Government kept USD 382 million of child support from families on cash assistance in FY2024 — above the federal-share waiver cap a state must remit the federal share out of its own funds

What is happening?

A family applying for TANF cash assistance in the United States must assign its right to child support to the state. From that point the money a noncustodial parent pays is collected by the child support program as a government receivable rather than as income for the child. In fiscal year 2024 the federal Office of Child Support Services reported USD 607,637,737 collected on behalf of families currently receiving assistance, of which USD 382,483,602 was retained by federal and state government as assistance reimbursement instead of being delivered to the family. That is 62.9 percent of what was collected for those families — a ratio produced here by dividing two adjacent lines of the same federal table, and one that the report itself does not print.

Both governments keep the money, and the federal government keeps the larger half. Across current- and former-assistance cases the total retained in FY2024 was USD 807,846,134, split USD 451,887,390 federal and USD 355,958,744 state. Federal law waives the federal claim only on a capped amount — USD 100 per month for one child, or a state-set figure up to USD 200 per month for two or more children — and only where the state both pays that amount to the family and disregards it when calculating the assistance grant.

States are not barred from sending more. The option to pass through and disregard has existed since the Deficit Reduction Act of 2005 and took effect in October 2009, or October 2008 at state option. Above the cap the state must remit the federal share out of its own funds. In FY2024, 27 jurisdictions reported exactly zero pass-through dollars — 25 states plus Guam and the US Virgin Islands. All FY2024 figures here come from a report the agency labels preliminary.

Whose problem is this?

The people who decide and the people who lose are in different rooms. A child in an assisted family has no standing in a state appropriations debate, and the amount at stake per family is small enough that no organized constituency forms around it.

Where does this problem end?

The boundary here is not whether child support is collected but where the money goes after it is collected. Enforcement works well enough that the program reported collecting USD 4.24 for every USD 1 spent in FY2024. The question is who receives the fourth dollar when the family is on assistance.

What is the state now, and what should it be?

Now

Needs a new measurementthe target state: no source opened here names how much of the support collected for an assisted family should reach the child, nor a date by which it should. The statute sets a capped exception and stops. Federal budget documents describe a program whose aim has moved away from cost recovery without naming an endpoint, and the two congressional research products opened here state that any change would have fiscal implications without stating what the change should be. There is a cap and no objective.

How big is it?

Between about 82,000 and about 916,000 children, as of FY2024. The two bounds answer different questions and the population file carries the same figures.

The low bound counts children whose support demonstrably moved and demonstrably did not reach them. Three federal tables produce it. First, 236,949 current-assistance IV-D cases sit in the 27 jurisdictions that reported zero pass-through dollars in FY2024, which is 27.1 percent of the national 873,207. Second, the program reports 12,213,636 children across 11,646,025 cases, a national average of 1.049 children per case. Third, only 33.0 percent of current-assistance cases saw any collection at all in FY2024. Those three multiply to 82,024, rounded here to 82,000.

The high bound counts every child in a current-assistance case nationwide, 873,207 cases at 1.049 children each, or about 916,000. It is national rather than confined to the 25 states because the federal cap applies everywhere. Even in the state that goes furthest, the full pass-through is written into statute as contingent on an annual appropriation.

The gap between the two bounds is not measurement noise. It is the distance between having had money taken this year and being subject to the rule. A defensible middle reading is about 300,000, meaning all children in current-assistance cases with a collection, nationwide.

The largest weakness in the chain is the second step. No federal table publishes children by assistance category, so a national average has to be applied to a sub-population that may differ from it. If assisted families average 1.5 children per case, the low bound rises to about 117,000 and the high to about 1.31 million.

Under what conditions does it arise?

1. Assignment converts a private obligation into a public receivable. A family applying for cash assistance assigns its right to support for the period it receives assistance, capped at the cumulative unreimbursed assistance. Nothing about the child changes and nothing about the paying parent changes. Only the payee changes.

2. The federal waiver is capped, so delivering more than the cap has a price. Federal law waives the federal share only for an excepted portion of USD 100 per month, or up to USD 200 for two or more children, and only where that portion is also disregarded in setting the grant. Under the cap a state gives up only its own share. Above it the state must send the federal government its share from state funds, which means paying the federal government in order to route a payment to the child it was collected for.

3. A pass-through is an appropriation, and an appropriation is repealable. The first full pass-through in the country was enacted in 2015 and took effect in 2017. Five years later the same legislature conditioned it on an annual appropriation of at least 90 percent of the county share after the full federal share is paid. Another state discontinued a USD 50 pass-through in 2011 for budgetary reasons.

