[MUSIC]
[VOICEOVER]
National Data Archive on Child Abuse and Neglect 

[Clayton Covington]
Welcome everyone to the 2024 Summer Training Series hosted here at the National Data Archive On Child Abuse Neglect I'm going to offer a brief introduction before we get started with our presenter. As a reminder before we get started you're going to hear a presentation and you may understandably have questions while the presentation's ongoing. With the way that this is formatted please make sure to use the Q&A box at the bottom of your screen on Zoom. We'll answer all the questions that we can after reading them aloud and then we'll get those answers to you. And should you have any questions or need support via Zoom you can use the link on your screen or reach out to Andres Arroyo the Archiving Assistant. Next slide please. So as a reminder if you haven't joined us before the theme for this year's series is around the best practices and the use of NDACAN data. We've done a variety of themes in past but this one is making sure that as you know studied users as new users you all know the best approaches and some you know cautions to take when approaching our data sets. Next slide. So this particular presentation is going to focus on the AFCARS dataset focusing on both strengths and limitations. And if you weren't already aware of the National Data Archive and Child Collect is funded through the Children's Bureau which is under the office of Administration for Children and Families within the Department of Health and Human Services. Next slide. So just to give you a little idea of where we've been and where we're going. We're about at the midpoint of this year's series we have a kind of a recurring theme of our various datasets strengths the limitations which started with the NCANDS dataset during our first week. We then did some review of some of the reporting issues with the NCANDS and AFCARS datasets with our research associate Dr. Alexander Roehrkasse this week is going to focus again on the AFCARS strength limitation before we move on to survey design and using weights. We'll our penultimate session is going to be the NSCAW 3 dataset for both experienced and new users before closing out with the NYTD strengths limitations presentation. And without further ado I will pass over to our presenter NDACAN Statistician Sarah Sernaker.

[Sarah Sernaker]
Hello everyone I'm sure if you've come to one of these before you've heard my voice and heard my presentation style. So I'll just right in thanks for joining us today. As Clayton said this is our strengths and limitations series. I hope this is helpful even for people who have used AFCARS and maybe thinking about things you hadn't thought about before or just what else you can get out of it. So today we'll just do a quick background on AFCARS what it is and includes so that we're all on the same playing field and then we'll dive into some strengths and limitations thereof. So what is AFCARS? AFCARS is the Adoption and Foster Care Analysis and Reporting System and it's a federally-mandated data collection where all states have to report information on children in foster care. Other terms are removed from their home or placed in out-of-home care or those who have been adopted with a state agency's involvement. So not really private adoptions. And that is a federally-mandated collection system this has been in place for over 20 years now and so we have full state reporting for almost every single year. I think in the first few years one state didn't respond. But anyway so it's very complete for a long series of time. And the data are used to provide summaries and information to the public and for policy makers. They help policy makers and states monitor progress on improving child welfare programs. It helps them understand how implemented programs or statutes affect the child welfare system. You can see trends over time and you know observe real changes from the data. And this truly helps states and again policy makers lawmakers allocate money, funding, sponsorship for programs. So this is really a great data collection system and you know really a testament to our government's data collection functioning I think. So with the data collection it is annual annually submitted data for the United States but I would say to consider the data collection as an individual data collection at the state level. So each state as a sort of individual data collecting entity. But you know with that being said all states are required to submit data on sort of two prongs. The foster care population which is all children in foster care and For Whom the state child welfare agency has responsibility for placement case or supervision. So it's children in out-of-home care or who are receiving foster care services like Title IV-E Services. The other sort of prong is the adoptions data and so that states are required to collect information on all adopted children who are replaced by the state child welfare agency and etc. Etc.. It's mostly public adoptions. There private adoptions can voluntarily submit information and I've written you know the private adoptions right here but the majority are public adoptions facilitated by the child welfare agency. And so the AFCARS has been you know a fixture and constant since 1993. Congress literally mandates what variables need to be collected by each state. And so states submit to Children's Bureau they collect the data and they submit to Children's Bureau 37 data elements related to adoption. And there are 66 data elements related to foster care. And that's been regulated since 1993. And I'm kind of making a point here because there's been a new ruling what's called the AFCARS 2020 Final Rule and what while most of the data elements of the existing foster care you know AFCARS will remain in place more data is actually getting collected. And so actually the AFCARS moving forward from this time point out is going to look very different than the AFCARS