Zubair Khalid

Virologist/Molecular Biologist | Veterinarian | Bioinformatician

Conventional & Molecular Virology • Vaccine Development • Computational Biology

Dr. Zubair Khalid is a veterinarian and virologist specializing in conventional and molecular virology, vaccine development, and computational biology. Dedicated to advancing animal health through innovative research and multi-omics approaches.

Dr. Zubair Khalid - Veterinarian, Virologist, and Vaccine Development Researcher specializing in Computational Biology, Multi-omics, Animal Health, and Infectious Disease Research

Category: Guides

Measurement Bias in Research: Sources, Types, and Prevention

Measurement bias occurs when the tools, procedures, or conditions used to collect data systematically distort the values recorded, leading to conclusions that do not reflect the true state of the exposure, outcome, or relationship under study. For researchers in the life sciences, agriculture, and clinical fields, understanding measurement bias is essential because it threatens the internal validity of a study, meaning the study may not actually measure what it set out to measure. This article explains the major types of measurement bias, their sources, how they differ from selection bias, and practical strategies for prevention, including a checklist you can adapt for your own research protocols.

What Is Measurement Bias and Why It Matters

Measurement bias, also called information bias or observation bias, arises when the method of gathering data produces systematic errors in the measurement of exposure, outcome, or both. Unlike random error, which tends to cancel out across a sample, measurement bias pushes results in a consistent direction, either exaggerating or obscuring the true association between variables. Readers of medical literature need to consider two types of validity, internal and external. Internal validity means that the study measured what it set out to, while external validity is the ability to generalise from the study to the reader's patients. With respect to internal validity, selection bias, information bias, and confounding are present to some degree in all observational research. Information bias results from incorrect determination of exposure, outcome, or both, and its effect depends on its type. If information is gathered differently for one group than for another, bias results. By contrast, non-differential misclassification tends to obscure real differences. These distinctions matter because the direction of the bias determines how you interpret your findings and whether you can trust the magnitude of an observed effect.

Measurement bias is not a minor technical concern. Incomplete reporting has been identified as a major source of avoidable waste in biomedical research, and essential information is often not provided in study reports, impeding the identification, critical appraisal, and replication of studies. The Standards for Reporting Diagnostic Accuracy (STARD) 2015 statement was developed to address this problem, presenting an updated list of 30 essential items that should be included in every report of a diagnostic accuracy study. This update incorporates recent evidence about sources of bias and variability in diagnostic accuracy and is intended to facilitate the use of STARD. If you cannot describe exactly how you measured your key variables, you cannot expect others to replicate your work or to judge whether bias may have influenced your conclusions.

Measurement Bias vs Selection Bias

A common point of confusion is the distinction between measurement bias and selection bias. Selection bias stems from an absence of comparability between groups being studied. It occurs when the participants or samples included in a study are not representative of the target population, or when the groups being compared differ systematically in ways unrelated to the exposure or outcome of interest. For example, if you recruit volunteers for a nutrition study and only health-conscious individuals agree to participate, your sample is selected in a way that may distort the relationship between diet and health outcomes.

Measurement bias, by contrast, occurs after participants are enrolled, during the data collection phase. It concerns how information is obtained, recorded, or interpreted. A study can have excellent selection procedures and still suffer from measurement bias if the weighing scale is miscalibrated, if the questionnaire leads respondents toward certain answers, or if laboratory assays drift over time. Both types of bias can operate simultaneously in the same study, and both threaten internal validity. The key distinction is that selection bias relates to who is studied, while measurement bias relates to how they are measured.

At a Glance: Measurement Bias Types and Prevention

The following table summarizes the major types of measurement bias, their typical sources, and the primary prevention strategies you can implement.

