Dictionary of terms used in Advertising Principles
There are 61 entries in this glossary.
Correlations give an idea of how closely two variables vary, going from -1 (perfect agreement in the negative in the negative direction: if x is higher, y is lower) to +1 (perfect agreement in the positive in the negative direction: if x is higher, y is higher). It is widely confused with causality (it is not). A pattern of correlations may help to support a statement about causality, but the lack of a correlation does not mean that variables are not causally related. For example, it might occur because one variable did not vary, or because the relationship was masked by the variations in another variable. It is not an easy measure to understand because it is related to various properties of the data. For example, data on individuals tend to have smaller correlations (below 0.1) than for group data, and time series can have very high correlations (above .9) even for spurious relationships (such as two variable that rise over time). Unfortunately, it is difficult to infer importance or effect sizes from correlations. In addition, it is easy for statisticians to obtain higher correlations by using various “tricks.” Armstrong (1970) showed how a correlation of .92 was obtained by correlating two series of random numbers. I try to refrain from reporting correlations; however, in some cases, it is the only measure provided by the researchers, and can be used as a very rough measure of relative importance.