Costs
Part of Telling whether a niche social network is still alive or just dormant
Measuring niche social networks data
Niche social networks data: use counts and consistent windows, account for snowball sampling, set thresholds, and record each observation.
On a small network, almost every number moves for unrelated reasons. Two regulars take a holiday and posting halves. One thread gets shared and the month looks like a breakout.
A percentage based on forty people is a sentence about four people. It can swing twenty points next week because one of them was busy.
The instinct is to measure harder. The right response is to measure differently: fewer ratios, more counts, longer windows, and a few thresholds that matter at this scale.
This page covers how to measure niche social networks data on a network of a few hundred people, and how the sample got in front of you.
What to take away
- At small scale, report counts and windows, not percentages. A rate computed on a base under a few hundred is mostly noise wearing a decimal point.
- Widen the window until the number is stable, then keep that window forever. Changing it later ends the series.
- The small networks you know about are the ones somebody told you about, which is not a sample of small networks.
Why the arithmetic behaves badly down here
Large services get stable numbers for free, because the law of large numbers does the work: individual variation averages out and a weekly figure means something. A small network has none of that protection. Every individual is a visible fraction of the total, so ordinary human variation, one illness, one holiday, one argument, shows up as a trend line.
Three consequences follow, and they are worth stating in the entry rather than discovered by a reader.
Percentages mislead. Forty active members and a rate quoted to one decimal place is a false precision that invites comparison with a service where the same figure means something completely different.
Week-on-week movement is noise. At this scale the honest reporting period is a quarter, and even then a single event can dominate it.
Extremes revert. A very good month is usually followed by an ordinary one, not because anything went wrong but because the good month contained an unusual event. Reading that as decline, or reading the good month as a new baseline, is regression toward the mean mistaken for a story about the community.
Measures that survive a small base
| Measure | How to take it | Why it works at this scale |
|---|---|---|
| Distinct posters per quarter | Count names in the quarter, not messages | Whole people, no denominator, resistant to one loud member |
| Median days between posts in a room | The gap, not the volume | Answers whether the place is alive without needing a total |
| Newcomers who got a reply | Count both numbers, publish both | Two small integers beat one rate computed from them |
| Longest current silence | Days since the last post anywhere | A single number that catches death earlier than any average |
| Names appearing in two consecutive quarters | A count of returners | Continuity, which is what small networks actually run on |
| Rooms with any activity | A count out of a stated total | Concentration into one room is the standard decline pattern |
Every one of those is an integer with a stated window. None of them can be inflated by a busy fortnight, and all of them mean the same thing next year, which is the only property that matters in a series.
Measures for a Small Base
- Distinct posters per quartercount names
- Median days between poststhe gap
- Newcomers who got a replypublish both
- Longest current silencedays since last post
- Names in two consecutive quartersreturners
- Rooms with any activitycount of stated total
Thresholds instead of trends
A trend needs enough points to be a trend. A threshold needs one observation, and at this scale thresholds do more work.
Pick a few ahead and write them down: a silence longer than a set number of days, a quarter with fewer distinct posters than a set floor, a newcomer count of zero, an operator unseen for a stated period. Crossing one is the event.
Record it as an event with a date. That puts it in the same shape as everything else in the catalog. It also avoids arguing about whether a wobbling line is going down.
Thresholds also make the honest limits obvious. A network of two hundred with a floor of ten distinct posters a quarter is a network you can describe as alive or not without claiming to know anything about its size, which you do not.
The sample problem, which no measurement fixes
You did not find these networks at random; someone mentioned one, which led to another, then a third. That is snowball sampling, the only practical way to build a catalog in this category.
It favors networks connected to other networks, networks with someone willing to talk publicly, and networks in your language and subject area.
The quiet ones are missing. So are the ones whose members deliberately do not advertise them, which for some subjects is most of them.
Record How Each Entry Arrived
- Recommended by a member
- Found through a link from another entry
- Found in a directory
- Found by search
Publishing that field lets a reader see the shape of the collection, and it turns an invisible bias into a stated one. The same reasoning applies to what any of the entries claim about themselves, which is the subject of the notes on checking a small network's four standard claims.
What not to record
No member counts. Small networks display them, and the figure counts registrations, and registrations on a network that has existed for years include a great many people who left without saying so. It is the number most likely to be quoted back at you and least likely to mean anything.
No comparisons with large services. Nothing measured here is comparable with anything measured there, for the reasons the notes on how platform figures are produced set out at length. A catalog that puts the two in one table has invented a relationship.
No growth claims. Growth needs a stable definition, a stable window, and a base big enough to move for reasons other than chance. You will not have all three.
No predictions. Whether a network will still be here next year depends mostly on one or two people's circumstances, and no measurement reaches that. The failure modes in the overview of small networks are almost all personal rather than statistical.
Recording it the same way each time
Use one row per observation with the window, the method version, and the date, exactly as the three-tier observation discipline describes for larger services. The tiers apply unchanged: what you counted yourself, what the network published about itself, and what nobody outside can know.
The one addition worth making at this scale is a confidence note in plain words. "Counted by hand from the public timeline; three rooms were not visible without an account" is a better companion to a number than any interval. It also stops a later reader treating a hand count as an audit.
Common questions
Is a hand count really good enough?
For a few hundred people, a hand count is more accurate than most automated collection, because you can see what you are counting and notice the thing the script would have silently skipped. It stops being practical somewhere in the low thousands, which is well past where most of these networks live.
The operator offered me their dashboard figures. Should I use them?
As published figures, quoted with their definitions and dated, never mixed into a column of your own observations. An operator sharing numbers is doing you a favor and it does not turn their definitions into yours.
How do I compare two small networks?
On the measures above, taken by you, in the same window, with the same method, and reported as counts. Anything else compares two different measuring instruments and calls the difference a finding.
A network looks dead by every measure but the members say it is fine.
Ask them what they use it for and where else they talk. Usually the answer is that the conversation moved somewhere adjacent and the network is now a noticeboard, which is a real and useful thing for it to be. Record what it is now rather than scoring it against what it was.
