The Detention Buildout

Roughly double the immigration detention estate and the removal machinery that feeds it, using the funds appropriated in 2025.

AI evaluation · not yet reviewed by a human

This evaluation was produced and sourced by an AI model; a human review is still pending. Figures and conclusions may still change. The review log is at the foot of the page.How review works →

The 2025 reconciliation law appropriated more than 170 billion dollars over four years for border and interior enforcement, about 75 billion of it to the immigration enforcement agency and 45 billion to detention alone. The daily detained population has risen from roughly 40,000 before 2025 to 65,765 by July 2026, and the funded plan runs toward 100,000 beds. What the money buys is capacity rather than a change in the law: the same removal grounds, applied to far more people, with detention holding them available for it. This evaluation compares that buildout against the enforcement level of 2024 and looks four years ahead from 2026.

Balance

Much worse for the future · 0.01 previous scale

Balance on the previous scale. The Bilanz 2.0 simulation is not yet available for this evaluation. The category comes from the share of the debate on the pro side (r).

For 0.4 · 1 % Against 34 · 99 %

Size class: large Scale of this evaluation: Normalised Impact — unitless, calibrated to this topic. For comparison: one point here is worth roughly 2 billion euro per year. No price is set any more on a removal order that is carried out. Until this version it was valued at 10,000 euro and carried most of the case for the measure; since 24.09.2026 an order counts only through the concrete harm of leaving it unexecuted, and the studies that separate cause from coincidence find none that is not already on this page. What remains on the pro side are crimes not committed and deaths avoided at the border, and the second rests on a share of the fall in crossings — a tenth, in a range from none to three tenths — that nothing measures. A year of detention is valued at 40,000 euro, this site's rate for a year of deprivation of liberty in every debate; the American federal rate for wrongful imprisonment, about 43,000 euro, lies just above it. How we score →

Arguments for

Arguments against

7 arguments evaluated · Scoring v1.3 Δ absolute −33.6

Arguments — For

3 arguments

Crimes not committed by people removed

0.2of 100

Three in ten people in immigration detention have a criminal conviction, and removing them takes them out of the population that might offend again. Whether large-scale interior enforcement actually lowers crime has been measured. It did not.

Value 9 · Safety from crimeImpact 0.1Plausibility 2

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is offences that do not happen: assaults, burglaries, thefts and the smaller number of serious violent crimes behind them. This site places safety from crime in the same class as life and health, one step below the top of it here because the great majority of what is prevented is property crime and lesser assault rather than killing. The people spared are not identified in advance, which is normal for a stream of this kind and does not lower its weight. What is counted is the harm to victims, not the cost of policing or of prisons, which sits elsewhere. The value is high because the stream is safety from harm by other people, set one step below the maximum because most of what is prevented is not violence against the person.

The buildout raises removals by roughly 350,000 a year above the level of 2024. About three in ten people in immigration detention have a criminal conviction, which gives 103,000 people a year with a record among those removed ^[2]. Not all of them would offend again: a reoffending rate of 15 percent a year is used here, in a range from 5 to 30 percent, which is at or below what is observed for released populations generally. That gives about 15,000 offences a year that do not happen. The average harm to victims across the mix of offences involved — mostly property crime, drug offences and lesser assault — is put at 15,000 euro, in a range from 5,000 to 40,000. The result is 232 million euro a year. This is what the mechanism would deliver if it worked as described; whether it does is the subject of the next paragraph. The Impact is modest even taken at face value, because the share of those removed with any criminal record is under a third and most of their offending is not violent.

▸ Show calculation ▾ Hide calculation

	Removals a year above the 2024 level [1]		350,000 people a year
×	Share with a criminal conviction seven in ten people in immigration detention have no conviction [2]	29.4 %	103,000 people a year
×	Offences that would have followed in a year Setting, range 5 to 30 percent: at or below observed reoffending for released populations generally	15 %	15,450 offences a year
×	Average harm to victims Setting, range 5,000 to 40,000 euro: mostly property crime, drug offences and lesser assault	15,000 euro each	232 million euro
÷	Normalised Impact scale of this evaluation	2 billion euro a point	0.12

