Billion Dollar Babies

I sat on this post until after National Adoption Month because it seemed like the polite thing to do and because I didn’t have time to finish it until now. It’s about adoption. Not whether adoption is good or bad. My feelings about adoption are whatever my client’s feelings are — and it should be no surprise that some kids want to be adopted and others do not. Instead, this post is about adoption as a policy, a specific tool of the system to accomplish certain goals. It is specifically about how much that tool costs.

6 years of data

71,000 adopted children

1 billion dollars in subsidies

If you’re low on time, here’s the short version: between July 2012 and January 2019 there were about 71,000 children in Florida’s adoption system, depending on how you count. The monthly number has grown from 32,500 children at the beginning of 2013 to nearly 40,000 in 2019 and has grown by 2-5% per year. (By comparison, there are currently 23,000 children in out-of-home care, and 10,500 receiving in-home services.) Over that period, DCF paid out about $1 billion in adoption subsidies to care for those children.

Source: DCF public record. The children receiving an adoption subsidy in Florida. The number of children has increased 89 per month. The payment amount has increased $70,000 per month. The data was not complete for 2012; the source of the dip in November 2018 is unknown.

Whether you believe that is a lot of kids or money probably depends on how you feel about adoption more generally. Adoption subsidies make up about 22% of the total child welfare budget in a given year. About 47% of the subsidies come from federal grants, which means adoptions are a good draw for federal dollars (which I thought Florida was against?). If you’re interested in good governance, though, the amount raises a big question: could we have used that billion dollars to help children in other ways?

If you somehow found this post because you’re looking to adopt, I recommend checking out the North American Council on Adoptable Children for a lot of great information. If you want to know what DCF says about adoption, they publish an annual report on adoption incentives that is very good. Here’s the 2018 one. It’s worth a read, but it doesn’t cover the same ground as this post. If you want to hear about the costs of adoption, read on.

What’s the current state of adoptions?

Let’s start with some background. The chart below shows the number of adoptions per month in Florida since September 2003. One thing stands out immediately: it’s very cyclical.

Source: Florida DCF dashboards. Number of adoptions per month.

Those ups and downs are seasonal effects. Adoptions spike higher than the yearly average in June and November each year (the end of the fiscal year and National Adoption Month), and drop in January and July.

Seasonal factors in adoptions per month.

We can smooth out these seasonal effects to see how the adoption rates are changing more clearly. They’re rising. Don’t read too much into that dip at the end — DCF data is provisional for the last two months and may get adjusted upward later.

Source: Florida DCF Dashboards. The last two months of data are provisional and may be adjusted upward in the future.

The system has grown, so it’s not unusual that adoptions are going up. What is weird is that everything else is going down. It’s hard to overstate just how much the system has changed over the last 15 years. In 2006, DCF took in 1,800-2,000 children and reunified 1,200-1,400 per month. Today, the ratio is about 1,200 to 500.

Source: DCF Dashboards. Exits from Florida foster care system. Values are seasonally adjusted.

In the graph above, you can see a separation starting in 2016 when reunifications and guardianships began to drop, but adoptions kept climbing. (Note: the guardianship line looks like 0 in 2003-2006 because they were called something else back then and it doesn’t show up on the dashboards. They weren’t actually zero.)

The reunification rate is mostly synced to the number of children who entered care. That makes sense because half of kids go home, and the rest are split between guardianship, adoption, and aging out (if they make it to 18). The gap between reunifications and adoptions (using a six-month running average) is currently down to +149 children, the second lowest we have on record. The lowest was April 2019 at +139. I can’t say whether this narrowed gap is because we’re taking in harder cases or we’re being harder on the cases we take in.

Who gets adopted?

So who are the adopted kids? Comparing a 10% sample of kids from the FSFN database who exited care in 2006 and 2018, the adopted children were about the same median age at removal (2.4 years old) and at adoption (5.2 years old). They weren’t any more male or female, but they were a little whiter in 2018 (74.4%) than in 2006 (66.9%). Adoptions were faster in 2018 (median: 715.5 days) than in 2006 (median: 860 days), but the difference may not be that significant because there’s lots of variation in individual cases.

Another curious thing is that foster care got less crowded: in 2006 an adopted child could expect 6 children placed concurrently in their homes, but in 2018 it was down to 4. Speed and size might be explained by the fact that more adopted children in 2018 spent time in relative care (59.7%) than kids did in 2006 (47.8%).

