Ethics training programs metrics: definitions, what four measures establish, and what can be claimed

Four things get measured about ethics training. Three of them are measuring the delivery system and one of them is hard, which is why the first three are on every dashboard.

This article says what each actually establishes and what a training function can honestly claim.

What to take away

The four things people measure

Completion

What it establishes: the assignment system works and people can click.

Why it persists: it is available, it is a percentage, and it goes up. It also answers a question that gets asked, which is whether everybody was told.

What it cannot do: distinguish a workforce that understood from one that did not open the tab. Any program reporting completion as its headline is reporting on its administration.

Keep it, at the bottom of the page, labeled as an input.

Satisfaction

What it establishes: whether people enjoyed it.

The specific trap is that satisfaction rises with production values and with sessions that avoid discomfort, while the sessions that change behavior involve disagreement and are often rated lower. Optimizing this measure actively degrades the program.

If you collect it, collect one question instead: was this about work you actually do? That answer is useful and it is not a satisfaction score.

Quiz and assessment scores

What it establishes: that the correct answer is identifiable when it is sitting next to three wrong ones.

Nobody at work is presented with four options, one of which is right. The task is to notice that a situation requires a decision at all, and then to produce a response. A quiz tests neither, and a high pass rate on first attempt usually means the questions were transparent.

Behavior change

The thing you actually want, and it cannot be isolated. Reports, incidents and exception volumes all move for reasons unrelated to training, and they move for opposite reasons at the same time: a rise in reports can mean the channel started working or that conduct deteriorated.

That does not make the program unmeasurable. It means the claim has to be narrower.

The measure that works

Sample a dozen people, three to six weeks after a session, from teams that received it. Ask one situation from your scenario bank, in conversation, and ask them to describe what they would do. Then one follow-up: what if your manager told you otherwise?

Score three things: did they name a route, did they name a person, and did they know what to do when the manager is the problem.

Why it works: it is recall rather than recognition, it is in their words, there is no form to fill in, and it cannot be gamed by clicking. It is also cheap, at about two hours a quarter.

Why it is rare: it produces uncomfortable numbers early, and it requires someone to make twelve phone calls rather than export a report.

What can honestly be claimed

A training function can claim three things and should refuse the fourth.

People know the route. Measurable by the method above, and the most valuable thing training delivers.

People have practiced a decision once before making it for real. Evidenced by session records and the cases used.

The curriculum reflects the organization's actual questions. Evidenced by the intake log and the changes it caused. What the log should contain, and how the rest of the program is audited, is set out in ethics training programs checklist.

That conduct improved. Not claimable. Too many other things move it, and a function that claims it will be held to failures caused by incentives and leadership that it does not control.

Saying the fourth is out of scope, in writing, at the start, is the single most useful thing a program owner can do for themselves. Why the surrounding conditions dominate the outcome is set out in ethics training programs and the incentive half of it in leadership ethics.

Reading a number that moved

The same order that applies to any conduct measure applies here.

A program that runs a session on reporting and then sees reports rise has not proved the training worked. It has proved that visibility changed, which is a real effect and a smaller claim.

What to show a sponsor

One page. The delayed recall result with the three components separated. The intake log's count of questions and one rule that changed because of them. One thing that got worse. And an explicit line about what the program does not control, naming the incentive most likely to override it.

An outside reviewer asks whether training was tailored to actual risks and changed anything, not how many people completed it. That is the standard in the Department of Justice's guidance on evaluating corporate compliance programs.

A one-page report built around recall and the intake log answers that directly, while a completion dashboard does not.

When material teaches a reporting route, whether that route is safe is a separate question with a partly legal answer. the UK's whistleblowing framework shows that for one jurisdiction. No training metric can substitute for it.

The audit that tests the rest of the program is this site's checklist for these programs. What can be measured about individual conduct more generally is in employee ethics.

Common questions

Our board asks for completion rates. What do we give them?

Give them the rate, labeled as an input, and put the recall result above it. Boards ask for completion because it is what they have been given; most will prefer the better measure once they see it.

Is anonymous survey data useful at all?

For finding out whether people think the channel is safe, somewhat, and treat it as directional. For finding out what people would do, no: a survey with visible good answers measures reading comprehension.

How large should the recall sample be?

A dozen people a quarter, drawn from different teams. You are looking for whether a route is known, and that becomes obvious quickly.