Open government records

Why does a right to inspect records change official decisions before anyone inspects them?

An access law commences. On the first morning, no request has been made and no file has been read by any outsider. Yet officials describe the day as different — a meeting minuted differently, a submission redrafted, a phrase struck out.

Nothing was disclosed. So whatever changed cannot have been caused by disclosure. What is doing the work?

The obvious answer is that people behave when they might be watched. True, but too loose to be useful, because it predicts nothing about what changes. Tighten it by asking when an official actually decides how a record will read. Not when it is requested — by then it is written and filed. He decides at the moment of writing, so the only thing that can influence the record is his estimate, at that moment, of a future he cannot observe.

Put numbers on the estimate. Suppose one memo in fifty is eventually requested — chosen to make the arithmetic visible, not drawn from any register. For a single memo that is a small risk, easily ignored. But an official writes perhaps two thousand in a career, and two thousand divided by fifty is forty. Forty of his documents will surface. Notice what that does to the rational response: judging memo by memo is pointless, since no individual memo justifies the effort, while exposure is near certain in aggregate. So the sensible adaptation is a standing change in how everything is written — which is exactly what we observe, a general shift arriving before the first request.

Now the harder question, where honest accounts diverge from hopeful ones. In which direction does that shift run? The hopeful answer is candour, or at least defensibility: a man who expects to be quoted writes what he can defend. That does happen. But cheaper adaptations exist, and cheap ones get chosen. He can write less and brief orally. He can write somewhere the Act does not reach — a personal device, an informal message, a corridor. He can frame the content to fall inside an exemption, since access regimes protect deliberative advice for reasons that are themselves defensible. All three cost far less than improving the decision.

So the mechanism contains a feedback loop, and the loop is self-limiting. Decisions produce records; records produce exposure risk; exposure risk changes record practice; changed practice determines what exposure can reach. The more effective the right becomes, the harder it pushes content out of its own scope. That is the load-bearing claim: the anticipation effect operates on the location and form of the record at least as much as on the substance of the decision.

What would refute it? If, after a law commenced, internal records were unchanged in volume and candour and no migration to out-of-scope channels occurred, the anticipation channel would be doing nothing, and the whole effect would have to run through actual disclosure — appearing only where requests landed, and scaling with their volume. That test has been partly run, and the honest report is mixed: UK research on the 2000 Act found little hard evidence of a chilling effect on record-keeping, alongside plenty of officials who reported feeling one. Self-report and measurement disagree, and I will not pretend otherwise.

Two folk accounts should be set aside. Transparency deters wrongdoing is too narrow: most decisions touched by an access regime involve no wrongdoing, and the adaptations above are made by honest officials protecting ordinary candour. It made government accountable, full stop is a magnitude claim where magnitudes are contested and the counterfactual unobservable; Tony Blair's much-quoted regret at having legislated is an anecdote about one politician's temper, and I mention it only to decline to use it.

One thin figure deserves the same treatment. It is often said that most access requests come from journalists. I recall the claim but cannot confirm it, and it surely varies by jurisdiction. The argument needs only the weaker point that requesters are not a random sample of the public: because they are predictable in kind, the anticipated audience is specific, and officials guard against the requester they expect rather than the public at large.

Two effects are worth separating. The ex post effect is what a disclosure does once made: a story, an inquiry, a resignation. The ex ante effect is what the possibility does to the record before anything is disclosed. They are not the same size and need not run the same way, and it is the second that answers our question — which is why judging an access regime by what its disclosures revealed misses most of what it did.

Exemptions, too, are not simply holes in the law. A regime with no deliberative protection would drive candid advice out of the written record entirely, which is worse for the archive than one keeping that advice on file and withholding it for a period. Designing an access law is therefore not a matter of maximising what can be released, but of keeping content inside the recorded system, where a later release or inquiry can still reach it.

This piece stands almost exactly opposite a neighbour, Secret ballot, and setting them side by side sharpens both. That piece concludes that the ballot works by destroying evidence: make it impossible to prove how you voted and a vote cannot be bought or coerced. Access law works by creating evidence. Both are true, because both are engineering the evidence set rather than regulating conduct directly — and the fixed point of difference is who is being protected from whom. The ballot removes evidence to shield a weak party from a powerful observer; access law creates evidence to expose a powerful party to a weak observer. Their second-order effects then rhyme in a way that is genuinely useful: in each case the party who loses by being observed relocates the action beyond observation — vote-buying shifts to buying turnout, and candour shifts to the corridor.