Doing back-of-the-envelope calculations — and controlling for age, race, gender, and standard placement measures — here are some things that pop out of the data from 2018:

  • If you don’t look at anything else but gender, then race mattered. Black children had 0.68x (0.57x – 0.95x) the odds of adoption as white kids. If you add in other case stats, like age and placement stability, then race melts away and the other things become more predictive. This is probably because race correlates strongly with other (largely negative) experiences in care.
  • A child’s age at removal had a strong impact on the odds of adoption. Every five years older, a child’s odds of adoption got cut in half. By comparison, the odds of reunification also decreased with higher ages at removal; the odds of aging out, however, went through the roof as kids got older. Makes sense.
  • The longer a child was in out-of-home care, the higher their odds of adoption. The odds more or less doubled every year the child was in care. The opposite was true of reunification — every year in care lowered the odds of reunification by 4x. But the strongest effect was seen in the odds of aging out, which skyrocketed every year a child was in care.
  • Being in foster care (as opposed to another setting) also raised a child’s odds of adoption a little bit. Every 10% of their time extra that was spent in foster care raised their adoption odds 1.065x. That’s not huge, but it’s not nothing either.
  • A child who disrupted placements had lower odds of adoption and higher odds of reunification. I found this one interesting. Every 10% of placements more a child disrupted, their odds of adoption fell 1.31x and their odds of reunification rose 1.62x. That means a child who disrupts 30% of their placements has 2.26x lower odds of being adopted and 4.22x higher odds of being reunified. A child can’t necessarily struggle their way out of foster care, though: being less stable raised the odds of a child aging out.

These are back-of-the-envelope calculations because we know other factors also come into play and this is just looking at the system in 2018. We’re also not discovering any natural laws here. Any of these can be changed in 2020 if we put our mind to it, which the DCF Secretary appears intent on doing.

You may have already wondered about opioids, and so let’s talk about drugs. In 2006, 40% of adopted children entered care at least in part due to parental substance abuse. In 2018, that number was up to 57%. Some of the movement in the other categories is due to changes in how DCF handles investigations (call the Safety Methodology), but drug abuse is drug abuse under both investigation protocols.

Source: DCF public placement database. The reasons children who were adopted were originally removed. Maltreatments with significant increase are in colors; all else are grey.

How do maltreatment codes affect permanency outcomes? Let’s switch to talking about risk instead of odds (they’re related but different), because risk is easier to understand when there’s a simple y/n comparison. Controlling for age, race, gender and the other maltreatment categories — a child whose case came in for parental substance abuse had approximately a 45% higher chance of being adopted in 2018 than one whose case didn’t (with a plausible range of 20-73%). For comparison, the death of a parent increased the chance by 157% (85-206%) and relinquishment increased the chance by 166% (81-216%).

Focusing instead on reunification, a child exiting care in 2018 whose case came in for parental substance abuse had a 20% lower chance of reunification (8-31%) than one who didn’t. For comparison again, abandonment lowered the chance of reunification by 42% (24-59%) and the death of a parent lowered the chance 103% (53%-112%). The age, race, and gender of the child had no significant impact on reunification chances. And no other maltreatment codes had a significant negative impact on chance of reunification.

In 2018, parental substance abuse corresponded with:

+45% chance of adoption

-20% chance of reunification

This put parental substance abuse in the same class of cases where the parent was literally missing, unwilling, or dead.

These are only back-of-the-envelope calculations, and they’re just descriptive of the system in 2018 — you can’t generalize them to any other place or time. Moreover, the differences may even melt away if we control for other factors that we can’t see in this data and policy changes could wipe them out as well. But, someone with a Ph.D. and government money should really look into that. The primary goal of a case is supposed to be reunification, and if we start with a bias against reunification in drug cases, then we need to rethink what reasonable efforts looks like.

Who is adopting?

The trends we see in adoption are not just happening with foster parents. Relatives and non-relatives alike are adopting in higher number. The chart below shows the uptick in both types of adoptions. (The “blank” line is for FSFN entries that don’t state the relationship of the child to the adoptive parent.)

Source: DCF public placement database. Number of adoptions by relatives, nonrelatives, and unknown status persons.

The strongest factor predicting who adopts a child is not surprising: it’s where they were placed. (But the factors for where they get placed is a whole other post.) Seen below, children with lots of time in foster care were largely adopted by non-relatives, and vice versa. Note that only 1.7% of kids (4 in this sample) spent 80% or more time foster care but were adopted by relatives.

