What Is a Percentile? How to Read and Use It in SPSS

What Is a Percentile

You open the descriptive statistics output and see rows labelled Percentiles, 25, 50, 75, or 90. These numbers show where a value sits after all observations have been ordered from smallest to largest. That is the short answer to what is a percentile.

A percentile divides data into 100 relative positions. The pth percentile is the value at or below which approximately p% of the observations fall, while the remaining observations are at or above it. For example, the 75th percentile, usually written as P75, marks the point at or below which approximately 75% of the observations fall.

For a quantitative variable, the basic calculation starts by arranging the values in ascending order. The software then identifies the position corresponding to the required percentage and interpolates if that position falls between two observations. Because interpolation conventions differ, P75 can vary slightly between SPSS, Excel, R, and an online tool. In your thesis, state which software you used and apply it consistently throughout the analysis.

For example, if P90 for app usage time is 42 minutes, you can report that approximately 90% of respondents used the app for no more than 42 minutes. Saying that 90% of users had exactly 42 minutes of usage would be inaccurate.

Why Percentiles Matter in Quantitative Research

Percentiles help you see the shape of a distribution that the mean can sometimes hide. Two groups may have the same mean, while one group is tightly clustered and the other contains several very high or very low observations. Comparing P25, P50, and P75 gives you more information about the middle portion of the data.

In quantitative research, P50 is the median. P25 and P75 are commonly used to describe the 50% of observations in the central part of the data. The distance between P75 and P25 is the interquartile range, abbreviated as IQR. IQR is useful when data are skewed or contain outliers because it is less strongly pulled by a few extreme values than the range from the minimum to the maximum.

You can use percentiles for three main purposes. First, you can describe the sample, such as income, years of experience, usage time, or number of purchases. Second, you can compare one person's position with the whole sample, such as identifying a person's score as being at P80. Third, you can support the detection of unusual values with a rule based on Q1 and Q3, where Q1 is P25 and Q3 is P75.

If your variable is right-skewed, reporting the mean and standard deviation may not be enough. Check the median and percentiles as well. For Likert survey data, percentiles can describe the distribution of scores, but interpret them carefully because the scale has a limited number of response levels. If you are checking observations that differ substantially from the rest, you can also read the guidance on outliers in quantitative data.

Percentiles can also support data grouping. For example, you could divide respondents into low, medium, and high groups using P33 or P67 as cut points. Grouping removes some of the variable's continuous information. Use it only when you have a clear theoretical reason or analytical purpose, rather than grouping simply to make the results table easier to read.

Which Percentile Is Acceptable

There is no general acceptable percentile threshold for every thesis. A percentile is a descriptive measure of position in a distribution, not a coefficient used to decide whether a scale is good or a model fits. Therefore, P25 equal to 3 or P75 equal to 4 does not automatically show that the data pass or fail.

Evaluate percentiles according to your research question, scale, and the practical meaning of the variable. For example, a lower P90 for delivery time may be better if your goal is to reduce waiting time. By contrast, a high P90 for satisfaction scores may be a positive signal. The same number can lead to a different decision depending on the direction in which the variable is measured.

The table below helps you choose how to interpret each value. It is not a list of acceptable thresholds. P25, P50, and P75 are called quartiles under common descriptive statistics conventions. The practical interpretation must come from the variable and the context of your research.

Check the coding direction as well. If a higher score means greater customer satisfaction, a high P75 usually describes a group with more positive evaluations. If a higher score means a longer waiting time, a high P75 may instead indicate something that needs improvement. Do not conclude that a percentile is good or bad before reading the variable label and scale.

When you evaluate scale quality, use appropriate measures such as Cronbach's Alpha, Corrected Item-Total Correlation, EFA, or PLS-SEM measures. A percentile does not replace those tests. You can also read what is a scale if you are confusing the percentile of one variable with the evaluation of a scale containing multiple items.

How to Read Percentiles in the Output

In SPSS, you can obtain percentiles through two commonly used paths. For basic descriptive statistics, select Analyze > Descriptive Statistics > Frequencies, move the variable into the Variable(s) box, select Statistics, mark the Percentiles you want to inspect, then choose Continue and OK. If you also want a boxplot and statistics for checking outliers, select Analyze > Descriptive Statistics > Explore.

In the Frequencies output, the Statistics table usually displays Valid, Missing, Mean, Median, Std. Deviation, Minimum, Maximum, and the Percentiles you selected. In Explore, SPSS usually provides the Descriptives, Tests of Normality, and Extreme Values tables, along with a Boxplot. Table names can vary slightly by version or dialog-box settings, but the main rows and columns have the same meaning.

The table below is illustrative output, created so you can recognize the structure of the output. It is not the result of a real study and must not be copied into Chapter 4 as a finding from your research.

From the illustrative table, you could write that P25 for usage time is 18 minutes, the median is 27 minutes, P75 is 41 minutes, and P90 is 58 minutes. The correct interpretation is that approximately 50% of observations have a usage time of no more than 27 minutes, while the other 50% have a usage time at or above that point, depending on how the software handles tied values and interpolation.

The interquartile range in this example is 41 - 18 = 23 minutes. This value shows the width of the central 50% of the data. It is not the standard deviation, and it does not show the distance from the minimum to the maximum.

When reading the output, check the N column first. If Valid is much lower than the number of rows in the file, the data contain Missing values and the percentile was calculated from the valid observations only. Check Minimum and Maximum as well, so you can identify incorrect codes, such as a time variable that should range from 1 to 60 but contains the value 999.

In the Extreme Values table, do not automatically delete observations at the high or low end. An observation at P99 can still be genuine data. Review the original file, data-entry procedure, sample filters, and variable logic before deciding how to handle it.

What to Do When a Percentile Does Not Meet Your Expectation

Because percentiles do not have general acceptable thresholds, the phrase “the percentile does not pass” usually indicates that the measure is being used for the wrong purpose. Before changing the data, identify what your supervisor is asking you to evaluate. They may want you to check outliers, distributional skewness, sample size, or scale quality, rather than require P25 or P75 to exceed a fixed number.

Step 1: Check the coding and measurement units

Read the variable label, units, and codebook again. If half of the data are recorded in thousands of currency units and the other half in individual currency units, the percentile will be distorted. If a Likert variable uses codes from 1 to 5 but some rows contain 0 or 99, determine whether those values are Missing or valid data before running the analysis again.

Step 2: Check Missing values and sample filters

In SPSS, check the number of Valid and Missing observations in the Statistics table. Check Data > Select Cases to make sure you have not accidentally activated a filter. A major change in a percentile can sometimes result from analysing a subgroup rather than from any change in the original data.

Step 3: Check outliers and why they occur

If P75 is far from P25 or P90 rises sharply, open Analyze > Descriptive Statistics > Explore, select Plots, and mark Boxplots. Then compare the extreme rows with the questionnaire or data-entry source. You may retain an observation if it represents a real case and explain the treatment in your thesis. Deleting it simply because it makes the percentile look better is difficult to defend.

Step 4: Use a transformation or an appropriate method

If the data are skewed because of the nature of the variable, you can report the Median and IQR instead of the Mean and Std. Deviation when your goal is description. If the regression model has a problem, also check the residuals, linearity, homoscedasticity, and multicollinearity. Do not apply a log transformation or divide the data into groups simply to make the percentiles closer together.

You should not alter respondents' answers to reach a desired percentile. If the problem is a data-entry error, correct it using the questionnaire or original file and keep a record of the editing history. If the value is genuine, retain it or state your treatment criteria clearly. This is the kind of detail the committee may ask about when they see that observations were removed.

Percentile vs Mean, Median, and Quartile

A percentile is the broader concept. The median is a specific percentile, corresponding to P50. Quartiles are the points that divide data into four parts, usually Q1 equal to P25, Q2 equal to P50, and Q3 equal to P75. These terms should not be used as though they were completely interchangeable measures.

The arithmetic mean is calculated by dividing the sum of the values by the number of observations. It uses every value and can be affected by outliers. The median is based on the middle position after the data are ordered, so it is usually more stable when the distribution is skewed. Percentiles let you examine many different positions, such as P10, P25, P50, P75, and P90.

For example, a household income variable may have a mean of 15 million VND but a median of 10 million VND. This gap suggests that several high income values are pulling the mean upward. If P25 is 7 million and P75 is 16 million, you also know the width of the central part of the sample rather than seeing only one average value.

You also need to distinguish a percentile from the percentage of respondents. P90 for a variable is a value on the measurement scale, while 90% is the proportion of observations at or below that value. The two statements are related but do not have the same unit. P90 for income is expressed in currency, P90 for time is expressed in minutes, and 90% is a proportion.

If you are building a model with an interaction variable, percentiles are sometimes used to select low, medium, and high levels of a moderator. The cut points should be tied to the analysis design rather than selected mechanically at P25 and P75. You can read more about moderating in quantitative research and what is a moderator to distinguish a moderating variable from a variable that has simply been divided into percentile groups.

Common Errors

Calling P50 the mean

P50 is the median, not the Mean. The two values coincide only for some distributions or datasets. When you write up the results, use the exact name of the measure shown in the Statistics table.

Interpreting a percentile as a probability

P90 does not mean that a randomly selected observation has a 90% probability of taking exactly the value P90. It describes the position of the data after ordering. Your interpretation should refer to the proportion of observations at or below the point, while noting the software's interpolation convention when relevant.

Setting your own acceptable threshold for P25 or P75

Percentiles do not have general acceptable thresholds like Cronbach's Alpha or AVE. If your study has a practical threshold, such as a maximum delivery time, that threshold must come from the research requirement or business context, not from the concept of a percentile itself.

Ignoring Missing values and filters

SPSS calculates percentiles from the observations included in the analysis. If you activate Select Cases or have Missing values, the result may differ from what you would obtain by looking at the entire file. Report the number of valid observations.

Deleting extreme values without a reason

A value at P95 or P99 is not automatically an error. Check the data source, coding, and variable logic before removing it. If you remove it, record the criteria, number of rows, and effect on sample size.

Frequently asked questions

What is a percentile and what is it used for?

A percentile is a value that identifies the relative position of an observation in an ordered dataset. It is used to describe a distribution, compare positions, calculate the interquartile range, and support checks for extreme values.

Which percentile is acceptable?

There is no general acceptable percentile for every study. P25, P50, P75, and P90 are descriptive points. Interpret them according to the research question, measurement unit, and practical standard for the variable.

Is P50 the mean?

P50 is the median, not the arithmetic mean. The median is the middle value after the data are ordered, while the arithmetic mean is calculated by dividing the sum of the values by the number of observations. When data are skewed or contain outliers, the two measures can differ substantially.

Why is the percentile in SPSS different from the one in Excel?

SPSS and Excel may use different interpolation conventions or calculation methods, especially when the sample size is small or the percentile falls between two observations. Use one software package consistently, state which tool you used, and do not combine results from different methods without explaining why.

Should I delete data above P75?

No. P75 is the point at or below which approximately 75% of observations fall. Observations above P75 can still be valid data. Consider removal only when there is a data-entry error, a violation of the sample criteria, or a predefined and clearly defensible treatment rule.

How to write this in your thesis

Open your SPSS file again and run Analyze > Descriptive Statistics > Frequencies or Explore. Check Valid, Missing, P25, P50, and P75, then compare them with the variable label before writing the results. If you need to run the complete analysis workflow on your own .sav or .csv file, see M4 data analysis by DoThesis.