Level 7 · Debug: 60 departments for 60 people?

count, nunique and value_counts

Three methods sound alike and answer different questions about a column:

employees["department"].count()          # 60  non-missing values (one per row here)
employees["department"].nunique()        # 9   distinct values
employees["department"].value_counts()   # how many rows have each value

count() is about rows: how many cells in the column are filled. Since every employee has a department, it equals the headcount. nunique() is about values: it counts each different department once, whatever the number of people in it. value_counts() gives the detail behind nunique(): one line per department with its row count, Sales and Customer Support at the top with 9 people each, down to Executive with 2.

Reading a suspicious number

The CEO caught this bug without looking at code: a number equal to the row count is a classic sign that something counted rows instead of values. Whenever a “how many different” answer comes out exactly equal to len(df), check the method. The real figures are 9 departments and 31 titles.

The two counts also tell a story together. Nine departments but 31 titles means most departments have several roles: the head of the department, leads and staff. If the numbers had been 9 and 9, that would suggest one title per department, which is not how this company is organized.

When count and nunique agree

On a column of IDs, such as employees["employee_id"], count() and nunique() both give 60, because every row has a different ID. That’s a handy check that a key column really is unique: compare nunique() with len(df). If they differ, there are duplicates.

count() is still useful on its own. On a column with gaps it tells you how many cells are filled: employees["manager_id"].count() is 59, because the CEO has no manager.

Common mistakes

• Using count() or len() when the question says “how many different”.
• Using len(df["department"].unique()). It works, but nunique() says what you mean and skips missing values by default, which is usually what a report wants.
• Reading the length of value_counts() and the sum of value_counts() the wrong way round: the length is the number of departments (9), the sum is the headcount (60).