Source: DCF public placement database. Amount of time in foster care and who adopted the child. Data is 10% sample of all children adopted in 2018.

Conversely, only 0.9% (2 kids in this sample) were in relative care 80% or more and adopted by non-relatives. I feel like I need to show this graph, too, because people have very strong feelings about adoption interventions when they go from non-relative to relative. This chart shows they also go the other way.

Source: DCF public placement database. Amount of time in relative care and who adopted the child. Data is 10% sample of all children adopted in 2018.

Finally, how many kids do people adopt? Usually one, but sometimes a lot more. The data doesn’t tell us how many of these are sibling groups.

Source: DCF public records. Number of families who adopted the given number of children, and the percentage of all families who adopted that many children.

How often do adoptions fail?

Adoptions don’t fail very often, as far as we know. About 966 adoptions failed in Florida over the last 19 years according to the public placement database. That number could be higher if the families moved out of state, if DCF didn’t link the cases, or if the failures did not result in DCF involvement. The failure rates weren’t significantly different for relatives (1.4%) and non-relatives (1.5%).

Just doing back of the envelope calculations, black and older children had higher odds of a failed adoption than white and younger children (but note that black kids in general may have higher exposure to investigations than white kids). Kids who took longer to get adopted had slightly higher odds of failure as well. Kids who came into care for parental drug abuse had lower odds of adoption failure, but kids who came in for parental alcohol abuse had higher odds of failure. I have no idea why.

The largest predictor of adoption failure that we see in the database was the number of placement disruptions the children had before adoption. Every disrupted placement prior to adoption raised the odds of adoption failure by 1.6x. That means a child with five disruptions would have 1.6^5 = 10.5x the odds of adoption failure than a child with none. Comparative odds are hard to imagine, so just read that as “a lot higher.”

Of the adoptions that failed, the largest number of children were abandoned, emotionally abused, or physically abused by their adoptive parents. There were no significant differences between relative and non-relative adoptions in any of these categories.

Source: DCF public placement database. Maltreatment codes for children who came back into care after a failed adoption. Children can have multiple maltreatments, so the numbers do not add to 100%.

How much do adoptions cost?

Here’s what this post is actually about. I wanted to know the cost of adoptions compared to foster care, so I asked DCF for all data in the FSFN system that showed placement payments. I wanted board rates, subsidies, clothing allowances, emergency funds, and anything else they logged. And I wanted the payments for foster care, relative care, independent living, in-home care, and adoptions. I mostly got it.

The data they sent covers the period of July 1, 2012 to March 31, 2019, but doesn’t seem to be complete until the beginning of 2013. I’ve had it for a while, trying to make sense of it. I won’t say it’s a mess, but I have a lot of reservations about it. For example, the data doesn’t cross reference with the AFCARS ID in the placement database, so I can’t really connect the children — and my efforts to do so manually came up with inconsistent results.

The data also includes a lot of entries that were wrong and later got corrected by assessing a negative payment against a provider. For example, a foster parent may be cut a check for $3,450 instead of $345, and a subsequent payment entry will be for the -$2,105 difference. To account for the delay in corrections, I’m limiting the data to everything before January 31, 2019.

As with all DCF data, I assume it is accurate and interpret it literally because DCF has a legal obligation to, you know, keep its records correctly. If there are problems here, I’m sure someone will tell me.

So, here’s the big picture: over the time period covered by the data, DCF made $2,255,636,185 in payments. That’s 2.26 billion dollars. During the same period, they recovered $185,322,211 in over-payments. That’s 0.19 billion dollars or 8% of the total. That leaves the amount actually paid out at 2.07 billion dollars. Adjusting for annual inflation, that’s 2.19 billion dollars. From here on out I’m going to use the adjusted (i.e. real) dollar amounts. All amounts are in 2019 dollars.

Is adoption expensive?

In the six years from January 2013 to January 2019, the State of Florida paid over a billion dollars in adoption subsidies. By comparison, $222 million was paid out for group care and $85 million on foster care. (The rest of the money was spent on services and other administrative payment codes.) Adoptions therefore cost 13.7x as much to maintain as foster care placements, and almost 3.7x as much as the all the other placement types combined.

Source: DCF public records. Total amount paid in the data for each payment code. All amounts are in 2019 dollars.

The reason is simple — there are a lot more children in adoptive homes than in foster care at any given time. We tend to forget about the adoption system because it does not receive ongoing case management or court oversight, but it is huge and growing every day. As seen below, the adoption system, at around 70,000 children (plus or minus some depending on which payment codes you consider when counting), cared for 1.7x as many kids as foster care over the same period of time, but at 13.7x the placement cost. It’s not really fair to compare it to group homes because, as we saw, kids in group homes are rarely adopted.

