Abstract:
|
The counternull was proposed in Rosenthal and Rubin (1994) as the value of an estimand that is supported by as much evidence in the data as the null hypothesis, which asserts that there is absolutely no effect. One advantage of the counternull approach is that the strength of evidence can be defined by the Fisherian p-value, as we will illustrate here using real data examples. Although it can require some thought to formulate, we believe that this effort is often rewarded with consequential insights because extraneous assumptions are avoided, especially in randomized experiments.
|