. L. (1981). Meta-analysis in social research. Beverly Hills, CA: Sage Publications. Henderson, M. D., Fujita, K., Trope, J., & Liberman, N. (2006). The effects of spatial distance on social judgment. Journal of Personality and Social Psychology, 91, 845–856. Herzog, S. M., Hansen, J., & Wanke, M. (2007). Temporal distance and ease of retrieval. Journal of Experimental Social Psychology, 43, 483–488. Huffcutt, A. I., & Arthur, W. Jr., (1995). Development of a new outlier statistic for meta-analytic data. Journal of Applied Psychology, 80, 327–334. Hunter, J. E., & Schmidt, F. L. (2004). Methods of meta-analysis: Correcting error and bias in research findings. Thousand Oaks, CA: Sage Publications. Jain, S. P., Lindsey, C., Agrawal, N., & Maheswaran, D. (2007). For better or for worse? Valenced comparative frames and regulatory focus. Journal of Consumer Research, 34, 57–65. Janiszewski, C., Noel, H., & Sawyer, A. G. (2003). A meta-analysis of the spacing effect in verbal learning: Implications for research on advertising repetition and consumer memory. Journal of Consumer Research, 30, 138–149. Judge, T. A., & Piccolo, R. F. (2004). Transformational and transactional leadership: A meta-analytic test of their relative validity. Journal of Applied Psychology, 89, 755–768. Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47, 263–291. Kastenmüller, A., Fischer, P., Jones, E., Greitemeyer, T., Frey, D., Köppl, J., et al. (2010). Selective exposure: The impact of framing information search instructions as gains and losses. European Journal of Social Psychology, 40, 837–846. Khan, U., & Dhar, R. (2010). Price-framing effects on the purchase of hedonic and utilitarian bundles. Journal of Marketing Research, 47, 1090–1099. Kim, M., & Hunter, J. E. (1993). Relationships among attitudes, behavioral intentions, and behavior: A meta-analysis of past research, Part 2. Communication Research, 20, 331–364. Krashnikov, A., & Jayachandran, S. (2008). The relative impact of marketing, research-and-development, and operations capabilities on firm performance. Journal of Marketing, 72, 1–11. Krishnamurthy, P., Carter, P., & Blair, E. (2001). Attribute framing and goal framing effects in health decisions. Organizational Behavior and Human Decision Processes, 85, 382–399. Kühberger, A. (1995). The framing of decisions: A new look at old problems. Organizational Behavior and Human Decision Processes, 62, 230–240. Kühberger, A. (1998). The influence of framing on risky decisions: A meta-analysis. Organizational Behavior and Human Decision Processes, 75, 23–55. Labroo, A. A., & Patrick, V. M. (2009). Psychological distance: Why happiness helps you see the big picture. Journal of Consumer Research, 35, 800–809. Lee, A. Y., Keller, P. A., & Sternthal, B. (2010). Value from regulatory construal fit: The persuasive impact of fit between consumer goals and message concreteness. Journal of Consumer Research, 36, 735–747. Levin, I. P., & Gaeth, G. J. (1988). Framing of attribute information before and after consuming the product. Journal of Consumer Research, 15, 374–378. Levin, I. P., Gaeth, G. J., Schreiber, J., & Lauriola, M. (2002). A new look at framing effects: Distribution of effect sizes, individual differences, and independence of types of effects. Organizational Behavior and Human Decision Processes, 88, 411–429. Levin, I. P., Johnson, R. D., Deldin, P. J., Carstens, L. M., Cressey, L. J., & Davis, C. R. (1986). Framing effects in decisions with completely and incompletely