Source: DCF public records. Unique number of children for each payment code. Some children may be counted in multiple codes.

We have to be careful using child counts to compare things in the child welfare system, because kids can dip in and out of categories for a single day and drive up the numbers. Instead, we often use “bed days,” which is the number of days children spend in a category, to account for both population and duration. Doing that here, we see that children spent nearly 80 million bed days in adoptive placements but only 3.5 million bed days in foster care and around 1.8 million days in group care during the same period. The adoptive system, by this measure, is 22.6x larger than the foster care system.

Source: DCF public records. Number of bed days for each payment code. Some bed days may be double counted if children were paid for overlapping periods under different codes. A manual review of the data shows this is very rare.

So, adoption is…

a bargain:

22.6x the bed days at 14.7x the cost

or, a bust:

1.7x the kids at 14.7x the cost

People will very quickly tell you there are other costs to foster care, like case management, courts, lawyers, services, and so on. More on that later; we’re only talking about placement costs here. The point is that it’s a lot of money that we don’t think a lot about. One happy moment at a time, we release hundreds of children per month into the adoption system with a guaranteed monthly check for up to 18 years. That’s how you spend a billion dollars.

Sanity check

Just to make sure these payment amounts aren’t crazy, I checked the legislative appropriations for adoptions in 2017-18. Here’s what the legislature spent:

  • $197,397,316 for adoption subsidies under section 409.166. Forty-five percent of this comes from general revenue, 47% from federal grants, and 7% from the welfare transition fund.
  • $2,750,000 for special needs adoptions by state employees under section 409.1664.
  • $2,250,000 for adoption incentives to the CBCs under section 409.1662.

So, my numbers look reasonable. The appropriation went up to $199.7 million in 2018-19 with a 45/48% split between state and federal. (This is where I get the billion dollars over 5 years figure.) For comparison, the CBC appropriation for 2018-19 was $697 million to run the entire child welfare system. So ~22% of annual child welfare expenditures go to pay adoptive parents to care for their own kids.

[Edit 12/16/19: Immediately after hitting post, I realized that the numbers in fact are weird: the payment database can’t possibly be reflecting all of the foster care placements. So, I did a double-check of the placement database, and sure enough there were about 22.5 million bed days spent in relative care and 21.2 million bed days spent in non-relative (foster) care from January 2013 to January 2019. So, adoption would be 1.8x the bed days at 14.7x the cost. I have no idea why those 19 million foster care bed days have no payments listed in FSFN, but I am assuming that FSFN is accurate for the purposes of this post, so maybe foster parents are grossly underpaid for their time? Or they’re getting paid in some way that isn’t reflected in FSFN? This seems like a big problem with the data.

I’m also aware that relatives (and some non-relatives) get relative caregiver funds that are not reported in the FSFN data. I don’t know how many relatives actually received the payments, but let’s assume ALL of them did and they got the maximum amount of $298 per child per month ($9.93 per day). That would be $224 million extra. The appropriations bill seems to allocate $136 million per year for TANF payments, but I don’t know how much of that are relatives and how much are other recipients. Add that to the foster care payments and it’s 1.8x the bed days at 3.8x the cost. If there are enough unrecorded foster payments out there, you might be able to get the adoption amount lower than the foster care amount. It’s still a billion dollars, though.

Here’s what the chart would look like with the payments estimated at $15 per day for non-relative and $9.93 for relatives.

Source: DCF Public Placement Database and payment data.


What does the research say on subsidies?

I’m going to get hate mail for even asking whether adoption subsidies are worth the cost, so let’s look at more reputable authorities. Adoption subsidies do matter when it comes to permanency, both at micro and macro levels. Argys and Duncan (2012) found a connection between adoption subsidies and adoption rates. By their calculations, raising the adoption subsidy by $100 per month raises the adoption rate for boys by 4.6% and for girls by 5.9%. Hansen (2007) found that a state’s raising its adoption subsidy by $100 per month could raise its adoption rates by 1 per 1,000 births in the state.

Subsidies haven’t been found consistently effective across all subgroups of kids. Avery and Mont (1992) found that adoption subsidies did not affect adoption placement rate for special needs children except for those with mental disabilities. And Buckles (2013) found that changes in subsidies mostly affected foster parent adoption rates as opposed to relatives.

