Calculate Mean & SD from P10, P50, P90
Estimate a normal distribution’s mean and standard deviation using the 10th, 50th, and 90th percentiles. This tool assumes an approximately symmetric, bell-shaped distribution.
How the estimator works
For a normal distribution, percentile positions are tied to z-scores. The 10th percentile sits at approximately -1.28155 SD below the mean, while the 90th percentile sits at +1.28155 SD above it.
- Mean estimate: usually the median (P50) if the data are symmetric.
- SD estimate: (P90 − P10) ÷ 2.5631
- Symmetry check: compare P50 with the midpoint of P10 and P90.
This is a practical shortcut for reporting and modeling when only summary percentiles are available.
How to calculate mean and standard deviation from P10, P50, and P90
When analysts ask how to calculate mean sd from p10 p50 p90, they are usually trying to recover familiar summary statistics from limited percentile data. This happens all the time in real-world reporting. A dashboard may publish only the 10th percentile, median, and 90th percentile. A paper may show quantiles but not the raw sample. A compensation benchmark, wait-time summary, biological measurement, or engineering tolerance report may provide spread information in percentile form rather than as a mean and standard deviation. In those situations, it is often helpful to estimate a normal distribution that matches the available quantiles as closely as possible.
The most important assumption is that the underlying distribution is approximately normal, or at least reasonably symmetric and bell-shaped in the region covered by the 10th through 90th percentiles. Under that assumption, P50 is a natural estimate of the mean, because the median and mean coincide in a perfectly normal distribution. The spacing between P10 and P90 then gives you a direct route to the standard deviation, because those percentiles correspond to fixed z-score positions.
The key formula
For a normal distribution:
- P10 = mean + z0.10 × SD, where z0.10 ≈ -1.28155
- P50 = mean, because the 50th percentile is the center of a symmetric normal distribution
- P90 = mean + z0.90 × SD, where z0.90 ≈ +1.28155
Because the z-scores for P10 and P90 are symmetric around zero, their total distance is:
P90 − P10 = (1.28155 − (-1.28155)) × SD = 2.5631 × SD
So the standard deviation estimate becomes:
SD ≈ (P90 − P10) ÷ 2.5631
And the mean estimate is usually:
Mean ≈ P50
Why P10, P50, and P90 are useful together
Using only one percentile tells you almost nothing about spread. Using only two percentiles can estimate scale, but not the center with much confidence unless the percentile pair is symmetric. The trio of P10, P50, and P90 is especially informative because it combines:
- A lower-tail location from P10
- A central tendency estimate from P50
- An upper-tail location from P90
This means you can perform a simple sanity check for symmetry. First, compute the midpoint of P10 and P90. If that midpoint is almost identical to P50, your estimated normal distribution is likely to be a good approximation. If P50 differs meaningfully from the midpoint, the data may be skewed. In practical settings, that does not always invalidate the estimate, but it does tell you to interpret the inferred mean and SD cautiously.
Step-by-step example
Suppose you are given these percentiles:
- P10 = 42
- P50 = 55
- P90 = 68
First, calculate the midpoint of P10 and P90:
(42 + 68) ÷ 2 = 55
That matches P50 exactly, which is an excellent sign of symmetry.
Next, estimate the standard deviation:
SD ≈ (68 − 42) ÷ 2.5631 = 26 ÷ 2.5631 ≈ 10.14
Then estimate the mean:
Mean ≈ P50 = 55
So the reconstructed normal distribution is approximately:
Normal(mean = 55, SD = 10.14)
| Input percentile | Value | Normal interpretation |
|---|---|---|
| P10 | 42 | About 1.28155 standard deviations below the mean |
| P50 | 55 | Center of the distribution; estimate of the mean under symmetry |
| P90 | 68 | About 1.28155 standard deviations above the mean |
| Estimated SD | 10.14 | Derived from the P10–P90 span divided by 2.5631 |
When this method works best
This percentile-to-parameter conversion is strongest when the underlying data behave like a normal distribution. That usually means:
- The distribution is roughly symmetric around the center
- Extreme skewness is absent
- The reported percentiles are not heavily rounded or truncated
- The percentiles come from a reasonably large and stable sample
Domains where this can be useful include clinical measurements, quality control, educational assessment, some financial planning contexts, environmental monitoring, and operations reporting. However, whether it is appropriate depends on the variable. Heights and many biological measurements often approximate normality reasonably well. Income, latency, costs, and survival times often do not.
