Civil service tenure

Why does making officials nearly impossible to dismiss produce more honest advice than making them removable?

Everywhere else we assume accountability improves performance: an employee who can be dismissed works harder than one who cannot. Yet governments build the opposite into their permanent staff, and defend it on the ground that it makes the advice better.

Either that is a guild's self-serving story in constitutional dress, or job security does something ordinary employment cannot. Let us find out which by asking what kind of thing advice is.

Consider a minister who wants a policy and an official who has studied it. Suppose the official's estimate is three times what the minister has announced. Ask what happens to that estimate under two employment rules.

Under dismissal at will, the official weighs two options. Report the estimate: the minister may value the warning, or may resent it, and the resentment falls on a person whose livelihood the minister controls. Report something softer: no risk today, and the overrun arrives years later, attributable to a hundred causes. The safe option is soft, and it is safe in every case regardless of what the official believes — so the choice does not depend on the underlying fact at all.

That last point is the whole mechanism, and it is worth stating carefully. If the adviser always says the pleasant thing, then hearing the pleasant thing tells the minister nothing. The advice has stopped carrying information — not because the official is dishonest, but because the channel transmits the same signal whatever the input. An adviser who cannot safely disagree is an adviser whose agreement is worthless.

Now impose tenure. Dismissal requires cause, proved through a procedure the minister does not control. The cost of unwelcome advice drops toward zero, so the official's choice can once again turn on what he actually believes. Agreement now means something, because disagreement was available.

Notice what tenure is doing. It is not making officials braver, and it is not assuming they are more virtuous than anyone else. It is removing the penalty that made the message independent of the truth. This is the same logic that argues for insulating auditors, statistical agencies and central banks: the output is valuable only if the producer could have said something the principal did not want to hear.

There is a second, cruder function, historically the first. Where posts are rewards for political support, a change of government replaces the staff and the state loses whatever the departing officials knew. The nineteenth-century reforms that established permanent merit services — Northcote-Trevelyan in Britain, the Pendleton Act in the United States — were aimed at patronage first and candour second. The two purposes are distinct and can pull apart, which matters when reformers propose loosening tenure: they may be attacking the second while dismantling the first.

What would falsify this? If tenure works by making unwelcome advice safe, observable candour should track the strength of protection. We should see more written internal dissent, more published cost estimates that embarrass the government of the day, and more reported willingness to contradict a minister where protections are stronger — and see those decline after protections are weakened. The refuting observation would be equal candour under at-will arrangements: if political-appointee-heavy administrations produced just as much on-record disagreement, the mechanism would be dead, and tenure would be a benefit with a story attached rather than a design.

I should be honest that measuring candour is hard, and that the comparative work here is contested. What is better established is the patronage half: turnover and the loss of institutional knowledge at transitions are easy to observe, and they are worse where posts are political.

Two clarifications, then the price, which is large.

First, tenure is not immunity from direction. The permanent official is obliged to implement the policy chosen, including one he advised against; the protection covers the advice, not the outcome. Systems that blur this — where officials treat protection as licence to resist decisions rather than to inform them — lose the argument for tenure entirely, because the case rests on separating the two.

Second, it is not all-or-nothing. Arrangements differ in where they cut the line: some protect the entire hierarchy up to the department head, others put department heads on fixed-term performance contracts while protecting everyone below, others allow a thin political layer atop a protected service. These are genuinely different designs with different failure modes, and the comparison is the evidence — there is no natural arrangement here that others deviate from.

Now the cost. The same procedure that makes unwelcome advice safe makes poor performance safe. Dismissal for incompetence requires proving cause through a process designed to be hard, so managers avoid it, and the burden of a weak colleague falls on those around him. And an insulated service accumulates its own settled views; a government elected to change direction may find that the estimates, the options papers and the implementation timetables all quietly favour continuity. Whether that is expertise resisting a bad idea or an unelected body resisting a mandate is the contested question, and it is genuinely contested — one side calls it institutional memory, the other calls it an unaccountable veto, and the same episode can honestly be described either way.

There is also the point that should be said plainly. Any proposal to loosen tenure must be designed and administered largely by the people it would expose, which biases the process toward preserving it. That does not make the underlying argument wrong — an argument is not refuted by noting who benefits from it — but it does mean the survival of a particular set of protections is weak evidence that those protections are set at the right level.

• How auditors, statistical agencies and central banks solve the same credibility problem with different instruments — fixed terms, publication rules, or removal only for cause.
• Whether a thin political layer over a protected service captures most of the benefit of both.

Every term the collection defines is gathered in the glossary.