Level 5 · Does more ad spend bring more visits?

One number for “moving together”

The correlation coefficient r measures how closely two numbers follow a straight-line relationship. It runs from -1 to +1:

• the sign is the direction: + means they rise together, - means one rises as the other falls;
• the size is the strength: as a rough guide, above 0.7 is strong, 0.3 to 0.7 moderate, below 0.3 weak;
• 0 means no straight-line relationship (there could still be a curved one).

In pandas it’s one call on two Series of the same length, lined up row by row:

daily["paid_spend"].corr(daily["paid_sessions"])     # about 0.65

That’s pandas’ default, Pearson’s r. method="spearman" gives a rank-based version that is less sensitive to a few extreme days.

Lining the data up first

Correlation compares row 1 with row 1, row 2 with row 2. So both numbers must describe the same thing, here the same day. Spend has one row per day and channel; sessions have one row per visit. Count the visits per day, then merge them onto the spend days:

per_day = visits.groupby("day").size().rename("paid_sessions").reset_index()
spend_days = paid[["spend_date", "amount"]].rename(columns={"spend_date": "day", "amount": "paid_spend"})
daily = spend_days.merge(per_day, on="day", how="left").fillna({"paid_sessions": 0})

The left merge keeps every spend day, and fillna gives 0 to days with no paid visits. Those zero days are real observations; dropping them would hide the low end of the relationship.

Seeing it: scatterplot and regplot

sns.scatterplot(data=daily, x="paid_spend", y="paid_sessions") draws one dot per day. sns.regplot draws the same dots plus a fitted straight line and a shaded band around it showing how uncertain the line is. It’s the np.polyfit line from chapter 11 in one call:

ax = sns.regplot(data=daily, x="paid_spend", y="paid_sessions",
                 scatter_kws={"alpha": 0.35}, line_kws={"color": "#c0392b"})

With r around +0.65, days with higher spend tend to have more paid visits, but the dots are far from a line: plenty of high-spend days had few visits. Report it as “paid spend and paid visits are clearly associated”, with the number. Both also rise in November and December, so part of this link is the season itself, which the next levels deal with.

Common mistakes

• Correlating columns that aren’t aligned on the same unit (one row per day on both sides).
• Calling r = 0.65 “65% of visits are explained by spend”. r is not a percentage.
• Reading only the size and forgetting the sign, or the other way round.
• Writing “spend causes visits”. Correlation shows that they move together, not why.