Rules
Part of 5 things worth knowing about online status and identity
Online status and identity statistics are instrument readings, not proof
Online status and identity statistics are instrument readings: what a counter counts, why it stops describing anything, and why comparisons fail.
A counter beside a name looks like a measurement. It is closer to an instrument panel that the service designed, calibrates, and can recalibrate without telling anyone. Treating it as a fact about a person is the first mistake; treating two accounts' counters as comparable is the second.
This page is about what those numbers are, what changing them does to behavior, and how to keep a record of your own that survives the service changing its mind. No figures appear here, and the last section explains why none could.
What to take away
- A counter measures events the service defines. Change the definition and your history changes with it, retroactively and silently.
- The moment a number becomes a goal it stops describing anything, because everyone starts acting on it.
- Two accounts' counters are not comparable unless both were measured the same way, and you cannot check whether they were.
- If you want a record of your own standing over time, you have to keep it yourself, dated.
What a counter actually counts
Every visible number rests on a definition nobody publishes in full.
- A follower count counts accounts in a relation, including inactive ones, duplicates, automated ones and accounts whose owners have forgotten them.
- A view count counts an event whose threshold is a design decision: how much of a thing must be shown, for how long, and whether a repeat counts.
- A reaction count counts taps, which are cheap, and cheapness is the point of them.
- A score or rank counts a formula, which is the least stable of all, because a formula can be rewritten in an afternoon.
None of these is dishonest. Each is a real count of a defined event. The problem is that the definition is not visible to the reader and is not stable over time, so the number is a measurement whose units keep changing. What that does to the measurement is the ordinary problem of construct validity: the thing being counted and the thing people believe is being counted are not the same thing.
Why the number stops describing anything
The counter is public, so people act on it. Acting on it changes the behavior the counter was measuring. This is Goodhart's law, and it applies with unusual force here because the measure is displayed to the measured party in real time.
| Counter | What it measured before anyone optimized | What it measures once everyone does |
|---|---|---|
| Followers | Rough interest in a person | Willingness to trade a follow, plus history, plus luck |
| Views | Whether something got shown | How well it matched what the ranking rewards |
| Reactions | A weak signal of approval | A social obligation among people who reciprocate |
| Rank or level | Standing among peers | Time spent, mostly |
That last row is the honest summary of most rank systems: they measure persistence. The design literature is open about this, and the general account of adding scored progress to activities that had none is under gamification. Persistence is a real thing to measure. It is not what the number is read as meaning.
The retroactive problem
A service can change what it counts, and when it does, your past changes. A view threshold moves and a year of your history is suddenly larger or smaller. A cleanup removes inactive accounts and a follower figure drops overnight without a single person leaving. A formula is rewritten and a rank that took years is worth something different.
None of this is announced in a form that reaches the people affected, and none of it is reversible. That is the practical reason to keep your own record: the service's number is a current reading of a current definition, and it will not tell you when the definition moved. The general case for retaining prior states rather than overwriting them is made under platform timelines.
Keeping your own record is unglamorous and takes a minute a month. Note the counters, the date, and anything the service announced. What you get is a series measured by an instrument that changed, which is imperfect and is still infinitely better than a single current reading with no history at all.
Why comparisons between accounts fail
Two counters are comparable only if the same rule produced both. In practice the rule differs by account age, by region, by how the audience was gathered, and by whatever tests the service is running at that moment. So the comparison is between two measurements taken with two instruments, neither of which is described.
Add to that a selection problem: the accounts whose numbers you see are the ones the ranking chose to show you, which is not a sample of accounts. It is a sample of what performed.
Drawing conclusions about typical outcomes from it is survivorship bias in its purest form, and it is the mechanism behind almost every discouraging comparison anyone makes about their own standing. What the comparison does to the person making it is covered under critiques of social media.
Why this page carries no numbers
Any figure here would be one of three things. A number a service published about itself, under a definition it can change. A number from a study measuring a sample of a population nobody can define. Or a number somebody estimated. The first is not comparable to anything, the second cannot be generalized, and the third is a guess with a decimal point.
The same refusal, applied to counting attention rather than standing, is worked through under benefits of social media. The material worth having here is the mechanism, and the mechanism does not need a figure to be understood or used. What the underlying signals are and what each one buys is set out under online status and identity.
Common questions
Is a big number meaningless, then?
No. It reliably means something happened at scale. It does not mean the specific thing readers assume, and it cannot be compared with another account's number.
Should I stop looking at my own counters?
That is a personal decision and this page will not make it. The useful version is to record them and look at the series rather than the reading, because the series is at least about you.
Why do services not publish the definitions?
Publishing an exact definition of a scored quantity invites people to optimize against it directly, and the definitions also change often enough that a published one would be wrong quickly. The effect on readers is the same either way.
Can a third party measure this better?
Only from the outside, by sampling what is publicly visible, which reintroduces every problem above and adds a worse sampling frame. Outside measurements of these systems are estimates of a surface.
