Answers to Exercises in the Book
11.1 Dealing with interdependence
The detailed interpretation of the studies listed in Table 11.1 would depend on thorough knowledge of the academic literatures for the concepts which are addressed by the researchers. Some students will have that knowledge, while others may not.
Context for the study – would the relationship between the chosen DV and the chosen PV depend on the setting where the research was carried out? For example, the first study by Coyle-Shapiro and Kessler on psychological contract fulfilment might depend on conditions in the local labour market where the study was carried out. If there are few alternative jobs available to employees, then they may feel they have little choice other than to stay with the organisation. As noted in the text, organisational size was taken into account in several of the studies, because different findings would be expected for large and small organisations. If the researcher is interested in contextual factors, then they could become part of more complex theorising with respect to interaction effects or contingencies. If they are not, then the obvious options are either to simplify through selecting sub-samples (for example, only study companies in a specific country) or to include a variety of contextual factors and examine their effects through multivariate analyses.
Other possible PVs which may also be related to the DV. The issue here is one of confounding, especially if the variable selected as a PV is itself correlated with other potential PVs. A significant correlation between two variables does not mean that the relationship is real, especially where there is another variable which is correlated with both. For example, in the Thompson (2004) study it is unlikely that all possible predictors of decline in national competitiveness have been included. The researcher therefore has to assume that other potential PVs which were excluded are unrelated to those in the study. This may be a valid assumption, but the wise researcher makes strong efforts to consider all the PVs that theory suggests, and then uses multivariate analysis to tease out their relative effects.
11.2 Designing your own multivariate causal analysis project
In general, hypotheses are stated in the form of causal relationships among concepts which are of academic interest, and it is useful to do that as the first step in refining the logic of a causal analysis project. The second step involves being more specific about each of the following:
1. What is your dependent variable (DV) – here you could be guided by other studies in relevant literature. What variables do they include, and why do studies choose one DV rather than another? It can be a very fruitful way of advancing theory to work out why findings differ depending on the DV, since this could lead you in the direction either of a contingency theory (the effect of a PV is like this for one DV but like that for another DV).
2. What are your predictor variables (PVs) – here again you should be guided by existing literature, which will indicate predictors which are important to include in your study (perhaps because they have been shown to have important relationships with the PVs which you are interested in). Some variables therefore have to be included (we have already seen that organisational size is one of those), but of course, you make a contribution to existing knowledge by using theory or intuition to examine PVs which have not been studied before.
11.3 Your multivariate causal analysis project
1. Consider your DVs – what kind of measurement scale would you use for it (category / continual scale)? What choices are there for statistical tests of your research hypothesis?
2. Consider your PVs – what kind of measurement scale would you use for it (category / continual scale)? What choices are there for statistical tests of your research hypothesis?
As we discussed in chapter 9, the measurement scale is not an intrinsic feature of a concept or variable. Instead, it is a choice made by the researcher, and that choice may be decided by the quality of the data available, the precision by which it can be measured and by the analysis methods available. If you have only crude data, then you may feel that the only safe thing you can do is to treat it either as category data (pass or fail, yes or no) or as ordered continuous data (where you trust the rank ordering of data but not necessarily their precise values). With respect to variables measured on Likert scales, researchers differ in how they prefer to treat them. Some see them as no more than ordered category scales, while others see them as approximations to continuous category scales. Our view is that this is a matter of choice rather than principle. There is a much wider choice of analysis tools available to you as a researcher if you consider DVs as measured on continuous scales rather than ordered category scales. Your guide is Table 11.4 for deciding on appropriate statistical tests for your research hypotheses.
Note: Not all questions have answers, this is because some questions don't lend themselves to a clear correct answer - dicuss with you're tutor if you're not sure of your work on the questions not on this website.