요인분석의 응답순서효과

Abstract

During the past several decades, researchers have shown empirically that validity of survey reports is highly dependent on the cognitive processes that respondents use while answering survey questions. Context effects suggest that answers to survey questions can be affected by prior items, which may provide respondents with cognitive cues that are used to answer subsequent questions. Among the context effects of survey questions, response-order effects suggest that the order in which a response category appears will also affect its selection. Due to the previous studies, it is now well known that responses to questions measuring beliefs and attitudes may be significantly altered by apparently trivial changes in form and wording of questions or in the context in which they are asked. However, although much empirical researches have reported the effects of such changes in surveys, no study has focused on the response-order effects in factor analysis. This paper reviews previous studies and evaluates the effects in deriving latent variables (or factors) in factor analysis. Researches of the social sciences have commonly used a factor analysis in empirical studies to derive latent variables related to fundamental concept of consumer behavior, even though main purposes of researches may differ from each other. Factor analysis is not free from the response-order effects. Nevertheless, many researchers have performed a factor analysis, without considering the response-order effects, and have been using the questionnaire in which the response-order effects are not controlled. For example, some of previous studies apply factor analyses to the data obtained from the survey in which questions are presented in order of expected factor. This study compares the three forms of survey questionnaires that the response-order effects are expected to be different to each other. Subjects for this study were 188 university students in Korea. The selected service categories are family restaurant and fast food restaurant, and the selected measurement instrument is SERVQUAL that measures service quality with the 22 questions. Subjects were asked to evaluate the 22 questions of SERVQUAL that are presented as a form among the three forms of surveys (Category, Sequence, Random) for two service types (family restaurant and fast food restaurant). In the category form of surveys, the 22 questions were categorized into the 5 factors (reliability, assurance, tangibles, empathy, responsiveness) and they were presented sequentially in the order of the factors with the names of factors. In the sequence form of survey, the 22 questions were presented sequentially in the order of the factors without the names of factors. In the random form of survey, in contrast, the 22 questions were randomly presented. The correlations between the responses to questions (the items which are related to each factor) in the category condition were higher than those in the sequence condition, and those in the sequence condition were higher than those in the random condition. The results imply that researchers should be cautious of the response-order effects, and the random form of survey is the better than the sequence form of survey or the category form of survey to eliminate the response-order effects.