For organizational researchers employing surveys, understanding the semantic link between and among survey items and responses is key. Researchers like Schwarz (1999) have long understood, for example, that item order can impact survey responses. To account for “item wording similarity,” researchers may allow item error variances to correlate (cf. Rich et al., 2010, p. 625). Other researchers, such as Newman et al. (2010), have pointed to semantic similarity between items as support for the premise that work engagement is like old wine in a new bottle.
Recently, organizational researchers (e.g., Arnulf et al., 2014, 2018) have been able to use latent semantic analysis (LSA) and semantic survey response theory (SSRT) to quantify the semantic similarity between and among scales, items, and survey responses. Latent semantic analysis is a computational model that assesses similarity in language where the similarity of any “given word (or series of words) is given by the context where this word is usually found” (Arnulf et al., 2020, p. 4). Latent semantic analysis involves establishing a semantic space from a corpus of existing documents (e.g., journal articles, newspaper stories, item sets). The corpus of documents is represented in a word-by-document matrix and then transformed into an LSA space through singular value decomposition. The reduced LSA space can be used to assess the semantic similarity of documents within the space as well as new documents that are projected onto the space.
Patterns of semantic similarity resulting from LSA have accounted for a substantive amount of variability in how individuals respond to survey items that purport to measure (a) transformational leadership, motivation, and self-reported work outcomes (60–86%; Arnulf et al., 2014), (b) employee engagement and job satisfaction (25–69%; Nimon et al., 2016), and (c) perceptions of a trainee program, intrinsic motivation, and work outcomes (31–55%, Arnulf et al., 2019). It also appears that personality, demographics, professional training, and interest in the subject matter have an impact on the degree to which an individual's responses follow a semantically predictable pattern (Arnulf et al., 2018; Arnulf and Larsen, 2020, Arnulf et al., 2020). While being able to objectively access the degree to which survey responses are impacted by semantics is a great step forward in survey research, such research is often conducted with LSA spaces that are not open and therefore not customizable except by those that have access to the body of text upon which the LSA space is built. In this day of open science, researchers need access not only to the LSA space on which semantic survey research may be based but also to the underlying corpus of text to determine whether choices made in the generation of the LSA space have an impact on the results found.
Researchers may not be able to create their own LSA spaces for a number of reasons, including the fact that on some occasions it is difficult to collect a representative corpus of text (Quesada, 2011). However, building an LSA space allows researchers to customize the space including the application of weighting schemes and the level of dimensionality for the LSA space. As shown by Arnulf et al. (2018), the dimensionality of the LSA space is a factor when using an LSA space to predict empirical correlations from scale item cosines. To help address the barrier to creating an LSA space for use in the analysis of scale items in organizational research, this report provides a dataset of documents from measures reviewed in Taking the Measure of Work. In Taking the Measure of Work, Fields provided the items for 324 scales and subscales which cover the areas of job satisfaction, organizational commitment, job characteristics, job stress, job roles, organizational justice, work-family conflict, person-organization fit, work behaviors, and work values. The MOWDOC dataset presented in this manuscript provides the documents necessary to create a semantic space from the item sets presented in Fields's Taking the Measure of Work.
Copyright © 2021 Nimon. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY): http://creativecommons.org/licenses/by/4.0/.
Date of publication
Nimon, Kim, "MOWDOC: A Dataset of Documents From Taking the Measure of Work for Building a Latent Semantic Analysis Space" (2021). Human Resource Development Faculty Publications and Presentations. Paper 14.
Nimon KF (2021) MOWDOC: A Dataset of Documents From Taking the Measure of Work for Building a Latent Semantic Analysis Space. Front. Psychol. 11:523494. doi: 10.3389/fpsyg.2020.523494