The numbers everyone repeats · 6.7

What can actually be known about work

What can actually be known about work. What is actually the case, and how it compares with what is repeated.

For a practical software perspective on this topic, the Monitask overview provides a concise companion guide.

Forty-two entries in, the useful summary is not a list of conclusions but an account of what kinds of thing can be established about work, and what kinds cannot.

What the good sources are

Labour force surveys, which are large, repeated, methodologically documented, and comparable with themselves over time. Administrative data, meaning tax and social insurance records, which cover everybody and do not depend on recall. Linked employer-employee datasets, which are the best thing in the field and exist in a handful of countries. And randomised evaluations where somebody has managed to run one, which is rare in employment and increasingly less so.

What is genuinely well established

The composition of employment, which is measured directly and is the basis of the argument this site is built on.

That the gender earnings gap in countries with population registers opens at the first child and does not close.

That external hires are paid more and perform worse initially than internal promotions.

That the largest component of variance in performance ratings is the rater.

That structured interviews predict job performance considerably better than unstructured ones.

That schedule unpredictability has effects on household wellbeing independent of hours.

Each of those has been shown more than once, by different people, using different data.

What is genuinely contested

The employment effects of minimum wages at high ratios to the median. The causes of the productivity slowdown. The net employment effect of any particular technology. The direction of causation between engagement and performance.

On each of these, confident public statements are made in both directions by people who know the literature, which is the definition of a contested question rather than a settled one.

What cannot currently be known

Individual productivity in most jobs, for the reasons the part on measurement gives. The counterfactual of any particular organisational change, because nobody ran the control. And anything about the future of occupational structure beyond a few years, on the record of every previous attempt.

Saying so is not defeatism. A field that is clear about its limits is more useful than one that answers everything.

The three questions

Which jobs. The first question in this site and the last. Most claims about work are claims about a subset, and naming the subset usually rescues the claim while shrinking it.

What was measured. Not what the finding is called. The unit, the population, the period, the definition. The gap between what a study measured and what its title suggests is where most misreporting lives.

Who benefits from the number. Not an accusation. Vendor research is not worthless and it is research with an interest, and knowing whose it is changes how much weight the figure carries.

What this site has been doing

Applying those three to a subject where they are rarely applied, stating the share of employment each entry concerns before arguing anything, and naming the source behind each figure at the foot of each page.

None of that is difficult and very little writing about work does it. That gap, rather than any particular finding, is what the forty-two entries were for.

The last word

Work is the thing most people spend most of their waking life doing, and the public conversation about it is conducted largely about a minority of it, using figures that mostly do not survive being looked up.

Both of those are fixable by anybody, with a search engine and a quarter of an hour.

Where to look things up

National statistical agencies publish more than anybody reads, including the methodology notes that answer most of the questions in this part. International bodies publish comparable series with documented definitional caveats.

Almost everything cited across this site is publicly available without payment, which is worth knowing because the secondary coverage frequently is not, and is worse.

What would improve the field

More linked employer-employee data, more randomised evaluation of workplace interventions, and time-use surveys run on a regular cycle rather than when a budget allows.

All three are expensive, none is glamorous, and each would settle questions this site has had to leave open.

How to disagree with this site

Every entry names its sources and states what they measured. If a conclusion here is wrong, the route to showing it runs through those, and that is deliberate: an argument that cannot be checked is not worth making.

For broader context, consult IMF labour-market resources.