Bland-Altman Analysis: A Practical Guide to Assessing Agreement

When you need to know whether a new measurement method can replace an established one, correlation alone will not answer the question. Bland-Altman analysis compares two quantitative measurement methods by examining the differences between paired observations, estimating the mean bias and the limits of agreement within which 95% of differences fall. This guide explains how to perform, interpret, and report Bland-Altman analysis for method comparison studies in clinical, laboratory, and research settings.

The Bland-Altman approach was proposed in 1983 as an alternative to correlation and regression for assessing comparability between measurement methods. Correlation studies the relationship between one variable and another, not the differences, and is not recommended for assessing comparability between methods. The Bland-Altman plot analysis evaluates bias between the mean differences and estimates an agreement interval within which 95% of the differences of the second method compared to the first one fall. Data can be analyzed both as unit differences plots and as percentage differences plots. The method only defines the intervals of agreements, it does not say whether those limits are acceptable or not. Acceptable limits must be defined a priori based on clinical necessity, biological considerations, or other goals.

At a Glance

Why Correlation Is Not Agreement

Many method comparison studies have historically relied on correlation coefficients and regression analyses. These approaches answer a different question. Correlation examines the relationship between one variable and another, not the differences between paired measurements. Two methods can be highly correlated yet disagree substantially in their actual values. A new method that consistently reads 20% higher than the reference method will show perfect correlation with it, but the two methods do not agree.

The Bland-Altman analysis was developed specifically to address this limitation. It quantifies agreement by studying the mean difference between methods and constructing limits of agreement. This approach is now widely recommended in clinical laboratory practice and other fields where method comparison is required. Current guidelines recommend using Bland-Altman plots, also called difference plots, as part of method comparison evaluation in the veterinary clinical pathology laboratory. Analysis of differences can meaningfully augment linear regression techniques and allows fuller summarization of the performance of two methods relative to each other.

The validation of a new measurement method for application to medical practice requires comparison with gold standard techniques. The Bland-Altman analysis is a frequently applied technique in studies that investigate the agreement between two methods of the same medical measurement. Potential areas of usage span clinical viewpoints, and possible pitfalls in study designs merit statistical consideration.

Core Principles of Bland-Altman Analysis

The Mean Difference and Bias

The first step in Bland-Altman analysis is calculating the difference between each pair of measurements. For each subject or sample, subtract the value from one method from the value obtained with the other method. The mean of all these differences represents the bias, which is the systematic tendency of one method to read higher or lower than the other.

A mean difference near zero suggests no systematic bias. A positive mean difference indicates that the first method tends to produce higher values than the second. A negative mean difference indicates the opposite. The bias is expressed in the same units as the measurements themselves, which makes interpretation straightforward for clinicians and laboratory staff.

Limits of Agreement

The limits of agreement define the range within which 95% of the differences between the two methods are expected to fall. These limits are calculated as the mean difference plus or minus 1.96 times the standard deviation of the differences. The choice of 1.96 corresponds to the 95% confidence level for normally distributed differences.

The limits of agreement provide a practical summary of how much the two methods can differ for an individual measurement. If the limits are narrow enough to be clinically acceptable, the methods can be used interchangeably. If the limits are wide, the methods do not agree sufficiently for clinical purposes, even if the mean bias is small.

The Bland-Altman Plot

The Bland-Altman plot displays the difference between the two methods on the vertical axis and the mean of the two methods on the horizontal axis. Each point represents one paired observation. The plot includes a horizontal line at the mean difference and additional lines at the upper and lower limits of agreement.

The plot serves several purposes. It reveals whether the differences are consistent across the range of measurement values. It shows outliers that may warrant investigation. It allows visual assessment of whether the assumptions of the analysis are met, particularly the assumption of homogeneous scatter.

Setting Acceptable Limits Before Analysis

The Bland-Altman method defines the intervals of agreement, but it does not determine whether those limits are acceptable. Acceptable limits must be defined a priori based on clinical necessity, biological considerations, or other goals. This is one of the most important steps in the entire analysis, and it must occur before data collection or analysis begins.