Score 0.12 Impact × 9 Value × 2 Plausibility ÷ 10 = 0.2 of 100

This claim has been tested at national scale and the test found nothing. Secure Communities checked the immigration status of everyone arrested by local police and was rolled out county by county between 2008 and 2013, which lets the counties that got it early be compared against those that got it late ^[3]. Across that comparison the programme had no observable effect on rates of homicide, rape, robbery or aggravated assault. The confounder that would otherwise dominate — that counties adopting early differ from those adopting late — is what the staggered rollout is designed to absorb, and reverse causation cannot arise because the order of adoption was set administratively rather than by local crime rates. The usual reply is that Secure Communities screened arrestees while the current buildout removes people who have already been convicted, which is a real difference and is why this argument is not scored at zero. But it is a difference in target, not in mechanism, and the mechanism is what failed the test. The Plausibility is low because a close twin of this measure ran across the whole country and the effect it is supposed to produce did not appear.

evidence basis: Precedent · P ceiling 8 identification: Quasi-experimental · rung ceiling 8 band: Effect did not materialise · P 1.5–2

Counterfactual: counties that received Secure Communities later, compared against those that received it earlier. Design: quasi-experimental — difference-in-differences on a staggered administrative rollout (Cox and Miles, Journal of Law and Economics 2014 [3]). Confounder: early-adopting counties differing from late ones, absorbed by the staggered design. Direction: no reverse causation, adoption order was set administratively rather than by local crime rates. Ceiling: quasi-experimental 8.0 binds below the praezedenz ceiling of 8.5; the band binds far below it. Band: effect did not materialise — a close twin ran nationally and the crime effect did not appear.

Secure Communities ran across the country under a design that can measure it, and produced no observable change in violent crime rates [3]. Read back: roughly one time in five, an interior enforcement programme of this size lowers crime by about the amount assumed here; the rest of the time it does not.

Open: The current buildout is itself staggered across states and sectors. A comparison of crime rates in areas where removals rose earliest against those where they rose latest would settle it within two years and could carry P to 6, or take the argument out of the ledger.

Fewer people die attempting the crossing

0.1of 100

Encounters at the southwest border fell from 2.5 million in 2023 to 444,000 in 2025, most of it before the buildout. If some part of that fall is the signal that arriving now means being detained and removed, fewer people take a journey on which hundreds die every year.

Value 10 · LifeImpact 0.0Plausibility 3.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is lives: people who die in the desert, in rivers or in trucks while trying to cross, whom the rule deters from the attempt. It carries this site's weight for life, the highest on the scale. What is not counted is any benefit from the person's absence — their labour, their taxes, their use of services — because those are transfers or run the other way, nor the gain the person would have had from arriving, which is not booked anywhere on this page. Earlier versions valued the fall in crossings at 10,000 euro each as a sign that the rule holds; that price had no source, and the only harm of an attempted crossing that is documented is the risk of dying on it. The value is the highest on the scale: what is counted are lives.

Encounters at the southwest border fell by roughly 2.06 million between 2023 and 2025 ^[5]. Almost all of that came before the interior buildout and is attributed to what changed at the border and in asylum processing; a tenth is attributed here to the interior signal, in a range from none to three tenths, which leaves about 206,000 attempts a year that do not happen. The border agency recorded 704 migrant deaths in fiscal year 2023 and calls the count incomplete ^[9]; against about 2.5 million encounters that is a risk of death of roughly 0.028 percent per crossing. That gives about 58 deaths a year avoided, each valued at this site's rate for a life, with a tenth added for the grief of the families. On the card that is about 0.04 points. Penalties at the border are shown to lower the chance that the same person is caught crossing again ^[10], but no study measures whether detention inland deters new crossings, so the share stays a judgment. The Impact is small: a few dozen lives a year, resting on a share of the fall in crossings that nothing measures.

▸ Show calculation ▾ Hide calculation

2,060,000 encounters a year

Fall in encounters at the southwest border

[5]

× 10 % Setting, range 0 to 30 percent: almost all of the fall came before the interior buildout; no study measures inland deterrence of new crossings

206,000 attempts a year that do not happen

Share attributable to interior enforcement

× 704 recorded deaths ÷ 2.5 million encounters (FY2023) Border Patrol's own count, acknowledged as an undercount

58 deaths a year

Risk of death per crossing

[9]

× 1.4 million euro × 1.1 (families) this site's rate for a life, plus a tenth for the grief of relatives

