How Science Works (5): Study Bias

Study bias refers to a significant flaw in a study design or data collection that makes it more likely that the results will turn out a certain way. It is important to identify all possible sources of bias and address them in the interpretation of study results. There are two major types of study bias in epidemiology studies: selection and information.

Selection bias refers to the unintentional (or intentional) selection of study participants that would preferentially include or exclude certain results. For example, suppose a researcher was interested in determining whether income is associated with adverse health, and the researcher decided to recruit participants through primary care providers in private hospitals. Recruitment in private hospitals would yield a fairly homogenous study sample, without many low-income patients. The study may be biased from the start towards having a null result, if income’s affects on health are only evident at the lowest income levels.

Information bias refers to errors in measuring either the exposures or cases. In selection bias, suppose a researcher wants to determine whether income is associated with adverse health, and she decides to recruit participants through primary care providers in private hospitals. Recruitment in private hospitals would yield a fairly homogeneous study sample, without many low-income patients. The study may be biased from the start toward having a null result, if income's effects on health are only evident at the lowest income levels.

Suppose a researcher seeks to determine new risk factors for autism that occur during pregnancy. To measure various prenatal risk factors, she interviews mothers with and without autistic children. The mothers of autistic children would be motivated to find out what caused their child's condition and would attempt to recall every detail of their pregnancy. The mothers without autistic children might not be as motivated. Their motive for participating in the study might be economic; therefore, they might not think to mention certain of their behaviors during pregnancy. This could result in the mothers of the autistic children identifying what appear to be unique risk factors, even though the other mothers may also have been exposed to the same factors but failed to report them.


Note: This is an excerpt from my online course Epidemiology and Biostatistics for Medical Writers.