As one phase of this research is looking into the interaction effect of community
participation on the relationship between perceived value and loyalty, it was important that the analysis tool enabled this function, and provided more accurate results than in first generation tools. In SEM, and the SmartPLS program, interactive effects of the moderating variables are tested within one model using product term approach. The researcher used the
following background and guidelines for creating and interpreting the research model within the SmartPLS program.
Cortina (1993) writes that during the development of scientific disciplines such as the social sciences, the complexity of hypothesized relationships and models has increased steadily. Jaccard and Turissi (2003) demonstrated that there are basically 6 types of models which describe the relationship between dependent and independent variables, including those which measure moderating (or interaction) effects when a moderator variable
influences the strength of the direct effect between the independent and dependent variables (Henseler and Fassott, 2008). Baron and Kenny (1986) write that in general terms, a
moderator is a qualitative (e.g. sex, race, class) or quantitative (e.g. level of reward) variable that affects an independent or predictor variable and a dependent or criterion variable.
Chin, et al. (2003) support that previously, interaction effects might have been dismissed as the analytical methods used to describe these were insufficient. Measurement error, which SEM takes into account, has been proposed to result in diminished detection of interaction effects as well as underestimating the strength of such effects (Chin et al., 2003; Busemeyer and Jones, 1983). First generation SEM techniques lacked the integration of measurement error in the analysis and thus might have led to these phenomena (non-detection of
interaction effects or underestimation of effect strength). Henseler and Fassott (2008) state that although in the majority of structural equation models interaction effects are not taken into account, the literature supports the importance of understanding moderating effects in order to better understand complex relationships. For example, a researcher would not be surprised when customer satisfaction is correlated positively with customer loyalty, but understanding better the circumstances under which this relationship is either very strong or very weak would provide a significant progression in scientific knowledge. In this study, the moderating community variables are included in the model to understand better the circumstances under which perceived value and loyalty relate to each other in the news site
context, and the analysis of this interaction effect can be determined using SEM techniques and the SmartPLS program.
In PLS path modelling, there is a distinction made between two types of interaction effect calculation methods. The first is the product multiplier effect, where the product of the independent variable and the moderator variable is taken into consideration in the model.
Thus the simplified model above, with the addition of a moderator variable would be depicted as follows:
Figure 3.5 Interaction effects
The PLS product term calculation is then calculated as (Henseler and Fassot, 2008; Baron and Kenny, 1986):
Item 1
Item 2
Item 3
Exogenous LV
Item 1
Item 2
Item 3 Endogenous LV
Item 1 Item 2 Item 3
Moderator variable
Figure 3.6 Product Multiplier method for PLS Path Model interaction effects
This calculation therefore separates the effects of each the moderator, exogenous as well as the product term (moderator * exogenous) which describes the single effect of each of these variables on the endogenous variable. Only when the direct effects of all the components of the product term are taken into consideration, can the product term represent the interaction effect, otherwise, overestimation of the interaction effect could result (Carte and Russell, 2003: Henseler and Fassott 2008). The moderating variable should be represented by an interval scale (Henseler and Fassot, 2008), as the community variables are this research.
In analysing moderation effects, the level of effect of the independent variable on the dependent variable changes in strength and or direction depending on the level of the moderator variable. Typical problems with assessing interaction effects have been accurately assessing the effects and correctly interpreting main effects in the presence of interaction effects. Following Chin, Marcolin and Newsted (1996, 2003), Hubona suggests a three step approach to analysing interaction effects in PLS, and was adopted in this research:
Standardize the indicators for the main and moderating constructs
Create all pair wise product indicators where each indicator from the main construct is multiplied with each indicator from the moderating construct
Use the new product indicators to reflect the interaction construct
Endogenous LV Exogenous LV
Moderator variable
Exogenous LV * Moderator variable
The two ways to determine the significance of the moderating effect are to bootstrap the model and look at the significance of the path coefficient, as well as determining the difference in the R2 of the model with and without the moderating effect, and using
Cohen’s formula to determine the power of that effect, which will test the predictive quality of that moderating effect. This procedure was followed in the research and results are presented in chapter 5.
Baron and Kenny (1986) write that the presence of measurement error in either the moderator of the independent variable complicates the analysis of moderator variables.
Whereas there are methods for making adjustments for measurement error as proposed by Kenny and Judd, these assume a normal distribution of data. However, as the PLS method takes measurement error into account, and does not require normally distributed data, the method provides the researcher with a more flexible calculation of moderating effects. This was also an important consideration in the researcher’s selection of SmartPLS as an
analysis tool.