89 million euro a year

Value of the lives

÷ 2 billion euro a point scale of this evaluation

0.05

Normalised Impact

Score 0.05 Impact × 10 Value × 3.5 Plausibility ÷ 10 = 0.1 of 100

The mechanism is plausible and the timing is against it. The counterfactual would be an America that built the same border and asylum changes without the interior buildout, and no such comparison exists: everything changed within eighteen months of everything else. The fall in encounters is measured but cannot be apportioned, and the largest part of it predates the appropriation this measure spends. The chain has three links — enforcement inland, information reaching people considering the journey, decisions changing — and the middle one is exactly the sort that has been checked for other policies and often found weaker than expected, because information travels through smugglers with an interest in understating it. That check has not been done here. Reverse causation is not the issue; the confounder is, and it is unresolved because the border changes and the interior changes are the same eighteen months. There is also a counter-mechanism: enforcement can push crossings onto more remote and more dangerous routes, which is one reason the lower end of the range is zero. The Plausibility is low because the effect cannot be separated from everything else that changed at the same time, and the share attributed to this measure is a judgment with nothing behind it.

evidence basis: Mechanism · P ceiling 6 identification: Mechanistic · rung ceiling 6 band: Chain open · P 3–3.5

Counterfactual: none available — the border and asylum changes and the interior buildout occurred within the same eighteen months. Design: mechanistic — chain named (interior enforcement → information reaching intending migrants → fewer attempts), with no design separating it from concurrent border policy. Confounder: the border and asylum changes themselves, which almost certainly carry most of the fall; unresolved. Direction: no reverse causation. Ceiling: mechanistic 6.0 binds. Band: chain open, because the link carrying the quantity — the attribution share — is unmeasured and the concurrent-policy confounder is unanswered.

The chain is named but the attribution share carrying the whole quantity has nothing behind it, and the confounder — that border and asylum policy changed at the same time and probably carries most of the fall — is unanswered. Read back: about a third of the time, roughly a tenth of the fall in crossings is attributable to interior enforcement rather than to what changed at the border.

Open: Survey data on what intending migrants in Central America believe about interior enforcement, collected against the timing of the buildout, would put a number on the attribution share and could carry P to 6.

Removal orders that are carried out

0of 100

More than a million people in the United States hold a final removal order that has never been executed. An order nobody enforces is not a decision, it is a filing. Detention is what makes a person available on the day the removal happens.

Value 6 · Trust in institutionsImpact 0Plausibility 7

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is the correspondence between what the law says and what happens, which this site places in the class it uses for systems and institutions rather than in the constitutional core. It is not a claim that removal is good for anyone; it is the observation that a legal order routinely ignored costs something, and that the cost falls on every other order the same system issues. Nothing here weighs the interest of the person removed, which is counted on the other side. The value is not raised because immigration is a charged subject: what is priced is the enforcement of an adjudicated decision, the same as it would be for a tax assessment or a court judgment. The value sits in the middle of the scale, at the level this site uses for the working order of public institutions.

Without the buildout, a large share of people with a final removal order are never located; with it, roughly 350,000 more removals a year take place, most of them on orders already issued ^[1]^[5]. What the evaluation asked here is what goes wrong when such an order is not carried out, and the answer from the studies that could separate cause from coincidence is: nothing that is not already on this page. Public budgets gain or lose only through transfers, which move money between people rather than destroying it. The effect on the jobs of American-born workers runs the other way and is counted against the measure. Crime does not fall measurably when enforcement rises ^[3]^[8], and what remains of that channel is in the argument on crimes not committed. A price of 10,000 euro per executed order, used until this version, had no source and carried more than two thirds of the case for the measure. The Impact is zero: the harm this argument claimed is either not found or already counted elsewhere.

That detention produces removals is close to a matter of definition rather than a prediction: a person held in custody is available on the day a flight leaves, and a person who is not held may not be. The counterfactual is the non-detained docket, where a substantial share of people ordered removed are never located again, and both the detained and non-detained outcomes are recorded by the agency itself ^[2]^[6]. There is no confounder to speak of in the link from custody to availability, and reverse causation does not arise because the decision to detain precedes the removal. What is not definitional is whether the extra 350,000 removals materialise at all: the agency has never removed at that rate, staffing takes years to build, and destination countries must accept the returns. That operational doubt is reflected in the range around the removal figure rather than here. What is entirely unmeasured is the second half of the claim — that executing orders strengthens the system's other orders — and without evidence for it the argument is excluded rather than priced. The Plausibility is high for the mechanical half of the claim and the number itself rests on a stated price, which is where the doubt about this argument belongs.

