Regression to the Mean
AnalyticsAlso: Reversion to the Mean · Regression Effect
Quick definition
Regression to the mean is the statistical tendency for unusually high or unusually low results to move back toward average on the next measurement, even with no change in strategy. It explains why a record-breaking campaign week is rarely repeated and why a terrible month often self-corrects without intervention.
How it varies across Australia
Teams running frequent campaign reviews are the most exposed to this, because more measurements mean more chances to catch an extreme result and react to it. The businesses that avoid the trap are the ones judging performance against a rolling baseline rather than against last period's single number.
See how AU teams structure their reporting cadence →What it actually means
Picture a golfer who shoots the round of their life on Saturday. Ask them to play again on Sunday and they'll almost certainly shoot worse, not because their swing got worse overnight but because Saturday was an outlier stacked on top of their real ability. Sunday's score drifts back toward what they actually average.
Marketing performance behaves the same way. A campaign's best-ever week usually contains real improvement plus a slice of good luck: a competitor went quiet, the algorithm had a good day, seasonality tilted in your favour. The next period, that luck doesn't repeat, and the number falls back toward the underlying average. Teams read that fall as decline. Often it's just gravity.
This matters most in A/B testing and channel reporting, where small sample sizes make extreme results common. A landing page that spikes to a huge conversion rate on 40 visitors isn't a breakthrough, it's a small sample producing noise. Confidence intervals exist specifically to separate real signal from this kind of statistical drift, and statistical significance testing is the discipline that stops you acting on it too early.
The best week you ever had was probably part luck. Regression to the mean is just the universe collecting on that loan.
How it shows up
It shows up in the campaign that gets celebrated for a record week and then quietly gets blamed the following week for 'losing momentum.' It shows up in A/B testing when a variant leaps ahead early on a small sample and then flattens out as more data arrives. It shows up in cohort analysis when the highest-performing customer segment from one month looks average the next, purely because a handful of unusually high spenders happened to land in that window. Anywhere a small or extreme sample gets treated as the new normal, regression to the mean is waiting to correct it.
The Australian context
Seasonality makes this trickier in Australia because the calendar doesn't match the northern hemisphere cycle. A retailer who has an outlier week around EOFY sales or a one-off event like a Melbourne Cup promotion needs to separate the seasonal bump from genuine channel improvement before drawing conclusions. Comparing that week to the following ordinary week and calling it a decline is a version of the same mistake, just wearing a seasonality costume.
Where people get this wrong
Related terms
Common questions
Is regression to the mean the same as reverting to bad performance?
No. It's a statistical drift back toward the underlying average, not a judgement about quality. A weak period can regress upward toward the mean just as easily as a strong period regresses downward.
How do I know if a result is regression or a real trend?
Look at the sample size and the run of data around the extreme point. A single outlier surrounded by stable average results is likely regression. A sustained shift across several periods with a larger sample is more likely a genuine trend.
Does regression to the mean affect A/B testing results?
Yes, heavily. Early in a test, small samples produce extreme swings that often correct themselves as more data comes in. That's part of why tests need to run until they reach statistical significance rather than being called early on a promising lead.
How do I stop my team misreading regression as a real change?
Report on rolling averages instead of single-period snapshots, and require a minimum sample size before drawing conclusions from any spike or dip. If a result can't survive being compared against a longer baseline, treat it as noise until proven otherwise.
Debrief
Get the next one
No spam. No fluff. Just the next article, straight to your inbox.
Keep exploring
About New Rebellion
New Rebellion is a marketing intelligence consultancy. We build tools, score Australian businesses on how their marketing actually performs, and publish Debrief every day. This dictionary is part of how we work in the open.
How we think →