The first path analysis article introduced the path analysis framework for testing theories of cause and effect relationships, as well as the reduced-form path model as the basis for quantifying training impact at a macro-level. The purpose of this article will be to “flesh out” the path analysis framework in more detail and show how path models are created.  Subsequent articles will then show how to calculate path coefficients, construct reduced-form equations and demonstrate how to use the reduced-form equations to quantitatively predict the impact of training.

Creating Path Models

If you think of the purpose of a training impact study as demonstrating the relationship between a causal “intervention,” the training investment (usually the training event or program), and the development of skills that change performance and ultimately affect a business result then, in effect, a theory is created. In general terms, a theory is a set of cause and effect relationships that attempt to explain and predict behavior in particular (social) contexts.  It is a testable model of behavior that is subject to test by experiment or observation.  For example, a theory of training impact could posit a causal relationship among several variables. This can include the training event and how it enables the development of specific skills needed to do the job, the effect of these skills to on-the-job performance and how this performance influences a business result.  In terms of cause and effect, the theory states that the specific training investment causes the development of appropriate skills that in turn, causes a change in performance, which then causes a change in the business result.

The cause and effect relationship between these variables can be represented graphically. For instance, the relationship between the training investment and skill development would look like this:

Here the directed arrow points in the direction of cause to effect and can be interpreted as “training has a direct effect on skill development.”

Similarly, the relationship between skill development and work performance can be represented graphically as follows:

Below is the third relationship between work performance and business result:

Combining the three cause and effect relationships into a single graphic enables us to represent our theory of training impact:

This combined graph is called a path model for this theory of training impact.  A path model acts as a language for describing how you believe the social context you are studying works. It summarizes how the concepts of the theory relate to each other in both a direct and indirect way to account for the behavior that you observe. This means the effect of one variable on another is mediated by a third variable.

While training investment has a direct effect on skill development, it has an indirect effect on work performance mediated by skill development. Furthermore, training investment has an indirect effect on business result mediated by skill development, which impacts work performance, which then influences business results. In fact, this theory of training impact aligns with Kirkpatrick’s four levels of training evaluation.  Once there’s a path model defined, the analysis of the training impact can be done using the path analysis framework.

To further elaborate concepts from the path model of training impact, note that the model has four terms: training investment, skill development, work performance and business result. This can be referred to the training investment variable as an exogenous variable and the skill development, work performance and business result variable as endogenous variables, borrowing terminology from Econometrics. An exogenous variable is a variable whose values are determined by factors outside the causal system being modeled.  Typically, exogenous variables are thought of as independent variables in the experimental design context.  An endogenous variable is a variable whose values are determined by factors within the causal system being modeled.  Thus skill development, work performance and business result are considered endogenous variables in the path model. You can think of endogenous variables as dependent variables in the experimental design context.  Thus path models represent the relationships, or effects, among exogenous variables and endogenous variables.

A further point here is that path models do not have to be as “linear” as has been outlined above. That is, variables can have both direct and indirect effects. For example, benchmarking studies have shown that companies with high level of talent investment typically outperform companies with average level of talent investment. You can use this research to create a path model for the effect of talent investment on business result that could look like the following:

This path model shows that talent investment has both a direct and indirect effect on business result, the indirect effect mediated by the enhanced talent development resulting from the high level of investment and the effect of that talent on business results.

Exogenous variables do not have to be “funding” or investment related. Path models are used to represent theories in general, a theory of training impact being one such theory. For example, a review of employee satisfaction studies indicates that trust in leadership and relationship with immediate management increase employee engagement, which is a key driver of job satisfaction.

In this path model the exogenous variables are trust in senior leadership and relationship with immediate management. Both have direct and indirect effects on job satisfaction.

The next article will introduce the concept of path coefficients a way of quantifying the strength of cause and effect relationships and how the reduced form model will be derived. It is this reduced form model that is used to quantify the impact of investment on business results.