Excluded (axis = 0): Excluded on 24.09.2026 (booking rule ordnungsgut/1.0): an order being carried out is valued only through the concrete harm that follows when it is not, and the causal evidence finds none that is not already counted elsewhere. Taxes and benefits are transfers; the labour-market effect of enforcement runs against American workers and stands in con-3; the crime effect is nil and stands in pro-1 [3][8]; no effect on trust in institutions has been measured. The former price of 10,000 euro per executed order had no source.

evidence basis: Mechanism · P ceiling 8 identification: Definitional · no rung ceiling

Counterfactual: the non-detained docket, where a substantial share of people ordered removed are never located. Design: definitional for the link that carries the quantity — custody makes a person available for removal, which is a matter of physical fact rather than behaviour; the agency's own detained and non-detained outcome data record it [2][6]. Confounder: none applies to the custody link. Direction: no reverse causation, detention precedes removal. Ceiling: set at 8.0 rather than left open, because the second half of the claim — that executed orders strengthen other orders — has no evidence behind it at all and is carried by the stated price. The operational doubt about reaching 350,000 removals sits in the quantity band, not here.

Arguments — Against

4 arguments · top 3 shown

What the buildout costs to run

24of 100

Detention runs at about 220 dollars a person a day, and the estate is roughly doubling. Adding the officers, the transport and the contracts around it, the 2025 law gives the enforcement agency about 75 billion dollars on top of an ordinary budget of roughly 10 billion a year — close to 19 billion dollars a year more, for four years. The money is borrowed.

Value 5 · Public financesImpact 8.1Plausibility 6

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is federal money, priced at the middle of the scale like every other euro on this site. Public money is not treated as cheaper or dearer than private money, and a deficit is not a harm on top of the spending — borrowing shifts who pays and when, and that shift is inside the euro figure. What the money buys is counted in the three arguments above; this argument is the payment. That a large share of it goes to private detention contractors rather than to public employees makes no difference to the weight, since a euro leaving the budget is a euro either way. The value is the middle of the scale, the level this site uses for public money whatever it is spent on.

The 2025 law appropriated more than 170 billion dollars over four years for border and interior enforcement, of which about 75 billion goes to the enforcement agency and 45 billion to detention alone ^[1]. That money comes on top of the agency's ordinary appropriation, which was about 10 billion dollars in 2025 and continues alongside it; it is not a new total that replaces the old budget ^[1]^[6]. Spread over the four years in which it can be obligated, the addition is 18.75 billion dollars a year, or 16.2 billion euro at 1.16 dollars to the euro. The range around it runs from 9 to 17 billion euro, set by how fast the estate actually fills and how much of the appropriation the agency can obligate before it lapses — a bed that is built and not occupied still costs money, and the daily rate of about 220 dollars a person applies only when it is ^[2]. The euro carries the standard weight of one for public money. Not counted here is the border wall construction in the same appropriation, which buys something different and belongs to a different measure. The Impact is the largest single figure in this debate and it is close to three times everything the buildout is credited with delivering.

▸ Show calculation ▾ Hide calculation

	Appropriated to the enforcement agency over four years, on top of its ordinary budget [1]	of more than 170 billion for border and interior enforcement	75 billion dollars
÷	Per year the money can be obligated until September 2029; the ordinary appropriation of roughly 10 billion dollars a year continues alongside it and is not deducted [6]	4	18.75 billion dollars a year
÷	In euro Setting, range 9 to 17 billion euro, set by how fast the estate fills and how much the agency can obligate before the money lapses	1.16 dollars to the euro	16.16 billion euro
÷	Normalised Impact scale of this evaluation	2 billion euro a point	8.08

Score 8.08 Impact × 5 Value × 6 Plausibility ÷ 10 = 24 of 100

The appropriation is a matter of statute, so the ceiling on this figure is not in question ^[1]. The counterfactual is the agency's ordinary budget before the 2025 law, which is published and which continues to be appropriated alongside the new money. What is estimated is how much of the money is actually spent within the four years, and here the doubt runs downward rather than upward: agencies frequently cannot obligate appropriations of this size on schedule, hiring at the pace assumed has never been achieved, and unspent balances carry forward. That doubt is reflected in the 9 to 17 billion euro range. The one factor that could push the figure higher is that detention is more expensive per person than the planning rate when the estate is expanded quickly into temporary facilities, which has been documented. Nothing here depends on anyone's behaviour except the agency's own capacity to spend. The Plausibility is at the top of what a budget projection can carry: the money is appropriated by law, and only the pace at which it is spent is estimated.

evidence basis: Projection · P ceiling 6 identification: Definitional · no rung ceiling

Counterfactual: the enforcement agency's ordinary published budget before the 2025 law, which continues alongside the new money [6]. Design: definitional — the appropriation is set by statute; the only estimated element is the pace at which it is obligated. Confounder: none for an appropriation. Direction: not applicable. Ceiling: projektion 6.0 binds because the spend-out pace is forecast; definitional carries no ceiling of its own. The doubt runs downward and is carried by the 9 to 17 billion euro band.

