2000 HSS/PSA Program 1 - History of Science Society
2000 HSS/PSA Program 1 - History of Science Society
2000 HSS/PSA Program 1 - History of Science Society
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>PSA</strong> Abstracts<br />
reliable evidential conclusions on the basis <strong>of</strong> the outcome <strong>of</strong> their experiment.<br />
Sometimes such pre-trial planning fails, however. Experimenters then may try<br />
to use after-trial strategies for getting around the problems introduced by failure<br />
to adhere to pre-trial strictures. I discuss a general type <strong>of</strong> situation <strong>of</strong> this sort,<br />
in which failing to adequately predesignate the choice <strong>of</strong> test statistic leaves<br />
experimenters uncertain as to how they ought to characterize the experiment<br />
that they have done, and propose a method for potentially drawing reliable results<br />
from data even in the face <strong>of</strong> such uncertainty. In such a situation, the calculation<br />
<strong>of</strong> statistical significance becomes impossible to calculate reliably. Significance<br />
calculation requires that the experiment performed be specified, which in turn<br />
determines the reference class to be used in calculating the probability <strong>of</strong> the<br />
test result that obtained under the assumption <strong>of</strong> the null hypothesis. In the errorstatistical<br />
account <strong>of</strong> experimental reasoning, such significance calculations are<br />
central because <strong>of</strong> the close relationship between significance and the errorstatistical<br />
concept <strong>of</strong> severity. I propose an extension <strong>of</strong> error-statistical methods<br />
for assessing evidence for application to situations <strong>of</strong> uncertainty regarding the<br />
choice <strong>of</strong> reference class for calculating significance. The method uses<br />
calculations <strong>of</strong> apparent significance based on counterfactual choices <strong>of</strong> reference<br />
class. That is, experimenters using this method employ a single body <strong>of</strong> data,<br />
and ask what would have been the significance <strong>of</strong> the result that they would<br />
have obtained with that data, under the assumption <strong>of</strong> a variety <strong>of</strong> different test<br />
specifications. In this way, experimenters can determine how sensitive the<br />
apparent significance <strong>of</strong> the result obtained is to particular choices <strong>of</strong> test<br />
specification. I argue that this strategy can be used to determine whether the<br />
apparent severity with which the hypothesis in question was tested is an artifact<br />
<strong>of</strong> the failure <strong>of</strong> pre-trial planning. In particular, the method outlined may be<br />
appropriate to situations in which pre-trial precautions against biasing the test<br />
toward a particular result fail, but experimenters nevertheless wish to regain the<br />
ability to draw reliable conclusions from the data at hand. I illustrate this approach<br />
with an example from the search for the top quark conducted by the Collider<br />
Detector at Fermilab collaboration in the early 90s.<br />
242<br />
P.␣ Kyle Stanford University <strong>of</strong> California, Irvine<br />
Refusing the Devil’s Bargain: What Kind <strong>of</strong> Underdetermination Should We<br />
Take Seriously?<br />
Advocates have sought to prove that underdetermination obtains because all<br />
theories have empirical equivalents. However, algorithms for generating<br />
empirical equivalents simply exchange underdetermination for existing<br />
philosophical chestnuts, while the few convincing examples <strong>of</strong> empirical<br />
equivalents will not support the desired sweeping conclusions.<br />
Underdetermination does not, however, depend on empirical equivalents: our<br />
warrant for current theories is equally undermined by presently unconceived