The process of setting acceptable limits requires input from clinicians, laboratory professionals, or other stakeholders who understand the intended use of the measurement. For a laboratory analyte, the acceptable difference might be based on the biological variation of the analyte or on the clinical decision thresholds that guide patient management. For a physical measurement such as joint angle assessment, the acceptable difference might be based on the precision needed for treatment decisions.

Broad consensus exists for the a priori establishment of acceptability benchmarks in Bland-Altman agreement analysis. Reporting standards also call for estimation of repeatability of measurements, description of the data structure, visual assessment of the normality and homogeneity assumption, and plotting and numerically reporting both bias and the Bland-Altman limits of agreement including respective 95% confidence intervals.

Step-by-Step Workflow for Bland-Altman Analysis

Step 1: Define the Research Question

State clearly whether the study aims to validate a new method against a reference standard or to compare two methods where neither is established. The interpretation of results differs depending on the context. When neither method represents a reference standard, the analysis evaluates agreement instead of validation.

Step 2: Set Acceptable Limits A Priori

Determine the maximum acceptable difference between methods before collecting data. Document the rationale for these limits. The limits should reflect clinical necessity, biological considerations, or analytical goals. Without predefined acceptable limits, the analysis cannot answer the question of whether the methods agree sufficiently.

Step 3: Collect Paired Measurements

Obtain measurements from both methods on the same subjects or samples under comparable conditions. The sample should span the full range of values expected in clinical or research use. Consider the repeatability of each method, because poor repeatability in either method will widen the limits of agreement.

Step 4: Plot the Data

Create a Bland-Altman plot with the difference between methods on the vertical axis and the mean of the two methods on the horizontal axis. Examine the plot for patterns. The differences should be approximately centered around the mean difference and should not show systematic changes as the mean increases.

Step 5: Check Assumptions

Assess whether the differences are approximately normally distributed and whether the scatter is homogeneous across the range of means. If the differences increase systematically with the mean, the limits of agreement will not be valid across the entire range. Options include log transformation or variance function approaches.

Step 6: Calculate Bias and Limits of Agreement

Compute the mean difference and the standard deviation of the differences. Calculate the limits of agreement as the mean difference plus and minus 1.96 times the standard deviation. Report these values with their 95% confidence intervals.

Step 7: Interpret Against Predefined Limits

Compare the calculated limits of agreement with the acceptable limits defined in Step 2. If the calculated limits fall within the acceptable range, the methods agree sufficiently for the intended purpose. If the calculated limits exceed the acceptable range, the methods do not agree sufficiently.

Step 8: Report Transparently

Report the bias, limits of agreement, confidence intervals, sample size, data structure, and any transformations applied. Describe the repeatability of the measurements. Provide the Bland-Altman plot so readers can assess the data visually.

Handling Nonconstant Differences

An important assumption of Bland-Altman analysis is that paired method differences exhibit approximately constant scatter when plotted against pair means. This allows estimation of limits of agreement that retain validity across the entire range of mean values. In practice, pair differences often increase systematically with the mean.

Bland and Altman used log transformed data to achieve approximately homogeneous scatter. However, a logarithmic transformation fails when data are located near the detection limit of an assay, a region that is often of considerable clinical importance. A variance function estimated from pair differences can be used to transform problematic data into a form suitable for traditional Bland-Altman analysis. Following transformation by variance function, Bland-Altman results can be readily interpreted by back-transformation either to the original measurement scale or as percentage values. Limits of agreement are no longer horizontal straight lines, but their shapes simply reflect error characteristics that are familiar to laboratory analysts.

When differences increase with the mean, an alternative is to express the differences as percentages of the mean. The percentage differences plot can be more appropriate when the magnitude of disagreement scales with the size of the measurement. The choice between unit differences and percentage differences should be made based on the data structure and the clinical context.

Proportional Bias Assessment

Proportional bias occurs when the difference between methods changes systematically as the magnitude of the measurement increases. This pattern can be detected by regressing the inter-device differences against the mean of the two measurements. A significant slope indicates proportional bias.

