Week-over-week reacts fast but picks up cyclical noise; year-over-year is clean but slow to react

What it is

Period-over-period (this week versus last) and year-over-year (this year versus the same stretch last year) are the two workhorse baselines, and their fitness is exactly complementary: the former is sensitive but polluted by cyclic noise — a Sunday-versus-Monday dip is usually calendar effect, not business decay; the latter cancels the cycle (the same stretch last year had the same weekends, the same seasonality and promotions) but reacts slowly — aligning by year means recent changes take a long time to surface in a like-for-like comparison.

Why it happens

Seen as a time series, a metric is the superposition of trend, cycle, and noise: the period-over-period difference has a lag of one, preserving the full cyclic structure, so periodic swings pour straight into the reading; the year-over-year difference has a lag of exactly one full cycle, so the cyclic components cancel in the subtraction and what remains approaches a pure trend signal. This explains the two classic misuses: judging growth trend with period-over-period (reading the cycle), and running daily alerts on year-over-year (discovering too late). The correct posture lets each do its own job with both on screen for cross-checking — period-over-period scouts recent movement (read against its cyclic background), year-over-year sets the trend verdict; agreement strengthens the conclusion, and divergence (period falling, year rising) is precisely the signal worth a closer look.

Where it stops holding

Year-over-year is not a universal cycle remover: shifted holidays (Lunar New Year drifting in the Gregorian calendar), irregular promotion rhythms, and leap-day counts all break the "same stretch last year is comparable" assumption, and rigorous year-over-year needs working-day or holiday-calendar correction. Period-over-period has its own corrections — against the average of the previous four same weekdays, against the previous working day — chosen by the metric's cycle structure. Metrics with no natural cycle (cumulative totals) make the two indistinguishable, and the raw series is enough.

Applying it

• Show period-over-period and year-over-year side by side on metric cards, each with its scope labelled, period-over-period primary by default.
• For strongly cyclical metrics (weekday effects, seasonality), period readings must carry the cycle hint ("Sunday versus Saturday").
• Verification: replay three months of history and measure how many period-over-period alerts the cycle explains; over half means switch the baseline or add cycle correction.

Related

• Same group: U7.05.1 A single value means something only against a baseline · U7.05.3 The baseline choice can flip the same metric into a rise or a fall · U7.05.4 With a tiny base, percentage change inflates into distortion · U7.05.5 The baseline period's statistical scope must match the current period's
• Nearby: U7.06.4 When the time range changes, comparison baselines must update in sync · U2.03.1 Line charts suit trends over ordered variables
• Search terms: year over year · period over period · seasonality