A crucial part of research design is the ability to establish causality. In order to make a statement about causality, three conditions are necessary:
- the cause precedes the effect,
- the cause and effect covary, and
- there is no plausible alternative explanation for the covariation.
In order to make these inferences, Campbell and Stanley (1963) discuss that the Solomon four-group design is best suited to make statements of causality.
The true design 4 (pretest-posttest control design) can maintain internal validity. History, maturation, and testing are controlled. Regression is controlled in so far as mean differences are concerned. Selection is ruled out through randomization. It can tell whether mortality and instrumentation offer plausible rival hypotheses. The Solomon four-group design not only increases external validity by controlling the main effect of testing and interaction of testing and X, but it also replicates the effect of X. Among the pre-experimental designs, the one shot design is least suited for these inferences. This design totally lacks control and is of no scientific value. The process of comparison is required.
Correlation does not indicate causality. Correlation simply implies that the mean difference between groups show that the two groups are related on some attributes.
- Correlational data helps to disconfirm existence of causal relationships.
- Weak correlationional design: Consider the case in which two units of analysis (two groups for comparison) are being observed outside the laboratory setting. One group gets treated with X, the other doesn’t. However, these two groups differ on many other attributes other than the presence/absence of X. Each of these other attributes could create difference in Os and each therefore provides a plausible rival hypothesis that X had an effect.
- Stronger correlational design that could point to more causal relationships: You still have groups for comparison, but X is a naturally occurring event that varies with one group more so than it does with the others. The key is that X is not artificially implanted, but is naturally part of one group and not a part of the other. Example: heavy smoking and lung cancer.
A correlational study must meet the first, three of the kinds of validity (internal validity, statistical conclusion validity, and construct validity) (Mitchell, 1985). Then, based on Popper (1959, 1963)’s falsification orientation, causal hypotheses with correlation zero will be disconfirmed, otherwise the hypotheses survive and will be examined further by cross-validation or other methodologies.
Furthermore, in a correlational study, Campbell & Stanley (1963) suggest that if the correlation is zero, the hypothesis on causality can be disconfirmed, otherwise the hypothesis survives and the researcher can examine it further through the use of other settings.
The causal interpretation of a simple or a partial correlation depends upon both the presence of a compatible plausible causal hypothesis and the absence of plausible rival hypotheses to explain the correlation upon other grounds. Any third variable that could affect the signaling frequency of both pairs of drivers in a similar fashion becomes a plausible rival hypothesis.