In a comparison of anterior segment parameters measured with two imaging systems, inter-device differences varied substantially across parameters. Anterior chamber depth and the anterior chamber depth to white-to-white ratio demonstrated minimal mean bias, narrow limits of agreement, and no proportional bias, indicating high inter-device stability and reliability. In contrast, anterior chamber angle and anterior chamber volume exhibited larger bias, wider limits of agreement, and significant proportional bias. White-to-white corneal diameter and pupil diameter showed intermediate stability with evidence of magnitude-dependent discrepancies. These findings highlight the importance of parameter-specific and device-specific interpretation.

When proportional bias is present, the limits of agreement are not constant across the measurement range. Reporting a single set of limits of agreement would be misleading. The analysis should describe how the bias and limits change with the magnitude of the measurement.

Practical Examples Across Fields

Clinical Laboratory Method Comparison

In a contemporary clinical laboratory it is very common to have to assess the agreement between two quantitative methods of measurement. The correct statistical approach to assess this degree of agreement is not obvious, and correlation and regression studies are frequently proposed. However, correlation studies the relationship between one variable and another, not the differences, and it is not recommended as a method for assessing the comparability between methods.

A hematology analyzer validation study compared the Atellica HEMA 580 with the Sysmex XN-3100 following Clinical and Laboratory Standards Institute guidelines. Method comparison per EP09-A3 was supplemented by Bland-Altman analysis. Platelet method comparison demonstrated a modest negative proportional bias with a Deming slope of 0.95. High correlation coefficients across major hematology parameters did not preclude the need for Bland-Altman analysis to characterize the agreement.

Therapeutic Drug Monitoring

A study of minimally invasive therapeutic drug monitoring of immunosuppressants in children with kidney diseases evaluated whether fingerstick capillary sampling combined with liquid chromatography-tandem mass spectrometry provides results analytically comparable to those of conventional venous sampling. Seventy-four paired samples from 21 patients were analyzed. Strong correlations were observed between capillary and venous samples for mycophenolic acid, tacrolimus, and cyclosporine A. Hematocrit correction improved agreement for mycophenolic acid. Bland-Altman analyses demonstrated acceptable bias across analytes.

Physical Therapy Assessment

A study assessed the agreement between video-based expert visual estimation of shoulder abduction and three-dimensional motion capture as a reference standard. A total of 923 paired observations from 33 experts were analyzed. Visual estimation showed a consistent tendency toward overestimation with a mean difference of 14.7 degrees and limits of agreement ranging from -8.0 to 37.5 degrees. A cross-classified mixed-effects model estimated a mean bias of 14.5 degrees, confirming systematic overestimation. Most estimates exceeded the reference value. Variability was substantial with 27.3% of variance attributable to differences between video stimuli and 10.5% to differences between raters.

Surgical Blood Loss Estimation

A retrospective registry-based study included 828 patients undergoing elective primary total hip arthroplasty. Agreement between perioperative hemoglobin drop and hidden blood loss calculated using the Gross-Sehat method was assessed using Bland-Altman analysis and intraclass correlation coefficients. Agreement was limited with wide limits of agreement on Bland-Altman analysis. Although a positive association was observed, substantial interindividual variability persisted, particularly at higher blood loss levels. Hemoglobin drop demonstrated only moderate discriminative performance for identifying patients with high hidden blood loss.

Dietary Assessment Validation

A study assessed the relative and construct validity of the 14-point Mediterranean Diet Adherence Screener used in the Prevencion con Dieta Mediterranea study. A validated food frequency questionnaire and the screener were administered to 7146 participants. Using Bland-Altman analysis, the average screener Mediterranean diet score estimate was 105% of the food frequency questionnaire score estimate. Limits of agreement ranged between 57 and 153%. The screener was judged a valid instrument for rapid estimation of adherence to the Mediterranean diet.

Radiographic Measurement Comparison

A study evaluated six commonly used approaches for detection of radiographic angles in hallux valgus deformity patients. Postoperative radiographs were analyzed using six methods for drawing reference lines. The measurement results were subjected to Bland-Altman analysis and consistency evaluation. No statistically significant differences were found in the measurement values among the six methods.

