A video of a child being forcibly removed from his mother has been in the news lately. It’s brutal to watch. A group of police officers and security guards yank at the one-year-old while another swings a taser wildly around the room at anyone who gets too close. The woman is on the floor. Her sin is apparently trespassing, i.e,. sitting on the floor instead of standing when there were no seats available in the four-hour line. The charges are later dropped because she was trespassing at a government office (to extend daycare for her child) where people are actually allowed to be, and people sit on the floor in airports all the time and nobody rips their kids from them. The harm’s already done to the child, though. Nobody in the crowd intervenes — they don’t want to get arrested, shot, or lose their place in line — but they document it for the world to see.

(For added absurdity and likely thanks to the word “baby” in the title, the video I watched was preambled with an ad for Zales’s Enchanted Disney collaboration wherein a rich-looking white lady finds a diamond ring on a table, puts it on, and imagines she’s become a Disney Princess®. She has no idea whose ring it even is, but it’s hers now (hey she deserves it). Nobody arrests her for theft and wrestles her baby out of her arms. She does not spend a few nights in jail for having the audacity of self-worth. She, in fact, lives happily ever after.)

Children are removed from their parents every day, and it frequently looks just like that video. It is traumatic for everyone (especially the child) and is supposed to be only for very exigent reasons. In Florida, we know that anywhere from 1,000 to 1,500 kids enter the foster care system each month. The video got me thinking about when those removals happen.

I had a lot of assumptions. I’ve heard that removals go up around the holidays, and that they go down in the summers when kids are home. I felt sure that fewer removals would happen on the weekends, but also that there should be no reason for that because kids would be in more danger when not in school. Removals are supposed to happen when they are unavoidable, not when they are convenient to the investigator.

And I thought: how many kids are actually removed on Christmas? That would be horrible. 

As part of a project we are working on, our office came into possession of the entire Florida DCF placement database (anonymized and unforgivably massive). This database includes the removal dates and details of 280,839 kids going back to some who entered care in the 1980s. Looking only at the removals from January 1, 2007 to December 31, 2017, we have data on 156,357 kids. Some of those kids came into care multiple times, so the time period covers 181,799 removals. (Caveat: as with all real-world records, the data is only as good as it is. Given the large numbers here, it is reasonable to assume that errors are evenly spread out and not biased in any given direction.)

How many of those removals were on a holiday?

Since we have the dates for all the removals, it isn’t hard to count them. On a normal, non-holiday day, the average number of removals is about 46. For Christmas, the average is six. For Thanksgiving, it’s eight. Only 67 kids were removed on Christmas Day in Florida from 2007-2017.

There are a few takeaways there. First, holidays seem to suppress removal numbers, at least on the day of the holiday itself. The only holiday that appears above average is Columbus Day; but, with year-to-year variations, it — along with President’s Day and New Year’s Eve — is not statistically different than a normal non-holiday. There is no holiday where more kids are removed than average (though there are periods of the year when removals are up, discussed below).

So what about the theory that more kids are removed around the holidays? If you thought (like I did) that there would be a giant spike in removals before or after, say, Christmas — well, there isn’t. In fact, removals bump up slightly and then start dropping off around the week before Christmas (a.k.a. now). The slight bump isn’t enough to call a correction. A glance at DCF’s dashboard on investigations shows that Decembers are usually high points for closing investigations during the year, so these numbers are even more pronounced. Removals don’t pick up again until January when school is back in session.

Here is the year-round chart. This confirms that removals go down in the Summer and rise again in the new school year. If you thought sentimentality was what kept DCF from removing kids on Christmas, think again. You see similar holiday drops on the 4th of July and Veterans Day. The dips would be more pronounced for MLK Day, Labor Day and Memorial Day, except those holidays don’t happen on the same calendar day each year. Kids don’t get removed on holidays because investigators are on vacation.

So that’s the question: if only 15 kids had to be removed on the 4th of July, why did 48 have to be removed on the 7th of July? 

What about the weekends?

On average only 14 kids per day were removed on Saturdays and Sundays. The highest day of the week for removals was Thursday — probably because court is held one day after a removal and nobody wants to go to court on a Saturday, for sure. At 14 removals per day, the weekend was on par with the 4th of July.

