can be quite valuable or even essential. For example, if you think that variables X1, X2, and X3 may be disrupting your readings of D as a function of A, then temporarily keep A constant and record variations of D, X1, X2, and X3. At this reconnaissance stage, these problem variables need not be controllable. If they are controllable, however, factorial design is a more powerful experimental technique: it allows us to both isolate and quantify the influence of these variables. A related approach is the method of residuals (Chapter 3): measure variations caused by the dominant variable, remove its estimated effects, then compare data residuals to second-order variables.
Studies of the causes of spread of the AIDS disease long ago established that most U.S. cases are attributable to homosexual or intravenous transmission. But does heterosexual transmission occur, and if it does, how abundant is it? One technique to examine these questions is clearly biased, yet it is apparently the best available. Any AIDS instance that could be either homosexually or intravenously transmitted is attributed to those origins rather than to heterosexual transmission, regardless of the relative abundance of heterosexual versus other encounters. Only cases in which homosexual or intravenous transmission are impossible are attributed to heterosexual transmission. Because (we think) heterosexual transmission is much less likely per encounter than are other forms of transmission, this accounting bias toward the dominant variables is considered to be acceptable [Hilts, 1992].
Problem: the Noisy Widgetometer
You need to measure some widgets on your new high-precision widgetometer. Before starting, however, you prudently run some standard samples and find that the precision and accuracy are far below what is advertised. In desperation, you connect the widgetometer to a chart recorder and let it run for 24 hours, obtaining the record in Figure 21. How do you interpret this record, and what techniques and experimental designs could you use to deal with the problem?
