Demand vs Retention
Where what players ask for fails to keep them playing
One report, two readings joined on the same vocabulary: the share of public requests each design inclination carries, against the coefficient it earns on a title still being played. A demand can be loud and empty, or quiet and load-bearing — and the gap between the two is the whole point.
Titles scored
46
Asking measured
100.0 pp
Asked, not retained
2
Retained, never asked
0
The reading
Campaign & narrative takes 25.0 of every 100 points players ask for and still moves survival by r = -0.14 over 43 titles. What does travel with survival is mod support & tooling, on 17.9 points of asking (r = 0.31).
The quadrant map
Across, how much of the asking a factor carries. Up, how strongly it moves survival. The four quadrants are the four answers: ask for it and it works, ask for it and it does nothing, ask for nothing and it still works, or neither.
A hollow point sits at the left edge because the stated layer carries no counterpart to ask about — it is not a measured zero. Both bars are the platform's own: half the asking mass, and the coefficient at which a relationship is claimed at all.
The density table
Density is asked points divided by retention points: how much noise stands behind each point of measured association. A high figure is demand that costs the audience a lot of words and returns almost nothing.
| Inclination | Asked (pp) | Retention r | n | Evidence | Density | Verdict |
|---|---|---|---|---|---|---|
Campaign & narrative | 25.0 | -0.1436 vs 40 survival | 43 | measured, no relationship | —pp asked per point | Asked for, not retained |
Balance & competitive play | 21.4 | 0.0136 vs 38 survival | 43 | measured, no relationship | 21.40pp asked per point | Asked for, not retained |
Mod support & tooling measured as “Mod support” | 17.9 | 0.3144 vs 37 survival | 43 | medium | 0.58pp asked per point | Asked for and retained |
Factions & unit rosters measured as “Faction identity” | 14.3 | —no medians | 43 | too few titles | —pp asked per point | Not testable |
Base building & economy | 10.7 | 0.0741 vs 38 survival | 43 | measured, no relationship | 1.53pp asked per point | Quiet and flat |
Smarter opponents & AI measured as “AI quality” | 7.1 | 0.3161 vs 38 survival | 43 | medium | 0.23pp asked per point | Asked for and retained |
Co-op & multiplayer | 3.6 | 0.1539 vs 38 survival | 43 | measured, no relationship | 0.24pp asked per point | Quiet and flat |
Graphics & presentation | — | -0.2934 vs 41 survival | 43 | measured, no relationship | —pp asked per point | Asking unmeasured |
Retained, never asked
These travel with survival, but the stated layer never polled a matching inclination — so nobody asked, and nobody was asked. They are listed rather than plotted as zero demand.
| Factor | Retention r | n | Evidence | Verdict |
|---|---|---|---|---|
| Community content | -0.01 | 46 | measured, no relationship | Quiet and flat |
| Match tempo | — | 43 | too few titles | Not testable |
| Unit readability | — | 43 | too few titles | Not testable |
| Replayability (playtime depth) | — | 0 | too few titles | Not testable |
Method and limits
How the two readings are joined
- Asked — percentage points of observed public requests, from the Insight Engine's stated layer. The stated inclinations sum to 100 across what it could classify.
- Retained — the point-biserial coefficient (r) between a factor being present and the composite survival score, with its own n. Every coefficient is printed beside the number of titles behind it.
- Density — asked points divided by retention points (r × 100). Low means the asking is cheap to satisfy; high means a lot of noise for very little measured return.
- Bars — 12.5 pp divides the conversation in half, and r = 0.3 is the coefficient this platform requires before calling a relationship at all.
Stated layer read Oct 27, 2026 · retention cohort read Nov 9, 2026 · 46 titles scored
What this report cannot say
- Correlation is not cause. A factor that travels with survival can do so because both follow from budget, genre, era or marketing, and nothing here separates those.
- The two layers are read from different populations: what players say is read from public requests, what retains is read from measured RTS titles. A mismatch is a disagreement between two readings, not proof that anyone is wrong.
- The stated layer is a snapshot of observed requests, not a census of the audience. A quiet factor can be genuinely unwanted or simply unasked.
- Survival is 40% age and 60% current activity. Lifetime playtime depth is excluded this run because SteamSpy no longer publishes it, so very long play sessions cannot lift a title here.
- Factors are read from storefront tags and community mod copy. Tags are player-applied and overlap, so a title can carry a factor's tag without that factor describing its design.
Not plotted, because one of the two readings is missing: Factions & unit rosters, Graphics & presentation, Match tempo, Unit readability, Replayability (playtime depth).