Pages: 13

Quantitative behavioural science research (BSR) is research in the behavioural sciences that employs numerical data and statistical analysis to examine human behaviour. This research paradigm typically includes the development and logical justification of a causal model that likely offers an explanation for a phenomenon of interest and provides a theoretical contribution. In developing the causal model, it is typical for the scientist to explore existing theories from the knowledge base to generate and justify a set of hypotheses that together constitute a new, actionable causal model. Several approaches have been presented for developing the hypotheses of actionable causal models, some of which are the pure hypothetico-deductive (H-D) method, the traditional approach (T-A) to theory development, abduction based on qualitative data for BSR, and abduction based on quantitative data for BSR. However, each of these approaches presents challenges. For instance, hypotheses generation in the pure H-D method is limited by the reliance on human imagination and by the set of theories considered in the T-A. Data collection is typically necessary for the generation of hypotheses and the generated causal model may not be novel are the challenges of the abduction based on qualitative data for BSR and abduction based on quantitative data for BSR, respectively. Therefore, we propose that a new approach for generating hypotheses could be of value to the research community. A potentially promising starting point is to use results from meta-analytic reviews and some more recent studies to identify relevant, novel, parsimonious, causal models that are likely to have strong empirical support when evaluated.

Pages: 8

In this chapter, our new process for generating strong, novel, and parsimonious explanatory models is presented and logically justified. This process would take as its major inputs the following: 1) results from a recent meta-analytic review (MAR) that focused on studies in the specific domain of interest and 2) results from a study in the domain of interest that is even more recent than the given MAR, and which includes at least one statistically significant predictor variable that was not included in the MAR. The proposed process has as its major outputs one or more novel candidate causal models, each of which is likely to offer strong empirical support with respect to the given dependent variable. The number of novel causal models that are generated depends on or is determined by the research team. The process also has an optional feature that given a set of candidate models, the research team may be interested in determining what would be a promising subset of these models for which can data be collected on a questionnaire that will result in at least the required N Min usable observations via sampling the relevant target population.

Pages: 26

To illustrate the process for generating relevant, robust, novel, and parsimonious causal models, we select the phenomenon of information security policy compliance (ISPC). Based on a recent/relatively recent ISPC meta-analytic review (MAR), we selected top ranked empirical predictors/categories of ISPC, as identified in the MAR. The particular MAR was chosen because of its relative recency, extensiveness in the number of empirical studies included in the analysis, the range of theoretical bases covered, and the computation of the relative importance of antecedents/categories in predicting/confirming the explanatory power of security policy compliance in the presence of other predictors. We then supplemented this analysis with additional newer research to uncover emerging empirical predictors of the phenomenon. This led us to propose a causal model that includes both established and emerging ISPC factors. We provide theoretical justifications for the hypotheses in this new ISPC causal model.

Pages: 28

In this chapter, we evaluated the new causal model developed in Chapter 3, which was constructed by applying our relevant causal model generating process (as described in Chapter 2). This model included established hypotheses previously tested in other studies as well new hypotheses not previously tested. The model also investigated well-established and emergent latent variables influencing Information Security Policy Compliance Intention (SINT) among end users. Data was collected through an online survey with a pre-tested measurement instrument. Analysis using partial least squares path modelling revealed good model fit, a reliable and valid measurement model, and a strong structural model. While the model explained a significant amount of variance in SINT (R 2 = 0.784) and IS security attitude (SATT) (R 2 = 0.537), as is typical in many quantitative behavioural science research, some hypotheses were not statistically supported. However, the model achieved one of the highest reported R 2 values for SINT and SATT in the literature. Therefore, our proposed causal model generating process resulted in a robust, novel, and parsimonious model with strong explanatory power for SINT, validating its effectiveness in generating relevant, robust, novel, and parsimonious causal models.

Pages: 18

Many quantitative behavioural science studies on a given topic appear to involve the assumption that there is a single causal model that applies to all contexts. Thus, there is often talk about previous research providing conflicting results or absence of consensus. A more realistic assumption would be to assume that there is likely to be only a subset of all contexts to which each model is valid or applicable. A major consequence of the fallacious assumption is that there is inadequate reflection on context before the data collection phase of the study and often a relatively superficial analysis of context after the data analysis phase of the study. In this chapter, we present a new procedure for more adequately addressing context in the pre-data collection and post-data analysis phases of a given study. While focusing on the WHO, WHERE, and WHEN dimensions of context, we also consider the Modeling Environment of context. It can be reasonably expected that this procedure would be really useful for doctoral and other graduate students planning their empirical work and also for empirical scientists generally.

Pages: 17

Practitioners, while being interested in causal models that have strong predictive power, would also require that the statistically significant predictor variables be actionable. Yet often the guidance for practice that appears in published papers is vague and/or superficial. In this chapter, we present an analytical framework and associated process that could be used to expose actionable guidance that is offered by the results of a causal model. We present two illustrative examples that involve the application of our analytical framework and associated process.

Pages: 10

For several years, there has been the Relevance vs. Rigor debate, and many researchers have attempted to address relevant problems using rigorous methods. Still, often a major limitation of such studies is that it is often not clear how the relevant results can be made actionable by decision-makers in organizational settings where costs and benefits of actions are of major concern. In this chapter, we present a multi-criteria decision-making (MCDM)-based approach that offers a path to making research results actionable for studies that have causal models with high explanatory and predictive powers.