Bias Type Common Source Primary Prevention Strategy
Recall bias Participants remember past exposures or events inaccurately, often because of their current health status or outcome Use objective records or biomarkers when available, blind participants to study hypotheses where feasible, and keep recall periods short
Interviewer or observer bias Data collectors know the exposure or outcome status and unconsciously or consciously record or interpret responses differently Blind interviewers and outcome assessors to group assignment, use standardized scripts and protocols, and conduct periodic quality checks
Instrument or equipment bias Measuring devices are miscalibrated, drift over time, or have different accuracy across subgroups Calibrate instruments before and during data collection, use the same device for all participants, and document calibration dates
Measurement reactivity The act of measuring changes the behavior or responses of participants Use unobtrusive measures where possible, minimize the number of assessments, and account for reactivity in the analysis
Social desirability bias Participants give answers they believe are more acceptable instead of truthful Use validated instruments designed to reduce social desirability, assure confidentiality, and consider indirect questioning methods
Differential misclassification Errors in measurement occur differently between comparison groups Standardize measurement protocols across all groups, blind assessors, and use identical instruments and procedures

Sources of Measurement Bias in Research

Measurement bias can enter a study at many points, from the initial design of the measurement instrument to the final data entry. Understanding the sources helps you identify where your own protocols may be vulnerable.

Instrument and Equipment Error

The tools you use to measure variables can introduce bias if they are inaccurate, imprecise, or applied inconsistently. Exposure measurement error causes bias in estimates of association between an exposure and an outcome, and this bias can be substantial. A simple hand calculation, in conjunction with validation study data and a calibration equation, can be used to correct estimates for the bias caused by exposure measurement error. Correcting estimates of association for measurement error helps researchers appropriately assess effect size. For example, if a scale used to weigh animals is off by a fixed amount, every weight recorded will be systematically too high or too low, shifting the estimated relationship between weight and any outcome of interest.

Instrument bias can also vary across subgroups of the population being studied. Pulse oximeters overestimate arterial oxygen saturation in patients with darker skin tones, potentially increasing rates of occult hypoxemia. The emission of broad-linewidth light-emitting diodes has been proposed as a possible cause for this melanin-dependent bias. Simulations demonstrate that pulse oximeter calibration curves depend on melanin concentration when broad-linewidth LEDs are utilized, while calibration curves remain independent of melanin when employing monochromatic laser diodes. This example illustrates that a device can be accurate for one group and biased for another, which has direct implications for equity in research and clinical care.

Data Collector Effects

The people who collect data can introduce bias through their expectations, behavior, or interactions with participants. Interviewer bias occurs when data collectors know the exposure or outcome status and unconsciously or consciously record or interpret responses differently. Even subtle differences in tone of voice, phrasing of questions, or follow-up prompts can lead participants to respond differently. Standardized training, written protocols, and blinding of data collectors to group assignment are the primary defenses against this source of bias.

Observer bias is closely related and occurs in studies where outcomes are assessed subjectively. If an outcome assessor expects a particular result based on the treatment group, they may interpret ambiguous findings in a way that confirms their expectation. Blinding outcome assessors to treatment allocation is a core strategy in randomized controlled trials, and the revised Cochrane risk-of-bias tool for trials distinguishes five domains of bias that can affect the results of trials, including outcome measurement. Using the risk-of-bias framework helps researchers mitigate these sources of bias and ensure transparency in reporting so that users of research are aware of them.

Participant Factors

Participants themselves can be a source of measurement bias through inaccurate recall, intentional misreporting, or changes in behavior caused by being measured. Recall bias is particularly problematic in case-control studies where participants with a disease or outcome may search their memory more thoroughly for potential causes than healthy controls. The act of measurement can also affect the people being measured. Asking people to complete a questionnaire can result in changes in behaviour, and changes in measured behaviour and other outcomes due to this reactivity may introduce bias in otherwise well-conducted randomised controlled trials, yielding incorrect estimates of intervention effects. Measurement reactivity is not currently adequately considered in risk of bias frameworks, despite clear systematic review evidence that measurement can affect the people being measured.

Social desirability bias is another participant-related source. Respondents may underreport behaviors they consider undesirable, such as excessive alcohol consumption, or overreport behaviors they consider virtuous, such as physical activity. This type of bias is particularly challenging in nutrition and lifestyle research where self-report is the primary data collection method.

