What Would Change It

Every lesson in this Part has ended at the same place: countries with similar technology, similar productivity and similar attitudes have markedly different distributions.

That finding has been reached from six directions now — the measures (8.1.2), the mobility rates (8.3.1), the discrimination estimates (8.4.2), the child penalties (8.5.1), the poverty rates (8.6.1), and the comparative test against the dependency thesis (8.6.2).

It means the distribution is not a fact of nature. This lesson asks what actually changes it, on the best available evidence — and what the historical record says about when it has changed, which is more sobering than the policy literature.

Three factories, one product, three wage distributions.

Three plants in three countries make substantially the same product with substantially the same equipment and comparable productivity per worker.

In the first, pay is set plant by plant. The lowest-paid worker earns close to the statutory minimum; the plant manager earns around twenty times that; the chief executive of the parent company earns several hundred times it. Roughly one worker in ten is covered by any collective agreement.

In the second, pay for the whole sector is negotiated between an employers' association and a union federation, and the resulting agreement is legally extended to every firm in the sector, whether or not it belongs to the association and whether or not its workers are union members. Coverage is near-universal. The bottom of the scale is well above the statutory minimum. The ratio from the bottom to the plant manager is around eight.

In the third, coverage is high, there is a works council with legal rights to information and consultation, and worker representatives sit on the supervisory board.

Now the crucial observation. The differences between these three plants are in the distribution of what the firm pays out, before any tax is levied or any benefit paid.

No redistribution has occurred yet. The tax and transfer systems of the three countries will then do further and quite different amounts of work on top.

And this is the distinction the whole lesson turns on.

Redistribution takes a distribution as given and moves resources afterwards, through taxes and transfers.

Predistribution changes the distribution the market produces in the first place — through the institutions that set wages, the rules of corporate governance, the terms of employment, the distribution of assets, and the availability of the things people would otherwise have to buy.

Both work. They are measurable. And most public argument is about only one of them.

Where does the difference between countries actually come from?

Decompose it. Take two rich countries with very different disposable income inequality and ask how much of the difference is made before taxes and transfers and how much after.

The answer is: both, in proportions that differ by country and are frequently the opposite of what people assume.

Some countries with low final inequality get there mainly through enormous redistribution — their market distributions are not unusually equal, and their tax and transfer systems reduce the Gini coefficient by something like twenty to twenty-five points.

Others get much further before the state acts at all, through wage-setting institutions, and then redistribute less.

And countries with high final inequality typically do both less — their market distributions are more unequal and their systems reduce the Gini by nearer ten points.

Which means "should we redistribute?" is not the whole question and never was. The market distribution is itself an institutional product, and the institutions producing it are as changeable as tax rates — and considerably less visible, which is why they are argued about less.

Redistribution: what the evidence shows

Three findings, and two of them are counterintuitive.

One — transfers do more than taxes almost everywhere.

The intuition is that progressive taxation is the main redistributive instrument. In most rich countries it is not. Decompositions consistently find that cash transfers account for the larger share of the reduction in inequality, with direct taxes doing less — partly because tax systems are less progressive in practice than in schedule, and partly because transfers are targeted at the bottom by construction.

Which relocates the policy question: the size and design of transfers matters more, mechanically, than the top rate of income tax — a conclusion that fits neither of the standard political positions.

Two — the paradox of redistribution.

The intuition is that targeting benefits on the poor redistributes more per unit spent than paying them to everybody. Arithmetically that is true.

The finding is that the countries with the most universal systems achieve the most redistribution overall, and the reason is political rather than arithmetical.

Universal programmes have universal constituencies. Everyone receives them, everyone has an interest in their adequacy, and their budgets are large and defended. Targeted programmes have constituencies consisting only of the poor, who are the least politically powerful group; the budgets are smaller; the benefits erode; stigma attaches; take-up falls; and the middle class, receiving nothing, resists the taxation.

So a highly targeted system that redistributes efficiently per pound spent redistributes less in total, because it spends less. The efficiency of the instrument is dominated by the size of the budget, and the size of the budget is political.

