In 1998, Walter Korpi and Joakim Palme published “The Paradox of Redistribution and Strategies of Equality: Welfare State Institutions, Inequality, and Poverty in the Western Countries.” The paper aimed to tackle the argument that targeting welfare programs to those with low market incomes reduces inequality and poverty more than not targeting them in that way.

In their response, Korpi and Palme accept the basic math of the “targeting efficiency” claim, but argue that it fails to consider the political effects such targeting has on support for increasing the size of the welfare budget. If (1) big, non-targeted welfare states reduce inequality and poverty more than small, targeted welfare states, and (2) targeting causes welfare states to be small, then (3) targeting is actually worse for inequality and poverty than not targeting.

This paper started a fire in the welfare state discourse that still burns hot even today. A massive number of secondary papers have been written about the topic trying to prove or disprove its thesis with various cross-country regressions. And the basic argument has become boilerplate rhetoric for social democratic politics across the world.

But Korpi and Palme’s paper makes a key mistake that few have taken note of. In fact, once this mistake is understood, the whole argument of the paper and the discourse it spawned becomes moot.

Korpi and Palme accept the idea that, for a given welfare budget, targeted programs reduce inequality and poverty more than non-targeted programs. This claim is not true and the seemingly irrefutable math underlying it is based on accounting games.

The Greatest Trick in Welfare History

My preferred way to dissect the “targeting efficiency” argument is to imagine two people sitting at a table, one an advocate of non-targeted programs (aka universalism) and the other an advocate of targeted programs (aka means-testing). A piece of paper is given to each of them with the following challenge:

  1. Design a tax policy that raises exactly $100 billion of revenue. No additional taxes allowed.
  2. Design a cash welfare program for kids that costs exactly $100 billion. No additional spending allowed.
  3. The tax and welfare policy combination that reduces poverty and inequality the most wins.

Upon receiving the challenge, the means-tester and the universalist begin working on their $100 billion tax policies.

Both look at the criteria for winning — reducing poverty and inequality the most — and independently devise the same tax policy that optimally levies $100 billion from the richest people in society. After a few minutes, each reveals their tax policy to the other and have a laugh about the fact that they came up with the same thing.

Next, the means-tester and universalist begin working on their $100 billion welfare program for kids.

The universalist, being a universalist, naturally decides to take the $100 billion, divide it by the number of kids in the society, and then pay out the resulting dollar amount to every kid. Based on the universalist’s math, this is equal to $3,000 per kid.

The means-tester, being a means-tester, takes a different approach. Rather than providing $3,000 to each kid, the means-tester decides to provide $6,000 per kid except that kids living in families earning more than $50,000 per year will receive 5 cents less than the $6,000 benefit for every dollar they earn beyond $50,000.

After a few minutes, each reveals their welfare policy to the other. The means-tester starts to smile and states triumphantly that his welfare policy provides twice as much benefit to poor and low-income kids as the universalist’s welfare policy. Since the universalist and means-tester have the same tax policy, this difference in the welfare policy makes the means-tester the winner of the challenge.

After seeing the means-tester’s welfare policy, the universalist immediately begins to revise both of his policies. He adds a second line to his tax policy that says “5% surcharge tax on earnings beyond $50,000 for tax units with children, up to $6,000 of tax per child.” He changes his welfare policy so that the $3,000 per kid benefit is increased to $6,000 per kid.

Once the universalist reveals these modifications, a heated argument ensues:

Means-Tester (MT): The rules clearly state that taxes and spending cannot exceed $100 billion. Under your revised plan, you are raising $200 billion of tax revenue and spending $200 billion on the $6,000 universal child benefit. You are disqualified.

Universalist (UN): In your spending program, parents that earn beyond $50,000 of income lose 5 cents of benefit for every dollar they earn beyond $50,000. That “phaseout” is the exact same thing as my “tax.” Either both violate the rules that taxing and spending cannot exceed $100 billion or neither do.

MT: No, a phaseout is not a tax! It’s a spending reduction. I taxed and spent $100 billion to provide a means-tested $6,000 benefit. You taxed and spent $200 billion to provide a universal $6,000 benefit.

UN: No. We both taxed and spent $200 billion to provide a universal $6,000 benefit. It’s just that your 5 percent tax is applied at the point of benefit disbursal while my 5 percent tax is applied at the point of paycheck disbursal. Other than that administrative difference, our policies are exactly the same. In fact, now that I think about it, applying a tax on earnings at the point of paycheck disbursal is administratively easier than applying a tax on earnings at the point of benefit disbursal. This means my policy requires less spending on bureaucracy than yours does. I am amending my welfare policy. My lower administrative costs allow me to increase my benefit to $6,200 per kid. I now provide $200 more to poor kids than you do and am the winner of the challenge.

The means-tester and the universalist are unable to settle this dispute and so their two policy packages are submitted to the judge. Unfortunately, when selecting a judge, the organizers of the challenge looked for someone with extensive experience running tax-and-spending microsimulations. The judge is able to simulate the two policy packages and conclude that the math checks out in both: the means-tester really can fund a means-tested $6,000 benefit with his tax on the rich and the universalist really can fund a universal $6,200 benefit with his tax on the rich and 5 percent tax on families with earnings over $50,000. From the microsimulations alone, the universalist has won.

But the means-tester presses his case that the universalist’s policy violates the rule that taxes and spending cannot exceed $100 billion. The means-tester no longer argues that his policy is optimal in any welfare-maximization sense of the word, but instead argues that it satisfies the semantical requirements of the challenge rules. The judge has no idea how to settle this semantical dispute.

Eventually, it is decided that the only way to settle it is to determine how the national accounts would categorize the sums of money in the two policy packages. After a few hours poring through the national accounting guides, a clear answer emerges. Under the existing national accounting rules, the means-tester’s phaseout is counted as reducing spending while the universalist’s distributively identical and administratively superior tax is counted as increasing taxes. The universalist is disqualified and the means-tester is declared the winner.

Reject the Premise

Korpi and Palme’s mistake was failing to realize that “targeting” is just taxing by another name. Means-testers have not figured out how to better spend a fixed amount of tax revenue. Rather, in these debates, they use national accounting rules to allow themselves to tax more in order to spend more while preventing universalists from doing the same thing. Simply put: the whole idea of “targeting efficiency” and the simple math underlying it is just an accounting game.

Korpi and Palme’s argument that the politics of targeting result in smaller welfare states that are, on net, worse for the poor is a very elegant one and they muster impressive cross-country statistics in its favor. But it’s also not necessary at all. The premise that, relative to universal programs, targeted programs result in more poor relief for the same amount of spending is flatly wrong and should be rejected.

Amusingly, once the mechanics of “targeting efficiency” are actually understood, it is the means-testers, not the universalists, who ultimately have to rely upon Korpi and Palme type arguments about politics if they hope to make any kind of case for their technically inferior and costlier policy designs. Specifically, means-testers ultimately have to speculate that, despite their distributive similarities, the public responds differently to a phaseout than a tax, and that the political possibilities opened up by this public confusion enable so much more taxing (through phaseouts) and spending than universalist alternatives that it overwhelms the inefficiencies of targeting.

For now, nobody has had to make an argument quite like that because almost nobody who argues about these topics actually understands that “targeting efficiency” is an accounting trick. But maybe one day that understanding will be more widespread and we will get a hallmark paper about the paradoxical way in which implementing taxes as phaseouts, despite being administratively wasteful and reducing program participation among the poor, nevertheless results in more benefits for the poor than the more efficient universal designs.