Study Design and Protocol Factors

The way a study is designed can create conditions for measurement bias. If different measurement procedures are used for different groups, differential misclassification can result. If the timing of measurements differs between groups, the comparison may be biased. If the measurement instrument is administered in different settings or under different conditions, the responses may not be comparable. The number of response options on a rating scale can also affect results. A review of 60 papers using a Likert scale found that only 10 percent of studies use a measurement scale with an even answer choice category, while 90 percent use an odd number of response choices, with the five-point scale being the most popular. The presence of response bias and central tendency bias can affect the validity and reliability of the use of the Likert scale instrument. Researchers should consider whether the number of response options on their scales may be directing respondents toward certain answers.

Types of Measurement Bias in Detail

Beyond the general sources described above, several specific types of measurement bias deserve attention because they appear frequently across research domains.

Recall Bias

Recall bias arises when participants do not accurately remember past events, exposures, or behaviors. This is a form of information bias that results from incorrect determination of exposure, outcome, or both. The accuracy of recall can be influenced by the length of the recall period, the salience of the event, and the participant's current state. In retrospective studies, participants with a particular outcome may think harder about potential causes and report more exposures than controls, creating a spurious association. Using objective records, such as medical charts, production logs, or laboratory results, can reduce reliance on memory. When self-report is unavoidable, keep the recall period as short as possible and use validated questionnaires with clear, specific questions.

Interviewer and Observer Bias

Interviewer bias occurs when the person collecting data influences the responses or the recording of responses. This can happen through leading questions, differential probing, or selective recording of information. Observer bias occurs when the person assessing an outcome interprets subjective data in a way that aligns with their expectations. Both are forms of information bias that result from incorrect determination of exposure, outcome, or both. Blinding is the most effective prevention strategy. If the interviewer or observer does not know the exposure or outcome status, they cannot consciously or unconsciously favor one group. Standardized protocols and periodic monitoring of data collectors also help maintain consistency.

Instrumentation Bias

Instrumentation bias refers to systematic errors introduced by the measuring device itself. This includes miscalibration, drift over time, and differences in accuracy across subgroups. Staffing allocation methods in health services research induce substantial attenuation bias, and there are easily implemented estimation methods that overcome this bias. In this example, the method used to allocate nursing staff to acute care inpatient settings introduced measurement error that biased the estimated relationship between staffing levels and quality of care. The bias induced by the adjusted patient days method was smaller than for other methods, but the bias was still substantial, with the estimated coefficient for staffing level on quality of care expected to be one-third smaller than its true value. Instrumental variable estimation, using one staffing allocation measure as an instrument for another, addressed this bias, but only particular choices of staffing allocation measures and instruments were suitable. This example demonstrates that the choice of measurement method can have a large impact on the results, and that correction methods exist but require careful selection.

Measurement Reactivity

Measurement reactivity, also called the question-behaviour effect, occurs when the act of measuring changes the behavior or responses of participants. The usual methods of conduct and analysis of randomised controlled trials implicitly assume that the taking of measurements has no effect on research participants. Changes in measured behaviour and other outcomes due to measurement reactivity may therefore introduce bias in otherwise well-conducted randomised controlled trials, yielding incorrect estimates of intervention effects, including underestimates. The MERIT study was designed to promote awareness of how and where taking measurements can lead to bias and to provide recommendations on how best to avoid or minimise bias due to measurement reactivity in randomised controlled trials of interventions to improve health. If you are measuring a behavior that participants can consciously change, such as physical activity or dietary intake, consider whether the measurement itself may be altering the behavior you intend to study.

Differential vs Non-Differential Misclassification

Misclassification of exposure or outcome can be differential or non-differential. If information is gathered differently for one group than for another, bias results. This is differential misclassification, and it can bias results in either direction, either toward or away from the null. By contrast, non-differential misclassification tends to obscure real differences, biasing results toward the null. Non-differential misclassification occurs when errors in measurement are unrelated to the exposure or outcome status. For example, if a questionnaire about dietary intake is equally inaccurate for cases and controls, the resulting misclassification is non-differential and will tend to dilute the true association. Understanding which type of misclassification is present helps you predict the direction of bias and interpret your results appropriately.

Measurement Bias in Specific Research Contexts

Measurement bias manifests differently across research domains, and understanding these context-specific patterns can help you identify risks in your own work.

