Level 5 · Does traffic move with sales? CORREL

A correlation answers one narrow question: when one column is high, is the other one high too? CORREL turns that into a single number. It is quick to compute and easy to over-read, so this level is half formula, half interpretation. The sheet holds 52 weeks of 2025, Monday to Sunday, with web sessions, average pages per session, conversions and total ad spend.

CORREL in one line

=CORREL(array1, array2)
=CORREL(B2:B53,D2:D53)

CORREL returns the Pearson correlation coefficient, always between -1 and 1. The order of the two ranges doesn’t matter; their size does: both must cover the same rows, so that row 10 is the same week in both. Rows where either cell is empty or text are skipped as a pair.

Value	Reads as
close to 1	when one is high, the other is high, in a near-straight line
around 0.5	a clear but loose tendency
close to 0	no straight-line relationship
negative	when one is high, the other tends to be low

These bands are a rule of thumb, not a law. With 52 points, anything under about 0.3 either way is too weak to build a claim on.

What the four numbers say

Cell	Pair	2025
H2	sessions vs conversions	0.82
H3	ad spend vs sessions	0.91
H4	pages per session vs conversions	-0.29
H5	sessions vs conversions, Jan–Oct	0.18

On their own, H2 and H3 look like Susan’s pitch, proven. H5 is the reason you check. Remove the eight weeks from November 3 onward and the sessions-conversions correlation collapses from 0.82 to 0.18. Look at rows 46 to 53: the Black Friday week (row 49) had 986 sessions and 30 conversions, against 181 to 379 sessions and at most 14 conversions in any week from January to October. A handful of extreme weeks can manufacture a strong correlation on their own.

H4’s -0.29 is weak, and it also comes from the holidays: crowds of quick browsers in December lower the pages per session while conversions rise. For January to October that pair is +0.11.

Correlation is not causation

H3 is strong all year, 0.91, and still 0.70 without the holidays. It does not show that ad spend causes traffic. The store raises spend for Black Friday week because shoppers are coming then anyway; both columns follow the calendar. The same pattern links ice cream sales to sunburn.

Before anyone writes “because”, ask three questions:

• Could a third thing drive both? Here, the season.
• Which way could it run? More traffic could just as well trigger more spend, through budget rules tied to demand.
• Is it a few extreme rows? Recompute without them, as in H5.

What Susan can honestly say: “Traffic and ad spend rise and fall together through the year; to show the ads cause the extra traffic we’d need a test, such as pausing a channel in some weeks.”

A relationship the course example expects

The classic textbook pair is quantity vs line total, expected to be strongly positive. Try it on the order lines of level 1 and you get 0.22: at Harbor & Pine, one unit of a $1,599 bed frame outweighs 16 bud vases, so price drives revenue far more than quantity does. Expectations are not results; compute them.

Common mistakes

• Ranges of different sizes or offset rows: =CORREL(B2:B53,D3:D54) pairs each week with the next one’s conversions. Different sizes give #N/A.
• Calling 0.3 strong on 52 points.
• Saying “drives” or “because” on the strength of a correlation alone.
• Trusting one number without looking at the rows that make it, or at a chart. A scatter plot of B against D shows the eight holiday weeks sitting far from the rest.