4. The federal performance system does not record where the money went. The five incentive measures built by the 1998 incentive act count paternity establishment, cases with orders, current collections, arrearage cases and cost-effectiveness. A dollar retained by government and a dollar delivered to a child score identically in all of them.

5. The aggregate hides the sub-population. Cost recovery is 3.03 percent of all distributed collections in FY2024, and never-assistance collections are 72.1 percent of the total. Every official framing quotes a ratio computed over that whole denominator, which is dominated by families to whom assignment never applied.

What has been tried?

Twenty-one years after the federal option was created, 25 states still report zero. What has been tried at federal level since 2005 is not a distribution change at all.

What was found?

Why is it still unsolved?

Incentive inversion — the party that would deliver the money to the child is charged a price for doing so, and the price is set by the party that collects the larger half of what is kept.

The first movement is the cap. Federal law does not forbid a state from sending a family everything collected for it, and several states do. What federal law does is waive the federal claim on the first USD 100 per month, or USD 200 where there are two or more children, and no further. Above that line the state has to send the federal government its share out of state money. So the state faces a schedule in which a small delivery is nearly free and a full delivery is a recurring purchase. A cap is not a prohibition and that is exactly what makes it durable, because nothing in it can be pointed at as the thing blocking the outcome.

The second movement is that the two enacted state statutes on record confirm the price is real. Colorado built the first full pass-through in the country and then, five years later, wrote an automatic shut-off into its own law making the pass-through contingent each year on an appropriation of at least 90 percent of the county share after the full federal share is paid. Michigan ran the same mechanism backwards and discontinued a USD 50 pass-through in 2011 for budgetary reasons. Once delivery to the child is a line in an appropriations bill it competes annually with every other line, and it can be lost in a bad year in a way a statutory entitlement cannot.

The third movement is that nothing in the measurement system pushes the other way. The five federal performance measures score a retained dollar and a delivered dollar identically, so a program that keeps everything and a program that keeps nothing look the same on the federal scorecard. And the headline ratio everyone quotes is computed over the whole caseload, in which never-assistance collections are 72.1 percent of the money and assignment never applies. The aggregate is true and it describes a different population from the one the rule falls on. Where 25 states decline a waiver that is free up to the cap, the state share is evidently a live cost even inside the waiver, which is what makes this an inversion rather than an absence of authority.

What observation would mean it is solved?

Candidates — (a) every jurisdiction reports nonzero pass-through dollars in a federal fiscal year (b) the retained share of current-assistance collections falls toward zero (c) Congress removes the cap on the federal-share waiver, or ends assignment altogether.

(a) alone counts the wrong thing. The federal table that reports pass-through dollars does not split current-assistance from former-assistance pass-through by state, so a state can report a large number while assisted families still see little. It also misses delivery through other instruments — one state treats its USD 100 as an addition to the assistance payment rather than a separate pass-through, and correctly reports zero. A roster of nonzero rows would therefore be neither necessary nor sufficient.

(b) alone moves for the wrong reason. Total assistance reimbursement fell 55 percent between FY2020 and FY2024, driven mostly by a shrinking assisted caseload rather than by any policy change. A ratio whose denominator is collapsing will keep falling while every individual family is treated exactly as before. A denominator collapsing is not a remedy.

(c) alone is a statute, not a delivery. The one state that went furthest by statute then made its own pass-through contingent on an annual appropriation. Removing the federal claim would still leave the state share, and the state share is the cost that 25 states already decline to bear inside the free portion of the waiver. The three would have to be read together, and (b) would have to be read against the caseload it came from.

What is it connected to?