Diagnostic Accuracy Studies

Diagnostic accuracy studies are particularly vulnerable to measurement bias because they involve comparing a new test against a reference standard. The STARD 2015 statement was developed to improve the quality of reporting of diagnostic accuracy studies, presenting an updated list of 30 essential items that should be included in every report. This update incorporates recent evidence about sources of bias and variability in diagnostic accuracy. In these studies, bias can arise from the choice of reference standard, the interpretation of test results, and the handling of indeterminate results. A study of uterine measurements used a multirater reliability and diagnostic accuracy design, with the reference standard being the decision made most often by several independent experts. Fifteen representative experts examined anonymized images and provided their independent opinion, with the decision made most often for each case being considered the correct diagnosis. This approach illustrates how expert consensus can serve as a reference standard when no objective gold standard exists, and how interobserver reliability can be assessed using measures such as the concordance correlation coefficient.

Occupational and Environmental Epidemiology

Measuring exposures in occupational and environmental settings presents considerable challenges due to wide variations in intensity, frequency, and duration. Because exposure levels and conditions often vary by site, comparing results across studies is difficult. Key recommendations include reporting specific exposure levels within a study population and including quantitative bias assessment to address the impact of exposure measurement error on results within and across studies. These recommendations can improve the reporting quality and utility of occupational and environmental epidemiology, thereby strengthening its role in risk assessment. If you are studying farm workers exposed to pesticides or dust, for example, you need to consider how exposure levels vary across tasks, seasons, and individual work practices, and whether your measurement strategy captures this variability.

Precision Nutrition and Wearable Devices

The rise of wearable sensors and artificial intelligence in nutrition research has created new opportunities and new sources of measurement bias. Artificial intelligence can offer individualized dietary guidance based on multimodal data collected from various sources, including wearable sensors, high-dimensional multiomics and biomarker analyses, behavioral tracking, and self-reported dietary intake. However, the predictive power and fairness of these models rely on the quality of the data inputs, and measurement errors in any of these underlying data streams can introduce systematic bias, degrade model performance, and disproportionately affect underserved populations. Uncorrected measurement error can perpetuate demographic biases, compromise efforts toward personalized medicine, and exacerbate equity gaps when models are deployed in real-world settings. If you are using wearable devices to measure physical activity or heart rate, consider whether the device performs equally well across different skin tones, body sizes, and activity types.

Innovation and Patent Research

Measurement bias is not limited to clinical and biological research. A study of disruptive patents found that 88 percent of the decrease in the average CD index over 1980 to 2010 could be explained by the truncation of all backward patent citations before 1976, and that this truncation bias varied by technology class. The authors also accounted for a change in U.S. patent law that allows for citations to patent applications in addition to patent grants, and showed that the number of highly disruptive patents has increased since 1980, particularly since 2008. This example demonstrates that measurement decisions, such as the time window for counting citations, can fundamentally alter the conclusions of a study. The results suggest caution in using findings as a basis for research and decision-making in public policy, industry restructuring, or firm reorganization.

Internet Measurement Platforms

Even technical fields such as network measurement are subject to bias. Internet measurement platforms, such as RIPE Atlas, RIPE RIS, or RouteViews, are used for monitoring network performance, detecting routing events, topology discovery, or route optimization. To interpret the results of their measurements and avoid pitfalls or wrong generalizations, users must understand a platform's limitations. Bias exists due to the non-uniform deployment of the vantage points, and a generic framework can systematically and comprehensively quantify the multi-dimensional biases of these platforms, including across location, topology, and network types. This example illustrates that measurement bias can arise from the infrastructure used to collect data, beyond from the instruments or people involved.

Practical Workflow for Preventing Measurement Bias

Preventing measurement bias requires attention at every stage of the research process, from planning through data collection to analysis and reporting. The following workflow provides a structured approach.

Step 1: Define Your Measurement Protocol Before Data Collection

Write a detailed measurement protocol that specifies the instruments, procedures, timing, and personnel for each variable. Include calibration procedures, quality control checks, and rules for handling missing or unusual values. The Experimental Design Assistant from the NC3Rs is a free online tool that can help you design experiments and identify potential sources of bias before you begin. The Research Data Framework from the National Institute of Standards and Technology provides guidance on managing research data throughout its lifecycle, which can help you document your measurement procedures and maintain data quality.

