Let us consider the following experimental scenario (while the context is an experiment the application is to any empirical study).
Authors run an experiment in a topic area. They manipulate two variables (A and B) and measure a third (C). Theoretically all three variables make sense. They use a sample of participants. They find interesting results and get them published in journal %.
Under the pressure to publish, the authors run another study in a topic area. They manipulate three variables (A, D and E) with one (A) being the same as one of the variables in the previous experiment and measure the same third variable (C). The combination of the previous variable with the two new variables makes sense theoretically. Further, the two new variables are not conceptually related to the variable not used in this study (B). They use the same experimental materials as in the first study modified for the differences in two of the three variables manipulated. They also use as a contextual variable in the new experiment one of the two manipulations of the variable (B’) from the previous experiment that is NOT manipulated in the current study. They use a new sample of participants that are similar in characteristics to the previous study but based on the descriptive statistics they are clearly a different sample. They submit this study to another journal $ and the study is published.
Neither study refers to the other.
When you set it up abstractly like this it seems obvious to me that these are two studies that share some features.
1. Same basic experimental instrument
2. Similar subject populations drawn from but not the same participants
3. One variable (A) in common across the two studies that is manipulated.
4. A measured variable (C) that is measured in both experiments
5. A context factor (B’) in the second experiment that is a one of two treatment variables in the first experiment.
What is different?
1. Two variables (D and E) manipulated in the second experiment
2. Interactions between variable A and (D, E)
3. Interactions between D and E.
There appear to be to be all sorts of advantages of this research design. Partial replication of results across experiments. Partial proof that the first convenience sample was not atypical. Common experimental instrument so that if replication did not succeed it would be due to not replicating rather than to instrument changes.
The next blog entry considers what the issues are when you move from the abstract to the concrete.