Forty-five thousand more people held

5.3of 100

Immigration detention is civil, not criminal: nobody in it is serving a sentence. Seven in ten of the 65,765 people held in July 2026 had no criminal conviction of any kind. The buildout roughly doubles that number.

Value 9 · Basic rightsImpact 0.9Plausibility 6.5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is liberty: months during which a person cannot leave a building, work, see their family except through glass, or make any decision about their own day. This site places that in the class it uses for the constitutional core, below life and health and above the working order of institutions. It is not counted here as a wrong — the detention is lawful and the people in it are subject to removal proceedings — but as a cost, in the same way that a lawful tax is a cost. That seven in ten of those held have no criminal conviction does not change the weight of a detention-year; it changes how the reader should understand the stream, and it is stated for that reason. The value is high because the stream is liberty, set below life and health because it is recoverable.

The detained population stood at 65,765 in July 2026 against roughly 40,000 before 2025, and the funded plan runs toward 100,000 beds ^[2]. An average of 85,000 held across the four years is used here, giving 45,000 more people in custody on any given day than under the enforcement level of 2024, in a range from 25,000 to 60,000. That is 45,000 person-years of detention a year. A year of detention is valued at 40,000 euro, the rate this site uses for every year a person spends without command of their own time: all 5,840 waking hours of it at the value of an hour spent under compulsion, which makes the year weigh as much as a year of life. Two statutes that put a price on the same thing bracket the figure. American federal law compensates wrongful imprisonment at 50,000 dollars a year, about 43,000 euro, and German law at 75 euro a day, about 27,000 euro. The result is 1.8 billion euro a year. What is not counted here is the medical care, the transfers between distant facilities and the deaths in custody, each of which would be a separate stream and none of which has a figure good enough to use. The Impact is about a ninth of what the buildout costs to run, which is the ordinary shape of a detention estate: holding people is expensive and what it takes from them is not the largest part of the total.

▸ Show calculation ▾ Hide calculation

	People held in immigration detention on an average day, above the 2024 level Setting, range 25,000 to 60,000: the plan runs toward 100,000 beds and 65,765 were held in July 2026 [2]	85,000 against roughly 40,000	45,000 person-years a year
×	Value of a year in detention This site's rate for a year of deprivation of liberty: 5,840 waking hours at 6.85 euro, the value of an hour under compulsion — equal to one year of life. Range 27,375 to 43,100 euro: German compensation for wrongful detention (75 euro a day) up to the American federal rate (50,000 dollars a year)	40,000 euro	1,800 million euro
÷	Normalised Impact scale of this evaluation	2 billion euro a point	0.9

Score 0.9 Impact × 9 Value × 6.5 Plausibility ÷ 10 = 5.3 of 100

That detention deprives people of liberty is not a causal claim requiring evidence; it is what detention is. The counterfactual is the enforcement level of 2024, and the detained population under both is published monthly by the agency and reconstructed independently from its own records ^[2]. The only estimated element is how full the estate runs across four years, and that carries a wide band because construction and staffing schedules slip. The number of people held with no criminal conviction — 46,436 of 65,765 in July 2026 — is a count rather than an estimate ^[2]. Where judgment enters is the price of a detention-year, and the wrongful-imprisonment rate used here is a legislative valuation of a different thing: compensation for a wrong, not the cost of a lawful deprivation. That is a real objection to the number and it is why the band is wide in both directions. Reverse causation does not arise. The Plausibility is above the middle: the count is measured and only the price of a detention-year is a judgment, which is where the doubt about this argument sits.

evidence basis: Precedent · P ceiling 8.5 identification: Definitional · no rung ceiling

Counterfactual: the detained population under the enforcement level of 2024, published monthly [2]. Design: definitional — detention is the deprivation; no behavioural link carries the quantity beyond how full the estate runs. Confounder: none. Direction: not applicable. Ceiling: praezedenz 8.5 binds; definitional carries no ceiling of its own. P sits below the ceiling because the price of a detention-year is borrowed from a compensation statute written for a different situation, and the occupancy across four years is projected — both carried additionally by the 25,000 to 60,000 band.

