What is the difference between merge, join, and concat in pandas?

How to think about it

All three combine DataFrames, but they answer different questions. concat asks “stack these top-to-bottom (or side-by-side)” — no key matching. merge asks “align rows where this column’s values match” — SQL-style. join is a convenience alias for merge that aligns on the index. Name each one’s sweet spot and you’ve answered well.

A worked example — all three side by side

concat stacks without matching, lining up columns by name:

import pandas as pd

jan = pd.DataFrame({"month": ["Jan", "Jan"], "sales": [100, 200]})
feb = pd.DataFrame({"month": ["Feb", "Feb"], "sales": [150, 90]})
print(pd.concat([jan, feb], ignore_index=True))     # vertical stack

  month  sales
0   Jan    100
1   Jan    200
2   Feb    150
3   Feb     90

That’s appending monthly exports — pure stacking, no key. pd.concat([features, labels], axis=1) instead glues columns side-by-side by position. merge is the SQL join on a column, even when the key names differ:

orders = pd.DataFrame({"order_id": [1, 2, 3], "cust_id": [10, 20, 10], "amount": [50, 200, 75]})
customers = pd.DataFrame({"id": [10, 20, 30], "name": ["Alice", "Bob", "Carol"]})
print(orders.merge(customers, left_on="cust_id", right_on="id", how="left"))

   order_id  cust_id  amount  id   name
0         1       10      50  10  Alice
1         2       20     200  20    Bob
2         3       10      75  10  Alice

Each order is matched to its customer by value (cust_id ↔ id), and Alice appears twice because she has two orders — that’s key alignment, not stacking. join is the same machine when both frames already carry the right index:

info  = pd.DataFrame({"city": ["NY", "LA", "SF"]}, index=[10, 20, 30])
score = pd.DataFrame({"score": [88, 72, 95]},     index=[10, 20, 30])
print(info.join(score))     # index-on-index shorthand for merge

   city  score
10   NY     88
20   LA     72
30   SF     95

join aligns on the shared index with no on= needed — it’s merge(left_index=True, right_index=True) with prettier syntax and no performance difference. So the real choice is almost always concat (stack) vs merge (align on keys); join is merge for pre-indexed frames.