The deadlock
Why more evidence does not help
Five years of studies, both sides better armed every year, and no convergence. The reason is not that the evidence is bad.
Claims this part is about, and what each was answered with
Being trusted to choose is part of being treated as an adult
Value≠A studyInnovation happens in unplanned conversations
Empirical≠A principleAnswered with a conviction about how innovation works
Flexibility is what the last five years were for
Value≠A studyCollaboration tools have replaced what the office used to do
Empirical≠A principlefour claims · two empirical and two value · four were answered with the wrong kind of answer
Four crossed answers, two in each direction.
Every year there is more of it. Large-sample studies, natural experiments, payroll data, surveys of hundreds of thousands of people. Both sides are better supplied than they were, and the disagreement is exactly where it was.
If the dispute were empirical this would be very strange. It is not strange at all once you count what the evidence can reach.
What the evidence can reach
On this record, nine of twenty-four claims are empirical. Evidence can bear on those nine and on nothing else. Ten claims are about what matters and five are about who pays, and no study addresses either.
So the maximum possible contribution of the entire research literature, applied perfectly, is to settle nine claims out of twenty-four — and in fact rather fewer, because for four of the nine no evidence exists anywhere and for three the evidence cited measures something else.
Two claims. Out of twenty-four. That is the ceiling.
The three ways a study misses
Worth separating, because they fail differently and only one of them is a flaw in the study.
Wrong question. The study is good and is about something else. Most citations in this argument are of this kind: a paper about company-level output quoted at a claim about team-level collaboration.
Wrong population. Sound work on a group that does not resemble the one in the room. Sick-pay behaviour in one country quoted at a claim about absence in another with entirely different rules.
Wrong kind. The claim is not empirical at all, so no study of any quality could bear on it. This is the commonest failure in this argument and the one most invisible, because the study is genuinely good and the person quoting it is genuinely informed.
| How it fails | Is the study at fault? | What would fix it |
|---|---|---|
| Wrong question | No | A different study |
| Wrong population | Sometimes | A study of this population |
| Wrong kind | No | Nothing; stop citing studies here |
What to do instead of citing
When somebody produces a study at you, there is a reply that is not a counter-study and is much harder to answer:
What claim do you think that settles?
Not aggressive, and it forces the useful step. Three things can happen. They name a claim the study addresses, in which case you have a real empirical dispute and can look at the paper properly. They name a claim it does not address, which becomes obvious as they say it. Or they find they cannot name one, which is the most common outcome and ends the citation without anybody losing face.
The uncomfortable corollary
If evidence can only ever settle a small part of this argument, then getting better at evidence will not end it, and people who are good at evidence need to hear that.
The temptation for a numerate person in this discussion is to be the one who has read the literature properly. That is worth doing and it caps out at two claims. The remaining work is of a different sort entirely: naming values, naming costs, and trading. Those are not lesser activities, they are just not the ones a research-shaped mind reaches for first.
What would change if the evidence improved
A thought experiment that clarifies the ceiling.
Suppose a perfect study appeared tomorrow: randomised, large, long-running, measuring exactly the right things at exactly the right unit, with an unambiguous result in one direction.
Nine claims on this record would be affected. Of those, four are about mechanisms the study would not address unless designed for them. So perhaps five claims move.
The other nineteen are exactly where they were. Ten of them are about what matters, and five are about who pays, and a perfect study speaks to neither.
The corollary for research
This is not an argument against doing the research. The five claims are worth settling and the literature on early-career outcomes has already done real work.
It is an argument against the expectation attached to it. People waiting for the evidence to settle this are waiting for something that, even in the best case, settles a fifth of it.
The remaining four fifths need a different activity, and the discipline most useful there is not methodology but negotiation.
What would actually help
If evidence caps out at a fifth of the argument, the question is what addresses the rest.
For the value part: people stating their values plainly and then trading. That is a negotiation and it is available immediately.
For the cost part: somebody saying the figures. Also available immediately, and on this record never done.
Both are unglamorous, neither requires reading anything, and between them they cover four fifths of what is in dispute.
Why the research-shaped response persists anyway
Because it is the response available to a person acting alone.
Trading requires a counterparty. Stating costs requires standing or nerve. Reading a paper and bringing it back requires neither, and it feels like contributing.
So the energy in these arguments flows to the one activity that can be done unilaterally, and that activity is the one with the smallest possible ceiling.