American workers lose jobs alongside them

3.1of 100

The intuition is that removing an unauthorised worker frees a post for a citizen. The largest study of a comparable programme found the opposite: employment and wages of American-born men fell where enforcement rose. Construction and child care were where it showed most.

Value 6 · OutputImpact 0.8Plausibility 6

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is output that does not happen: houses not built, shifts not covered, businesses that shrink because the people around a job disappear as well as the job's holder. It belongs to the class this site uses for economic systems and prosperity. What is counted is the production lost net of the time the displaced worker gets back, not their wage, which is a payment inside the economy rather than a loss to it. What the people removed themselves lose — the earnings and the lives they had here against what awaits them — is not inside this argument; it is a separate stream that this evaluation has not yet priced, and it would stand on the same side as this one. The value sits in the middle-upper part of the scale, at the level this site uses for economic output.

Over four years the buildout removes roughly 1.4 million people who were working, 350,000 a year. The loss builds up as they go: 350,000 fewer workers in the first year, 1.4 million fewer by the fourth, an average of about 875,000 across the four years. Against an unauthorised workforce of about 8.3 million in a labour force of 168 million, that lowers their share of American employment by about 0.52 percentage points on average, 0.83 at the end of the period ^[5]. The measured relationship from the largest study of a comparable programme is that a one percentage point fall in the employment share of likely undocumented men goes with a 0.12 percentage point fall in the employment rate of male citizens ^[4]. Applied to the average that is 0.06 percentage points, or about 75,000 American-born men out of work in an average year, in a range from 25,000 to 200,000. What each represents in lost output, after subtracting the value of the time they get back, is put at 22,500 euro a year — half of a typical full-year wage in the sectors concerned. The result is 1.7 billion euro a year. The Impact is about a tenth of what the buildout costs and it runs in the opposite direction to what the measure is usually expected to do for American workers.

▸ Show calculation ▾ Hide calculation

	Workers removed, averaged across the four years the loss accumulates: one year's removals are still missing in every later year of the horizon [1][5]	350,000 a year, building to 1.4 million by the fourth year	875,000 people
÷	Fall in the unauthorised share of American employment [5]	875,000 of a labour force of 168 million	0.52 percentage points
×	Effect on the employment rate of male citizens the measured rate of exchange from the staggered rollout of a comparable programme [4]	0.12	0.06 percentage points
×	Male citizens of working age	120 million	75,000 jobs
×	Output lost per job, net of the time the worker gets back Setting, range 15,000 to 30,000 euro: about half a typical full-year wage in the sectors concerned	22,500 euro a year	1,690 million euro
÷	Normalised Impact scale of this evaluation	2 billion euro a point	0.85

Score 0.85 Impact × 6 Value × 6 Plausibility ÷ 10 = 3.1 of 100

The counterintuitive direction here is the measured one. Secure Communities was rolled out county by county between 2008 and 2013, which allows early counties to be compared against late ones, and across that comparison employment of male non-citizens with at most a high school education fell about 7 percent while employment and hourly wages of American-born men fell as well ^[4]. The confounder that would otherwise dominate — that counties adopting early had different labour markets — is what the staggered rollout absorbs, and reverse causation cannot arise, because the order of adoption was administrative rather than a response to local employment. The mechanism the authors give is complementarity: construction crews and child care arrangements need all their parts, so removing one worker shrinks the work available to the others. Two limits apply. The current buildout is larger and faster than Secure Communities, and the relationship need not be linear that far out; and some low-skilled Hispanic citizens did gain employment as substitutes, which the aggregate figure already nets. The Plausibility is above the middle: the direction and the rate of exchange are both measured under a design that can support them, and what is extrapolated is the scale.

evidence basis: Study · P ceiling 8 identification: Quasi-experimental · rung ceiling 8

Counterfactual: counties that received Secure Communities later. Design: quasi-experimental — difference-in-differences on a staggered administrative rollout (East, Luck, Mansour and Velásquez, Journal of Labor Economics 2023 [4]). Confounder: early-adopting counties having different labour markets, absorbed by the staggered design. Direction: no reverse causation, adoption order was administrative rather than a response to local employment. Ceiling: quasi-experimental 8.0 binds below the studie ceiling of 9.0; a deduction of 2.0 applies because the current buildout is several times larger and faster than the programme measured, so linearity is assumed rather than shown. The size doubt sits in the 25,000 to 200,000 band.

