Dictionary of Evidence-based Medicine.pdf
Dictionary of Evidence-based Medicine.pdf
Dictionary of Evidence-based Medicine.pdf
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
F<br />
Factor cost (see under Gross domestic product)<br />
Fail-safe number<br />
For a variety <strong>of</strong> reasons, notably the preference <strong>of</strong> authors to submit studies<br />
with positive results rather than negative results for publication, studies in<br />
the former group are more likely to be published. Therefore estimates <strong>of</strong><br />
effect <strong>based</strong> on the results <strong>of</strong> studies in the public domain may be biased.<br />
Therefore when undertaking meta-analyses, a critical assessment <strong>of</strong> the<br />
potential threats to the validity <strong>of</strong> the estimates <strong>of</strong> effect is necessary.<br />
Rosenthal suggested calculating the number <strong>of</strong> studies needed to nullify<br />
any effect observed (Rosenthal R (1979) The 'file drawer problem' and<br />
tolerance <strong>of</strong> null results. Psychological Bulletin. 86: 638-41) and Cooper<br />
suggested calling this number the 'fail-safe sample size' (Cooper H (1979)<br />
Statistically combining independent studies: a meta-analysis <strong>of</strong> sex differences<br />
in conformity research. Journal <strong>of</strong> Personality and Social Psychology. 37:<br />
131^46). This number, which is now generally referred to as the fail-safe<br />
number, is calculated as follows:<br />
Suppose that there are k published studies and the significance level for<br />
these studies are p v p v py-,p^ Postulate the null hypothesis that the mean<br />
effect is zero and that each p i is uniformly distributed on [0,1]. Let Z ; be the<br />
standard normal variate associated with each p { . Z { is then given by Z ; = O" 1<br />
(1 - p.) where O is the standard normal cumulative distribution function.<br />
Let S k = Zj + Z 2 + ... + Z n and z a be the critical value for a one-sided test at<br />
a significance level <strong>of</strong> a for a normally distributed mean effect. The failsafe<br />
N is then calculated by solving the following equation.<br />
Fallacy <strong>of</strong> composition<br />
The fallacy <strong>of</strong> composition is the error made by wrongly generalizing to<br />
groups the results which apply to individuals. In economic theory, two