And, while we’re at it, what about time of day? Kids get removed during business hours. The later in the day, the higher the number of removals.  Below you can see three distinct peaks in the following hours: 9:00am, 1:00pm, and 5:00pm. Only 416 kids in this dataset were removed in the 6:00am hour. Maybe children are safer before dawn? (Those 5,580 kids removed at midnight include entries without a valid date — i.e., “00:00:00”. Don’t read into that spike.) I’m not including a graph because it’s messy, but if you look at the whole week the highest rates of removals happened during the 5:00pm hour on Thursdays. No surprise.

So removals happen all the time, except holidays, weekends, and usually not outside of business hours. And they especially happen when investigators first get to work, after lunch, and right before they go home for the day. Something feels very wrong about that. It’s as if “risk” is also a product of convenience, which is not how child protection is supposed to work.

If the averages hold, around six (and up to twelve) kids will be removed on Christmas Day this year. Let those removals be necessary and kind.

The First DCA published statistics on its caseloads and decisions. But notably (as appellate judges like to say), the length of time they take to resolve cases was not reported. It motivated me to update the How Long do Appeals Take tableau.

The answer? Probably 120 to 170 days for a Dependency case, 260 to 576 days for a Criminal case, and 345 to 603 days for a Civil case.

You can explore each district and case type at the tableau here.

In July of this year, the Florida foster care system did something unseen since February 2014: it shrunk. For the first time in over 50 months, the year-over-year (YOY) change in out-of-home care numbers went down by 45 children. By August it was down 118, and the reports out this month for October show a contraction of 165.

While a reduction of 165 kids does not seem like much in a system with over 24,000 children in it, the slowing actually started in January 2016, when the system was growing at a staggering YOY rate of 2,540 kids.  Just the month before held a YOY increase of 2,683 — the fastest growth since data is available in 2003.

There’s no official definition of a contraction period, or any way to tell if one is real or a blip. I actually sat on this post for a few months to make sure the trend was stable — we’ve had hurricanes, elections, resignations, and other unusual events recently, so I wanted to let those pass.

In reality, though, changes from positive to negative OOHC growth (expansions to contractions) do not happen quickly and appear largely driven by intentional policies and not outside events. The expansion under Secretary Hadi in 2004-2006 lasted 19 months and ended abruptly with the Secretary’s resignation from office in December 2006. The subsequent contraction during the Butterworth and Sheldon administrations lasted 50 months and never wavered until three months into Secretary Wilkins’ term. That change of direction occurred in March 2011, right in the middle of the public hearings and media frenzy on the Barahona case, though the contraction had been slowing since 2009 and was well on the way to reversing course even without the public outrage to speed it along. (That is, media frenzy tends to reinforce — not set — existing child welfare policy positions.)

Oddly, Secretary Wilkins’ DCF changed its expansionary course by August 2012 and entered a contraction period that continued sharply until the month that he resigned in July 2013. (I’ve never heard a good explanation for that period.) The tide immediately turned back toward expansion, continuing through Interim Secretary Jacobo and halfway through Secretary Carroll’s tenure. Growth peaked in December 2015 and then precipitously fell, flattened, and then fell again. (Note that steady growth is still growth — the chart above shows change. The charts below show the actual counts.)

Even though the system as a whole tends to move in unison, not every geographic area shifts course at the same time. The current contraction has been driven largely by sharp decreases in OOHC in three circuits — 17 (Broward), 11 (Miami), and 18 (Brevard) — which shrunk a total of 670 children over the previous year in October.  The top three growth circuits — 1 (Pensacola), 7 (DeLand), 9 (Orange/Osceola) — only grew by 127 kids in all.

Decreases were clustered largely, but not exclusively, in the southern regions. Here are the changes by county.

The contractions appear driven largely by reductions in removals.  (I’ve chosen to use seasonal trends below to make the changes over time more clear. The actual numbers for removals and discharges have large but regular oscillations month to month due to seasonal effects like summer and national adoption day. The raw numbers are much harder to read.)

You can see the same decreases in removals statewide. Here’s the statewide seasonal trend graph.

Here are the seasonal trend graphs for all circuits. If you notice anything interesting or know why any of these charts look the way they do, let me know.

