Median vs average views: which baseline to use
Use the median view count as your channel baseline, not the average: a single breakout video drags the average far above what the channel normally does, while the median stays anchored to the typical upload and keeps outlier scores honest.
- View counts are skewed, not normal, so the mean sits well above the typical upload.
- One 10x video in a set of 20 can lift the mean by around 45% and suppress every later outlier score.
- Use a median over at least 20 same-format uploads from the last 6 to 12 months.
- Means are right for forecasting volume; medians are right for detecting over-performance.
View counts are not normally distributed. A channel's uploads cluster in a narrow band with occasional very large results, which is exactly the shape that breaks averages. Choosing the baseline is the most consequential decision in outlier detection, because every score downstream is divided by it.
Published July 29, 2026 · Updated July 29, 2026
How to build a reliable baseline
- Filter the channel's uploads to a single format: long-form or Shorts, never mixed.
- Restrict to the last 6 to 12 months, and require at least 20 videos.
- Exclude uploads younger than 7 days, whose view counts have not stabilised.
- Sort the remaining view counts and take the middle value.
- Recompute monthly, or whenever the channel's upload cadence or format mix changes.
The shape of the data decides the statistic
Averages describe symmetric data well. View counts are the opposite: a dense cluster near the typical result with a long tail of rare large ones. For that shape, the median is the standard summary because it reports the middle upload rather than a value pulled toward the tail.
This is not a preference. Using a mean on skewed data reports a number that few of the channel's videos ever achieved.
- Median: the middle upload. Unmoved by one breakout.
- Mean: the balance point. Moved by every breakout, permanently.
The self-defeating loop of a mean baseline
A channel produces a breakout. The mean rises. The next genuine over-performer is now measured against an inflated baseline and fails to flag. The detector concludes the channel has gone quiet at the exact moment it started working, and the creator stops studying the format that just succeeded.
Keeping the baseline honest over time
Recompute the median on a rolling window rather than freezing it. On a growing channel a static baseline flags almost everything after six months; on a declining one it flags nothing. Split by format before computing, because Shorts and long-form are different distributions that should never share a denominator.
Worked examples
The same channel, two baselines
Nineteen uploads sit between 8,000 and 12,000 views, and one reached 120,000. The median is about 10,000; the mean is about 15,500. A new video at 24,000 views scores 2.4x against the median and flags, but only 1.55x against the mean and is silently discarded.
Where the baseline shows up in practice
Every outlier score, threshold and weekly research queue depends on this one number, so an error here propagates into every decision you make about what to publish next.
When a channel's flagged videos suddenly stop appearing, check the baseline before concluding anything about the channel. In most cases the median has drifted, not the content.
Frequently asked questions
Related reading inside Outlieo
How the baseline, window and threshold combine into a flag.
Choosing between 2x, 3x and 5x once the baseline is right.
The channel-relative definition this baseline serves.
Ship a week of content, not a to-do list.
Join the Outlieo waitlist. Track the outliers, script in your voice, and repurpose to LinkedIn.