Step 2: Validate Your Instruments

Before using a measurement instrument in your study, validate it against a reference standard or established method. If you are developing a new questionnaire, pilot test it with a small sample to identify ambiguous questions and assess reliability. If you are using a laboratory assay, run calibration samples and control samples to verify accuracy and precision. Validation study data and a calibration equation can be used to correct estimates for the bias caused by exposure measurement error, so collecting validation data as part of your study design is valuable even if you do not anticipate measurement error.

Step 3: Blind Data Collectors and Outcome Assessors

Whenever possible, ensure that the people collecting data and assessing outcomes do not know the exposure or outcome status of participants. Blinding prevents conscious and unconscious bias from influencing data collection and interpretation. In randomized controlled trials, blinding of outcome assessors is a key component of the risk-of-bias assessment. The revised Cochrane risk-of-bias tool for trials distinguishes five domains of bias, including outcome measurement, and provides a framework for recommendations to help researchers mitigate these sources of bias.

Step 4: Standardize Data Collection Procedures

Use identical procedures for all participants and all groups. This includes the same instruments, the same timing of measurements, the same environment, and the same instructions. If different data collectors are involved, train them together and assess inter-rater reliability. If measurements are taken at multiple sites or over an extended period, implement procedures to ensure consistency, such as periodic calibration checks and refresher training.

Step 5: Monitor Data Quality During Collection

Implement ongoing quality control procedures to detect measurement problems early. This may include running control samples with each batch of laboratory analyses, checking instrument calibration daily, reviewing data entry for errors, and computing inter-rater reliability statistics periodically. If you detect a problem, document it and take corrective action. The EQUATOR Network provides reporting guidelines for many study types, which can help you identify the information you need to report about your measurement procedures.

Step 6: Assess and Report Measurement Error in Your Analysis

Even with careful prevention, some measurement error is inevitable. Assess the potential impact of measurement error on your results and report this in your paper. Quantitative bias assessment can address the impact of exposure measurement error on results within and across studies. If you have validation data, you can use calibration equations to correct your estimates. If you do not have validation data, you can conduct sensitivity analyses to explore how different assumptions about measurement error would affect your conclusions.

Step 7: Report Your Measurement Methods Transparently

Transparent reporting of measurement methods is essential for readers to assess the risk of bias in your study. The STARD 2015 statement provides a checklist of 30 essential items that should be included in every report of a diagnostic accuracy study, and similar reporting guidelines exist for other study types through the EQUATOR Network. Include details about instrument calibration, data collector training, blinding procedures, and any quality control measures you implemented.

Records and Measurements to Maintain

Good record keeping is essential for preventing and detecting measurement bias. The following records should be maintained for any research study.

Record Type Purpose Frequency
Instrument calibration log Documents calibration dates, results, and any adjustments made to measuring devices Before each use and at regular intervals specified by the manufacturer
Data collector training records Documents who was trained, when, and on which procedures At study start and whenever new personnel join
Quality control sample results Tracks the performance of laboratory assays or other analytical methods over time With each batch of analyses
Blinding documentation Records who was blinded to what, and when blinding was broken if applicable At study start and whenever blinding status changes
Protocol deviation log Documents any deviations from the measurement protocol and the reasons for them As deviations occur
Data entry verification records Documents checks performed to verify the accuracy of data entry During and after data entry

Common Failure Patterns in Measurement Bias Prevention

Even experienced researchers make mistakes in preventing measurement bias. The following failure patterns are common and worth checking in your own protocols.

Failure to Calibrate Instruments

Researchers often assume that instruments are accurate because they were calibrated when purchased or at the start of the study. Instruments can drift over time, and calibration can be affected by environmental conditions such as temperature and humidity. A scale that is accurate at the start of a six-month study may be significantly off by the end. Regular calibration checks with documented results are essential.

Inadequate Blinding

Blinding is often described in the methods section of a paper but not actually implemented in practice. Data collectors may know the treatment assignment because of the nature of the intervention, or blinding may be broken inadvertently. If blinding is not feasible, the limitations should be acknowledged and the potential impact on results discussed.

Inconsistent Data Collection Procedures

Different data collectors may administer questionnaires or take measurements slightly differently, even with training. This can introduce systematic differences between groups if data collectors are not randomly assigned to participants. Periodic monitoring and inter-rater reliability checks can help identify and correct these inconsistencies.

