Cross-Cultural Patterns in Emotion Recognition: Highlighting Design and Analytical Techniques
Total Page:16
File Type:pdf, Size:1020Kb
Emotion Copyright 2002 by the American Psychological Association, Inc. 2002, Vol. 2, No. 1, 75–84 1528-3542/02/$5.00 DOI: 10.1037//1528-3542.2.1.75 BRIEF REPORT Cross-Cultural Patterns in Emotion Recognition: Highlighting Design and Analytical Techniques Hillary Anger Elfenbein Manas K. Mandal Harvard University Indian Institute of Technology, Kharagpur Nalini Ambady Susumu Harizuka Harvard University Kyushu University Surender Kumar Chikushi Women University This article highlights a range of design and analytical tools for studying the cross-cultural communication of emotion using forced-choice experimental de- signs. American, Indian, and Japanese participants judged facial expressions from all 3 cultures. A factorial experimental design is used, balanced n × n across cultures, to separate “absolute” cultural differences from “relational” effects char- acterizing the relationship between the emotion expressor and perceiver. Use of a response bias correction is illustrated for the tendency to endorse particular mul- tiple-choice categories more often than others. Treating response bias also as an opportunity to gain insight into attributional style, the authors examined similarities and differences in response patterns across cultural groups. Finally, the authors examined patterns in the errors or confusions that participants make during emotion recognition and documented strong similarity across cultures. The long-standing debate between psychologists turally specific is ceding to attempts to integrate evi- who argue whether emotions are universal versus cul- dence for both perspectives (e.g., Fiske, Kitayama, Markus, & Nisbett, 1998; Markus & Kitayama, 1991; Hillary Anger Elfenbein, Program in Organizational Be- Mesquita & Frijda, 1992; Mesquita, Frijda, & Scher- havior, Harvard University; Manas K. Mandal, Department er, 1997; Scherer & Wallbott, 1994). Researchers of Humanities and Social Sciences, Indian Institute of Tech- studying emotion recognition have extensively docu- nology, Kharagpur, Kharagpur, India; Nalini Ambady, De- mented both cultural similarities and differences. partment of Psychology, Harvard University; Susumu Hari- zuka, Faculty of Education, Center for Clinical Psychology Cross-cultural studies of emotion recognition have and Human Development, Kyushu University, Fukuoka been one of several central sources of evidence in City, Japan; Surender Kumar, Department of Preschool favor of emotional universality. Classic studies (e.g., Education, Chikushi Women University, Fukuoka City, Japan. Ekman, 1972; Ekman, Sorenson, & Friesen, 1969; Preparation of this article was supported by a National Izard, 1971) demonstrated that facial photographs of Science Foundation Graduate Student Fellowship, a Rand Americans expressing basic emotions could be recog- research grant, Presidential Early Career Award for Scien- nized at above-chance accuracy in literate and prelit- tists and Engineers Grant BCS-9733706, and Indian Insti- erate cultures. Despite this evidence for universality, tute of Technology Scheme for Innovative Research and the same studies also provide evidence for cultural Development Grant IIT-Accts. 05-19-05. We thank Susan differences, given that American samples generally Choi, Heather Gray, Abby Marsh, and Kevyn Yong for their outperformed others when viewing these American helpful comments, and Sanaz Ghazal, Maiga Miranda, and Julie Weintraub for research assistance. stimuli. For example, in Izard’s (1971) large-scale Correspondence concerning this article should be ad- study, American and European groups correctly iden- dressed to Hillary Anger Elfenbein, Harvard Business tified 75%–83% of the facial photographs, whereas School, Baker Library 477, Soldiers Field, Boston, Massa- the Japanese group scored 65% and the African group chusetts 02163. E-mail: [email protected] correctly identified only 50%. 75 76 BRIEF REPORT “Forced-Choice” Experimental Designs above” option. Third, perhaps the likeliest reason forced-choice designs remain is convenience. Dimen- The majority of studies on the cross-cultural com- sional ratings can be tedious for participants, and ana- munication of emotion have used “forced-choice” de- lyzing open-ended participant responses can be time- signs, in which participants view expressions and en- consuming and ambiguous for experimenters. ter multiple-choice responses from a predetermined list. The relative merits of such studies have been hotly debated (e.g., Ekman, 1994; Frank & Stennett, Analytical Possibilities With 2001; Haidt & Keltner, 1999; Russell, 1994). Russell Forced-Choice Experiments (1993, 1994) argued that these studies prime partici- pants to interpret stimuli as expressions