Trump is lying about dead people and Social Security

But we should fix this problem and have a central federal birth/death registry

Last month, Elon Musk complained that the Social Security Administration’s database contains millions of Social Security numbers for people who seem implausibly old but who are not marked as dead in the official files.

“Maybe Twilight is real,” he quipped, “and there are a lot of vampires collecting Social Security.”

This is the kind of thing that often happens when a person stumbles into an unfamiliar dataset. For example, I once tried to compute a “new homes per new people” ratio on FRED and noticed that there were consistently weird jumps in the ratio every January. I thought this might have something to do with the seasonality of home building, but when I looked into it, I learned that the Current Population Survey does a revision every January that creates a discontinuity in the population figures.

Literally, this chart says there was a mass death event in January of 2024. But we all know that this didn’t happen. By the same token, the US population did not increase by 2.8 million people in January of 2023. Instead, the population level estimate was revised upward by 2.8 million, and that is recorded in the database in a way that is annoyingly unhelpful.

By the same token, it’s true that the United States of America does not maintain a comprehensive federal death registry. As I have frequently complained, national data about the number of murders committed in the United States is incredibly laggy. The numbers on drug overdoses aren’t quite as delayed, but the most recent release (from late February) told us how many overdoses we had back in September.

In the United States, the tracking of births and deaths of livestock is much more timely and comprehensive than the tracking of births and deaths of humans, because livestock data is recorded by the Department of Agriculture, while collecting data on humans is a disaggregated state and county function. As a result, the CDC has one effort to figure out how many people are dying of drug overdose, while the FBI has another to figure out how many people have been murdered, and the Social Security Administration has a third to figure out how many people are ineligible for benefits due to death.

I find this to be a genuine government efficiency problem, and I think it would be great to create a centralized federal vital statistics database that could share relevant information with other agencies. This would reduce duplicative efforts, give decision-makers access to more up-to-date information, improve benefits administration, and constitute a genuine public good.

I was prepared to leave it at that, but Donald Trump repeated Musk’s claims in the State of the Union Address, again suggesting that money is being paid out to super-centenarians on a scale that is relevant to long-term fiscal policy.

This isn’t true, Trump is not proposing any solutions to the actual problem, and I don’t understand what move he’s trying to set up with this claim.

Super-centenarians and Social Security

The best source of accurate information on the number of active Social Security numbers held by implausibly old people is the SSA Inspector General Report, “Numberholders Age 100 or Older Who Did Not Have Death Information on the Numident.” (The Numident is the name of the file where they list every SSN that has ever been issued.)

The IG report expresses quite a bit of frustration with the Social Security Administration.

They note that, as of 2020, the Numident contained 18.9 million SSNs issued to people born in 1920 or earlier who are marked as alive, even though the Census Bureau says only about 86,000 people over the age of 100 reside in the United States.

Of these SSNs, 1.7 million are issued to people who have more than one SSN and who have already been marked as dead with the other number. In hundreds of thousands of other cases, the death information is available to the Social Security Administration through the payment files or Medicare records, and the Numident simply hasn’t been updated. Beyond that, fully 13.3 million people listed as alive on the Numident are older than the oldest known living person. The IG believes, strongly, that the Social Security Administration should try harder to update the Numident — that they should mark as dead those 13.3 million presumptively dead people and the 1.7 million duplicates.

In most cases, precisely because these people are old, there are no electronic records of their death, which means deaths are hard to identify. But the reports says that in addition to Social Security benefits, the SSA has electronic access to all information about “SSI payments, Medicare benefits, Black Lung benefits, or Railroad Retirement Board benefits since the 1970s.” The IG’s view is that if you are over 100 years old and have not been collecting federal benefits of any kind, you should be marked as dead in the Numident.

The SSA has been slowly improving this document over the years, but keeps rejecting advice to conduct a large-scale purge of the people who are most clearly dead.

They claim that they lack statutory authority to do this. There also isn’t a very strong incentive to do it, because the IG’s whole point is that you can tell these people are dead precisely because they aren’t drawing any federal benefits. In fact, the IG’s report says that in “Tax Years 2016 through 2020, employers and individuals reported approximately $8.5 billion in wages, tips, and self-employment income using 139,211 SSNs assigned to individuals age 100 or older.” In other words, the main fraudulent use of undead SSNs is not to cheat the government out of money; it’s undocumented immigrants paying into the system.

This is the kind of situation in which an alternate universe DOGE could actually be extremely useful.

There is a known problem of inaccuracies in the Numident that the Social Security Administration is dragging its feet about correcting, because the problem doesn’t have any real downside for Social Security — it’s a problem for other agencies, primarily those charged with enforcing immigration laws. Without impetus from the top to change things, bureaucracies tend to move at the pace of their most lead-footed member, so high-level executive attention to these kind of long-simmering problems could be constructive. Across the whole government, there are lots of inspector general and GAO reports with various program integrity recommendations that are not being acted on with sufficient alacrity.

