Research Bias – Definition, Types & Examples

Example: Selection Bias

Selection bias in research is a study that examines the effects of a new medication on a particular disease but only includes patients willing and able to participate in the study. This may result in a biased sample not representative of the broader population of patients with the disease, and the study findings may not be generalisable to the larger population.

For instance, let’s say that the study only includes patients with mild or moderate symptoms of the disease but excludes those with severe symptoms or co-morbid conditions that would make them ineligible or unwilling to participate. This could lead to overestimating the medication’s effectiveness, as the study only includes patients more likely to respond positively to the treatment.

Alternatively, let’s say that the study only includes patients from a single geographic area or medical centre and does not include patients from other regions or medical centres. This could lead to a biased sample not representative of the diversity of patients with the disease, and the study findings may not be generalisable to other populations.

To control selection bias, researchers should use appropriate sampling methods that ensure a representative sample of the population and should strive to recruit a diverse range of participants who reflect the diversity of the population under study.

Example: Measurement Bias

Measurement bias in research is a study that measures the effectiveness of a new treatment for depression using self-reported measures of mood and symptoms but does not use any objective measures or clinical assessments.

For example, if the study relies solely on self-reported questionnaires completed by the participants, there may be a risk of bias due to social desirability or recall bias. Participants may be inclined to provide responses they think the researchers want to hear or may have difficulty recalling their symptoms accurately.

Alternatively, suppose the study uses observer-rated measures of mood and symptoms. In that case, there may be a risk of bias if the observers are not blinded to the treatment group or if their ratings are influenced by their expectations or preconceptions.

To tackle measurement bias, researchers should use validated and reliable measures of the outcome variables and ensure that the assessors are blinded to the treatment group or intervention. They should also consider using multiple measures or assessments to triangulate the results and increase the validity and reliability of the study findings.

Example: Reporting Bias

An example of reporting bias in research is a study that finds no significant difference between two treatments for a particular condition. Still, the authors only report the positive findings and do not mention the non-significant results.

For example, a study compares two different surgical techniques for a particular condition and finds no significant difference in outcomes between the two groups. However, the authors only report the positive findings, such as the fact that both techniques effectively reduced pain and improved function, and do not mention that there was no statistically significant difference between the two groups.

This reporting bias can lead to a distorted view of the study findings. It can create the impression that one treatment is superior to the other, even though the evidence does not support this conclusion.

To deal with reporting bias, researchers should report all of the study results, including both positive and negative findings. They should avoid selectively reporting only the findings that support their preconceptions or hypotheses. They should also follow established reporting guidelines, such as the Consolidated Standards of Reporting Trials (CONSORT) statement, to ensure that their study is reported transparently and completely.

Example: Publication Bias

Publication bias in research is when studies that report positive or statistically significant results are more likely to be published than those that report negative or non-significant results.

Let’s consider that a pharmaceutical company conducts multiple clinical trials of a new drug for a particular condition. Suppose some trials find positive results and some negative or non-significant results. In that case, the company may only choose to publish the positive results, not the negative or non-significant ones.

This publication bias can lead to a distorted view of the overall evidence, as it creates a situation where the published studies only represent a subset of the total studies that have been conducted. This can lead to an overestimation of the drug’s effectiveness and can lead to inappropriate or unsafe clinical decisions.

You can control publication bias by striving to publish all of their study results, regardless of whether they are positive, negative, or non-significant. Journals and publishers can also reduce publication bias by adopting policies and practices that encourage the publication of all types of studies and discourage selective reporting or publication.

Example: Confounding Bias

An example would be a study examining the relationship between coffee consumption and the risk of heart disease while ignoring other factors influencing the results, such as age, smoking status, or physical activity level.

Let’s say the study finds that people who drink more coffee have a higher risk of heart disease; this could be due to a confounding variable such as age, as older people may be more likely to drink more coffee and also more likely to have heart disease. The study findings could be biased and misleading if the researchers did not control for age in their analysis.

Researchers should identify potential confounding variables in advance and control for them in their study design and analysis. This can be done through stratification, regression analysis, or matching techniques. Researchers should also use appropriate statistical methods to assess the potential impact of confounding variables on the study results and report any limitations or caveats related to the study findings.

Example: Recall Bias

One example of recall bias in research is a study that asks participants to recall their past behaviours or experiences, such as their diet or physical activity level, but the participants may not accurately remember or report their behaviours due to memory bias.

Let’s consider an example of a study that asks participants to recall their dietary habits over the past year, including how often they ate certain foods. If some participants have difficulty remembering their past behaviours or overestimating or underestimating their consumption, this could introduce recall bias into the study results.

This recall bias can lead to an inaccurate or incomplete picture of the true relationship between diet and health outcomes and can compromise the validity of the study findings.

To minimise recall bias, researchers can use different strategies such as conducting a validation study to assess the accuracy of participants’ recall, using objective measures or biomarkers to corroborate the self-reported data, and reducing the recall period to a shorter time frame that participants are more likely to remember accurately.

Example: Observer Bias

An example of observer bias in research is a study that uses a subjective measure, such as a rating scale or a questionnaire. The observer’s expectations or preconceptions influence their interpretation of the data.

For example, let’s say that a study examines the effectiveness of a new therapy for children with autism. The study involves an observer who rates the children’s behaviour using a rating scale before and after the therapy. However, the observer may have preconceptions about the therapy, and their expectations could influence their interpretation of the children’s behaviour, leading to observer bias.

This observer bias can lead to an overestimation or underestimation of the true effect of the therapy and can compromise the validity of the study findings.

To overcome observer bias, researchers can use different strategies such as blinding the observer to the treatment group (i.e., not revealing which children received the therapy), using objective measures whenever possible, and using multiple observers to ensure inter-rater reliability.

