Level 2 · Average shipping fee per channel

Split, apply, combine

groupby does three things in one call:

1. Split the rows into groups that share a value (one group per channel).
2. Apply a calculation to each group (the mean shipping fee).
3. Combine the answers into a new Series or DataFrame, one row per group.

orders.groupby("channel")["shipping_fee"].mean()

channel
app            ...
marketplace    ...
web            ...
Name: shipping_fee, dtype: float64

Read it left to right: group orders by channel, take the shipping_fee column, average it. The group keys become the index of the result, sorted A to Z by default.

Common aggregations

method	per group
.sum()	total
.mean() / .median()	average / middle value
.min() / .max()	extremes
.count()	non-missing values in the column
.size()	rows, missing or not
.nunique()	distinct values

count and size look alike but answer different questions:

orders.groupby("channel")["coupon_code"].count()   # orders WITH a coupon, per channel
orders.groupby("channel").size()                    # all orders, per channel

count skips missing values, so it’s a count of coupons used. size counts rows. Mixing them up produces numbers that look plausible and are wrong, which is the worst kind.

Series or DataFrame?

Selecting one column after groupby gives a Series indexed by the group. To get an ordinary two-column table, which is usually what you want for display, merging or export, use reset_index() or as_index=False:

orders.groupby("channel")["shipping_fee"].mean().reset_index()
orders.groupby("channel", as_index=False)["shipping_fee"].mean()

Both give a DataFrame with columns channel and shipping_fee.

Reading the result

The average fee per channel mixes free-shipping orders (fee 0) with paid ones. That’s a fair answer to “what does shipping bring in per order?”, but not to “what do people pay when they pay?”. For the second question, filter first:

paid = orders[orders["shipping_fee"] > 0]
paid.groupby("channel")["shipping_fee"].mean()

Same code, different rows, different meaning. Always ask which population the question is about before you group.

Sorting the output

The result is sorted by the group key. To rank groups by the value instead, sort afterwards: .sort_values(ascending=False).

Common mistakes

• Forgetting to select a column: orders.groupby("channel").mean() tries to average every column and fails on text columns in pandas 2. Select what you need, or pass numeric_only=True.
• Using count() when you mean “number of rows”.
• Comparing averages without the group sizes. A channel with few orders can have an extreme mean by chance; show size() alongside.

If you know SQL

This is SELECT channel, AVG(shipping_fee) FROM orders GROUP BY channel.