Percentile calculator
Paste a column of values. Calculate the mean, median, standard deviation, quartiles and any percentiles you need (P5, P95, P99…) with the method of your choice; see the distribution and outliers in a box plot.
Type one number per line or copy a column from Excel. If there are several columns they must be separated by tab, ; or |, and you pick the column. Decimal comma and point are supported; a header row is skipped and non-numeric cells are skipped.
The calculation runs in your browser; the data you paste is not sent anywhere.
Paste the values or load the sample.
Let us look together at which percentiles of your measurements should be tied to a target or SLA and how to monitor them.
Request a call01
How to use
A
Paste the values (you can copy a column from Excel) or load the sample data; if there are several columns, pick the column.
B
Type the percentiles you want (for example 5, 50, 95, 99) and choose the method.
C
Read the summary statistics, the percentile table and the box plot; copy the result.
02
What is a percentile?
The p-th percentile is the value below which p% of the data fall. P50 is the median, P25 and P75 are the first and third quartiles. With skewed distributions such as response times the mean is strongly affected by outliers; P95 and P99 answer the question "what is the longest most users wait?" much better than the mean.
Percentiles are used in service level objectives ("95% of requests under 200 ms"), in quality control and in lead-time planning.
03
Why are there different methods?
A percentile can fall between two values in the sorted data, and there is no single right way to handle that. This tool offers three common methods. R-7 (the default of Excel PERCENTILE.INC, NumPy and R) fills in linearly between the two neighbouring values: h = (n−1)·p + 1. R-6 (Excel PERCENTILE.EXC) uses h = (n+1)·p; it is undefined at extreme percentiles, for example P99 with 20 values. The nearest-rank method returns the actual value at rank ⌈p·n⌉ and does not interpolate.
For example, for the data 15, 20, 35, 40, 50 the P40 is 29 with R-7, 26 with R-6 and 20 with nearest rank. With large data the differences shrink; with small data they matter. When comparing results with other software, choose the same method.
04
Box plot and outliers
The box runs from Q1 to Q3 and the line inside is the median; the width of the box is IQR = Q3 − Q1 and covers the middle half of the data. The whiskers extend to the most extreme values within Q1 − 1.5·IQR and Q3 + 1.5·IQR (Tukey's rule); values beyond the fences are drawn as separate points.
An outlier does not mean an error: it can be a sensor fault or a real event. Investigate the cause before removing it; the mean and standard deviation are strongly affected by outliers, the median and quartiles are not.
FAQ
- What does P95 mean?
- It is the value below which 95% of the data fall: 95 of 100 measurements are equal to or smaller than it. For response time, P95 = 200 ms means that 95% of the requests completed in 200 ms or less.
- Is the median the same as P50?
- With linear interpolation (R-7) and the exclusive method (R-6), yes. With the nearest-rank method and an even number of values, P50 is the lower middle value and can differ from the conventional median (the average of the two middle values).
- Why does my result differ from Excel?
- Excel's PERCENTILE and PERCENTILE.INC functions are R-7 and PERCENTILE.EXC is R-6. If you match the tool's method to Excel's, the results agree. QUARTILE.INC and QUARTILE.EXC make the same distinction.
- Which method should I choose?
- R-7 for general use and agreement with most software. Nearest rank if the result must be a value that actually occurs in the data (for example an SLA report). R-6 is the method of some statistics textbooks and of Excel's exclusive function.
- How many values do I need?
- For P99 you need at least a few hundred values; with 20 values P99 rests on only the few largest values and is very unstable. P50 and the quartiles are reasonable with far less data.
- In what format should I paste the data?
- One number per line or a column copied from Excel. If there are several columns they must be separated by tab, ; or | and you pick the column. Decimal comma or point is accepted; a header row and non-numeric cells are skipped.
Tie your measurements to a target
Let us look together at which percentiles in your process, latency or quality data should be monitored and how to turn that into a dashboard and an alert.