Selection Bias


Terms in this set (...)

Selection Bias
People included in study are unrepresentative due to sampling or selection factors

-Non Participation; agree or decline to participate

-Attrition; Person stay in study until end differs from those who drop out during study

*Cohort study; Major source of bias
Measurement Bias
Flaws in process of gathering research information distort and bias the information
Hawthorne Effect*
Presence of investigators distorts responses and data collected. Modifies behavior of people being studied
i.e clinicians conducting studied or subjects being watched.
Observation Bias*
Observers prior knowledge of subjects treatment condition biases observations.

i.e African Americans more likely to have HTN. Clinicians know more African Americans in study and automatically assign number for Vascular dementia
Lead-Time Bias
Early detection interval on a screening test is misinterpreted as an increase in interval.

Decreases onset of diseases by picking up on risk factors and intervening.
Recall Bias
i.e Case control study, comprehensive recall and detail of past exposure vs. controls
Unanticipated variables obscure the true cause and effect relationship.

uncritical analyses show relationship btwn con founder and disease.

Solution; stratify data analyses on levels of co-founder

i.e Diabetes rate in obese/non obese in regards to cholesterol level. Given RR. By stratifying we see Obesity likely cause of DM II, not cholesterol.

i.e Smokers, drinkers, and cirrhosis. Stratify and look at high/low alcohol use and determine Alcohol (not smoking) is cause of cirrhosis.
Selection Bias; Inappropriate Sampling
Select non-representative people limit generalization of findings to larger population
Selection Bias; Berkson's Bias
Over matching control subjects with cases in a case control study
i.e smoker a risk for ulcer disease
Measurement Bias;
Asking leading questions, Hawthorne effect (i.e participants like the proctors)

Solution; include control or placebo group
Observation Bias;
Investigators know treatment assignments of subjects prior to.

Solution; Double Blinding (masking)
Experimenter Expectancy
Subjects "tipped off" to the study goals or desired outcomes

i.e "i'm president of ABC pharmaceuticals. This is going to be a great medication. You will get to try this new med for free"

-Pygmlian effect

Solution; Dbl Blinding (masking) and give clear, informed consent statement
Lead Time Bias (#1 for screen test)
Screening test allow early detection and diagnosis but early detection cant be claimed as an extension of survival time (live longer)

Screen test; Major source of bias
Solution; Compute true survival time; onset to death
Recall Bias;
Cases have better recall of disease, because they have experienced it.

Case control; Major source of bias
Solution; Multiple confirmatory information sources (ask others to give history)
Late Look Bias
Earliest and most sever cases are lost to follow up.
-i.e case die prior to interview/surveillance

New disease investigation; Major source of bias
Solution; Stratify cases by severity