Circuit 20

Circuit 19

Circuit 18

Circuit 17

Circuit 16

Circuit 15

Circuit 14

Circuit 13

Circuit 12

Circuit 11

Circuit 10

Circuit 9

Circuit 8

Circuit 7

Circuit 6

Circuit 5

Circuit 4

Circuit 3

Circuit 2

Circuit 1

The ACLU of Florida did a fantastic (and super data-heavy) study of racial and ethnic disparities in the Miami criminal justice system called Unequal Treatment. It’s amazing and you should check it out. The study reminded me that DCF publishes its own statistics on race, but they are buried in the Trend Report excel graveyard. This weekend I decided to dig them up for folks to see.

All of the diagrams in this post are in tableaus here:

• Racial Disparity by Case Stage
• Racial Disparity by Placement Type

The analysis is based on data from May 2017 to April 2018.

The gist: DCF’s out-of-home care population is racially disparate. You start with the hypothesis that child abuse is equally likely across all racial populations and the system will treat everyone the same, therefore the OOHC population will mostly look like the general population. It doesn’t. Black kids are over-represented by 33.1% in OOHC. So-call “Other” kids (which are mostly mixed race and Asian kids) are over-represented by 37.4%. White kids, on the other hand, are under-represented by 15.5%. If you divide those numbers to get the ratio, you get approximately 1.59. This means non-white kids are 1.59x represented over white kids.

The differences aren’t uniform across the state. So your next hypothesis might be that whatever is causing the differences would be systemic across the state. It’s not. Racial disparity in OOHC varies greatly among the counties, with some even having a bias towards White kids. The map below shows the disparity index (i.e., the ratio of non-white to white bias in the system). Orange counties have a Non-white bias. Blue counties have a White-bias. (Counties with no statistically significant difference are shaded a neutral taupe color.)

What does a White-bias county look like? Dixie County has the out-of-home care numbers most biased toward White kids (it’s the dark blue county in the map above). The county has approximately 16,000 people, skews slightly Democrat, and has about 14.5% of its population below the poverty line. It is 77% rural and approximately 9.0% Black. It is the third-whitest county in Florida. Based on the race demographics, you would naively expect about four Black kids and 48 White kids in its OOHC population. What you get is 0 Black kids and 51 White. It’s not huge, but it is statistically significant. Compare the next example to see why.

What does a Non-white bias county look like? Miami.  Miami is obviously huge and Latin — it has 2.7M people, and is 65% Hispanic (any race). It is 17.1% Black (non-Hispanic) and 15.4% White (non-Hispanic). About 51% of its population was foreign-born. It voted 63% Democratic in the 2016 elections.  It’s racial disparity is extreme: Non-white kids are over-represented by 140%, while White kids are under-represented by 45%. You would expected about 1,400 white kids in foster care in Miami — you get around 775. Meanwhile, you would expect 435 Black kids, and you find about 1,050. The racial disparity index is 4.25.

Racial disparity generally increases the deeper into the system you get. Your next hypothesis may be that once kids are in the system they are treated by the same rules and same players, and should therefore have similar outcomes. No again.  DCF breaks its numbers down by the stage of a case: Investigation, Verification, Removal, OOHC, Spending more than 12 months in OOHC, and Discharge from care.  Racial disparity tends to rise the farther into a case you get.

The disparity index numbers go something like this. Remember that a positive number means that Non-white kids are represented that many times more than White kids. A negative number is biased towards White kids. A (*) indicates no statistically significant value.

If you look at the Statewide column, you can see that Investigations have a stronger bias than Verifications. Once a child is in care, Discharges tend to be less racially biased than Removals, which actually increases OOHC and 12+  bias over time. The pattern is on steroids for Miami where non-White kids are 4.44x more represented in the 12+ population than White kids.

What about placements? If the process itself has racial bias in it, then it may be safe to bet that placements have a similar bias. This time we assume that the breakdown of kids in a given placement type will be the same as the general OOHC numbers. It’s not. Statewide, Non-white kids are over-represented in the Runaway, Facility, and Other populations, while White kids are slightly over-represented in the Relative and Foster Care populations. The non-Relative caregiver placement did not show any statistically significant differences, possibly because it’s a smaller population and therefore requires more difference to be significant.

The expected vs. actual values for Facility placements look like this.

