Evidences — the generational half of the floor claim: that raw emotional payload is not just what wins at an engagement-tuned gate today, but a want the medium trains into cohorts raised inside it. Theory in Chapter 7 and the medium note; the mechanism is medium-shapes-want. Status — open (logged 2026-09-18). Design only; no data gathered. Cheapest first move — an age-period-cohort decomposition of existing long-run survey series on affective polarization and news-sharing preferences.
The claim that needs evidence: Chapter 7 argues raw emotion is the floor of what spreads at scale, and the medium note pushes further, that a medium run for a generation may be a co-writer of wants, not just a casting director selecting among standing ones. The note flags the co-writer claim as unearned, because the within-campaign evidence (Mercier’s mass-persuasion failures) cannot see decade-scale shaping. This doc tracks the analysis that could earn it.
Why age-period-cohort
Any change in what people want over calendar time is a sum of three things that a raw trend cannot separate:
- Age effects — people may reach for emotional, identity-flagged content more at some life stages than others, independent of when they were born.
- Period effects — a given year’s events (a war, an election, a pandemic) shift everyone at once, which is the casting-director reading the medium note already earns: *Psychology of Virality* shows the viral slot flipping from out-group animosity to in-group solidarity around the 2022 invasion.
- Cohort effects — people born into and formed by a particular media regime carry wants no prior generation had, because what got amplified through their formative years is what they learned to want.
The co-writer claim is precisely a cohort claim. The casting-director claim (the safe one) predicts period effects; the co-writer claim (the strong one) predicts a cohort gradient: generations raised inside engagement-tuned media hold a lower floor, reaching for raw-emotional and identity content at a higher baseline that persists as they age. An age-period-cohort decomposition is the standard instrument for pulling those three apart, and it is the shape of evidence the floor claim owes.
The measurement move
- Assemble a long-run repeated cross-section with a stable measure of the outcome: affective polarization thermometers, self-reported sharing of high-arousal vs. considered content, or coded emotional content of what respondents say they pass on.
- Fit an age-period-cohort model (with the usual identification caveats below), reading off the cohort term.
- The strong prediction: a monotone cohort gradient, with cohorts whose formative years fell after the engagement-tuned feed became dominant (roughly post-2010 adolescence) sitting at a higher emotional-sharing baseline, net of age and period. A flat cohort term with all the action in the period term supports only the casting-director reading, and the medium note should retreat to it.
Evidence in the wild (no new collection)
- ANES / GSS carry decades of affective-polarization and media-use items, enough for a cohort decomposition of the polarization outcome directly.
- Boxell, Gentzkow & Shapiro (2017) is the sharpest existing adversary and asset at once: they found polarization grew fastest among the cohorts least exposed to the internet, which cuts against a naive “the feed did it” story and must be explained by any cohort model here, most likely as an age or period effect swamping a still-forming cohort signal.
- Brady et al. (2017) and platform-shared retweet corpora give a behavioral (not self-report) emotional-content measure that could be cohort-tagged where age is observable.
- Cross-country timing. Engagement-tuned feeds arrived at different dates in different countries; the co-writer claim predicts the cohort gradient shifts with the local arrival date, a difference-in-differences the single-country series cannot give.
Confounds / threats to validity
- APC is not identified without a constraint. Age, period, and cohort are linearly dependent (cohort = period − age), so the decomposition rests on an assumption (a constrained term, a nonlinearity, a proxy for the mechanism rather than raw cohort). The result is only as good as that constraint, and the write-up has to defend it, not bury it.
- Self-reported wanting is not wanting. The paradox of virality is that people say they do not want high-arousal content to spread while it spreads anyway, so a survey measure of stated preference may move opposite to behavior; a behavioral outcome is safer but harder to cohort-tag.
- Reverse causation within the cohort. A cohort gradient is consistent with the medium shaping wants and with differentially emotional people selecting into heavy feed use; the cross-country timing test is what separates shaping from selection.
- The outcome may be measuring polarization, not the floor. Affective polarization is downstream of the emotional floor but not identical to it; a clean design needs a measure closer to “what emotional register does this person reach for,” not just “how much do they dislike the out-group.”
Status & next step
Open. Next: pick the outcome series (start with ANES affective-polarization thermometers, which have the longest clean run), specify the identification constraint before looking at the data, and run the decomposition. If a cohort gradient survives Boxell’s counter-pattern and the cross-country timing test, the medium note can promote its co-writer claim from plausible to earned; if not, it holds at casting-director and Part IV’s cross-cohort prescriptions inherit that limit.