When you should be cautious
There are several cases where trying to calculate mean and SD from P10, P50, and P90 can be misleading:
- Highly skewed data: If the right tail is much longer than the left tail, the mean will often exceed the median, so P50 may understate the actual mean.
- Bounded variables: Percentages, rates, and scores with hard lower or upper bounds may not follow a normal pattern.
- Mixtures: If your data come from multiple subpopulations, the combined distribution may not be bell-shaped.
- Rounded summaries: If published percentiles are rounded to coarse units, the inferred SD can drift.
- Small samples: Quantile estimates can be noisy when sample sizes are limited.
In these cases, the formula still gives an approximation, but it should be framed clearly as an estimate based on a normal assumption, not as a mathematically exact back-calculation.
Mean versus median in this context
A common source of confusion is the distinction between mean and median. P50 is the median by definition. In a perfectly normal distribution, the mean, median, and mode are equal, which is why P50 serves as the estimated mean here. But outside normality, they diverge. That is why the phrase “calculate mean sd from p10 p50 p90” really means “estimate the mean and SD assuming a normal or near-normal shape.”
If your midpoint check shows that P50 is close to the midpoint of P10 and P90, that supports the assumption. If not, you may want to report both values:
- Median-based center: P50
- Symmetry midpoint: (P10 + P90) ÷ 2
If these differ, that difference itself becomes a useful diagnostic signal.
Quick reference table for the calculation
| Quantity | Formula | Interpretation |
|---|---|---|
| Estimated mean | P50 | Best simple estimate if the data are approximately normal |
| Estimated SD | (P90 − P10) ÷ 2.5631 | Uses the known z-score distance between the 10th and 90th percentiles |
| Midpoint | (P10 + P90) ÷ 2 | Symmetry check against P50 |
| Symmetry gap | P50 − midpoint | Near zero suggests a better normal approximation |
Interpreting the graph produced by the calculator
The interactive chart on this page plots the implied normal curve from your estimated mean and standard deviation. It also overlays markers for P10, P50, and P90. This visual is more than decoration. It helps you answer practical questions:
- Do the percentile markers appear well-centered around the inferred mean?
- Is the spread narrow or wide relative to the magnitude of the variable?
- Does the observed median align with the visual center of the distribution?
In many applied settings, stakeholders understand a picture faster than a formula. If you are summarizing a KPI, service metric, or lab measure for a nontechnical audience, graphing the reconstructed normal distribution can make the relationship between percentiles and standard deviation much easier to explain.
Connections to standard statistical references
If you want to review the foundations behind percentile positions and z-scores, trusted educational and government resources are helpful. The NIST Engineering Statistics Handbook provides excellent practical guidance on probability distributions and data analysis. For a rigorous academic treatment of normal distribution concepts, many university materials such as Penn State’s STAT resources explain z-scores, percentiles, and inference in an accessible way. For health and public reporting contexts where percentile interpretation is common, the CDC also publishes statistical and surveillance materials that illustrate how summary distributions are used in practice.
Practical reporting language you can use
If you are documenting the result in a report, use wording like this:
- “The mean and standard deviation were estimated from the reported 10th, 50th, and 90th percentiles under a normal-distribution assumption.”
- “The estimated SD was computed as (P90 − P10) ÷ 2.5631, based on the spacing of the 10th and 90th percentiles in a normal distribution.”
- “Because the median closely matched the midpoint of P10 and P90, the normal approximation appeared reasonable.”
This kind of transparent language is especially important in technical documentation, scientific writing, internal analytics notes, and benchmark studies. It tells readers exactly how the estimate was produced and what assumptions were involved.
Bottom line
If you need to calculate mean sd from p10 p50 p90, the simplest and most defensible approach under normality is straightforward: take P50 as the mean and compute SD as (P90 − P10) ÷ 2.5631. Then compare P50 to the midpoint of P10 and P90 to check whether the distribution appears symmetric enough for that approximation to be credible.
This method will not replace a full analysis of raw data, but it is an efficient and practical tool when all you have are percentile summaries. Used appropriately, it can convert sparse reporting into a more interpretable statistical model and help you compare distributions, estimate probabilities, and communicate spread in a more familiar way.