Regulatory ossification

Why does an agency hedged about with safeguards against bad rules end up making almost no rules at all?

Take each safeguard on rulemaking one at a time and it is hard to object to any of them. An agency should publish what it proposes, take comment, answer the serious objections, estimate the costs and the burden on small firms. A court should be able to set aside a rule the agency cannot justify on the record it built.

Every one of those was adopted after a real failure. And the aggregate is an agency that, on the matters that matter most, produces very little. The interesting question is not whether the safeguards are good — it is why good safeguards compose so badly.

Start with the base. The United States Administrative Procedure Act of 1946 asked for something modest: publish a notice, take comments, publish the rule with a concise general statement of its basis. That is a few pages of obligation.

Now watch what accumulated on top, each layer answering a genuine grievance. Courts moved from checking that the steps were taken to checking whether the agency had actually weighed the alternatives and the evidence — the "hard look", of which the Supreme Court's 1983 State Farm decision is the usual landmark (I state these landmarks as recalled). Analytic requirements arrived separately: environmental assessment, small-business impact analysis, central cost-benefit review inside the executive, which in the American case dates from an executive order in 1981. Each arrived on its own timetable and none replaced anything.

Here is the compounding. If a rule must clear n stages and may be delayed, sent back, or abandoned at each, its chance of emerging is the product of the stage-by-stage survival chances — so adding stages does not add cost linearly, it multiplies survival downward. And because the last stage is a court reviewing a written record, every earlier stage must be performed to litigation standard: not "consider the alternatives" but "document consideration of every alternative a future plaintiff might name". The record is built for the adversary, not the decision.

Now the question that turns this from a complaint into a mechanism. What does an agency do when the front door becomes expensive? It does not stop acting. It has other instruments: guidance documents, interpretive rules, enforcement priorities, licence conditions, case-by-case adjudication — and inaction. Notice what those share. They are cheaper because they carry less notice, less comment, thinner review. So procedures adopted to make agency power accountable end by moving that power into forms that are less accountable. The safeguard did not reduce discretion. It relocated it.

That is the first price. The second is symmetry: deregulation is ossified too. Withdrawing a rule is itself a rule, and on the American understanding it must survive the same reasoned-explanation review as making one. Whoever likes the status quo — and someone always does — has the delay working for them, whichever direction reform runs.

Who keeps it this way? Largely the parties advantaged by it. A regulated incumbent that would lose from a new rule funds the litigation that makes the record requirement expensive; the delay is not a side effect of its strategy, it is the strategy. Legislators often prefer constraining an agency to legislating themselves, since a procedural requirement is cheap to enact and its costs land on someone else's budget. Each safeguard has an attentive constituency and its removal has none.

The comparison is evidence rather than decoration. A British statutory instrument is typically laid before Parliament and takes effect unless objected to, with no general obligation to take public comment and much thinner judicial scrutiny of the reasoning — and rules issue far more quickly. The speed is bought with less participation and less external checking, not with better agencies. Neither arrangement is the natural case; each is a different settlement of the same conflict.

Three refinements hold the account together.

First, ossification is a claim about major rules, not all rules. Most rules are technical, uncontested, and move quickly. The burden concentrates where the stakes are large enough for someone to litigate — exactly where the safeguards were meant to bite, and exactly where substitution into softer instruments is most tempting.

Second, this is not the same phenomenon as a rule that will not die once made. That mechanism — diffuse costs, concentrated defenders, nobody willing to sign for repeal — is worked through in persistent-rules.md. The fixed point of difference is which act is being taxed. There, the expensive act is removal; here, the expensive act is promulgation, and the diagnostic symptom is not an old rule surviving but a new one never appearing in the form the statute contemplated.

Third, the size of the effect is genuinely contested. The strong thesis, associated with McGarity's writing from the early 1990s, holds that procedure has made significant rulemaking nearly impossible. Empirical work by Yackee and Yackee around 2010 challenged it, finding that most rules complete in unremarkable time. Both readings are live, and the disagreement is partly about which rules count. I state the attributions as recalled and give no completion-time figures: published counts vary sharply with which dockets are included.

That gives a testable claim. If procedural burden is doing the work, agencies facing heavier analytic and review requirements should shift a larger share of their output into guidance and adjudication than agencies exempt from them, and the shift should track the requirements rather than the subject matter. The refuting observation: if agencies exempt from the heavy machinery show the same substitution and the same output rate, then something else — budget, staffing, political attention — is the binding constraint.

• Why guidance documents attract their own procedural requirements once they become the main instrument, and whether the cycle restarts.

Every term the collection defines is gathered in the glossary.