Breaking the data down by county makes it harder to find statistically significant values. For example, only eight counties show significant differences in their facility placement numbers.

Four counties had significant disparities in their foster home placements, and three of those were White-biased.

This isn’t to say that the other counties are perfectly balanced. When we parse the numbers down to the tiny levels of “the four kids on runaway in Dixie county” then differences have to be more pronounced to distinguish a real difference from just random noise and the techniques I’m using here aren’t very good at small numbers. This data says “we can’t see a difference with the tools we’re using,” not “there is no difference.” 

We can’t tell why from this data. This is also important: this type of observational data does not show causation or even hint at underlying causes. A lot of writing has been done on systemic racism in the child welfare system, and the expert consensus is that the disproportionalities we see here are a consequence of (1) interplay between poverty and race at the individual and community level, (2) heightened governmental surveillance and intervention in non-white communities (like the ACLU report highlights), and (3) personal bias in individual decision-makers (for example the family that only wants to adopt a child of their own race or the judge who is less likely to approve the removal of a child of their own race).

Even if these effects may be undetectable in an individual case (or, more likely, they’re one of a hundred other things going on in a case), when you multiply them across tens-of-thousands of kids and decades, you can start seeing the cumulative impact. You only have to remove one more kid than you discharge each month to grow a population over time. If racial factors increase removals and suppress discharges even marginally, that can explode into real differences that must be addressed.  For a full discussion see Shattered Bonds: The Color of Child Welfare by Dorothy Roberts.

Our office has been handling more appeals lately, and I am learning the rhythm of the process a little better each day. Appeals seem to go like this: (1) you lose or win at trial and feel really emotional about it, (2) you file your appeal or get noticed that someone filed one on you, and (3) you wait until you don’t feel anything at all anymore. Somewhere in there you file a brief. Then you wait some more and file other briefs. Sometimes a court reporter loses your transcripts and tells you your trial never happened. That can rouse some feelings, but they pass. Because mostly you just wait.

And while you’re waiting, everyone is constantly asking you how much longer they’ll have to wait. I haven’t yet mastered delivering earnest but vague statements of reassurance, such as “waiting is good because it means you haven’t lost yet.” I’ve heard that’s what appellate lawyers do. The people waiting don’t think waiting is good, because it means they haven’t won yet either.

I wanted a real answer to the question how much longer? I looked all over the internet. There were reports (cited below) on dependency and TPR apppeals from 2010 and 2015, but no follow-ups or ongoing data on whether those reforms were successful. There were also lengthy reports on trial court clearance statistics. There was nothing (that I could find) on the district courts. So I decided to create something.

But first, an answer to How long do I have to wait on my appeal?

Probably at least 122 days for a dependency or TPR case.

Probably at least 293 days for anything else. 

Probably a little longer if your case is in the Second DCA.

There. Quit asking.

I put it all in a tableau so you can play with it.

The details are really interesting, if you’re into numbers. I put it all into a tableau, a quick version of which should appear here:

A full version with more stats is available here. (You can also use the link if the embedded tableau above didn’t show.) The full version breaks things down by DCA, case type, wins and losses, and originating divisions. I will commit to updating it for a few months to test for stability. I can’t promise after that.

The process – also, why didn’t this already exist?

My plan was basically to dive in, coming up for air every now and then to run the same “florida district court statistics” google search to see if I missed something. If anyone wants to recreate (or check) my work, here’s how it went.

Step 1 to finding an answer was to see what information I even had access to. All of the DCAs report their opinions on their websites. Three of them use a searchable system that creates spreadsheets by month. Two publish weekly text lists that you have to go through on your own.  All of the DCAs use an online docket system that has a very convenient URL interface for going right to the case you want, unless that case is a dependency case.

Step 2 was figuring out how many cases they’re even putting out. My curiosity knows no bounds, but my actual time to spend on this was limited to a week or so.  The answer was about 200 cases per month per DCA. That wasn’t bad.  I planned to do a 10% sample of three months anyway, so 60 cases per DCA felt reasonable.

Step 3 was dealing with the fact that dependency cases are restricted from the public, so they are not available on the online docket system. Instead, I had to look them up on Westlaw and pull out their appellate case numbers and the outcomes. Fortunately, all of the DCAs use a linear case numbering system (for example 15-001 was filed earlier than 15-055 in the year 2015). Once I had case numbers and filing dates of known cases, I could interpolate the dependency filing dates to within a few days. That was good enough for these purposes.