Physics and Dosimetry Applications

The Bland-Altman approach has been applied beyond clinical medicine. A comparative study calculated Bethe-Bloch results with Bichsel-Sternheimer values for density-dependent energy loss of protons in lead and beryllium targets, employing the Bland-Altman analysis within 95% limit of agreement. The calculated mass-stopping power in view of the normalized percent difference was consistent with the Bichsel-Sternheimer results for the same projectile of higher energy in both targets. Results deviated from the Bichsel-Sternheimer results for high atomic number materials for the same projectile of lower energy.

The question of whether Bland-Altman analysis has a role in assessing radiation dosimeter performance relative to an established standard has been raised in the literature. The method can be applied whenever two quantitative measurement approaches need to be compared.

Records and Measurements

What to Record

For each Bland-Altman analysis, maintain a complete record of the following items. The sample size and the number of paired observations. The measurement units and the range of values covered. The mean difference and its standard deviation. The limits of agreement with their 95% confidence intervals. The predefined acceptable limits and the rationale for choosing them. The repeatability of each method if this was assessed. Any transformations applied to the data. The software and version used for the analysis.

Data Structure Documentation

Reporting standards for Bland-Altman agreement analysis call for description of the data structure. This includes whether the data are cross-sectional or longitudinal, whether multiple observations were obtained from the same subject, and whether the measurements were obtained under identical conditions. The data structure affects the validity of the analysis and the interpretation of the results.

Repeatability Assessment

Estimation of repeatability of measurements is a recommended reporting item for Bland-Altman analysis. If either method has poor repeatability, the limits of agreement will be wide even when the methods agree on average. Assessing repeatability separately for each method helps interpret the limits of agreement and identify the source of disagreement.

Common Failure Patterns

Using Correlation Instead of Agreement

The most common failure in method comparison studies is relying on correlation coefficients to claim agreement. High correlation does not imply agreement. Two methods can be perfectly correlated while differing by a large and clinically important amount. The Bland-Altman analysis was developed specifically to address this problem.

Failing to Set Acceptable Limits A Priori

The Bland-Altman method only defines the intervals of agreements, it does not say whether those limits are acceptable or not. Acceptable limits must be defined a priori based on clinical necessity, biological considerations, or other goals. Without predefined limits, researchers may rationalize any observed limits as acceptable after seeing the results.

Ignoring Nonconstant Differences

When pair differences increase systematically with the mean, the limits of agreement are not valid across the entire range of measurement values. Log transformation or variance function approaches can address this problem. Ignoring the pattern produces limits that are too narrow at high values and too wide at low values.

Inadequate Sample Size

The precision of the limits of agreement depends on the sample size. With small samples, the confidence intervals around the limits of agreement are wide, making it difficult to judge whether the methods agree sufficiently. The sample should be large enough to provide stable estimates of the mean difference and its standard deviation.

Poor Repeatability in Either Method

If either method has poor repeatability, the limits of agreement will be wide regardless of how well the methods agree on average. Repeatability should be assessed separately for each method before interpreting the limits of agreement.

Overinterpreting the Mean Bias

A small mean bias does not guarantee good agreement. The limits of agreement can be wide even when the mean difference is near zero. Both the bias and the limits of agreement must be considered when judging whether methods agree sufficiently.

Limitations of Bland-Altman Analysis

The Bland-Altman method defines the intervals of agreements but does not determine whether those limits are acceptable. The judgment of acceptability requires clinical or analytical context that the statistical method itself cannot provide.

The method assumes that the differences are approximately normally distributed. When this assumption is violated, the limits of agreement may not capture 95% of differences as intended. Transformations can address some violations but not all.

The method assumes homogeneous scatter of differences across the range of means. When differences increase with the mean, the limits of agreement are not constant. Log transformation or variance function approaches can address this, but each has limitations.

The method does not account for the repeatability of the individual methods. Poor repeatability in either method widens the limits of agreement. The limits of agreement reflect the combined effects of bias, random error in both methods, and any interaction between the methods.

The method does not identify which method is more accurate when neither method is a reference standard. It describes the agreement between the methods but cannot establish which method produces values closer to the true value.

Safety and Regulatory Context

Method comparison studies are often conducted as part of analytical validation required before a new measurement method is introduced into clinical or laboratory practice. The Clinical and Laboratory Standards Institute provides guidelines for method comparison, including EP09-A3. These guidelines describe the recommended statistical approaches and reporting requirements.

