Figure 18.2 Non-linear correlation.

Correlation and prediction

One of the interesting (and potentially powerful) features of correlational designs is that it is possible to move beyond describing the relationship between variables towards the modelling of the comparative impact of different variables on others, often in complex interactions. In these situations, researchers use statistical techniques to specify one or more of the variables as predictor variables, which essentially involves assigning to such variables the role of an independent variable which can be used to predict responses on the other variables. The assignment of predictor variables can be changed around to see which ones best account for the changes in the others. Again, returning to our earlier example, we might regard alcohol intake as the predictor variable and see how well it predicts smoking, diet and exercise, or we might regard exercise as the predictor and see how well it predicts alcohol intake, smoking and diet. Two things should be stressed here. First, our assignment of one variable as a predictor variable once again tempts us to regard it as causal, but in correlational research, as in other non-experimental research, there is always the danger that unconsidered variables are actually responsible for the relationships we think we see between the variables being examined. Thus, our degree of confidence in the causal link is considerably lowered. Second, sometimes it is possible to appeal to theoretical constructs or to other aspects of the research design (e.g. the care with which participants have been chosen, or the unlikelihood of causal explanations other than the one suggested by the observed relationships) to increase our confidence in the integrity of the causal chain. However, the assertions of causality on the basis of non-experimental research require sensitivity and experience and are never free from the possibility of alternative explanations.

One special case of non-experimental research is the survey. It is special because it often yields large amounts of observational descriptive data which can be combined by the researcher in many different ways to give rise to a whole range of tests, some of which represented comparative designs whilst others reflect a correlational approach, all within the same study. The practicalities of survey research are examined in the next chapter.

Stay updated, free articles. Join our Telegram channel

Join