Correlational design in research methodology

Correlational Design

Problems in Correlational Research Design

In an attempt to determine how school children develop an interest in science -- more specifically, what factors might work to influence scientific interest in children -- a group of researchers has created and validated a measure of actual scientific interest. Equipped with this means of measuring their dependent variable, the researchers must now go about the task of determining what independent variables should be included in their model, thus answering their central research question. To this end, their plan is to use a variety of instruments to measure as many variables as possible as quickly as they possibly can, with the assumption that each additional variable they measure increases the probability that they will find variables that indeed have an influence on scientific interest. Their reasoning and the research that would result from it are flawed in many ways, and significant improvements could be made.

First, the timeline is an issue. Interest, especially interest in academic subjects and especially among children of school age, is something that changes over time, and thus a static measure of scientific interest followed by a rapid (and again static) measure of other cognitive, psychological, and environmental variables would not really reveal what the researchers would want it to reveal, namely how scientific interest changes/is influenced by other variables. Examination would have to take place over time, with changes in the independent variables measured along with changes in scientific interest in order to determine a truly correlative relationship between any of the variables tested. The fundamental design of this research is flawed, then, however slightly, and a single valid measure of interest is not enough to create a strong baseline.

Beyond this, the assumption the researchers make that measuring as many variables as possible would increase their odds of finding strong correlations is completely backwards. The more independent variables the researchers identify the more difficult it will be to determine precisely which of these independent variables, if any, actually have an impact on the dependent variable of scientific interest. Even if measurements are taken more than once over a period of time so as to track changes and correlations between the independent and dependent variables in a more accurate and meaningful fashion than with a single measurement, there wouldn't be an easy way to determine the strength of any potential correlations due to the dilution of all variables by the presence of so many others. The most basic knowledge of statistical research would reveal the inadequacy of the proposed approach to the researchers, and would recommend that they select a handful of variables to test that are suggested by previous research in the area or other rational connections.