This is one of the most useful findings in social policy, and it has held up reasonably well since it was formulated, with the qualification that the relationship has weakened somewhat as systems have changed.

Three — in-work benefits work and have an incidence problem.

Wage subsidies and in-work tax credits have well-identified effects: they substantially increase employment among single parents, and they raise the incomes of low-earning households.

And they have a documented incidence question. By subsidising low-wage employment, part of the benefit is captured by employers through lower wages than would otherwise have been paid — so the subsidy is shared between worker and employer in a proportion determined by how the labour market is structured. Estimates of the employer share are meaningful.

Which is precisely why predistribution and redistribution are not alternatives. An in-work benefit combined with weak wage-setting institutions subsidises low pay. Combined with a binding wage floor, it does not, because the floor prevents the wage from absorbing the subsidy.

Predistribution: the institutions that explain the variation

Four instruments, and the first is the most under-discussed.

One — collective bargaining coverage, and specifically extension.

Coverage varies across rich countries from around one worker in ten to nearly all of them — and the variation is not principally about union membership. It is about institutional design.

Where sectoral agreements are legally extended to all firms in a sector, coverage can approach universality with union density of a quarter or less. Where bargaining is firm-by-firm, coverage tracks membership and both have collapsed together.

The evidence linking this to inequality is strong. Century-long analyses using historical survey data find a substantial household income premium associated with union membership, concentrated among workers without degrees, and a clear inverse relationship between union density and top income shares over time. Compression of the wage distribution is what collective bargaining does, and the decline of coverage is among the best-supported contributors to rising wage inequality in the countries where it fell most.

Two — minimum wages, where the evidence changed.

The textbook prediction was that a wage floor above the market rate destroys jobs. The empirical literature has moved substantially, and the best modern designs — using the fact that a minimum wage increase creates a detectable displacement of jobs from below the new floor to just above it, so that the number of jobs lost and gained can be counted directly — find minimal employment losses for the increases actually observed, with the missing jobs below the floor almost exactly matched by new jobs at or just above it.

Two honest qualifications. The evidence is strongest for minimum wages at moderate levels relative to the local median — the range where most of the variation has occurred — and much thinner at very high ratios, where confident claims in either direction are not supported. And the theoretical explanation matters: where employers have wage-setting power because workers cannot easily move, a floor can raise wages without reducing employment, which is a specific model rather than a general licence.

Three — the education answer, and why it is necessary and insufficient.

The most influential account of rising wage inequality frames it as a race between education and technology: technological change raises demand for skills, and if the supply of educated workers does not keep pace, the skill premium rises. On this account the remedy is educational expansion.

It has real support — the timing of the American skill premium's rise does track a slowdown in the growth of educational attainment.

And it is insufficient for a reason this Part has established repeatedly. Educational expansion has been enormous and relative mobility has barely moved (see 8.3.1), because advantage relocates from whether to which as access equalises (see 8.3.3). Countries with similar educational attainment have very different wage distributions. And the largest recent increases in inequality have been at the very top, where the skill premium account has least to say.

Education is necessary, it acts on the wrong margin for the top of the distribution, and it does not by itself change relative positions.

Four — wealth and inheritance, where the instruments are weakest.

Wealth is the most concentrated and most heritable dimension (see 8.1.2, 8.3.2), and it is where policy does least. Inheritance and estate taxes raise small and generally falling shares of revenue in most rich countries, with extensive exemptions and avoidance. Taxes on wealth stocks have been tried and largely abandoned, on grounds of valuation difficulty, mobility of assets and administrative cost — with a few exceptions.

And in the short run, wealth inequality is driven substantially by asset prices, particularly housing. Which means housing policy is wealth policy, whether or not anybody says so — and the transmission channel from 8.3.3 runs directly through it.

The historical record, which is the sobering part

The argument that peaceful levelling is rare, and the case against it.

