Use the median to compare a typical post when one viral result pulls a creator's average upward. Keep the arithmetic mean beside it to show the contribution of every post. Add the post count and observed range. Together, these numbers let you compare routine performance without hiding a genuine breakout.
The choice depends on your question. For a single planned placement, start with the median as a description of past typical performance. For total results across a completed set of posts, retain the total and mean. Neither number guarantees what the next sponsored post will deliver.
What the mean and median answer
The arithmetic mean is the sum of the observations divided by their count. The median is the middle observation after sorting. With an even count, average the two middle values. NIST's statistical handbook explains why extreme values pull the mean while the median is less sensitive to their size.
For creator selection, use them this way:
| Question | Number to lead with | What to keep beside it |
|---|---|---|
| What did a middle-ranked post deliver? | Median | Post count and full list |
| What was the average contribution per post? | Arithmetic mean | Total and median |
| How far apart were the smallest and largest results? | Minimum, maximum and range | The middle of the distribution |
| What will my next placement deliver? | No historical summary answers this alone | Comparable past work and a test plan |
The median does not erase viral posts. It keeps them in the sorted list without letting their magnitude determine the middle. It also gives limited information about exceptional upside, which is why the mean remains useful.
A reproducible comparison
The following data are hypothetical. Assume ten comparable videos per creator, using the same platform's view metric, format and observation age, with no paid distribution. Each column is sorted independently. A row does not represent matched publication dates.
| Sorted position | Creator A views | Creator B views |
|---|---|---|
| 1 | 8,000 | 14,000 |
| 2 | 9,000 | 15,000 |
| 3 | 9,000 | 16,000 |
| 4 | 10,000 | 17,000 |
| 5 | 10,000 | 18,000 |
| 6 | 11,000 | 19,000 |
| 7 | 12,000 | 20,000 |
| 8 | 13,000 | 21,000 |
| 9 | 14,000 | 22,000 |
| 10 | 104,000 | 23,000 |
Here are the calculated summaries of that hypothetical dataset:
| Measure | Creator A | Creator B |
|---|---|---|
| Posts | 10 | 10 |
| Total views | 200,000 | 185,000 |
| Mean views | 20,000 | 18,500 |
| Median views | 10,500 | 18,500 |
| Minimum | 8,000 | 14,000 |
| Maximum | 104,000 | 23,000 |
| Range, maximum minus minimum | 96,000 | 9,000 |
To reproduce the figures, enter the two columns into spreadsheet cells B2:B11 and C2:C11. Use =AVERAGE(B2:B11), =MEDIAN(B2:B11) and =MAX(B2:B11)-MIN(B2:B11), then repeat for column C.
Creator A's mean is 200,000 divided by 10. Its median is the average of 10,000 and 11,000. The largest post supplies 52% of A's total views.
If your brief prioritizes a higher typical result, B has the stronger historical evidence in this example. If you are examining total past delivery, A leads. A also has the larger observed spread. None of those statements establishes which creator will sell more products.
A sensitivity check shows the effect of one value. In a second hypothetical version, replace A's 104,000 with 14,000 while leaving every other value unchanged:
| Hypothetical version of A | Mean views | Median views |
|---|---|---|
| Original dataset | 20,000 | 10,500 |
| Largest value changed to 14,000 | 11,000 | 10,500 |
This change illustrates the arithmetic. It is not permission to replace or remove a real post in your report.
Define the sample before ranking creators
A median calculated from selected highlights still describes selected highlights. Modash's profile-assessment article points out that creator-supplied screenshots can leave you without poorly performing sponsored content. Ask for a defined set rather than the creator's best examples.
Choose one sampling rule before comparing results:
- Fixed publication period. Include every eligible post within a shared date window. This keeps the calendar comparable, but prolific creators contribute more posts.
- Fixed post count. Include the latest eligible posts up to a declared count and maximum lookback. This keeps counts closer, but less active creators may contribute older work or fewer posts.
A hypothetical review rule could be the latest ten eligible videos published within the previous 90 days. Ten posts is a workflow choice here, not a statistically proven minimum. If a creator has only four eligible posts, show four. Do not quietly extend their window until their results improve.
Record the format, content purpose, paid-distribution treatment, metric label, data source, extraction date and missing values. Keep sponsored work separate from unrelated organic content when the brief requires a sponsored-content comparison. Apply the same rule to every creator.
For a dated reporting specification, use the guide to choosing a reporting window before seeing results. For Instagram comparisons that require equal time to accumulate results, use the guide to comparing Reels at the same age.
The source of a number matters too. YouTube's counting documentation says realtime estimates can differ from watch-page counts, metrics can change during verification, and public view counts can include advertising views. A public count alone therefore cannot establish the no-paid-distribution assumption used in the example. Ask for supporting information or mark that condition unknown.
Investigate unusual posts without rewriting the record
NIST's guidance on outliers distinguishes erroneous observations from unusual but real ones. Correct a confirmed recording error. Investigate a large result before deciding how to analyze it.
Use these rules for a creator review:
| Situation | Treatment |
|---|---|
| A value was entered twice or mistyped | Correct the record and retain a correction note |
| A post belongs to an excluded format | Apply the predeclared format rule to every creator |
| Paid distribution is confirmed | Follow the same paid-distribution rule across the sample |
| A post performed unusually well or poorly | Keep it unless an independent eligibility rule excludes it |
| A value is missing | Mark it missing and disclose coverage; do not enter zero |
Deleting weak posts makes the retained sample look stronger. Deleting successful posts makes it look weaker. In both cases, choosing exclusions because of the outcome changes the question your summary answers.
If you calculate a result without the largest post to examine sensitivity, label it as a secondary calculation. Keep the full-sample mean, median and count visible. A ranking that reverses deserves further inspection of the post and campaign fit.
Put the decision on the shortlist
For each candidate, save the inclusion rule, eligible count, missing count, mean, median, minimum and maximum. Add one sentence explaining which statistic your brief prioritizes and why.
For the hypothetical example, that sentence could be: "Advance B for a brief prioritizing typical post delivery; review A's breakout topic before considering a separate placement."
Keep rates separate from counts. If your shortlist uses engagement rate, choose and record its denominator first. A median cannot repair inconsistent definitions.
Before contacting the next creator, write your sample rule and calculate both summaries on the same eligible posts. Investigate any creator whose ranking changes when you switch between mean and median.