Step 4 was pulling all of the data on 459 cases and punching it into a spreadsheet. I then crunched some probabilities, ran some ANOVAs, generated a few survival reports, and made some tableaus based on what was statistically relevant. Some people have other hobbies, I guess.

What I could tell you about appellate cases would not fill a book

The sample size of three months was enough to get a big picture number, but not enough to do a lot of fine parsing of the data. As I add months in the future, maybe things will stand out. In the meantime, here is what I can say with a reasonable amount of confidence.

More people won than I expected, but still not that many. About 11% of the cases were “wins.” I defined win very broadly to include anything that wasn’t a straight affirmance or dismissal of a petition.

The DCAs were surprisingly similar.  I was concerned that a 10% sample would result in garbage. It didn’t. All of the samples were roughly normal. The 1st, 3rd, 4th, and 5th all had numbers that were statistically indistinguishable. (A bigger dataset may eventually tease them apart, but this one didn’t.) Only the 2nd DCA stood out as statistically higher than the rest. For example, the 2nd DCA processed half of its cases in 282 days (+/- 18), while the statewide average was 208 days (+/- 11).

Below is a survival graph. Imagine the top left corner as the starting line, and each district racing to the bottom. The cumulative survival of 1.0 equals 100% of cases still open (“surviving”), and 0.4 would equal 40% of cases still open. The first to the bottom (measured in days across the bottom) is the fastest. As you can see below, four of the DCAs reach the bottom at about the same time.  The 2nd DCA stands out as statistically different, in large part because it was slower off the line and struggled with its last 20% of cases compared to other districts.

There wasn’t much variation among the types of cases, except for dependency. The average of 208 days also applied to case types, but dependency stood out as significantly faster. It took the DCAs only 121 days (+/- 2) to process half of their dependency cases. Civil and criminal were indistinguishable in this dataset, though more info later may tease them out as well. There weren’t enough probate, worker’s comp, family, or administrative appeals to say much about them individually yet.

You can see below that dependency cases resolved much faster than anything else. Civil, criminal, and family are pretty consistent in the middle. (Civil starts out slower, but eventually catches up to criminal.) The jaggy curves are probate and worker’s comp cases, which only had a few examples of each.

There was no measurable difference between writs and appeals. Again, a larger dataset may tease out a difference, but the line for writs and appeals were indistinguishable in this one.

Dependency “wins” follow the curve, but exaggerate it a little.  Again again, there aren’t that many dependency wins either. But in this dataset at least, they tended to come out faster at first, then move closer to the win curve above after a case has already taken about 150 days. This is a slight exaggeration of the full win curve above, which also flips somewhere around 150 days.

You can’t predict a win based solely on amount of time open.  Again, I want to stress that there are very few wins in general (11%) and they are scattered across the timeline. Knowing that an appeal has been open for 600 days doesn’t tell you much about its eventual outcome because the last 10% of the “loss” line accounts for far more cases than the last 10% of the “win” line.

Even though wins are a little faster or slower as a group, you can only know that after you know the outcome of the case. I ran the numbers — if you only know how many days the appeal took, you can predict a win with 5% accuracy. Adding in the DCA, appeal type, and division only gets you to 11% accuracy. That’s worse than guessing.

The takeaway

The good news is that this data supports a claim that the Court’s previous efforts (below) to speed up dependency appeals actually worked. Only time will tell if that is a stable finding or if I just happened to look at a particularly fast few months.  Stay tuned.

More info on the appellate courts

Recommendations for Dependency and Termination of Parental Rights Appeals (Florida Supreme Court – Commission on District and Trial Court Performance & Accountability – April 2016)

Dependency and Termination of Parental Rights Appeals (Florida Supreme Court – Commission on District and Trial Court Performance & Accountability – June 2014)

A Review of the Florida District Courts of Appeal Boundaries and Workload (OPPAGA – February 2017)

Examining the Work of State Courts: An analysis of 2010 state court caseloads (Court Statistics Project – December 2012)

Clearance Rate Dashboards (flcourts.org – September 2009 to present)