Skip to main content icon/video/no-internet
icon backReturn to Entries

Selection Bias

Selection bias, also known as sampling bias, usually refers to groups (e.g., experimental, control) that are systematically different prior to experimental manipulation or intervention due to the assignment of participants to groups. In other words, variations detected during a study are attributable to group differences due to selection bias or the independent variable (e.g., manipulated variable). Selection bias can occur during participant selection, assignment, and/or during the study. The bias that occurs during participant selection is generally identified as a threat to external validity, whereas bias that occurs during assignment is known as a threat to internal validity. During a study, if a significant number of participants withdraw without completing the study, selection bias can also occur. This entry examines the context in which selection bias may arise, how to avoid selection bias, and the limitations of ensuring selection bias.

Context in Which Selection Bias May Arise

Selection of a sample for a study leading to selection bias may occur when the researchers attempt to generalize their observations beyond the sample to other populations. Participants assigned based on cost, convenience, or to conditions in a manner intended to disperse participant characteristics (e.g., demographics and diagnosis) as opposed to an unbiased selection process may introduce selection bias. For example, researchers who are interested in evaluating a national school-based program may only solicit participants from their region due to cost considerations. The question then becomes whether the results obtained in the study can generalize to participants attending schools in other geographic regions.

Nonrandom assignment of participants in experimental designs (quasi-experimental) are likely to contribute to outcome differences that are not due to the intervention effect; but rather, certain characteristics of the groups being compared. This design is likely to occur especially when random assignment is unavailable to the researchers.

Even if there is a random assignment prior to the beginning of the study, participant attrition, especially in a nonrandom fashion, can end up with a selection bias problem. Participants may withdraw for various reasons that often times are unknown to the researchers. One should be wary when participants withdraw from a study.

How to Avoid Selection Bias

Random assignment of participants to groups is a commonly used procedure to guard against selection bias (as a threat to internal validity). Random assignment minimizes the likelihood that groups will be systematically different prior to introducing the independent variable. With random assignment and when sample size is sufficiently large, it is more likely to produce group equivalence before the independent variable is applied. Another method that may reduce selection bias is random assignment of participants in matched sets to ensure groups are equivalent based on key variables (e.g., age, income, gender, and geographic region). Pretesting participants may also be employed, which provides the opportunity for researchers to evaluate the presence, possible size, and direction of bias. However, even if no pretest differences are found, it does not guarantee the absence of selection bias.

Researchers are generally cautioned against overgeneralizing their conclusions in the face of promising results, even if they are confident in having sufficiently addressed the threat to internal validity, as doing so can still pose a threat to external validity. Ensuring all participants complete the entire study may be impossible without compromising the ethical treatment of participants. Researchers can try to generate high interest and increase motivation of participants to complete the study, while keeping in mind that there is nothing to stop a participant from withdrawing once the study begins. Researchers can employ blind or double-blind designs and provide detailed debriefing after the study. Blinded designs can aid in establishing strong baseline measures and reduce dropouts, but researchers should be wary of potential ethical issues.

...

  • Loading...
locked icon

Sign in to access this content

Get a 30 day FREE TRIAL

  • Watch videos from a variety of sources bringing classroom topics to life
  • Read modern, diverse business cases
  • Explore hundreds of books and reference titles

Read next

More like this

Sage Recommends

We found other relevant content for you on other Sage platforms.

close

Have you created a personal profile? Login or create a profile so that you can save clips, playlists and searches

close
close
close Signed In to profile You are signed in as:
close
Logged in Institution Institution