The familiar 'before-and-after' type of experiment is modified so that the duration of the 'baseline' period, rather than being fixed before the experiment, is made contingent on the experimenter's prior estimate of the experimental impact and on the baseline data as they appear. At the conclusion of each day of the baseline period, a decision is made as to whether to terminate the baseline at that time, weighing the costs of extending the baseline by one day against the benefits of better estimating the experimental impact. An analytic framework is proposed for making this decision and for comparing the contingent design against an alternative having a baseline period of fixed duration. Equations, figures, and three references are supplied. (Author abstract modified)