Ignoring Measurement Reactivity

Researchers often assume that the act of measuring does not affect the participants. However, asking people to complete a questionnaire can result in changes in behaviour, and this measurement reactivity can introduce bias in otherwise well-conducted randomised controlled trials. If you are measuring a behavior that participants can consciously change, consider whether the measurement itself may be altering the behavior you intend to study.

Overreliance on Self-Report

Self-report is convenient and inexpensive, but it is vulnerable to recall bias and social desirability bias. Whenever possible, use objective measures to supplement or validate self-report data. If self-report is the only option, use validated instruments and consider whether the wording of questions may be directing respondents toward certain answers.

Failure to Assess Measurement Error in the Analysis

Many researchers collect data without considering whether measurement error is present, and then interpret their results as if the measurements were perfect. This can lead to overconfident conclusions, particularly when the true effect size is small. Quantitative bias assessment should be a routine part of data analysis, not an afterthought.

Limitations of Measurement Bias Prevention

Despite best efforts, measurement bias cannot be completely eliminated. Some limitations are inherent to the research context and should be acknowledged in your reports.

Blinding Is Not Always Feasible

In many studies, blinding is impossible. For example, in a study comparing a surgical intervention to a medical intervention, the outcome assessor may be able to tell which treatment the participant received. In such cases, the potential for observer bias should be acknowledged, and objective outcome measures should be used whenever possible.

Self-Report Is Sometimes the Only Option

For many exposures and outcomes, such as dietary intake, physical activity, or pain, self-report is the only practical measurement method. These measures are inherently subject to recall bias and social desirability bias. Researchers should use validated instruments, keep recall periods short, and consider using multiple measurement methods to triangulate the true value.

Instruments Have Inherent Limitations

Every measuring device has limitations in accuracy and precision. Even well-calibrated instruments have measurement error, and some instruments are more accurate for certain populations than for others. The skin-tone bias in pulse oximetry illustrates that a device can be accurate for one group and biased for another, and similar issues may exist for other devices.

Measurement Error Can Be Difficult to Correct

While methods exist to correct for measurement error, they require validation data and make assumptions about the nature of the error. Classical measurement error and the omission of a true interaction effect are not independent sources of bias, but become entangled such that their combined effects cannot be recovered by applying standard corrections for either problem alone. The residual contributions from measurement error and interaction omission can partially cancel each other out, so that prediction error reduces marginally while leaving coefficient bias large. This means that models with more biased estimates can exhibit lower estimator variance, making misspecified models difficult to distinguish by standard fit metrics, even in large samples.

Welfare and Safety Context

Measurement bias has direct implications for animal welfare and human safety in research settings. If measurements of animal weight, feed intake, or health status are biased, management decisions based on those measurements may be incorrect. For example, if a scale consistently overestimates animal weight, you may overfeed or underfeed animals, affecting their health and welfare. If a diagnostic test has differential accuracy across subgroups, some animals or patients may receive incorrect diagnoses, leading to inappropriate treatment or missed conditions.

In clinical research, measurement bias can have serious consequences for patient safety. The skin-tone bias in pulse oximetry can lead to occult hypoxemia being missed in patients with darker skin tones, potentially delaying life-saving treatment. Researchers and clinicians should be aware of the limitations of their measurement tools and consider whether the tools perform equally well across all populations.

Professional Escalation Criteria

Knowing when to escalate a measurement problem to a supervisor, statistician, or institutional review board is important for maintaining research quality. The following situations warrant escalation.

Instrument Failure or Persistent Calibration Problems

If an instrument cannot be calibrated to within acceptable limits, or if calibration drifts repeatedly, stop using the instrument and escalate the issue to your supervisor or the equipment manufacturer. Continuing to use a faulty instrument will compromise the validity of your data.

Evidence of Differential Measurement Between Groups

If you detect that measurements are being taken differently for one group than for another, escalate the issue immediately. Differential misclassification can bias results in either direction and is difficult to correct after data collection is complete.

Unexplained Patterns in Quality Control Data

If quality control samples show unexpected variation or trends over time, escalate the issue to the laboratory supervisor or statistician. This may indicate a problem with the assay, the reagents, or the equipment that needs to be investigated before more data are collected.

