Severe Tests, Arguing from Error, and Methodological Underdetermination

A major problem that has been thought to stand in the way of an adequate account of hypothesis appraisal may be termed the alternative hypothesis objection: that whatever rule is specified for positively appraising H, there will always be rival hypotheses that satisfy the rule equally well. Evidence in accordance with hypothesis H cannot really count in favor of H, it is objected, if it counts equally well for some (perhaps infinitely many) other hypotheses that would also accord with H. This problem is a version of the general problem of underdetermination of hypotheses by data: if data cannot univocally pick out hypothesis H over alternatives, then the hypothesis is underdetermined by the data. Some have considered this problem so intractable as to render any attempt to erect a methodology of hypothesis appraisal impossible. No such conclusion is warranted, however. There is no general argument showing that all rules of appraisal are subject to this objection. At most the argument has been sustained against certain specific rules (e.g., the straight rule, simple hypothetico-deductivism, falsificationist accounts). A more adequate account of hypothesis testing, I will argue, gets around the underdetermination challenge. In this account, evidence is to be taken as a good test of (or good grounds for) a hypothesis only to the extent that it can be seen as the result of passing a severe test of that hypothesis. My task in this paper is first to sketch an account of severe tests, and second, to show how it answers the alternative hypothesis objection. To anticipate two of my main theses, I shall be arguing: