Pandas Descriptive Statistics
last modified November 17, 2026 Python 3.15
Descriptive statistics summarise a data set with a handful of numbers: a
typical value, a measure of spread, the smallest and largest observations, and
so on. Pandas makes this very convenient with the describe method,
which produces many of these summaries at once, as well as with individual
methods for each statistic.
All the examples below were run with pandas 3.1.1. We will use small Series and DataFrames so that every number in the output can be checked by hand.
Describing numeric columns
Calling describe on a DataFrame examines every numeric column. The
result is a table whose rows are the statistics and whose columns are the
original numeric columns. A numeric column is one that holds numbers,
so it can be added, averaged and ordered. Note that std is the
standard deviation, a measure of how far the values typically sit
from the mean, and the 25%, 50% and 75%
rows are quartiles.
import pandas as pd
df = pd.DataFrame({
'age': [23, 27, 31, 35, 39],
'salary': [40000, 50000, 60000, 70000, 110000]
})
print(df.describe())
For both columns count is 5 because there are five observations
and no missing values. The average age is 31 and the average salary is 66000.
The 50% row is the median, the middle value once the data is
sorted; here it equals the mean for age because those values are
evenly spaced.
$ python describe_basic.py
age salary
count 5.000000 5.000000
mean 31.000000 66000.000000
std 6.324555 27018.512172
min 23.000000 40000.000000
25% 27.000000 50000.000000
50% 31.000000 60000.000000
75% 35.000000 70000.000000
max 39.000000 110000.000000
Individual statistics
When we need just one summary, each statistic has its own method. The
mean is the arithmetic average, the median is the middle
value, min and max are the extremes, and sum
is the total. These methods work on a single Series and return one number.
import pandas as pd
s = pd.Series([10, 20, 30, 40, 50])
print('mean ', s.mean())
print('median', s.median())
print('std ', s.std())
print('min ', s.min())
print('max ', s.max())
print('sum ', s.sum())
The five values add up to 150, so the mean is 30, which is also the median.
Pandas computes the sample standard deviation by default, dividing by the
number of values minus one; that is why std is about 15.81 rather
than 14.14. The methods that return an integer do so here because the input
contains integers.
$ python describe_numeric_stats.py mean 30.0 median 30.0 std 15.811388300841896 min 10 max 50 sum 150
Non-numeric columns
By default describe skips columns that do not hold numbers. Text
columns can still be summarised by passing include='object'. Their
summary has different rows: count of the non-missing values,
unique for the number of distinct values, top for the
most common one and freq for how often it appears. To keep this
example free of pandas version differences, the text column is created with an
explicit object dtype.
import pandas as pd
df = pd.DataFrame({
'city': pd.Series(
['Bratislava', 'Prague', 'Bratislava', 'Vienna', 'Prague'],
dtype=object
),
'temperature': [12, 15, 14, 16, 13]
})
print(df.describe(include='object'))
Only the city column is shown, because the numeric
temperature column does not match the requested type. There are
three distinct cities; Bratislava is the most frequent and it
occurs twice.
$ python describe_object.py
city
count 5
unique 3
top Bratislava
freq 2
Counting values with value_counts
The value_counts method counts how many times each distinct value
occurs and returns a Series sorted from the most common to the least common.
It is the go-to tool for a quick frequency table of a categorical column.
import pandas as pd s = pd.Series(['apple', 'banana', 'apple', 'orange', 'banana', 'apple']) print(s.value_counts())
There are three apple entries, two banana entries and
one orange. The index of the result holds the distinct values and
the values hold their counts; the total, 3 + 2 + 1, is six, the length of the
original Series.
$ python describe_value_counts.py apple 3 banana 2 orange 1 Name: count, dtype: int64
Correlation and covariance
Two methods describe how columns vary together. Covariance measures the direction of a joint relationship: a positive value means the columns tend to rise together. Correlation is covariance scaled to a value between -1 and 1, which makes the strength of the relationship easier to compare. Both return a square table with one row and column per numeric column; the diagonal is 1 for correlation because a column is perfectly correlated with itself.
import pandas as pd
df = pd.DataFrame({
'hours': [1, 2, 3, 4, 5],
'score': [2, 4, 5, 7, 11]
})
print('corr:')
print(df.corr())
print()
print('cov:')
print(df.cov())
The correlation between hours and score is about
0.97, a strong positive relationship: more hours go with higher scores. The
covariance is positive too, at 5.25, but its size depends on the units of the
two columns, which is why the correlation is usually easier to interpret.
$ python describe_corr_cov.py
corr:
hours score
hours 1.000000 0.970725
score 0.970725 1.000000
cov:
hours score
hours 2.50 5.25
score 5.25 11.70
Quantiles
A quantile is a value below which a given fraction of the data
falls. Passing a fraction from 0 to 1 to quantile returns that
point; the fraction 0.5 is the median. Passing a list of fractions returns one
value for each, which is a convenient way to get quartiles at once.
import pandas as pd
s = pd.Series([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
print('median (0.5):', s.quantile(0.5))
print('quartiles:')
print(s.quantile([0.25, 0.5, 0.75]))
There are ten values from 1 to 10, so the middle falls halfway between 5 and
6, giving 5.5. The first quartile is 3.25 and the third quartile is 7.75;
these are the same numbers that appear in the 25% and
75% rows of describe.
$ python describe_quantile.py median (0.5): 5.5 quartiles: 0.25 3.25 0.50 5.50 0.75 7.75 dtype: float64
Selecting columns with include and exclude
The include and exclude arguments choose which column
types take part. The special value 'all' includes every column, so
we see numeric and non-numeric summaries side by side. Cells that do not apply
to a column are shown as NaN, the marker Pandas uses for a missing
value. Using exclude drops one type instead.
import pandas as pd
df = pd.DataFrame({
'name': ['A', 'B', 'C'],
'score': [10, 20, 30],
'passed': [True, False, True]
})
print(df.describe(include='all'))
print()
print(df.describe(exclude='bool'))
With include='all' all three columns appear. The name
column gets the text-style rows and score gets the numeric ones;
the boolean passed column gets count, unique, top and freq. With
exclude='bool' the passed column disappears and only
name and score remain.
$ python describe_include_exclude.py
name score passed
count 3 3.0 3
unique 3 NaN 2
top A NaN True
freq 1 NaN 2
mean NaN 20.0 NaN
std NaN 10.0 NaN
min NaN 10.0 NaN
25% NaN 15.0 NaN
50% NaN 20.0 NaN
75% NaN 25.0 NaN
max NaN 30.0 NaN
name score
count 3 3.0
unique 3 NaN
top A NaN
freq 1 NaN
mean NaN 20.0
std NaN 10.0
min NaN 10.0
25% NaN 15.0
50% NaN 20.0
75% NaN 25.0
max NaN 30.0
describe versus agg
Both describe and agg produce summaries, but they
serve different purposes. describe is a fixed, ready-made report:
it always returns the same set of statistics for the columns it selects, which
makes it ideal for a first look at data. agg is a flexible tool
where we choose exactly which functions to run, give the results our own names,
and can apply different functions to different columns.
In short, reach for describe when exploring and for
agg when building a specific result. The
Pandas aggregation tutorial covers
agg in detail, including named aggregations and its use with
groupby.
Source
Pandas DataFrame.describe Documentation, Pandas Series.value_counts Documentation, Pandas DataFrame.corr Documentation, and Pandas Essential Basics Documentation
In this article we have summarised data with describe on numeric
and non-numeric columns, computed individual statistics such as
mean and std, counted categories with
value_counts, measured relationships with corr and
cov, found quantile values, selected columns with
include and exclude, and compared describe
with agg. For building our own summaries see
Pandas aggregation, and for grouping rows before
summarising see the grouping data tutorial.
Author
List all Pandas tutorials.