The EQUATOR Network provides reporting guidelines for health research, including studies that assess measurement agreement. Following these guidelines improves the transparency and completeness of reporting.

The National Institute of Standards and Technology supports research data frameworks that promote rigorous data management and analysis practices. The NC3Rs Experimental Design Assistant helps researchers design experiments that are statistically sound and minimize the use of animals.

When a method comparison study is part of a diagnostic accuracy study, clinical trial, or epidemiological survey, the Bland-Altman analysis is often limited to brief descriptions in the main report. Reporting standards have been developed to improve the quality of these descriptions. Seven proposals were identified from a MEDLINE and PubMed search, three of which were derived by reviewing anesthesia journals. Broad consensus was seen for the a priori establishment of acceptability benchmarks, estimation of repeatability of measurements, description of the data structure, visual assessment of the normality and homogeneity assumption, and plotting and numerically reporting both bias and the Bland-Altman limits of agreement including respective 95% confidence intervals.

Professional Escalation Criteria

Consult a statistician or methodologist when any of the following conditions apply. The data structure is complex, such as repeated measurements from the same subjects or clustered data. The differences show nonconstant scatter that does not respond to log transformation. The measurement range includes values near the detection limit of an assay. The study involves multiple raters or observers. The analysis will be used to support regulatory approval or clinical implementation decisions.

Consult a clinician or subject matter expert when setting acceptable limits of agreement. The acceptable limits must reflect clinical necessity, biological considerations, or other goals that require domain expertise. A statistician can calculate the limits of agreement but cannot determine whether those limits are clinically acceptable.

Escalate to a journal editor or regulatory body when reporting standards are unclear or when the study findings will inform clinical practice. Transparent reporting of Bland-Altman analysis requires specific items that may not be familiar to all reviewers.

Software Options

Several software options are available for Bland-Altman analysis. Statistical packages such as R, Stata, and SPSS can perform the calculations and generate the plots. An R function called BA.plot has been developed specifically for Bland-Altman analysis because statistical software packages do not have menu-driven operation for this method.

A computer program is available that performs the necessary calculations for variance function approaches to Bland-Altman analysis. This program handles the transformation of problematic data into a form suitable for traditional analysis.

For researchers interested in Bayesian approaches, a tutorial provides guidance on Bayesian Bland-Altman analysis. One suggested approach addresses the objective of Bland-Altman analysis via the posterior predictive distribution. This allows estimation of the probability of an acceptable degree of disagreement, fixed a priori, for the difference between two future measurements. An interface applet is provided with a guideline to ease mathematical and computational complexity.

Frequently Asked Questions

What is the difference between correlation and agreement?

Correlation measures the strength of association between two variables. Agreement measures how close the values from two methods are to each other. Two methods can be perfectly correlated yet disagree substantially. Bland-Altman analysis was developed specifically to assess agreement by studying the mean difference and constructing limits of agreement.

How do I calculate the limits of agreement?

Calculate the mean difference between the paired measurements and the standard deviation of the differences. The limits of agreement are the mean difference plus and minus 1.96 times the standard deviation. These limits define the range within which 95% of the differences are expected to fall.

What does a small mean bias tell me?

A small mean bias indicates that the two methods produce similar average values. It does not indicate good agreement for individual measurements. The limits of agreement can be wide even when the mean bias is near zero. Both the bias and the limits of agreement must be considered together.

When should I use percentage differences instead of unit differences?

Percentage differences are appropriate when the magnitude of disagreement scales with the size of the measurement. If the differences increase systematically as the mean increases, expressing the differences as percentages of the mean can produce more homogeneous scatter. The choice should be based on the data structure and the clinical context.

What should I do if the differences increase with the mean?

Log transformation is the traditional approach to achieve approximately homogeneous scatter. However, log transformation fails when data are located near the detection limit of an assay. A variance function estimated from pair differences can be used to transform problematic data into a form suitable for traditional Bland-Altman analysis.

How many samples do I need for a Bland-Altman analysis?