of emotion, portray choices as mutually exclusive categories, and Given the popularity of forced-choice designs, it is inflate agreement across cultures by constraining helpful to consider analytical tools that offer richer choices. He demonstrated that forced choice can show perspectives on the data that are collected. Often, re- consensus on clearly incorrect categories when rel- searchers using forced-choice studies restrict them- evant choices are missing from the list. Participants selves to examining whether recognition accuracy is may choose a particular label when prompted by a greater than that predicted by chance guessing alone, multiple-choice list, even if they would not have gen- or differences in accuracy across samples. These re- erated the same label under free-response conditions strictions have notable exceptions, and the goal of the (Rosenberg & Ekman, 1995; Wagner, 2000). Further- current article is to highlight a range of analytical more, response choices are often generated by North tools researchers have used to gain greater insight American researchers, whose lists may or may not from forced-choice designs. correspond to emotion constructs in other cultures (Haidt & Keltner, 1999; Russell & Yik, 1996). However, there are also reasons for researchers to Interactions, in Addition to Main Effects continue using forced-choice designs. First, there are theoretical considerations. Forced-choice designs fit Broadly speaking, two different types of theories theory that emotions are categorical in nature because can explain cross-cultural differences in emotional individual emotions may have evolved separately to communication. The first type is absolutist, examin- solve distinct problems for human survival (Ekman, ing the fixed attributes of the groups expressing emo- 1992; Izard, 1994). By contrast, Russell (1980) main- tion or of the groups perceiving emotion. For ex- tained that emotions are qualitatively similar to each ample, Matsumoto (1989, 1992) and Schimmack other, yet vary quantitatively along underlying dimen- (1996) have argued that collectivistic cultural groups sions such as valence and arousal. Forced-choice de- are less accurate at expressing and perceiving nega- signs also facilitate the testing of recognition rather tive emotion, because of the risk of disrupting social than production of emotion-related constructs, which order. These arguments predict absolute differences in are different abilities evolving separately (Izard, communication accuracy across groups, or “main ef- 1994). Theoretical clarity may also decrease the vul- fects” in terms of an analysis of variance (ANOVA). nerability of forced-choice studies to methodological By contrast, relational theories characterize the rela- artifact, as Izard (1994) argued that theory-based lists tionship between the cultures of the emotional expres- of responses are less likely to miss desired choices sor and perceiver rather than the fixed attributes of and artificially boost consensus on an incorrect one group or the other. They focus on match or simi- choice. A second consideration in favor of forced- larity rather than on absolute characteristics of cul- choice designs is empirical, as findings using forced- tures. In terms of an ANOVA, relational theories pre- choice designs generally replicate those using alter- dict an interaction between expressor and perceiver native methods such as free-response and dimensional cultural groups. Recent empirical work has provided ratings (e.g., Boucher & Carlson, 1980; Ekman, 1994; initial support for relational perspectives on emotional Haidt & Keltner, 1999; Izard, 1994). Haidt and Kelt- communication (Elfenbein & Ambady, 2001), includ- ner (1999) and Frank and Stennett (2001) replicated ing a meta-analysis of the cross-cultural literature on high agreement levels with modified forced-choice emotion recognition suggesting the presence of an designs featuring improvements such as an expanded in-group advantage (Elfenbein & Ambady, 2002a, list of emotion words as well as a “none-of-the- 2002b; but see also Matsumoto, 2002). That is, emo- BRIEF REPORT 77 tional communication may be more accurate when the Elfenbein and Ambady’s (2002b) recent meta- expressor and the perceiver are both members of the analysis of emotion recognition did not provide the same cultural group rather than members of different confusion matrices necessary to calculate unbiased hit cultural groups. rates. However, the majority of these studies fit Wag- Both absolutist and relational perspectives can con- ner’s (1993) criteria for relatively low impact from tribute toward a valuable understanding of emotional response bias: using posed expressions and stimuli communication. However, only factorial experimental preselected for high accuracy rates. The current study designs balanced