Path models are a way of relating theoretical knowledge about a data-set to correlation-based data analysis (such as PLS) in order to arrive at causal relationships. Path models consist of three things. First are problem-specific
entities or latent variables, which are items of direct interest but which are not directly measurable. In this project, these variables will be based on the abstract frameworks used to explain suicide bombing. So, for example, “social factors” and “economic factors” would both be latent variables.
Second, are Attributes or manifest variables which are direct measures of indirect items of interest and which determine the level of importance that these unobservable might have. For example, to determine the status of the latent variable “economic factors,” we might have the manifest variables “Poverty level,” “Employment Prospects,” and “Financial Reward for Family.” These three manifest variables would give some insight into the strength of the economic motivations for a suicide bomber. For this analysis effort, a questionnaire is the basis for measurement. Details, such as what questions to ask, who will answer these questions, what set of answers are permitted, and so on, are addressed later in this work.
Finally there are causal paths, also known as relationship or influence paths, linking these constructs together. The nature of the manifest variable to latent variable linkages is straight-forward; a direct connection of one or more manifest variables to each latent variable, with manifest variables linking to only one latent variable. The latent variable interconnections are more complicated and must be determined from the results of a suicide bomber focused literature search, as well as expert opinion.
There are two stages to analyzing and interpreting PLS path models. The first assesses the validity of the measurement model—the way that the manifest variables relate to their latent variables. The second stage assesses the structural model—the way that the latent variables related to each other and the dependent variable of interest. Experts have suggested that the measurement model can be accessed on the basis of three metrics: individual item reliability, internal consistency and discriminant validity.
Once the measurement model of the manifest variables has been confirmed as valid, the structural model of the latent variables can be evaluated. For the structural model, PLS calculates indirect and total (direct and indirect) effects to establish the relative importance of the various latent variables. A measure of the predictive power of the model is the traditional R2 residual value for the final dependent variable.
It should be noted that different disciplines have differing thresholds of what is an acceptable R2 value. For example, in engineering or the hard sciences, an R2 value of less than 0.8 is rarely regarded as significant. In contrast, in many social sciences, R2 values of 0.4-0.6 are standard, and in some circumstances of cases of behavioral research R2 values as low at 0.2 can be significant.
