Internal Validity and Research Designs

An evaluation that establishes a causal relationship with great certainty is said to possess internal validity.

Theats to internal validity could be classified as follows:

1. Non-equivalence (or selection bias);
2. Attrition;
3. History;
4. Maturation;
5. Testing;
6. Instrumentation; and
7. Regression to the mean.

In order to eliminate these threats to internal validity, evaluation can adopt different designs, such as, experimental, quasi-experimental, and correlational designs.

Experimental design is quite similar to designs of scientific research in laboratory settings, which secure a control group against which effects occurred in treatment groups will be measured. This design is quite effective to eliminate those threats to internal validity. However, this often suffers from problems of practicality and ethical issues especially in assigning certain individuals to the control group. Also, depending on objectives of research, it may be difficult to avoid the issues of attrition or maturation.

Quasi-experimental designs will establish substitutes for actual control group (in this case it is often referred to as "comparison group.") with variety of methods.

Correlational designs will use statistical treatment to secure internal validity of data derived from treatment groups.

None of the designs mentioned above can completely eliminate threats to internal validity. Therefore, it is ultimately necessary to be explained by researchers about causal relationships between outcomes of data analysis and reality that they intend to find out.

One of the popular explanation is sensitivity analysis. This refers to analysis whether the final conclusion of research are sensitive to assumptions made by the researcher.