Finally, Barth et al. (2006) looked at whether adoptions are cost effective compared to long term foster care. The answer was yes, but a lot of assumptions went into that answer. The study looked at cases that had been in the system for three or more years and very explicitly did not compare adoptions to reunifications or attempt to make any statement on reunification at all. The point is that the fiscal trade offs can be calculated if you have access to the right information.

I could not find any study on the effect of adoption subsidies on reunification rates. In my experience, a caregiver’s willingness to adopt is one factor in the state’s decision to seek termination of a parent’s rights or give parents more time. If subsidies affect willingness and willingness effects case plan length, then someone should look into that.

How much are they really paid?

Now that I’ve shocked you with a billion dollars, let’s look at individual cases. The payments per child aren’t extravagant. The median day rate for adoption subsidies appears to be between $14 and $15 per day in 2019 money, which is fairly even with the board rate. That doesn’t include any one-time payments for medical or adoption related expenses (the latter of which is about $1,000 per child). That means a typical child would have an 18-year subsidy of $98,550 plus any medical stipends. Since the median child is about 5 years old at adoption, it’s more like $71,175 total.

DCF is supposed to find people willing to adopt without a subsidy, but I can only guess how often that happens. There were about 17,000 kids in the database that had payments totalling $2,000 or less from June 2013 through July 2018. That date range would have given them plenty of time to receive other payments if they were going to. Those are likely the non-subsidy cases, and they were about 24% of all kids.

At the other extreme, the twenty children below had the highest payment amounts in the database. The database only covers about six years, so I’ve also projected the amount out over 18 years for comparison. (Note: the database doesn’t have the ages, so we don’t know what the actual max payment would be for each child.) There were 978 children who had been adopted for at least a year and had a projected 18-year payment of $250,000 or more. That’s 2% of the 48,000 children with at least a year’s worth of payments. The adoptive family of the top child below was paid nearly $50,000 per year and would gross over $800,000 dollars if the child was in their care for 18 years. That deserves some public discussion.

Source: DCF public records. Children with the highest subsidy amounts in the database. Each line is a child.

And below are the twenty families that received the highest payments in the database. The top family’s day rate of $61.34 comes out to $90,000 per year for four kids. That would be over $1.5 million for the four children over 18 years.

Source: DCF public records. The adoptive families who received the largest adoption subsidy payments. Each line is an adoptive family.

I would love to know the justification for these highest subsidy amounts, and whether the standard is being applied evenly across all adoptive families. I suspect that it is not. If I were a pre-adoptive parent, I would print these charts out and walk them into every adoption subsidy negotiation meeting I have.

What’s the future hold?

I try to stay away from predicting the future, since the child welfare system can be taken in sharp turns by new leadership or public events. But, the adoption rates have been very consistent for the last decade, so let’s try this.

First off, it would be simple to just predict that adoptions are going to keep going up. Their year-over-year change has been positive since 2014 and has actually been rising even faster since 2016. It’s very notable that the adoption payments (orange line) have consistently grown faster than the number of adoptions (blue line).

Source: DCF public records. Year to year change in children count and monthly payment amount. Adoption payments have consistently grown faster than the number of children in the adoption system.

But that doesn’t take into account changes in the system that may actually impact the rate. Here are a few things we know from the data that could be helpful in modeling it:

  • Adoptions take about two years from removal to finalization.
  • Foster and relative care have different adoption patterns.
  • Younger children are adopted at higher rates than older children.
  • Drug cases have higher rates of adoption.

Using just those facts, we can model the number of annual adoptions since 2006 with about 85% accuracy. The predictors in the model, in order of importance, are the values two years prior for (a) the number of babies removed due to parental substance abuse [positive], and (b) the number of entries into care [negative]. That’s all you need.

Source: DCF dashboards/calculations by me. Projection of number of adoptions over the next two years. Inputs to the regression were number of babies removed for parental substance abuse and total number of children removed two years prior.

The model predicts that we’ll see the number of adoptions start to fall in 2020, largely because the number of drug babies that came into care dropped in 2018 and everything else stayed pretty much the same. As of November 2019, Florida had 4,068 adoptions. That’s 543 more adoptions than at the same point in 2017, but only 50 more than 2018. So, maybe things are slowing down already or maybe the dashboard updates next month will adjust the number by a significant amount. Or maybe DCF’s new policies will somehow generate even more adoptions. We’ll have to wait and see.


Leave a Reply

Your email address will not be published. Required fields are marked *