Children whose parent is taken

1.8of 100

Several million American citizen children live with a parent who has no lawful status. When that parent is detained or removed, the child stays. What follows is documented: worse sleep, worse school results, and a household that loses most of its income overnight.

Value 9 · HealthImpact 0.4Plausibility 5

▸ Show reasoning & sources ▾ Hide reasoning & sources

The stream is the health and development of children who are American citizens and who remain in the country after a parent is removed. It belongs to the class this site uses for life and health, one step below the top of it because what is lost is recoverable in principle. The children are identified individually rather than statistically, and their loss is not incidental to the measure but a direct consequence of it. The household's loss of income is not counted here, because that would be a second stream on the same family and the larger of the two is what happens to the child. The value sits one step below the maximum: the stream is the health and development of children, and both can be regained.

The buildout removes roughly 350,000 more people a year than the 2024 level. A quarter of them have American citizen children in the household, averaging about 1.5 each, which gives 131,000 children a year who lose a parent to removal or prolonged detention, in a range from 60,000 to 250,000. What each loses is set at 0.15 quality-adjusted years in that year and the ones following, in a range from 0.05 to 0.4 — a figure built from what is documented about sleep, school attendance and anxiety in these households rather than measured directly. Across 131,000 children that is 19,650 quality-adjusted years a year. The household's collapse in income is not counted here; it belongs with the family, and the child is the larger loss. The Impact is the smallest on the contra side and still larger than the crime reduction the measure is meant to buy.

▸ Show calculation ▾ Hide calculation

	Removals a year above the 2024 level [1]		350,000 people a year
×	American citizen children in those households Setting, range 60,000 to 250,000 children [7]	a quarter of removals × 1.5 children	131,000 children a year
×	Quality-adjusted years each loses Setting, range 0.05 to 0.4: built from what is documented about sleep, school attendance and anxiety in these households rather than measured directly [7]	0.15 a year	19,650 quality-adjusted years
×	Value of the years the value of a healthy life year used across this site	40,000 euro each	786 million euro
÷	Normalised Impact scale of this evaluation	2 billion euro a point	0.4

Score 0.4 Impact × 9 Value × 5 Plausibility ÷ 10 = 1.8 of 100

That children are harmed when a parent is removed is documented in a substantial literature, and the direction is not seriously disputed. What that literature almost entirely lacks is a comparison group chosen in a way that separates the removal from everything else about a household living without status. The studies compare children whose parent was removed against children in similar households whose parent was not, and the households differ in exactly the ways — prior encounters with enforcement, employment sector, neighbourhood — that also affect a child. The one design that could separate them is the staggered rollout used for the labour market findings, and nobody has applied it to child outcomes. Reverse causation does not arise: a child's school results do not cause a parent's removal. The chain from removal to lost income to worse outcomes is short and every link in it is visible, and the counter-argument that some families reunite abroad is answered by counting only children who remain. The Plausibility is at the top of the range for a claim whose steps are all visible and whose size has never been measured against a clean comparison.

evidence basis: Mechanism · P ceiling 6 identification: Mechanistic · rung ceiling 6 band: Chain closed, unevidenced · P 4–5

Counterfactual: children in similar households whose parent was not removed — a comparison the existing literature makes but cannot clean. Design: mechanistic — chain named (removal → household income collapse and separation → worse child outcomes); no quasi-experimental design has been applied to child outcomes. Confounder: prior enforcement contact, employment sector and neighbourhood, all of which differ between removed and non-removed households and also affect children; unaddressed by the literature. Direction: no reverse causation, a child's outcomes do not cause a parent's removal. Ceiling: mechanistic 6.0 binds. Band: chain closed but unevidenced — every link is named, the reunification counter-argument is answered by counting only children who remain, and only the measurement is absent.

Nothing measured argues against the claim; what is missing is any study with a comparison group clean enough to size it. The one counter-mechanism, families reuniting abroad, is answered by counting only the children who stay. Read back: a child who loses a parent this way loses roughly the amount of healthy time assumed here about as often as not.

