**4. Conclusions**

The first major section considered NHST, *p*-values without NHST, and CIs. All were found wanting. Consequently, the second major section focused on a new way to think: the APP. The APP differs from the others because all the inferential work is performed before data collection. This is not to say that the others involve no work, whatsoever, before data collection. The setting of threshold levels, for instance, is work that is done before data collection when one uses NHST. But there nevertheless is a strong difference. With traditional procedures, once the data have been collected, it is still necessary to calculate *p*-values or CIs. In contrast, using the APP, the only inferential work that needs to be done after data collection is to acknowledge the results of the descriptive work. The researcher can be assured that the descriptive statistics have acceptable probabilities of being acceptably close to corresponding population parameters. After all, it is the researcher who decided what constitutes an acceptable probability and an acceptable degree of closeness and collected the requisite sample size to meet specifications. And if reviewers or editors believe that the investigator was too liberal in setting specifications, they have the option to reject the manuscript or insist that the researcher augmen<sup>t</sup> the sample. For example, if the researcher uses *f* = 0.4 and the editor favors *f* = 0.1, it is transparent how to calculate how much the researcher needs to augmen<sup>t</sup> the sample to reach the editor's specification.

A reasonable person might agree that the APP is a good thing; but also argue that NHST, *p*-values without NHST, or CIs are good too. As was stated earlier, there is nothing about APP calculations performed before data collection that renders impossible the calculation of *p*-values or CIs after data collection. Thus, it is possible to use all the procedures for the same study. This possibility need not be inconvenient for setting the APP apart from other procedures. If researchers were to routinely use the APP, they also would become accustomed to APP thinking. In turn, this would result in their eventually perceiving just how barren *p*-values and CIs are if one wishes to advance science. This is not to say that scientists should not test hypotheses. They should. But they should not depend on automatized decision-makers such as *p*-values and CIs to do it. Instead, researchers should perform much of their thinking up front and make a priori specifications for closeness and confidence. Then they should take their descriptive results seriously; with such seriousness being warranted by a priori specifications of acceptable closeness at acceptable probabilities. Of course, there are other factors that also influence the trust researchers place in descriptive statistics, such as the worth of the theory, the validity of the auxiliary assumptions, and so on. Whether or how much to believe substantive hypotheses, or the larger theories from which they are derived, is a process that cannot be automated. There will always remain an important role for expert judgment. The APP recognizes this explicitly.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The author declare no conflict of interest.