But the actual DOGE is, as far as we can tell, not doing any of this. Instead, they’re conducting arbitrary layoffs that will make it harder to fix anything. And in the specific case of Social Security, they are lying about the upshot of the undead SSN problem to make it seem like huge sums of money are going out the door when they’re just not.

Very little money is improperly paid to dead people

If you want to learn about improper Social Security payments, you should check out another Inspector General report titled “Preventing, Detecting, and Recovering Improper Payments.”

According to the IG, the SSA does, in fact, make billions of dollars in improper overpayments every year. They also make a smaller number of improper underpayments. (In the chart below, OASDI is Old-Age, Survivors, and Disability Insurance — the main Social Security program — and SSI is supplemental income for the disabled.)

Obviously, you’re going to sound like an asshole if you stand up in front of the voters and say, “Hey, what’s the big deal? It’s only $10 billion a year in overpayments!”

But the government paid out around $1.5 trillion in Social Security benefits in 2024, and the IG estimates an improper payment rate of about 0.84 percent. That’s not bad for government work! And it’s worth noting that improper payments are made in both directions, and to an extent, there are tradeoffs between Type 1 and Type 2 errors. If your sole focus is colonizing Mars, you may be obsessively focused on reducing overpayments to zero with no concern about improperly underpaying benefits owed to the elderly, the disabled, widows, or orphans. But normal people can see why the people running this agency may be wary of making mistakes in the wrong direction.

At any rate, the IG report seems to indicate that benefits improperly paid to people who are dead account for less than ten percent of erroneous payments.

The main source of overpayment is people whose checks are supposed to be adjusted downward not reporting their own situations accurately. Some of this has to do with weaknesses in SSAs automatic benefit calculation systems, an area where some smart engineers could probably help. Some of this the IG says the SSA could remedy if they had direct access to IRS income data, which would Congress would need to pass new legislation for. And some of it has to do with the complicated calculations around the Government Pension Offset and Windfall Elimination Provision rules. Congress largely eliminated those rules in an unwise bipartisan bill that passed in December, so there will be much less need to try to do the calculations correctly. Note that this law, the Social Security Fairness Act, costs nearly ten times as much as could be saved by eliminating all improper overpayments.

Which just goes to show that while better program administration is always desirable, the real reason Social Security spending is so high is the actual laws on the books.

The government should track people better

Long story short:

• It’s not true that the government could save significant amounts of money by eliminating improper Social Security overpayments.

• It is doubly untrue that Social Security overpayments are largely a matter of money going to people who are actually dead.

• Eliminating the undead SSNs that Musk and Trump are complaining about might actually leave Social Security with less money, because many of them are being used to illegally pay into the system!

Beyond the fact that the administration is full of extremely dishonest people who like to lie about things, I genuinely don’t understand what they’re trying to accomplish by misleading people about this.

Social Security benefits are either going to get cut or they’re not. If they get cut, people will notice. If they don’t get cut, no money will be saved and the bond market will notice. If Trump were running for office, the point of the lying would be to put the incumbent president on the defensive about why he’s not fixing this. But he is the incumbent president, and he’s not going to be able to fix the problem because the problem is fake.

All that said, I do think this whole saga is a case study in why the government both needs better administrative data and should make better use of the data that it does have.

Social Security numbers are sort of odd. They’re the closest thing that we have to a national identification system, but when Social Security was created, only 56 percent of the workforce was covered by the system. The program wasn’t designed to function as a comprehensive national identification. It expanded over time, but it still doesn’t actually cover everyone. And it also doesn’t have the features you would expect a national identification system to have. For example, when a baby is born in a hospital, they’re issued a birth certificate. But the birth certificate does not come with a Social Security Number attached. Instead, parents use the birth certificate to apply for an SSN, which is then issued separately. Similarly, the SSA needs to piece together information about who is and is not dead based on data from state governments, banks, Medicare, and other stakeholders in the system.

It would be better, I think, to put privacy worries aside and actually create a canonical federal database of who is alive and who is dead and where they live.

This would solve a variety of problems. It would be easier to get people signed up for benefits they’re eligible for (and easier to cut people off when they’re ineligible). The FBI and CDC could publish timely statistics on topics of interest, like homicide and overdoses. It would be harder for people to work illegally. It would be easier to obtain accurate information about income and poverty, rather than relying on surveys. We could even have (shudder) a national photo ID card system instead of separate state driver’s license systems. And, of course, if we had a national system like that, we could require photo ID to vote without concerns about excluding anyone.

But this mires us in a timeless (though to me bizarre) debate in the United States, where we let privacy concerns stop us, even though in practice, we end up hacking together a bunch of substitute ID systems, like driver’s licenses and Social Security cards. I think people often don’t even know why they’re opposed to this kind of thing. Federalism has some real virtues in terms of decentralizing decision-making, but vital statistics tracking seems like an odd thing to decentralize. It’s just a legacy of historical happenstance that creates small-bore problems and lets dishonest people lie about Social Security fraud.