Human subjects research is essential across all disciplines. As it involves direct or indirect interaction between the researcher and participant, the researchers are tasked to gather as much information as possible while conducting their studies. However, it is important that all materials, activities, and questions are free of bias throughout the research process. Bias is having partiality, typically for a desired result. Bias is not always conscious nor controllable, and it can significantly influence data and may prevent researchers from finding the results they seek.
Bias in Written Protocols and Procedures
Language can be biased. Though assessments may cater to a particular audience, some researchers seek to involve participants of different backgrounds and cultures to create a generalized population. Of course, it is difficult to make a new procedure for each person, so it is best to aim for a neutral common ground. If there is a specific age or gender group a researcher wishes to study, this population must be included when writing an IRB protocol. In the study itself, it is best to address participants as “individuals'' or “persons.” Being aware that labels may cause different effects and reactions are essential to note as well. For example, some studies use the term, “minorities” to represent non-White racial and ethnic persons. However, using “minorities” may come with a negative connotation as “minority” typically refers to being less than or powerless over the “majority.” Though the meaning of the term may be understood by the participants based on the context of its usage, it is generally safer to use “people of color” or “ethnic minority groups” to address these populations.
For more information, please visit: General Principles for Reducing Bias.
Response Biases and Demand Characteristics
In in-person studies, bias effects can also influence participants to act differently than they would naturally. Most participants are aware that the researcher’s role is to collect data in order to answer a research question and some may be perceptive that their behaviors, feelings, and answers are being observed and analyzed. As a result, the study may be subject to the experimenter effect, in which participants react to researchers’ physical characteristics and behaviors rather than the stimuli presented in the experiment. Martin T. Orne pioneered the concept of demand characteristics, in which participants will try to be a “good” subject. Usually, this happens when the participant believes they know the hypothesis of the experiment. This may involve the participant trying to impress the researcher, help the researcher, or obey the researcher in situations when they normally would not. Ultimately, these demand characteristics may also cause the participants to act unnaturally, thus skewing the results of the study.
Non-Response Bias in Online Surveys
Online data collection is a popular method that allows the research to distribute questionnaires to many participants. Qualtrics, Amazon Mechanical Turk (MTurk), REDCap are a few of many study distribution platforms. Though convenient, there are factors which researchers may have difficulty controlling. When distributing studies online, non-response bias is stronger compared to in-person paper surveys. When participants are not directly interacting with the researcher, they may skip questions or leave a survey unfinished. Data may be skewed if only a small number of participants from the sample complete the surveys. As a result, this may affect the generalizability of the research.
How to Reduce Research Bias
Bias is part of human nature and thus, it cannot be completely eliminated. However, there are tactics that expert researchers use to reduce bias. Thankfully, there are many accessible sources that help novice researchers with conducting their research successfully. When formulating research protocols and questions, researchers can reduce the risk of bias by using inclusive language while conducting their studies. As previously stated, terms like “minorities” to describe ethnic groups can generate negative feelings and cause an increased risk for bias. Thus, using neutral and appropriate language allows the study to be understood by participants of all backgrounds. For example, there are many different gender identities. In studies that involve these topics, researchers must be aware of their meanings and differences. Asking participants to answer demographics about their sex may limit a person to “male,” “female,” or “intersex” options. Asking for their gender and pronouns through open-ended questions allows participants to answer comfortably with their own terms.
Bias can also be reduced in in-person conducted studies. Orne (2009) suggests the following methods to reduce the probability of demand characteristics:
- If there is intentional deceit in the study, debrief the participant after the study is completed;
- Double-blind the study to prevent the experimenter from giving cues to participants; or
- Eliminate the experimenter entirely and distribute the study in written/online form.
Even online studies may encounter non-response bias. Effective ways to reduce this bias include:
- Arrange a time for the researcher to share the study with the participant to take online;
- Provide appropriate incentives or compensation for participants to complete the study;
- Ensure the study meets best practices for accessibility; and
- Use short versions of survey instruments when appropriate.
The world of conducting research continues to expand every day. Scientists are using Artificial Intelligence (AI) to collect large amounts of data. As data is a valuable source of information, AI uses it to create technologies in hopes to solve current world problems. However, these models are taking data full of racial, gender, and cultural bias. As a result, scientists are developing inclusive data sets to reduce and hopefully remove bias. Hopefully, there will be more information on AI data collection tools that may be revealed in the future. Many types of biases exist in research, but thankfully, there are many accessible sources that help student researchers conduct their research successfully. TC IRB is also here to assist with research compliance questions and developing protocols. We hope that by understanding these biases, researchers can conduct more effective studies and become experts in their fields.
References
American Psychological Association. (2021). Inclusive language guidelines.
https://www.apa.org/about/apa/equity-diversity-inclusion/language-guidelines.pdf.
American Psychological Association. (2022, March). Bias-free language.
https://apastyle.apa.org/style-grammar-guidelines/bias-free-language.
American Psychological Association. (2022, July). General principles for reducing bias.
https://apastyle.apa.org/style-grammar-guidelines/bias-free-language/general-principles.
Orne, M. T. (2009). Demand characteristics and the concept of quasi-controls. In Rosnow R. L.
& Rosenthal R. (Eds.), Artifacts in Behavioral Research: Robert Rosenthal and Ralph L.
Rosnow's Classic Books, A Re-issue of Artifact in Behavioral Research, Experimenter Effects in Behavioral Research and the Volunteer Subject (pp. 110-137). Oxford University Press.
Simonite, Tom (2023, February 8). The WIRED Guide to Artificial Intelligence. WIRED.