Level 2 · Which return reasons cost the most?

Bars compare categories

Return reasons have no natural order, and Angela wants to compare their size. That’s a bar chart. The data is one number per reason:

refunds = returns.groupby("reason")["refund_amount"].sum().round(2).sort_values(ascending=False)
# changed_mind    132949.23
# damaged         128729.46
# late             57859.97
# wrong_item       51735.20

Always sort the bars by value. An alphabetical order makes the reader hunt for the biggest; a sorted one reads like a ranking at a glance.

Vertical or horizontal

ax.bar(x, heights) draws vertical bars, ax.barh(y, widths) horizontal ones. Pick horizontal when the labels are long (product names, job titles, reasons): they stay readable without tilting.

There’s one trap. barh draws the first item at the bottom. With data sorted largest first, the largest bar ends up at the bottom. Two fixes:

ax.barh(refunds.index, refunds.values)
ax.invert_yaxis()                     # flip: first item on top

# or sort ascending just for the chart
s = refunds.sort_values()
ax.barh(s.index, s.values)

With pandas it’s one line, and it returns the Axes for the finishing touches: ax = refunds.sort_values().plot(kind="barh").

Values on bars and one highlight color

ax.bar_label writes each bar’s value at its end, so nobody has to read the axis:

bars = ax.barh(refunds.index, refunds.values, color=colors)
ax.bar_label(bars, labels=[f"${v / 1000:.0f}k" for v in refunds.values], padding=3)

Color is for meaning, not decoration. Make every bar gray and give one bar a strong color: the one your title talks about. A list comprehension builds the colors in the same order as the bars:

colors = ["#c0392b" if r == "damaged" else "#b0b0b0" for r in refunds.index]

What the chart says

Changed minds lead with $132,949.23 from 698 returns. Damaged items cost almost as much, $128,729.46, from only 477 returns, so each damaged return is more expensive. Together the two make up 71% of the $371,273.86 refunded. A packaging fix would attack a cost as large as the one Angela was worried about, and it is far easier to act on than customers changing their minds. That is the kind of finding the title should state.

Common mistakes

• Forgetting that barh draws bottom-up, so the “top” item lands at the bottom.
• Leaving the bars unsorted. Readers compare neighbors; a ranking needs sorted bars.
• Giving every bar its own color. It looks busy and hides the one bar that matters.
• Starting the value axis above zero. A bar’s length is its value; cut the axis and the lengths lie.