Example: Sampling Bias

An example of a sampling bias would be a study evaluating the prevalence of a certain disease in a particular region, but the study only recruits participants from hospitals or clinics and does not include people who may have the disease but have not sought medical care. This sampling bias can lead to an overestimation or underestimation of the true prevalence of the disease in the population, and the study findings may not be generalisable to the broader population.

Another example of sampling bias could be a political survey conducted only on social media, which excludes people who are not active on social media, leading to a biased sample.

Researchers should use appropriate sampling methods to ensure that the sample is representative of the target population and that all eligible individuals have an equal chance of being included in the study in order to minimise sampling bias. This can be done through random, stratified, or other probability-based sampling methods. Researchers should also report any limitations or potential biases related to the study sample in their study findings.

Example: Performance Bias

Let’s talk about performance bias in a study that compares the effectiveness of two treatments, where the participants or researchers are not blinded to the treatment assignment, leading to biased or ungeneralisable results.

For example, let’s say that a study compares the effectiveness of a new medication versus a placebo for treating depression. However, the participants or researchers may know which treatment the participant receives, which can lead to performance bias. The participants who receive the new medication may expect to feel better and therefore report feeling better, regardless of the actual effectiveness of the medication.

This performance bias can lead to an overestimation of the true effect of the treatment and can compromise the validity of the study findings.

Researchers can control performance bias by using different strategies ,such as blinding the participants and researchers to the treatment assignment, using placebo treatments, or using objective measures to assess the outcomes. Blinding can be single-blind (where the participants are unaware of their treatment assignment) or double-blind (where the participants and researchers are unaware of the treatment assignment).

Example: Researcher Bias

An example of researcher bias in research is when the researcher’s expectations, beliefs, or opinions influence the study design, data collection, analysis, or interpretation of the study findings, leading to biased or ungeneralisable results.

A researcher may strongly believe that a certain treatment is effective, leading them to overemphasise the positive results and downplay the study’s negative results. Alternatively, a researcher may have a personal or financial interest in particular findings of the study, which may influence their interpretation of the data.

Researchers can control the bias by employing strategies such as blinding themselves to the treatment assignment, using objective measures, using standardised protocols, and involving multiple researchers to ensure inter-rater reliability. Additionally, researchers can disclose any potential conflicts of interest or biases in their study findings to ensure transparency and accountability.

Example: Response Bias

For example, if participants in a study about drug use are afraid of being judged or punished for their behaviour, they may under-report or deny drug use altogether. Alternatively, participants may over-report or exaggerate their drug use to appear more socially desirable. This response bias can lead to inaccurate estimates of drug use prevalence or patterns in the study population.

Another example of response bias could be a study on voting behaviour where participants may want to keep their true political affiliation private or may overstate their voting intention to please the researcher.

Researchers can control response bias by applying techniques such as anonymous surveys, using validated questionnaires, framing questions objectively and non-judgmental way, and emphasising the importance of honest and accurate responses. Additionally, researchers can collect data on non-response and attempt to adjust for it in their analysis to ensure that their results represent the target population.

Example: Information Bias

In a study on alcohol consumption and heart disease, information bias could occur if participants do not accurately report their alcohol intake or if the study relies on self-reported alcohol consumption rather than objective measures such as biomarkers. This could lead to inaccurate estimates of the relationship between alcohol consumption and heart disease.

Another example could be a study on the effectiveness of a medical intervention that relies on medical records to determine outcomes. If the medical records are incomplete or inaccurate, the study results could be biased or misleading.

You can deal with information bias by using validated measures, minimising missing data, training data collectors, conducting pilot studies to refine measurement protocols, and using objective measures when possible. Researchers can also conduct sensitivity analyses to assess the impact of measurement error on their study results.

Example: Cognitive Bias

Cognitive bias refers to systematic errors or deviations from rational thinking and judgment due to research limitations in human cognition. A specific example of cognitive bias is confirmation bias, which is the tendency to search for, interpret, or remember information to confirm pre-existing beliefs or hypotheses while ignoring or rejecting information that contradicts them.

For example, if a researcher believes that a certain treatment is effective, they may unconsciously seek evidence that supports their belief while dismissing or ignoring evidence that contradicts it. This can lead to biased interpretations of the data and confirmation of pre-existing beliefs rather than objectively evaluating the evidence.

Similarly, confirmation bias can occur in everyday life when individuals seek information supporting their beliefs or opinions while rejecting information that challenges them. For instance, in a political debate, a person may selectively interpret or remember evidence supporting their political views while ignoring or dismissing evidence contradicting them.

Individuals can seek out diverse sources of information, actively consider alternative perspectives, and engage in critical thinking and self-reflection to overcome the problem of cognitive bias. In research, researchers can use strategies such as blinding themselves to the treatment assignment and using objective measures to assess outcomes to the impact of confirmation bias.

Example: Regression to the Mean (RTM)

Let’s consider a study that measures the academic performance of a group of students. The students are selected based on their low scores in a preliminary exam. In the subsequent exam, even if the students do not receive any intervention, some will improve their scores, and some will perform worse. However, the students who performed extremely poorly on the preliminary exam are likelier to have had measurement errors or other random factors that lowered their scores. On the subsequent exam, these same factors may not be present, and thus their scores are likely to improve, moving closer to the mean.

This phenomenon can also occur in natural weather patterns or biological processes. For example, a plant that has experienced extreme growth in a particular season may experience a decline in the subsequent season due to the influence of random fluctuations in environmental factors.

Regression to the mean can be a confounding factor in research studies, especially those that use extreme values as selection criteria. To minimise the impact of regression on the mean, researchers can use a control group, random assignment, or repeated measurements of the same variable to identify the true effects of the intervention or variable of interest.