The sample size should be large enough to provide stable estimates of the mean difference and its standard deviation. With small samples, the confidence intervals around the limits of agreement are wide. The required sample size depends on the expected variability of the differences and the desired precision of the limits of agreement.

Can Bland-Altman analysis tell me which method is more accurate?

No. Bland-Altman analysis describes the agreement between two methods but cannot establish which method produces values closer to the true value. When neither method is a reference standard, the analysis evaluates agreement instead of validation. Determining accuracy requires comparison with a reference standard.

What should I report when publishing a Bland-Altman analysis?

Report the bias, limits of agreement, and their 95% confidence intervals. Describe the data structure and the repeatability of the measurements. State the predefined acceptable limits and the rationale for choosing them. Provide the Bland-Altman plot so readers can assess the data visually. Follow the reporting standards recommended for Bland-Altman agreement analysis.

Related Articles

• Statistical Analysis Methods
• Statistical Analysis Methods
• Statistical Analysis Methods
• Statistical Data Analysis
• Statistical Data Analysis

References and Further Reading

• Research Data Framework. National Institute of Standards and Technology.
• EQUATOR Network. EQUATOR Network.
• Experimental Design Assistant. NC3Rs.
• NCBI Literature Resources. National Center for Biotechnology Information.
• PubMed. National Library of Medicine.
• Understanding Bland Altman analysis.. Biochemia medica, 2015.
• Bland-Altman analysis: A paradigm to understand correlation and agreement.. Turkish journal of emergency medicine, 2018.
• Reporting Standards for a Bland-Altman Agreement Analysis: A Review of Methodological Reviews.. Diagnostics (Basel, Switzerland), 2020.
• A review of Bland-Altman difference plot analysis in the veterinary clinical pathology laboratory.. Veterinary clinical pathology, 2024.
• A Tutorial of Bland Altman Analysis in A Bayesian Framework.. Measurement in physical education and exercise science, 2021.
• A short screener is valid for assessing Mediterranean diet adherence among older Spanish men and women.. The Journal of nutrition, 2011.
• Using the variance function to generalize Bland-Altman analysis.. Annals of clinical biochemistry, 2019.
• Bland-Altman Analysis of Different Radiographic Measurements of the Hallux Valgus Angle and the Intermetatarsal Angle After Distal Osteotomy.. Orthopaedic surgery, 2020.
• Parameter-Dependent Agreement and Proportional Bias Between a Domestically Developed Scheimpflug-Based Imaging System and Pentacam HR.. 2026.
• Is full automation in ESR measurement clinically reliable? Revisiting a classic test.. 2026.
• Accuracy of visual estimation of shoulder abduction by expert physiotherapists: a comparative study using 3D motion capture.. 2026.
• Analytical Verification and Method Comparison of the Atellica HEMA 580 Hematology Analyzer for Routine Clinical Laboratory Implementation in a Tertiary Care Setting.. 2026.
• Hidden blood loss estimation using perioperative hemoglobin drop: agreement with the Sehat formula.. 2026.
• Minimally Invasive Therapeutic Drug Monitoring of Immunosuppressants in Children with Kidney Diseases: Validation of Fingerstick Sampling Using LC-MS/MS.. 2026.
• Utilization of third-trimester fetal transcerebellar diameter measurement for gestational age estimation: a comparative study using Bland-Altman analysis. AJOG Global Reports, 2024.
• BA.plot: An R function for Bland-Altman analysis. 2021.
• Density-dependent Energy Loss of Protons in Pb and Be Targets and Percent Mass-Stopping Power from Bethe-Bloch Formula and Bichsel-Sternheimer Data Within 1-12 MeV Energy Range: A Comparative Study Based on Bland-Altman Analysis.. Journal of Medical Imaging and Radiation Sciences, 2019.
• Use and reporting of Bland-Altman analyses in studies of self-reported versus measured weight and height. International Journal of Obesity, 2020.
• Comparison of two methods of measurements: The Bland-Altman analysis. Annales Francaises D Anesthesie Et De Reanimation, 2000.
• The Bland-Altman analysis: Does it have a role in assessing radiation dosimeter performance relative to an established standard?. Radiation Measurements, 2010.

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