The following post is a review/critique of :
Hofmann, D.A. & Stetzer, A. (1996).
A cross-level investigation of factors influencing unsafe behaviors and accidents.
Personnel Psychology, 49(2), 307-339.
Hoffmann & Stetzer (1996) offer an in-depth multi-level HLM based analysis of the factors underlying safety related behaviors. Probably among the first comprehensive studies to employ HLM, the study offers significant contributions in both theoretical and empirical aspects demonstrating how context affects and interplays with individual level variables in the specific content of safety behavior. The article is well articulated, simple to follow, doing an impressive job of building the theoretical arguments and clearly conceptualizing the model using a new-at-the-time empirical approach. While there are a few points that I would comment on, it is important to state that I find this article to be of very high academic standards, especially for that time.
My first and outmost concern with some of Hofmann’s work, including this paper and the more recent article on safety climate as moderator of LMX to OCB relationship (Hoffman et. al, 2003) has to do with their conceptualization of the group-level climate construct. Climate, by itself, is a somewhat elusive construct that requires a careful discussion, especially in definition, level of analysis and measurement. Sometimes, climate discussions shift from “climate perceptions” to “climate behaviors” to “climate outcomes” all considered as climate, incorporating either managers, employees or both, and looking at various units of analysis. While the authors refer to Zohar’s (1980) safety climate construct and questionnaire and Schneider’s (1990) work on climates, the conceptualization in this study includes safety climate at a team-level, rather than the originally conceptualized organizational level. While there have been a few conceptualizations of “climate” to a group level smaller than an organization, these studies have usually looked at sub-divisions or departments, constituting of a large number of people, whereas here we’re looking at small groups of 10 (split into 2 groups of 5 for reducing common method bias). The question then would be, what does the group-level analysis represent and whether that relates to the construct of climate or perhaps it reflects something else that is better articulated for small groups, like leader related variables (LMX etc.) as suggested by the 2003 article. The perhaps ironic point in conceptualization of climate at a group level that shows significant between-team variance is that it’s almost contradictory to the concept of organizational level climate suggested by the climate theorists. If there are so many climates at the group level, what is the sense in the organizational/divisional climate? If there is sense – are they the same thing that can be grouped into “climate”.
Further exploring climate and the group level variables, the empirical conceptualization of climate evaluates manager’s commitment to safety and worker involvement in safety activities (differing from both the conceptualization done in the Hoffman et. al 2003 paper and Zohar’s 1980 in both number of items, and dimensions used), which make less sense at the group level allowing for between group variance, even if the groups are relatively self-managed and independent. Taking a closer look at the constructs of group process, climate and even approach intensions, there is reason to believe the three share some conceptually similar dimensions. Group process and climate share aspects of employee involvement, while approach intention and group process share confidence, trust and sharing. That, by itself, may prove a challenge to some of the hypotheses like the suggested full mediation.
Interestingly, the article offers a faceoff between the HLM mediation test and OLS, showing that the HLM analysis fails to show statistically significant support for the full mediation, while OLS does. The article attributes this to the research design and number of teams participants. The authors argument for reduced statistical power in detecting partial corrections is intriguing in light of the later 2003 study showing a strong cross-level mediation effect (within group) for a sample much smaller consisting of less than a 100 participants with 3-5 members per team. Personally, with a partial mediation HLM support and full OLS mediation support, I find the mediation model empirically convincing, and the authors somewhat conservative in their report.
The questions raised by the authors in regards to causality and interventions is theoretically interesting and in my view – important. It is difficult enough to determine causality for individual level variables, so a cross-level investigation is even more complex. What forms a group process, or a climate, of a team level perceptions and whether that precedes or follows the individual level affect, perhaps both, is a question worth examining, yet the assumptions of causality in cross-level models should be drawn with care. It would make sense that both directions are happening at the same time in a interactive two way dynamic model, which would make the practical arguments for intervention policies slightly more difficult to design and apply. Once again, since we’re talking about team-level, suggested practical measures and interventions at the organizational level might prove less relevant.
Overall, an impressive paper, rich in meaning, theory and empirics, showing rigor and providing real-life relevance.
