Modeling Attitudes Toward Science: Development and Validation of the Credibility of Science Scale

ABSTRACT

We present the Credibility of Science Scale (CoSS), an efficient 6-item scale demonstrating excellent reliability and validity. CoSS scores exhibit criterion validity in predicting beliefs across a host of contemporary science topics, over and above previously documented predictors (Study 2). Further, we present evidence for the convergent and discriminant validity of the instrument, which correlates as expected with several variables previously implicated in scientific belief (e.g., political ideology, religious identity, conspiracist thinking; Study 2) and also with measures of cognitive ability (Study 3). Finally, we discuss possible uses of the CoSS as a tool for understanding science-related beliefs, behavior, and communication.

Acknowledgments

Approved for Public Release; Distribution Unlimited. Case Number 15-2885. ©2015 The MITRE Corporation. All Rights Reserved. We thank Cory Clark for comments on an earlier draft of this manuscript.

Notes

The cited work has used different strategies to assess evaluation of science, some of which are generalized measures (i.e., people’s attitudes toward scientists in general), but many are topic specific (e.g., people’s trust in scientists in a particular domain). All of these measurement strategies are linked by a focus on respondents’ evaluations of scientists, scientific disciplines, or their methods in favorable terms—as having or lacking credibility, expertise, trustworthiness, and so on. As described next, however, past measures are subject to a number of limitations.

Note also that there is an extensive literature on students’ attitudes toward studying science or in pursuing science-related careers, particularly among children and adolescents (e.g., Osborne, Simon, & Collins, 2003). Although this literature often adopts similar terms to those we use, the focus is quite different.

A common strategy aimed at minimizing response biases is to use both positively and negatively worded items. However, there is little evidence that such a strategy is effective, and the mixing of positively and negatively keyed responses may create psychometric method artifacts that can prevent scales from achieving unidimensionality (Conrad et al., 2004). We specifically chose to utilize negatively worded statements because a preliminary study suggested that these items had better psychometric properties (e.g., higher discrimination coefficients in exploratory IRT models).