Open: The buildout is staggered across states and sectors, and school and health records are held at state level. Comparing child outcomes where removals rose earliest against where they rose latest would size this within three years and could carry P to 6.

Summary

Two of the three things a detention buildout is supposed to deliver have been measured on a close twin of it, and neither appeared: interior enforcement at national scale did not lower violent crime, and it lowered rather than raised the employment and wages of American-born men. Earlier versions put a price on the enforcement of removal orders themselves, and that price carried more than two thirds of the case; it had no source, and the studies find no harm from leaving orders unexecuted that is not already counted, so it is gone. What is left on the pro side is small: some crimes not committed, and a few dozen deaths a year avoided if the buildout deters a share of the crossings. Against it stand some sixteen billion euro a year of additional running costs, forty-five thousand more people held in civil custody at any moment, seven in ten of whom have no criminal conviction, and the children who stay behind. The balance comes out clearly negative. Readers who think an unenforced legal order costs the system something beyond what can be measured would have to set that price themselves; this evaluation no longer does.

Outlook — effect over time

Much worse for the future · 0.01 previous scale

Sources

American Immigration Council: What's in the Big Beautiful Bill? Immigration and Border Security Unpacked. americanimmigrationcouncil.org
Transactional Records Access Clearinghouse: Immigration Detention Quick Facts. tracreports.org
Cox and Miles, Journal of Law and Economics 57(4): Does Immigration Enforcement Reduce Crime? Evidence from Secure Communities. journals.uchicago.edu
East, Luck, Mansour and Velásquez, Journal of Labor Economics 41(4): The Labor Market Effects of Immigration Enforcement. journals.uchicago.edu
Migration Policy Institute: Frequently Requested Statistics on Immigrants and Immigration in the United States. migrationpolicy.org
Department of Homeland Security: U.S. Immigration and Customs Enforcement Budget Overview, Fiscal Year 2027. dhs.gov
Migration Policy Institute: Large-Scale Deportations May Have Unintended Consequences. migrationpolicy.org
Hines and Peri, IZA Discussion Paper 12413: Immigrants' Deportations, Local Crime and Police Effectiveness (2019). docs.iza.org
U.S. Government Accountability Office, GAO-25-107548: Southwest Border: Border Patrol's Missing Migrant Program — 704 migrant deaths recorded in fiscal year 2023, acknowledged as an undercount. files.gao.gov
Bazzi, Burns, Hanson, Roberts and Whitley, NBER Working Paper 25100: Deterring Illegal Entry: Migrant Sanctions and Recidivism in Border Apprehensions. nber.org

Last reviewed by Claude Opus 5.5 · September 24, 2026 · 4× AI, not yet reviewed by a human

September 24, 2026AI reviewClaude Opus 5.5re-scored
Regel ordnungsgut/1.0: pro-2 (10.000 € je vollstreckter Anordnung, ohne Quelle) mit zero_reason ausgeschlossen — kein kausal belegter Schaden des Nichtvollzugs; pro-3 auf vermiedene Todesfälle beim Grenzübertritt umgestellt (V 10, rund 89 Mio €); r 0,22 → 0,01, schlechter → deutlich schlechter.
September 20, 2026AI reviewClaude Fable 5.1re-scored
Neuer Anker freiheitsentzug-jahr (Julian 20.09.): Haftjahr 43.000 → 40.000 € (5.840 Wachstunden × 6,85 €, Band 27.375–43.100 €); con-2 i 1,94 → 1,8, Score 11,35 → 10,53; r 0,219 → 0,221 (schlechter, unverändert). Mechanik-Satz con-2 korrigiert (ein Neuntel statt ein Viertel der Betriebskosten, seit v2 überholt).
September 14, 2026AI reviewClaude Fable 5.1re-scored
Gesamtprüfung 08.09. umgesetzt: con-1 75 Mrd $ sind Zusatzmittel neben dem Jahresetat (i 8,4 → 16,16, k 0,5); con-3 Jobverluste auf Jahresmittel 875.000 statt Endstand (i 2,7 → 1,69); Scope-Satz zu Abgeschobenen gestrichen; r 0,28 → 0,22 (schlechter).
September 6, 2026AI reviewClaude Opus 5First evaluation
Created for the English side: crime and labour-market claims placed against the two Secure Communities studies; removal-order price pinned beside the balance.

Evaluations are produced with AI support and reviewed on a schedule for new developments; human passes are marked separately.How we review →