A prominent historical argument holds that large reductions in inequality across recorded history have followed four things: mass-mobilisation warfare, transformative revolution, state collapse, and catastrophic pandemic. On this account, violent shocks are what have compressed distributions, and periods without them have seen inequality rise or persist.

The evidence assembled for it is genuinely impressive in scope, and the twentieth-century case is strong: the great compression in Western countries coincided with two world wars, the destruction of capital, and the political settlements that followed.

And the criticisms are substantial and should be weighed.

It under-weights the institutional settlements themselves. The post-war compression was sustained for three decades by taxation, bargaining institutions and welfare states that were built deliberately and could have been built otherwise — and their subsequent dismantling, which was also deliberate, is what the reversal tracks. Attributing the compression to the war rather than to what was constructed afterwards is a choice about how to allocate causation.

It handles the Nordic cases poorly, where substantial compression was achieved through wage-setting and public provision without state collapse.

And the strongest counterexample is recent and regional. Across most of Latin America between roughly 2000 and 2014, income inequality fell substantially — an unusual and well-documented regional reversal in some of the most unequal countries on earth. The decompositions attribute it roughly equally to two things: a falling premium on skilled labour, following a large expansion of education, and expanded and better-targeted cash transfer programmes.

No war. No revolution. No collapse.

And the honest coda: the compression stalled and partly reversed after the commodity boom ended, which tells you something about how much of it depended on favourable conditions. The reversal is part of the evidence too.

Where this leaves the argument. Violent shocks have compressed distributions, and they are not the only thing that has. The historical claim is a useful corrective to the assumption that inequality declines naturally with development, and it overstates its case where institutional settlements did the work.

And the political economy, which is why any of this is difficult.

The standard economic model predicts that as inequality rises, the median voter — falling further below the mean — will demand more redistribution. This is not what has generally happened, and the failure of that prediction is one of the more interesting puzzles in political economy.

Four candidate explanations, all with some support.

People do not know the distribution. They underestimate concentration substantially and place themselves near the middle wherever they are (see 8.1.1), so the preference the model assumes is formed on a false picture.

The advantaged are organised and the disadvantaged are not. Opportunity hoarding operates politically as well as economically (see 8.1.1), and the constituency for defending an existing arrangement is always better resourced than the constituency for changing it.

Beliefs about desert differ, and they differ between countries in ways that track the policies. Where more people believe outcomes reflect effort, support for redistribution is lower — and the belief and the policy are mutually reinforcing.

And there is evidence that policy tracks the preferences of the affluent more closely than those of the poor when the two diverge — although this literature is contested, with critics pointing out that the preferences of income groups usually coincide, so the cases of divergence are relatively few and the estimates rest on them.

The proportionate reading: the median voter model fails, several mechanisms plausibly explain why, and the strongest of them is the simplest — most people do not know how unequal their society is, and the correction of that belief changes attitudes less than one would hope.

Because "nothing works" and "one thing works" are both wrong, and the evidence supports a specific and unglamorous position.

Five things that are well supported.

Transfers do most of the redistributive work, and universal systems redistribute more in total than targeted ones because their budgets survive.

Collective bargaining coverage compresses wage distributions, and coverage is determined by institutional design rather than by membership alone.

Moderate minimum wages raise pay without the employment losses predicted, and in-work benefits raise employment while partly subsidising employers unless a floor prevents it.

Public provision changes what a given income can achieve — which is 8.1.3's argument, and it is why the same income buys different lives in different countries.

And place has causal effects on children, established by converging designs (see 8.3.2).

Three that are not.

That education alone equalises. It has expanded enormously and relative mobility has not moved.

That any single instrument is sufficient. Advantage relocates when a channel is closed (see 8.3.3), which is why packages outperform levers.

And that a policy transplants. An instrument that works within one institutional configuration frequently does not in another — which is 7.4.2's external validity problem, and the reason for the third factory in the story.

Two questions to carry out of this Part.

Predistribution or redistribution? Most arguments about inequality are conducted entirely about the second, while the first accounts for a large share of the difference between countries.