that I've kind of built this presentation on and that you're used to seeing. And really it's a really great thing but we don't have the data yet so I can't really speak or show examples using it. We have the data yet it's been a huge undertaking for states to sort of shift gears. They have to update their their computing like their data collection systems. And so long story short it says 2020 Final Rule that was when you know Congress mandated this sort of expansion of data collection but it's still being implemented is my point. So we're still kind of transitioning to this and we'll have new data probably next year in this new format. And it's just going to get better through the years as states adapt to this new change. But I'll circle back to that at the very end just to touch on it again. What you need to know is the AFCARS that's out there right now has been the AFCARS that's been that'll be talked about in this presentation and has been in place for over 20 years. So it's like the existing data that we have and the longstanding data. So who's included in the reporting population? This is just to show sort of the granularity of the children captured in the Foster Care File. This is strictly or mostly relates to the Foster Care File. And so this is 50 states plus D.C. In Puerto Rico they submit information on children who meet the following criteria. So if you're in a Title IV-E agency's responsibility think like your child welfare the state's child welfare agency so runaways, other public agencies, children whose removal episode is more than 24 hours. So I don't want to go through this list in verbatim. You guys can read it and it's also in the users's guide but this is just to show again like the granularity of the children who are showing up in the foster care system. And it is a sort of broad population of those receiving public services. This is additionally again to show sort of the expansive nature of the foster care and child welfare system this was to touch on the fact that children so zero age 0 to 17 may be placed in out-of-home care and for a while that was you know the majority of the Foster Care File but actually states have expanded laws so that children up to the age of 19, 20, or 21, in some states are still receiving Title IV-E foster care payments or services. And this was to help increase the outcomes of these kids and just provide them with transitional services as they enter adulthood. So again this is kind of to show the breadth of what's in the Foster Care File and this is mostly just Foster Care File-related. But let's take a little bit step back and talk about the two prongs I had kind of mentioned. When we talk about AFCARS data there are two distinct datasets you know you could come and order from us. One is called the Foster Care File. This is an annual record, there's one record per child each year and this is the most recent record of the child's experience in foster care. So it's sort of like a year-end summary of a child's history through the foster care system that year so how many times were they removed there's basic demographics etc., Etc.. And there's Statewide and national information on the status of the foster care population type of placement, case plan goals, services received. So a sort of broad look at the case itself. I will say when we talk about Foster Care File we are most often almost always referring to what's called the annual file and it's kind of you know in the same breath the Foster Care File is the annual Foster Care File but I will say there are what's called six-month file versions of the Foster Care File. And so this instead of being an annual look and one row per child per year, there are six-month records which can provide some more granularity and maybe a little more nuance and understanding multiple placement settings. I felt the need to mention it because they do exist. It's not as long-lasting or has not been in place as long as the Foster Care File it only goes back to 2010. And the other thing with these six-month versions is they're not as thoroughly cleaned. Like the annual file goes through rigorous cleaning, conforming, we create additional variables in there and that just is not the case with the six-month file. And so while it can add like a little some additional benefit we generally do not recommend using it and we generally tell people to stick with the annual file and that's almost always sufficient for research purposes. But we're also happy to talk about you know if you're trying to use that and you need help doing so. I just with caution and you just need to use it more carefully than the annual file. So that's the foster care holdings and then we have what called the Adoption File and so that's been around since 2000 as well. States submit data for each child with a finalized adoption, again where a public child welfare agency was involved or if like private agencies are voluntarily submitting. I say that and it's a truth it's a matter of fact of the data but I don't actually know that there's a way to identify such cases in the Adoption File. But in any case it's individual-level data by fiscal year and this includes information on the number and characteristics of adoptive and foster care children and their parents. So demographics of the child and their adoptive parents themselves. And there's a few sort of case-related information and services information there. But I will say when we talk about AFCARS and if someone mentions AFCARS quickly in passing it's most often the case they're talking about the annual Foster Care File. People and I will use it from here on out probably synonymously AFCARS with the annual Foster Care File. But before we you know dive full in on the Foster Care File, I just wanted to talk briefly about the Adoption File and what's included there. And as mentioned adoptions so what gets submitted to AFCARS are adoptions that states are legally adoptions that states are legally required to