Concerns About Participant Safety

If you believe that measurement procedures are causing harm or distress to participants, stop the procedures and escalate the issue to the institutional review board or ethics committee. Participant welfare takes precedence over data collection.

Uncertainty About How to Handle Measurement Error in Analysis

If you are unsure how to assess or correct for measurement error in your analysis, consult a statistician or methodologist. The Research Data Framework from the National Institute of Standards and Technology and the Experimental Design Assistant from the NC3Rs can provide guidance on research data management and experimental design.

Frequently Asked Questions

What is the difference between measurement bias and selection bias?

Selection bias stems from an absence of comparability between groups being studied, meaning the participants or samples included in a study are not representative of the target population or the groups being compared differ systematically. Measurement bias, also called information bias, results from incorrect determination of exposure, outcome, or both, and occurs during data collection. Selection bias relates to who is studied, while measurement bias relates to how they are measured. Both threaten internal validity and can operate simultaneously in the same study.

How does recall bias affect case-control studies?

Recall bias occurs when participants do not accurately remember past events or exposures. In case-control studies, participants with a particular outcome may think harder about potential causes and report more exposures than controls, creating a spurious association. This is a form of information bias that results from incorrect determination of exposure, outcome, or both. Using objective records, keeping recall periods short, and using validated questionnaires can help reduce recall bias.

What is non-differential misclassification and how does it affect results?

Non-differential misclassification occurs when errors in measurement are unrelated to the exposure or outcome status. It tends to obscure real differences, biasing results toward the null. By contrast, differential misclassification occurs when information is gathered differently for one group than for another, and it can bias results in either direction. Understanding which type of misclassification is present helps you predict the direction of bias and interpret your results appropriately.

Can measurement error be corrected after data collection?

Yes, in some cases. Validation study data and a calibration equation can be used to correct estimates for the bias caused by exposure measurement error. However, correction methods require validation data and make assumptions about the nature of the error. Classical measurement error and the omission of a true interaction effect can become entangled such that their combined effects cannot be recovered by applying standard corrections for either problem alone. Consulting a statistician is recommended if you are considering correction methods.

How does the number of response options on a rating scale affect measurement bias?

The number of response options on a rating scale can affect results. A review of 60 papers using a Likert scale found that only 10 percent of studies use an even answer choice category, while 90 percent use an odd number of response choices, with the five-point scale being the most popular. The presence of response bias and central tendency bias can affect the validity and reliability of the use of the Likert scale instrument. A scale with an even number of responses may be more suitable if the researcher wants to direct respondents to one side.

What is measurement reactivity and how can it be minimized?

Measurement reactivity, also called the question-behaviour effect, occurs when the act of measuring changes the behavior or responses of participants. Asking people to complete a questionnaire can result in changes in behaviour, and this can introduce bias in otherwise well-conducted randomised controlled trials, yielding incorrect estimates of intervention effects. Measurement reactivity can be minimized by using unobtrusive measures where possible, minimizing the number of assessments, and accounting for reactivity in the analysis.

What reporting guidelines are available for measurement methods?

The EQUATOR Network provides reporting guidelines for many study types. The STARD 2015 statement presents an updated list of 30 essential items that should be included in every report of a diagnostic accuracy study, incorporating recent evidence about sources of bias and variability in diagnostic accuracy. The revised Cochrane risk-of-bias tool for trials distinguishes five domains of bias, including outcome measurement, and provides a framework for recommendations to help researchers mitigate these sources of bias and ensure transparency in reporting.

How can I assess whether my measurement tools perform equally across different populations?

Consider whether your measurement tools have been validated in the populations you are studying. The skin-tone bias in pulse oximetry illustrates that a device can be accurate for one group and biased for another. If you are using devices or assays that may perform differently across subgroups, consider conducting validation studies in each subgroup or using statistical methods to assess differential measurement error. The Research Data Framework from the National Institute of Standards and Technology can provide guidance on managing research data and documenting measurement procedures.

Related Articles

References and Further Reading

This article is educational and does not replace institutional policy, professional advice, or applicable safety and regulatory requirements.