Bland–Altman Plot Generator

Figran

Paste two sets of paired measurements to get a Bland–Altman plot with bias and limits of agreement.

  • Bias and 95% limits of agreement
  • Confidence intervals included
  • Free SVG and 300 DPI PNG

Data

Row
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17

Plot options

200300400500600700−150−100−500501001501234567891011121314151617Upper LoA 73.9Bias −2.1Lower LoA −78.1Mean of Wright and Mini Wright (l/min)Mean of Wright and Mini Wright (l/min)Wright − Mini Wright (l/min)Wright − Mini Wright (l/min)

Results

StatisticEstimate95% CI
n17—
Bias−2.1−22.0 to 17.8
SD of differences38.8—
Lower LoA−78.1−112.9 to −43.3
Upper LoA73.939.1 to 108.6
Outside LoA1 (5.9%)—

Bland–Altman plot examples

200300400500600700−150−100−50050100150Bias −2.1 · LoA −78.1 to 73.9Mean of Wright and Mini Wright (l/min)Mean of Wright and Mini Wright (l/min)Wright − Mini Wright (l/min)Wright − Mini Wright (l/min)

Peak flow meters

First readings from Wright and mini Wright peak flow meters.

Bland & Altman 1986, Table 1

50100150200250−150−100−50050Bias −16.3 · LoA −54.7 to 22.1Mean of Observer J and Machine S (mmHg)Mean of Observer J and Machine S (mmHg)Observer J − Machine S (mmHg)Observer J − Machine S (mmHg)

Blood pressure: observer vs machine

First systolic blood pressure readings for 85 people.

Bland & Altman 1999

406080100120140051015Bias 9.88 · LoA 5.64 to 14.12Mean of Nadler and HurleyMean of Nadler and HurleyNadler − Hurley (%)Nadler − Hurley (%)

Percentage difference

Plasma volume measurements by Nadler and Hurley.

Bland & Altman 1999

Illustrative
050100150−100102030Bias 14.2 · LoA 2.8 to 25.5Mean of Method A and Method BMean of Method A and Method BMethod A − Method BMethod A − Method B

Proportional bias

Differences increase with measurement size.

Illustrative
50100150200250300−10−5051015Bias 3.0 · LoA −5.2 to 11.2Laboratory (mg/dL)Laboratory (mg/dL)Meter − Laboratory (mg/dL)Meter − Laboratory (mg/dL)

Reference method on the x-axis

A portable glucose meter compared with a laboratory method.

Illustrative
406080100120140160−10−50510Bias 0.9 · LoA −3.5 to 5.3Acceptance +5.0 / −5.0Mean of Wearable and ECG (bpm)Mean of Wearable and ECG (bpm)Wearable − ECG (bpm)Wearable − ECG (bpm)

Acceptance limits

Compare wearable heart-rate readings with ECG.

How to do a Bland–Altman analysis

  1. Method A　 Method B
    494　　　 512
    395　　　 430
    516　　　 520

    1Paste paired measurements

    Paste a table or edit the values directly.

  2. Y axis
    Difference　⌄

    X axis
    Mean of both　⌄

    2Choose axes and direction

    Set the difference direction and horizontal axis.

  3. Bias　　　 −2.1
    Lower LoA　 −78.1
    Upper LoA　 73.9

    3Read the results

    Compare the bias and limits with your prespecified criteria.

  4. Download PNGSVGExport ▾

    4Download the figure

    Save the plot or copy the results and figure legend.

How to interpret a Bland–Altman plot

Use the bias and 95% limits of agreement to read the center, spread and pattern of paired differences.

200300400500600700−150−100−50050100150Upper LoA 73.9Bias −2.1Lower LoA −78.1Mean of Wright and Mini Wright (l/min)Mean of Wright and Mini Wright (l/min)Wright − Mini Wright (l/min)Wright − Mini Wright (l/min)123456
200300400500600700−150−100−50050100150Bias −2.1 · LoA −78.1 to 73.9Mean of Wright and Mini Wright (l/min)Mean of Wright and Mini Wright (l/min)Wright − Mini Wright (l/min)Wright − Mini Wright (l/min)123456
  1. 1Bias is the average difference.
  2. 2The upper limit is bias + 1.96 SD.
  3. 3The lower limit is bias − 1.96 SD.
  4. 4Shaded bands show confidence intervals.
  5. 5Points beyond the limits are counted in the results.
  6. 6Inspect the pattern for changes with measurement size.

