Academic journal article Journal of the Association for Information Systems

Controlling for Lexical Closeness in Survey Research: A Demonstration on the Technology Acceptance Model

Academic journal article Journal of the Association for Information Systems

Controlling for Lexical Closeness in Survey Research: A Demonstration on the Technology Acceptance Model

Article excerpt

1 Introduction

In this research perspective paper, we present a new lexical aspect of the technology acceptance model (TAM) (Davis, 1989) and argue that there is more to TAM than just its current theory base: empirical research supports the theory extensively across IT contexts also because it represents connections among words as they are used in English. While we recognize that questionnaires not only measure the theory they pertain to but also may be influenced by a host of prejudices and priming related to the subjects, the researchers, and unrelated covariances introduced by the data-collection methods (see extensive discussion in Shadish, Cook, and Campbell (2002)), we present another, unrelated type of significant wording influence on the results. This significant influence pertains to the lexical closeness information embedded in language itself and that can be derived by analyzing word co-occurrences. Lexical closeness is the degree to which two terms or combinations of terms (including questionnaire measurement items and sentences) relate to each other as revealed through term co-occurrence in societal usage of the language1. This lexical closeness can be extracted through tools such as latent semantic analysis (LSA). LSA treats lexical closeness across documents as revealing shared inferences among the authors of those documents about the meaning of words (Landauer, Foltz, & Laham, 1998; Wild, Haley, & Bülow, 2011)2.

In this paper, we demonstrate the power of such lexical closeness by replicating TAM results based solely on the lexical closeness of the keywords in the TAM measurement items as derived through LSA. In Section 2, we present LSA in more detail. In Section 3, we discuss the lexical closeness information that we derived from two newspaper corpora discussed.

In Section 4, we show that the phenomenal success of TAM may plausibly in part result from the lexical closeness of its measurement items. We do not do so to challenge TAM; rather, we do so to show that there is more to TAM than currently considered. TAM is by far the most cited theory in the management information systems (MIS) discipline. The analyses show that simply analyzing the co-occurrence of the keywords in its questionnaire as they appear in magazine and newspaper articles statistically supports both its measurement model (how questionnaire items load into constructs) and even partly the correlations among its constructs are supported statistically. Each corpus produced adequate factorial validity in a principal components analysis (PCA) and supported TAM through linear regressions on those PCA factors-as done in the original TAM study.

In Section 5, qualifying the conclusion that only relying on how its questionnaire keywords relate to each other in newspaper articles can support TAM, we show that, nonetheless, empirical questionnaire data can provide a significantly better model. To do so, we again replicate the analysis method in the original TAM but also add covariance-based structural equation modeling (CBSEM) analysis. The above analyses support previous findings that LSA can in some cases be applied to sort questionnaire items into groups by analyzing the lexical relationships among questionnaire items (Larsen & Bong, 2016). Arnulf, Larsen, Martinsen, and Bong (2014) have shown as much for several influential theories on leadership, which Nimon, Shuck, and Zigarmi (2015) have independently replicated. We extend those previous findings by showing that the constructs derived from those items at least partly correlate with each other as theory predicts. In Section 6, we reassuringly show how an existing CBSEM method can be applied to statistically control for lexical closeness covariance when examining empirical data. That analysis shows that TAM is not based solely on lexical closeness.

Taking a broader methodological perspective, that lexical closeness revealed by analyzing corpora that does not deal with TAM studies can produce specific expected theory-based patterns is a departure from classical measurement theory3. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.