مفهوم رضایت از زندگی در سراسر فرهنگ ها: تجزیه و تحلیل IRT

The present study examined measurement equivalence of the Satisfaction with Life Scale between American and Chinese samples using multigroup Structural Equation Modeling (SEM), Multiple indicator multiple cause model (MIMIC), and Item Response Theory (IRT). Whereas SEM and MIMIC identified only one biased item across cultures, the IRT analysis revealed that four of the five items had differential item functioning. According to IRT, Chinese whose latent life satisfaction scores were quite high did not endorse items such as “So far I have gotten the important things I want in life” and “If I could live my life over, I would change almost nothing.” The IRT analysis also showed that even when the unbiased items were weighted more heavily than the biased items, the latent mean life satisfaction score of Chinese was substantially lower than that of Americans. The differences among SEM, MIMIC, and IRT are discussed.

Kitayama and Markus (2000) presented a theoretical analysis of cultural differences in well-being, and argued that (a) well-being comes from cultural participation, and (b) to the extent that cultural participation requires different forms across cultures, well-being feels different and means something different across cultures. For instance, the Item Response Theory (IRT) analysis of the positive affect (PA) subscale of the Positive and Negative Affect Schedule (PANAS: Watson, Clark, & Tellegen, 1988) showed that “pride” was not endorsed by Chinese who endorsed other “positive” emotions, whereas it was endorsed by Americans who endorsed other “positive” emotions (Oishi, in press). This measurement discrepancy indicates that “pride” is not conceived as “positive” among Chinese and reveals the conceptual difference of “positive” emotions between Chinese and Americans (see Huang, Church, & Katigbak, 1997 on anxiety between Philippines and Americans). A main implication of Kitayama and Markus’ theoretical analysis for culture and personality research is that it is crucial to examine not only mean-level differences in a construct (e.g., self-esteem) and the nomological net of this construct across cultures, but also the deeper structure of the construct because the traditional questions of mean-level difference across cultures presuppose conceptual equivalence. The Satisfaction with Life Scale (SWLS; Diener, Emmons, Larsen, & Griffin, 1985) has been one of the most widely used scales for the measurement of global life satisfaction. Life satisfaction is one of the central constructs of well-being (Diener, 1984) and has been of great interest to both cultural and personality psychologists (Diener et al., 2003 and Diener et al., 1999 for review). Its psychometric properties have been well-established in the United States (Pavot & Diener, 1993). In contrast, the psychometric properties of the SWLS in non-American samples have not been extensively examined (see Vittersø, Røysamb, & Diener, 2002, however, for an initial effort in this direction). Therefore, although previous research found large international differences in the mean levels of life satisfaction (e.g., Diener, Suh, Smith, & Shao, 1995), it is unclear exactly how these mean differences can be interpreted because of the lack of information concerning measurement equivalence. The present study examines measurement equivalence of the SWLS between Chinese and American college student samples, using the structural equation modeling (SEM), multiple indicator multiple cause (MIMIC) modeling, and Differential Item Functioning (DIF) analysis. 1.1. DIF analysis To examine measurement equivalence of the SWLS among Chinese and American college students, I employed the IRT analysis with a model-testing approach (Thissen, Stenberg, & Gerrard, 1986) using the Multilog 7.03 program. IRT is different from classical test theory (CTT) in several important ways (see Embretson and Reise, 2000 and Hambleton and Swaminathan, 1985 for details). The most significant difference between CTT and IRT in the present context is concerned with the standard error of measurement. Whereas the standard error of measurement is assumed to apply to the whole sample in CTT, the standard error of measurement in IRT varies depending on the latent trait score (typically, there is less reliability for those with extreme latent scores). In other words, whereas the source of errors in CTT is either occasion (in the case of test–retest reliability) or item sampling (in the case of internal consistency), additional sources of error can be considered in IRT (as in Generalizability theory by Cronbach et al., 1972 and Shavelson et al., 1989), such as a person’s latent score and person-by-item interaction. Traditional reliability indices such as Cronbach’s α and test–retest reliability coefficient do not provide information about person-by-item interaction, namely, whether some items measured some individuals better than others. In IRT, this interaction is considered. In addition, classical item parameters (e.g., item–total correlation) are sample-specific, whereas IRT parameters are not sample dependent. The score computed in IRT, therefore, can be readily compared across different test forms. By and large, IRT parameters have a greater degree of generalizability than classical item parameters. Second, in CTT, if two individuals answered the same number of items “correctly” (or gave the same number of “yes” responses), these two individuals would have the same total score. In contrast, in IRT, even if two individuals answered the same number of items “correctly” (or gave the same number of “yes” responses), the person who correctly answered more difficult items (or those who said “yes” to the items less frequently endorsed) would receive a higher total score than the other who correctly answered less difficult items in IRT. Virtually, all previous cross-cultural research in well-being has neglected item difficulty parameters. Thus, it is of great interest whether the scoring method of IRT, which takes into account the item difficulty parameters, would reveal a different result than the conventional scoring method.

The DIF analyses provided a different perspective from the traditional approaches to the measurement issue in culture and well-being research. Equally important, IRT analyses revealed a substantive level of mean differences between Chinese and Americans, even when the biased items were weighted less in scoring. Thus, previously found mean differences between Americans and Chinese (e.g., Diener et al., 1995) might not be due simply to item biases. Finally, IRT analysis provided invaluable information concerning the concept of life satisfaction. The present research illuminates the importance and benefit of employing DIF analyses in other constructs (e.g., self-esteem, depression). I hope that DIF analysis and other IRT models will be utilized in the future in various research topics in culture and personality.