How to interpret HLM analyses looking at work group and individual worker effects?

Liden et al (2004, PDF) is an example of a study examining employees in work groups in an organisation. Thus, measures are obtained of both individual workers and of the work groups that they belong too. The data is analysed using multilevel modelling (or what the paper calls HLM).

• How can the HLM results in the paper be explained in a way that is meaningful for someone with only a basic level of statistical knowledge?
• What resources could help someone with only a basic level of statistical knowledge understand the paper?

Note: The above questions are based on an email I received asking about how to interpret the results of Liden et al (2004). I'll probably add an answer. But I thought others might also have answers, and that the answer might be relevant to others.

The abstract:

Social loaﬁng was investigated by testing a multilevel model among 23 intact work groups comprised of 168 employees representing two organizations. Results demonstrated that as hypothesized at the individual level, increases in task interdependence and decreases in task visibility and distributive justice were associated with greater occurrence of social loaﬁng. At the group level, increased group size and decreased cohesiveness were related to increased levels of social loaﬁng. Of particular interest was the ﬁnding that group member perceptions of perceived coworker loaﬁng was associated with reduced social loaﬁng, opposite of our predictions. We suggested that this unexpected ﬁnding may provide evidence of a social compensation effect.

References

• Liden, R. C., Wayne, S. J., Jaworski, R. A., & Bennett, N. (2004). Social loafing: A field investigation. Journal of Management, 30(2), 285-304. PDF
-

Tutorials on HLM

Ultimately the paper assumes that you understand HLM. HLM analysis is a bit more complex than multiple regression because it deals with multiple levels of variance.

There are various tutorials and readings that can bring you up to speed:

• Woltman et al (2012) in particular provide a quantitative psychology tutorial.
• For a textbook treatment you could check out Raudenbush and Bryk (2002).

Interpreting Liden et al (2004)

The study uses a bunch of variables some at the individual level and some at the group-level. There are 168 participants in something like 23 groups (although there may have been some deletion of groups with insufficient interrater reliability).

The study reports individual and group level correlation matrices of these variables. Group level measures are presumably the mean of the individual group members.

The study often mentions $r_{wg}$, ICC(1) and ICC(2). These are all indices of within group rating agreement. A general assumption of multilevel research is that if you are going to aggregate individuals into group level scores, then individuals within groups should be more similar to each other than across groups. Thus, these three indices are used to justify aggregation into groups. See LeBreton and Senter (2008) for a discussion of how interrater agreement indices can be used in organisational research.

The main HLM predicts social loafing from both group and individual level variables. They estimate that 49.99% of the variance in social loafing can be explained by group differences. So in summary, some groups engage in more social loafing than others. Nonetheless, there is still variation in social loafing within groups (e.g., loafers in non-loafing groups, and non-loafers in loafing groups).

The authors then go on to analyse what predicts variation within groups using individual level measures. In Table 3 shows a model with four predictors. For example, it shows that individuals with more task interdependence are more likely to engage in social loafing as indicated by the positive coefficient and the significant p-value. The authors note that they were able to explain 21.81% of the within-group variance in employee social loafing.

Then they report a group level analysis which predicts the variation in average social loafing in groups from other group-level variables. They find for example that larger, less cohesive groups have more group level social loafing. Not surprising, they also find that groups which perceive that others engage in social loafing do in fact engage in more social loafing. They find that 30.72% of the variance in group level variance is explained by these group level variables.

The HLM approach is thus a nice fit to studying workers who are situated in different work groups. You are able to examine both individual and group-level effects. This is consistent with a theory of behaviour in the workplace that says that behaviour is influenced by both individual characteristics as well as group dynamics.

References

• LeBreton, J. M. & Senter,J. L. (2008). Answers to 20 Questions About Interrater Reliability and Interrater Agreement. Organizational Research Methods,11, p 815-852.
• Woltman, H., Feldstain, A., MacKay, J. C., & Rocchi, M. (2012). An introduction to hierarchical linear modeling. Tutorials in Quantitative Methods for Psychology, 8(1), 52-69. PDF
• Raudensbush, S.W. and Bryk, A.S. (2002). Hierarchical Linear Models, Sage Publications.
-