Fills with researchthe adequacy of TANF cash grants, arrears policy and the stock of debt owed to government rather than to families, incarceration and driver license suspension used as enforcement tools, and child support treatment in other social insurance systems. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• No current roster of which states pass through. Two congressional research products dated January 2026 both rest the state count on a policy page last updated 2023-05-30. No federal agency publishes a current inventory of which states pass through, how much, or under what authority — even though the agency collects the dollar figure from every state every quarter on a standard form line.
• Nobody reconciles the policy roster against the money. One research report cites the policy roster and the federal dollar data in the same document and does not notice that they name different states. The comparison made here finds at least seven jurisdictions where the two disagree, and it appears in no source opened.
• No source publishes the retained share for assisted families. The 62.9 percent figure is a division of two adjacent lines of a table the agency publishes itself, and no publication performs it. The official framings quote the aggregate instead — a budget justification says 96 percent of collections go to children and families, and research reports say cost recovery has fallen below 4 percent. Both are computed over a denominator in which never-assistance collections are 72.1 percent of the total.
• No target of any kind. Nothing in the statute, in the budget justification, in agency guidance, or in the research reports states how much of the support collected for an assisted family should reach the child. The cap exists and no objective does.
• The cap has never been indexed. USD 100 and USD 200 were set in 2005 and have not moved in 21 years, and the statutory text carries no escalator. Nothing opened here proposes indexing it, which is a statement about these documents and not about the wider literature.
• No count of affected children exists. The agency publishes cases by assistance category and children in total, but never children by assistance category. The number of children whose support was collected and kept cannot be read off any table and has to be constructed, which is why the estimate here spans an order of magnitude.
• The professional association of the program itself does not raise it. Its August 2023 federal legislative agenda, four categories of asks to Congress, contains zero occurrences of pass-through, distribution, cost recovery or assignment across the whole document. The organized voice of the program asks for more collection power and is silent on who receives what is collected.
• No cost estimate for ending it. The bipartisan paper calls the forgone revenue immaterial without producing a figure, and the research reports say such changes would have fiscal implications without producing one either. No budget score and no aggregation of state fiscal notes appears in anything opened. The number that would settle the argument is the one nobody has published.

See the evidence

One primary federal dataset was read directly and it anchors every money figure here. The FY2024 preliminary report was downloaded and its tables read line by line, and the statutory text and the implementing agency instruction were read in full, so the mechanism described in block 9 rests on primary sources rather than on secondary description. The congressional research products were opened and are used for framing and for the legislative history, but repeated retrievals of one of them returned inconsistent extractions of the same distribution table, and its government-retention figure appears only as a residual, which is why every dollar figure above is anchored to the agency report instead. Two disagreements are left visible rather than resolved. The first is between the policy roster and the money data — the roster lists Alaska, Georgia, Montana and South Carolina as having a pass-through while the FY2024 dollars show zero for all four, and lists Michigan, Mississippi and Wyoming as having none while the dollars show money moving in all three. Only one of the seven has a documented mechanical explanation, which is the Montana treatment of its payment as an addition to the assistance grant, and that explanation in turn means the headline count of 26 includes at least one state delivering through a different instrument. The second is smaller and internal to the roster, whose own page describes 52 jurisdictions over a 54-row table. Three derived ratios in this document — 62.9 percent, 3.03 percent and 55.9 percent — are divisions performed here between lines of the same federal table, and they are labelled as such wherever they appear. This is a Path A output, so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 82,000–916,000 As of FY2024

Sensitivity The width of the range is not measurement noise. It is the distance between having had money taken this year and being subject to the rule that would take it. The low bound moves with the economy, since the share of current-assistance cases with any collection fell from 41.5 percent in FY2020 to 33.0 percent in FY2024, and it moves with state policy rosters. The high bound moves only with the assisted caseload, which fell 21 percent over the same four years. A defensible middle reading is about 300,000, meaning all children in current-assistance cases with a collection, nationwide. What this number fails to count is larger than the number itself in at least three directions. It excludes deterrence, meaning parents who never applied for assistance because of the assignment and cooperation requirements, which no opened source measures and no federal instrument could. It excludes former-assistance families, from whom USD 425,362,532 was retained in FY2024, which is more than the USD 382,483,602 retained from currently assisted families, against a caseload of 4,389,696 cases. It excludes noncustodial parents paying into a debt owed to government rather than to their child, whom no federal table counts at all. In the opposite direction the count is too generous in one respect and contested in another. It says nothing about magnitude, because the average retained per assisted case with a collection was USD 1,336 in FY2024 and no source publishes a distribution, so a counted child may have lost tens of dollars or thousands. And seven jurisdictions are contested between the FY2024 money data and a policy roster last updated in May 2023, so Alaska, Georgia, Montana and South Carolina are counted here as passing through nothing on the basis of their reported zero, and Michigan, Mississippi and Wyoming are excluded on the basis of their reported dollars; if the roster is right and the reporting is wrong, the low bound moves in both directions at once. All figures are drawn from a report the agency labels preliminary.

Regional breakdown Two federal tables give state-level current-assistance caseload and state-level pass-through dollars, so a jurisdiction can be classified as reporting zero or nonzero, but the chain that converts cases into children rests on a national children-per-case average that no table publishes by state or by assistance category. Applying a national ratio state by state would produce regional figures that look measured and are not. No apportionment of the national figure by population share was performed, and no source opened this round publishes children affected by this rule for any state.