Characteristics of Single-Item Measures in Likert Scale Format

1. IntroductionThe dominant paradigm of multiple-item scale development in marketing, as advanced by Churchill Jr. (1979), has been challenged by several authors (Drolet and Morrison 2001; Rossiter 2002; Bergkvist and Rossiter 2007). The main argument is that multiple-item measures are not always necessary and can be substituted by single-item measures in many cases. The C-OAR-SE procedure suggested by Rossiter (2002) has become the focal point of the recent debate. C-OAR-SE is a comprehensive methodology for the development of multiple- and single-item measures claiming to surmount some of the current pitfalls in scale development in marketing. The focus in C-OAR-SE is content validity. Rossiter (2005; 2008) claims that this is the only validity needed in scale development, and the typical item purification through statistical procedures is unnecessary because it can change the meaning of the measured concepts. Therefore, according to Rossiter (2005; 2008), if a scale has a precise definition, there is no need to examine other types of validities. An important practical assertion advanced by C-OAR-SE is that most concepts (e.g., purchase intentions) are concrete and understood unequivocally by raters and there is no need to use multiple-item scales to measure them; a single-item measure is sufficient.Although the C-OAR-SE procedure is a solid argument in defense of single-item measures, it opens the door for a possible misconception. Specifically, Rossiter (2002; 2008) argues that the only appropriate scale for single-item measures is the semantic-differential scale. The Likert scale was called problematic and its use was discouraged because of the lack of a neutral point. According to Rossiter (2008, p.383), the Likert scale "produces hopelessly fuzzy scores." The danger with this assertion is that it annihilates a whole category of scales from the marketing research. It seems that the new attempt to relax scale development was restricted again. Furthermore, Rossiter does not provide comprehensive empirical or theoretical justification for his recommendation, except the lack of "psychological zero" and conceptual meaning.Addressing some of the raised concerns, the purpose of this study is to clarify the use of single-item measures in Likert scale format. The Likert scale is one of the most popular scales in marketing and its status is unlikely to change. Therefore, as more researchers may decide to use it as a single-item measure, it becomes necessary to examine the characteristics of the Likert scale more closely. The intention of this study is not to avow the superiority of the Likert scale to any other scales, but to describe its behavior and provide practical recommendations. The research question this study answers is: Are single-item measures in Likert format usable?2. Review of positive-negative asymmetryNunnally (1978) suggested that positively-worded items in Likert scales can be transformed into negatively-worded items and their scores can be reversed symmetrically afterwards. This practice continues even today, although it has been known that negatively-worded items introduce problems in multiple-item scales. Negatively-worded items often form a separate factor, independent of the main factor, and change the dimensionality of the construct (Herche and Engelland 1996; Mook et al. 1991; Tomas and Oliver 1999). Factors based on negatively-worded items have strong method effects and exhibit longitudinal invariance (Motl and DiStefano 2002; Horan et al. 2003). Negatively-worded items tend to lower the reliability of multi-item scales as measured by Cronbach's alpha by as much as 20% (Schriesheim et al. 1991; Barnette 2000), and confound measures in cross-cultural research, hampering measurement invariance (Wong et al. 2003). All of the above contribute to the positive-negative asymmetry, which is reviewed in the following section.Positivity and negativity are not symmetrical: negative information weights more than positive information (Anderson 1965; Rodin 1978), and positive and negative affective states have low correlation (Diener and Emmons 1984; Watson et al. …