Wartime turnover of commanders

Why do the generals who begin a war so rarely finish it?

The Union's principal eastern army went through McDowell, McClellan, Burnside, Hooker and Meade inside two years. The British Eighth Army in North Africa churned through commanders before Montgomery. The US Army of 1941 relieved a great many division commanders in its first year of fighting, and the men who ended the war in senior command were frequently unknown when it started.

The instinctive reading is that peacetime armies promote the wrong people, and war exposes the mistake. But hold that up to the light. These were the men their institutions had spent thirty years assessing, on exercises, in staff colleges, under close observation by peers who had every professional reason to judge accurately. Are we really to believe the assessment was simply careless — and careless in the same direction, in every army, in every century?

Let us try a different frame. Instead of asking whether the peacetime army chose badly, ask what it was choosing for.

Think about what a peacetime army actually does. It trains, it administers, it manages budgets and garrisons and equipment programmes, it maintains standards, and it does all this for decades without the thing it exists for ever occurring. What traits does that reward? Reliability. Administrative competence. Not producing incidents. Doing well on the assessments the institution can actually run. Notice that last one — an organisation can only select on tests it is able to administer, and no peacetime test can administer the real one.

Now ask what a war rewards. Willingness to act on bad information. Tolerance for casualties in one's own command. Adaptation when the doctrine one was taught stops working. Nerve at three in the morning after four days without sleep. Are these the same traits? They overlap — competence is not irrelevant — but they are plainly not identical, and nothing in the peacetime evaluation observes them directly.

So we have two different selection processes with different criteria, and the second one begins abruptly on the first day of the war. Put that way, high turnover stops looking like a failure and starts looking like the arithmetic you would predict. If the peacetime tests correlate only loosely with wartime performance, then the people at the top when the war starts are a roughly average draw from the pool as measured by the new criterion, and most of them will be outperformed by someone.

Push once more, because the second process has a property the first lacks. War is a test that runs continuously, on everybody, with unambiguous scoring and no waiting for the next appraisal cycle. What does that do to the rate of change? An army that will relieve a commander after one bad fortnight samples its officer corps far faster than one that promotes on a five-year board — so the turnover is not only the correction of an old error, it is the visible surface of a selection process running at high speed.

That also explains the direction of the churn. Relief creates vacancies, vacancies are filled from below, and the men filling them have already been observed in combat. Each round the pool of candidates is better measured than the last. The turnover slows not because armies become sentimental but because the information improves.

Two honest caveats. First, not every relief is selection. Some are scapegoating — a commander sacked for a failure that belonged to the plan, the intelligence, or the resources — and from outside the two are extremely hard to tell apart, which is a real limit on the tidy account above. Second, turnover has a genuine cost: a relieved commander takes accumulated local knowledge and unit trust with him, and an army that relieves too readily can churn away more than it gains. Where that balance sits is argued about, not settled.

The point is not that peacetime armies are foolish or that "fighting generals" are simply a different breed. It is that selection can only run on the tests available, and the available test changes character overnight. That reframing does two useful things.

It corrects the idea that better peacetime evaluation could eliminate the churn. It can narrow it — realistic force-on-force exercises, hard umpiring, and deliberate observation under stress all improve the proxy — but a proxy that fully captured the real test would have to be the real test.

And it corrects the mirror-image error: that a general relieved early was therefore incompetent. Frequently he was excellent at the job he had been selected for, in an institution that had no way of knowing whether that job resembled the one about to arrive. The failure sits in the correlation between two tests, not in the man.

• How armies try to narrow the gap in advance, through free-play exercises and deliberately unscripted umpiring.
• Relief as scapegoating, and whether the two functions can be separated even in retrospect.
• The same pattern in other fields where the real test is rare: emergency services, crisis leadership, and the selection of political leaders.

Every term the collection defines is gathered in the glossary.