Bias
By Rebecca J. Klemm and Donna F. Stroup
Edited By Eric Y. Drogin, J.D., Ph.D.
Bias
Statisticians can separate a pattern from other random forces through different techniques including a consideration of bias.
"Bias" is:
A particular tendency or inclination, especially one that prevents unprejudiced consideration of a question; prejudice; or A systematic as opposed to random distortion of a statistic as a result of sampling procedure.
Statisticians define different kinds of bias on the basis of the following considerations:
Sample selection: Are certain elements of the population given preference? Data collection: How are the data being collected or the questions asked? Were any interviewees nonresponsive or did any interviewer error occur? Population: What are the demographics of the population being studied? Statistical analysis: Are the assumptions satisfied in practice?
Confounding
A particular type of bias of interest in legal proceedings is confounding. Consider the simple example of an investigation to determine whether or not having gray hair is related to mortality risk. Two facts are clear from almost any data source:
Persons with gray hair have a higher death rate than others. Those with gray hair are often older than others.
Because of fact 2, the interpretation of fact 1 is unclear. The possible association of gray hair with mortality risk is entangled with the effect of age on mortality risk. This is confounding (i.e., a situation in which a variable, age, is related to the disease or condition under study and to the risk factor, gray hair, being investigated).
Statisticians can apply different techniques to adjust for confounding.
These include adjustment (i.e., removing the effect of the confounding variable through statistical manipulation), stratification (i.e., analyzing the effect of the risk factor on the outcome for different values of the confounding variable), and more complex model-based methods.
Sponsoring
Entity:
The ABA Section of Science and Technology Law
|