And where does the advantage go when you close this channel? If the answer is "somewhere else", the policy will disappoint — not because it was wrong, but because it was alone.

One closing thing, and it is what the whole Part has been for. The strongest single fact in it is comparative: societies with the same technology have very different distributions, mobility rates, discrimination levels, child penalties and poverty rates. Every one of those differences is produced by arrangements that people made and can unmake.

That is not an argument for any particular politics. It is the removal of the argument that no politics is possible — and it is what a sociological education is for (see 1.5.1).

Three factories with the same product and productivity have different wage distributions before any tax is levied — because wage-setting institutions differ. Predistribution changes what the market produces; redistribution moves resources afterwards. Both work and most argument is about one.

On redistribution, three findings. Transfers do more than taxes almost everywhere, which relocates the policy question away from top rates. The paradox of redistribution: universal systems redistribute more in total than targeted ones, because targeting produces small constituencies, small budgets and eroding benefits — the efficiency of the instrument is dominated by the size of the budget, and the budget is political. And in-work benefits raise employment substantially while being partly captured by employers, unless a wage floor prevents the wage absorbing the subsidy.

On predistribution, four instruments. Collective bargaining coverage — determined by whether sectoral agreements are legally extended, not by membership alone — with strong evidence linking density to compressed wages and lower top shares. Minimum wages, where modern bunching designs find minimal employment loss at the levels actually observed, with honest uncertainty at very high ratios. Education, which is necessary, acts on the wrong margin for the top, and has expanded enormously without moving relative mobility. And wealth and inheritance, where instruments are weakest, revenues small and falling, and housing policy is wealth policy whether or not anyone says so.

The historical argument that large levelling has followed war, revolution, collapse and pandemic is impressive in scope and overstates its case: it under-weights the deliberate post-war institutional settlements, handles the Nordic compression poorly, and is contradicted by Latin America between roughly 2000 and 2014, where inequality fell substantially through education-driven falls in the skill premium and expanded cash transfers — with the honest coda that it stalled and partly reversed afterwards.

The median voter prediction fails, and the candidate explanations are that people do not know the distribution, that the advantaged are better organised, that beliefs about desert differ and reinforce policy, and — contested — that policy tracks affluent preferences where they diverge.

Five things are well supported: transfers, universality, bargaining coverage, moderate wage floors, public provision, and the causal effect of place. Three are not: education alone, any single instrument, and policy transplant across institutional configurations.

Predistribution / redistribution — changing the distribution the market produces; moving resources after it.

Paradox of redistribution — universal systems redistributing more in total than targeted ones, through larger and better-defended budgets.

Bargaining coverage / extension — the share of workers under collective agreements, and the legal mechanism extending sectoral agreements to non-member firms.

Bunching design — counting jobs displaced from below a new wage floor to just above it, to estimate employment effects directly.

Monopsony — employer wage-setting power, under which a floor can raise pay without reducing employment.

Incidence — who ultimately bears or captures a tax or a subsidy, regardless of who receives it.

Skill premium / race between education and technology — the wage return to education, and the account framing inequality as a contest between skill supply and technological demand.

Great compression — the mid-twentieth-century fall in inequality in Western countries.

Median voter model — the prediction that rising inequality produces demands for redistribution; empirically unsupported.

Policy transplant — the assumption that an instrument effective in one institutional configuration will work in another.

One — decompose your country. Find its market-income and disposable-income Gini coefficients. The difference is what the tax and transfer system does; the level of the first is what its labour market institutions do.

Two — check the coverage. Look up collective bargaining coverage in your country and in two others. Then look up union density in the same three, and note that the two figures are not the same thing.

Three — test the paradox. Take a targeted benefit in your country and ask who defends it politically. Then take a universal one and ask the same.

Four — follow the subsidy. For any in-work benefit, ask what would happen to wages in its absence, and who is therefore paying part of it.

Five — apply the closing test. Take one inequality you care about and find two countries where its magnitude differs substantially. Then identify what differs institutionally. If you can complete that sentence, Part 8 has done its work.