Level 2 · Check and compare the price rise in pandas

What you are checking, and why

Before you compare anything, prove that you are looking at the right products and the right date. Thomas says “about 30 products, March 1”. The data can confirm both, because order_items stores the price paid on every line, while products["unit_price"] only holds today’s list price.

A product whose price changed has two different prices across its order lines. Named aggregation gives both in one pass, and a boolean mask keeps the products where they differ:

prices = order_items.groupby("product_id", as_index=False).agg(
    old_price=("unit_price", "min"), new_price=("unit_price", "max")
)
prices = prices[prices["old_price"] != prices["new_price"]]
len(prices)   # 29

Step 1 then labels each of their lines twice, by date and by price, with np.where, and counts the combinations. If the change happened cleanly on March 1, only two combinations exist: before with the old price, after with the new one.

Like for like, with a control group

Step 2 answers the agreed question. Three choices make the comparison fair:

• Same months, a year apart. March to August 2025 against March to August 2024.
• Like-for-like products. Only products launched before January 1, 2024, so every product was on sale for the whole of both windows.
• A control group. Every other product is labelled unchanged. Its growth is what the store did anyway.

isin labels the groups in one line, and pivot_table puts each year in its own column, the pandas version of SUM(CASE WHEN ...):

lines["price_group"] = np.where(lines["product_id"].isin(raised), "raised", "unchanged")
u = lines.pivot_table(index="price_group", columns="yr", values="quantity", aggfunc="sum")

The columns of u are the years as integers, so u[2025] - u[2024] subtracts row by row. Compute the percentages on these unrounded totals and round last.

Reading the result

price_group	units_change_pct	revenue_change_pct
raised	21.0	44.3
unchanged	9.2	13.4

If the rise had hurt demand, raised products would grow less than unchanged ones. They grew more. Revenue grew faster than units because each unit earned about 6% more and because sales shifted towards the dearer items in the group.

Pitfalls

• Filter with < on the day after the window ends: d < "2025-09-01" keeps all of August 31.
• Forgetting .copy() after filtering and then adding columns gives a SettingWithCopyWarning.
• pivot_table sorts its index, so the result is already in price_group order; the explicit sort_values makes that a promise rather than a coincidence.