Do the methods agree?

Judge agreement against a clinically or practically acceptable difference set before analysis, not from the plot alone. Limits inside that threshold support agreement; limits outside it do not.

406080100120140160−6−4−20246Mean (bpm)Difference (bpm)5.3 > 5.0

The upper limit 5.3 bpm exceeds +5.0 bpm; agreement is not shown at ±5.0 bpm.

  • Bias 0.9 bpm
  • Limits of agreement −3.5 to 5.3 bpm
  • Acceptable difference ±5.0 bpm

The acceptance limits are illustrative.

Does the pattern break the limits?

Constant limits assume independent pairs, roughly normal differences and a stable spread. Check your plot for these patterns:

  • Bias
  • Limits of agreement
  • Regression line

Stable differences

Limits hold
50100150200250300−10−5051015MeanMeanDifferenceDifference

The spread stays steady; keep Y axis set to Difference.

Proportional bias

Limits misleading
0501001500102030MeanMeanDifferenceDifference

Differences change with size; turn on Proportional bias regression.

Widening spread

Limits misleading
40608010012014005101520MeanMeanDifferenceDifference

Spread grows with size; set Y axis to Percentage difference.

How do I report it?

Report the number of pairs, bias and its 95% confidence interval, lower and upper limits of agreement and their 95% confidence intervals, and the prespecified acceptance threshold with its rationale.

How the values are calculated

Let d be the mean difference and s the sample standard deviation of differences.

di = Ai − Bi
95% limits of agreement = d ± 1.96s
Bias 95% CI = d ± t0.975, n−1 · s/√n
Limit 95% CI ≈ limit ± t0.975, n−1 · s√(1/n + 1.962 / [2(n − 1)])

Example methods sentence

Use Copy results in the editor to get this sentence for your data.

Mean difference −2.1 l/min (95% CI −22.0 to 17.8); 95% limits of agreement −78.1 to 73.9 l/min (n = 17). Lower limit 95% CI −112.9 to −43.3; upper limit 95% CI 39.1 to 108.6 l/min, using the Bland and Altman (1999) approximation.

References: Altman & Bland 1983 · Bland & Altman 1986 · Bland & Altman 1995 · Bland & Altman 1999 · Bland & Altman 2007 · Krouwer 2008 · Carkeet 2015 · Giavarina 2015

Method comparison: agreement is not correlation

Methods can rise together and still disagree. Correlation describes association; the differences and limits show how far paired measurements can be apart.

Measurements against the reference
100100200200300300400400500500600600700700WrightMini Wright
Differences against the mean
200300400500600700−150−100−50050100150Bias −2.1 · LoA −78.1 to 73.9Mean of Wright and Mini Wright (l/min)Mean of Wright and Mini Wright (l/min)Wright − Mini Wright (l/min)Wright − Mini Wright (l/min)

Bland–Altman plot FAQ

What is a Bland–Altman plot?

It plots each pair’s difference against its mean, showing bias and limits of agreement.

How do I interpret the limits of agreement?

For approximately normally distributed differences, about 95% of differences are expected to lie between the limits. Decide what difference is acceptable beforehand using clinical or practical criteria.

Why not just use correlation?

Correlation describes linear association, not agreement. Two methods can correlate strongly while differing substantially. ICC measures a different property and is not calculated here.

Should the x-axis be the mean or the reference method?

The mean is the default. If one method is a reference, plotting against it is an alternative; Bland and Altman and Krouwer discuss the tradeoffs.

Why 1.96 SD and not 2 SD?

The 1986 paper used 2 SD; the 1999 approach uses 1.96 SD. The difference is small.

What if differences grow with measurement size?

Inspect the regression slope and consider percentage differences. This tool does not calculate log ratios or regression-based limits.

Is this the same as a Tukey mean-difference plot?

Yes. Both display paired differences against their means.

Can I make a Bland–Altman plot in Excel? Do results match R, SPSS or Prism?

You can calculate differences, means and limits in Excel. Confidence interval formulas vary across software; this tool uses the 1999 approximation shown above.

Is my data uploaded? What can I download?

Measurements stay in your browser. Download SVG, PNG or CSV, or copy the results and figure legend.

Compare paired measurements

Compare paired measurements and export your figure.