described alternatives. Organizational Behavior and Human Decision Processes, 38, 48–64. Levin, I. P., Schneider, S. L., & Gaeth, G. J. (1998). All frames are not created equal: A typology and critical analysis of framing effects. Organizational Behavior and Human Decision Processes, 76, 149–188. Levin, I. P., Schnittjer, S. K., & Thee, S. L. (1988). Information framing effects in social and personal decisions. Journal of Experimental Social Psychology, 24, 520–529. Levin, I. P., Snyder, M. A., & Chapman, D. P. (1989). The interaction of experiential and situational factors and gender in a simulated risky decision-making task. The Journal of Psychology, 122, 173–181. Liberman, N., Sagristano, M. D., & Trope, Y. (2002). The effect of temporal distance on level of mental construal. Journal of Experimental Social Psychology, 38, 528–534. Liberman, N., & Trope, Y. (1998). The role of feasibility and desirability considerations in near and distant future decisions: A test of temporal construal theory. Journal of Personality and Social Psychology, 75, 5–18. Liberman, N., Trope, Y., & Wakslak, C. (2007). Construal level theory and consumer behavior. Journal of Consumer Psychology, 17, 113–117. Linville, P. W., Fischer, G. W., & Fischhoff, B. (1993). AIDS risk perceptions and decision biases. The social psychology of HIV infection (pp. 5–38). Hillsdale, NJ: Lawrence Erlbaum Associates. Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis. Thousand Oaks, CA: Sage Publications. Liviatan, I., Trope, Y., & Liberman, N. (2008). Interpersonal similarity as a social distance dimension: Implications for perception of others’ actions. Journal of Experimental Social Psychology, 44, 1256–1269. Lynch, J. G. Jr., (1982). On the external validity of experiments in consumer research. Journal of Consumer Research, 9, 225–239. MacKenzie, S. B. (1986). The role of attention in mediating the effect of advertising on attribute importance. Journal of Consumer Research, 13, 174–195. Magee, J. C., Milliken, F. J., & Lurie, A. R. (2010). Power differences in the construal of a crisis: The immediate aftermath of September 11, 2001. Personality and Social Psychology Bulletin, 36, 354–370. Maheswaran, D., & Meyers-Levy, J. (1990). The influence of message framing and issue involvement. Journal of Marketing Research, 27, 361–367. Marks, L. J., & Kamins, M. A. (1988). The use of product sampling and advertising: Effects of sequence of exposure and degree of advertising claim exaggeration on consumers’ belief Clarity, belief confidence, and attitudes. Journal of Marketing Research, 25, 266–281. Marteau, T. M. (1989). Framing of information: Its influence upon decisions of doctors and patients. British Journal of Social Psychology, 28, 89–94. Palmatier, R. W., Dant, R. P., Grewal, D., & Evans, K. R. (2006). Factors influencing the effectiveness of relationship marketing: A meta-analysis. Journal of Marketing, 70, 136–153. Pany, K., & Reckers, P. M. J. (1987). With- vs. between-subjects experimental designs: A study of demand effects. Auditing: A Journal of Practice & Theory, 7, 39–53. Perner, J., Gschaider, A., Kühberger, A., & Schrofner, S. (1999). Predicting others through simulation or by theory? A method to decide. Mind & Language, 14, 57–79. Peterson, R. A. (2001). On the use of college students in social science research: Insights from a second-order meta-analysis. Journal of Consumer Research, 28, 450–461. Raudenbush, S. W., Becker, B. J., & Kalaian, H. (1988). Modeling multivariate effect sizes. Psychological Bulletin, 103, 111–120. Riley, R. D. (2009). Multivariate meta-analysis: The effect of ignoring within-study correlation. Journal of the Royal Statistical Society, 172, 789–811. Riley, R. D., Thompson, J. R., & Abrams, K. R. (2008). An alternative model for bivariate random-effects meta-analysis when the within-study correlations are unknown. Biostatistics, 9, 172–186. Rosenberg, M. S., Adams, D. C., & Gurevitch, J. (2000). MetaWin: A statistical software for meta-analysis. Sunderland, MA: Sinauer Associates. Rosenthal, R. (1979). The ‘file drawer problem’ and tolerance for null results. Psychological Bulletin, 86, 638–641. Rosenthal, R., & Rosnow, R. L. (2008). Essentials of behavioral research: Methods and data analysis. Boston, MA: McGraw Hill. Sharpe, D. (1997). Of apples and oranges, file drawers, and garbage: Why validity issues in meta-analysis will not go away. Clinical Psychology Review, 17, 881–901. Shiv, B., Edell Britton, J. A., & Payne, J. W. (2004). Does elaboration increase or decrease the effectiveness of negatively versus positively framed messages? Journal of Consumer Research, 31, 199–208. Shu, S. B., & Gneezy, A. (2010). Procrastination of enjoyable experiences. Journal of Marketing Research, 47, 933–944. Smith, R. E., & Swinyard, W. R. (1983). Attitude-behavior consistency: The impact of product trial vs. advertising. Journal of Marketing Research, 20, 257–267. Smith, P. K., & Trope, Y. (2006). You focus on the forest when you’re in charge of the trees: Power priming and abstract information processing. Journal of Personality and Social Psychology, 90, 578–596. Sultan, F., Farley, J. U., & Lehmann, D. R. (1990). A meta-analysis of applications of diffusion models. Journal of Marketing Research, 27, 70–77. Todorov, A., Goren, A., & Trope, Y. (2007). Probability as a psychological distance: Construal and preferences. Journal of Experimental Social Psychology, 43, 473–482. Trope, Y., & Liberman, N. (2000). Temporal construal and time-dependent changes in preference. Journal of Personality and Social Psychology, 79, 876–889.

T.H. Freling et al. / Organizational Behavior and Human Decision Processes 124 (2014) 95–109 Trope, Y., & Liberman, N. (2003). Temporal construal. Psychological Review, 110, 403–421. Trope, Y., & Liberman, N. (2010). Construal-level theory of psychological distance. Psychological Review, 117, 440–463. Trope, Y., Liberman, N., & Wakslak, C. (2007). Construal levels and psychological distance: Effects on representation, prediction, evaluation, and behavior. Journal of Consumer Psychology, 17, 83–95. Ulkumen, G., & Cheema, A. (2011). Framing goals to influence personal savings: The role of specificity and construal level. Journal of Marketing Research, 48, 958–969. Vallacher, R. R., & Wegner, D. M. (1989). Levels of personal agency: Individual variation in action identification. Journal of Personality and Social Psychology, 57, 660–671. Vinchur, A. J., Schippmann, S. S., Switzer, F. S., III, & Roth, P. L. (1998). A metaanalytic review of predictors of job performance for salespeople. Journal of Applied Psychology, 83, 586–597.

109

Wakslak, C. J., Nussbaum, S., Liberman, N., & Trope, Y. (2008). Representations of the self in the near and distant future. Journal of Personality and Social Psychology, 95, 757–773. Wakslak, C. J., Trope, Y., Liberman, N., & Alony, R. (2006). Seeing the forest when entry is unlikely: Probability and the mental representation of events. Journal of Experimental Psychology, 135, 641–653. White, K., MacDonnell, R., & Dahl, D. W. (2011). It’s the mindset that matters: The role of construal level and message framing in influencing efficacy and conservation behaviors over the long-term. Journal of Marketing Research, 48, 472–485. White, K., & Peloza, J. (2009). Self-benefit versus other-benefit marketing appeals: Their effectiveness in generating charitable support. Journal of Marketing, 73, 109–124. Zhao, M., & Xie, J. (2011). Effects of social and temporal distance on consumers’ responses to peer recommendations. Journal of Marketing Research, 48, 486–496.