report to AFCARS include children placed for adoption by the child welfare agency who have been in public foster care system and were placed for adoption by a private agency but under contract with the public child welfare agency, or children in whose adoption the public child welfare agency was involved. It's the common theme is the public child welfare agency is involved with the adoption. The adoptions where states are not legally required to report to AFCARS are those involving children who are not in the public foster care system, placed for adoption by tribal agencies, or placed for adoption by private agencies. Again they may voluntarily report it to AFCARS but they do not involve a state agency and they're not included in the publicly available version of the data. So from here on out I think the majority of the focus is the Foster Care File. So we just dive into this more. So the Foster Care File as I mentioned is data on the individual child level and sort of their experience through foster care. There's child demographics such as sex, birth date, race and ethnicity, number of stays in foster care, service goals, dates of the original removal from the home, dates of the most recent removal, dates of discharge from foster care. And so you know one of the strengths we'll talk about is over all the years you can really track a child moving in and out of the system and sort of create a timeline over their whole life of their experience with the child welfare system. And data are submitted for the Foster Care File to the Children's Bureau who review, clean, compile and then abstract it into the annual file. And they do this using the six-month file which I'll get into in a second but this is really why the annual file is the dataset you should be using for your research because Children's Bureau and then they submit to NDACAN and NDACAN does additional processing. That is the cleanest file and the most like conformed data, the annual Foster Care File that is. And once they do all that the the nice the sort of nice thing about the Foster Care File in you know comparison to NCANDS is you only have one record per child in the foster care database so it kind of simplifies things. I won't get into NCANDS but just not the case with NCANDS and there's a lot of stuff in there. So that can kind of simplify things in some analytic cases I think. And as I mentioned since there is only one record per child in the annual databas,e you know if a child has been in and out throughout the year, the record in the annual database will be the most recent record received for the child search. So it's just the most recent series the most recent episode most recent removal. And just to kind of elaborate on the reporting in relation to the six-month files and the annual files. So as we've mentioned states have to submit an AFCARS record for every report, so every child in foster care, and they actually report report on every what's called reporting periods and those are six-month periods. So these are directly like related to the six-month files. So states submit what's called an "A" period which is six months the first six months of a fiscal year then they submit a "B" period which is the second six months of the fiscal year. And when Children's Bureau goes through cleaning they say if a child has a record in both six months periods the record from the second period will be retained. So as I just mentioned you know if a child has multiple episodes in a year, the record the most recent record is what's being retained in the annual. So the record from the first period will not be included in the annual file. There are some cumulative measures in the annual file that account for you know any removals or experience in the first period, but again the record from the first period is not retained in its original state. So as I keep saying a child who is in or comes into care at the start of the year or exits care and then later returns to care so basically re-entering they will have one record in the annual database the one submitted in the second report period when they re-entered. So that is something to just keep in mind sort of the functioning of how the annual file is put together and how you know maybe this is something I should have tied to a limitation. Sometimes you lose the granularity if a child was really active in the first period and but they reenter in the second period and sort of that data overwrites it. Like I said we have some cumulative measures to capture a little bit of that but you're not getting the straightup record within that period. So that's sort of the reporting period. Another thing to keep in mind with the Foster Care File reporting is some foster care entries and discharges for the most recent fiscal year, so you're using you know this fiscal year, the entries or discharges might not be submitted until after the close of the fiscal year. So if someone's entering foster care and like September 29th of 2014 let's say, if we're processing the data that record is just not going to get captured in time for the data release of the 2014 fiscal year. And so we would put out the data but we do get data resubmissions so V1 version one of the data files when they go out are almost wholly complete but there are some cases that we are missing and it's just something to be aware of as you use the data. It's most relevant in you know newer years as we just don't have data resubmissions and we don't have the actual data records yet. So kind of a big thing to keep in mind as you're working with recent data. Okay so that's the background let's get into the strengths of AFCARS. So AFCARS is all about foster care and adoption it's the best available data on the subject. I mean this is literally data that's mandated to be collected by Congress. This is you know like the the best information on the subject of children being removed from the home. It's statewide and national information there's characteristics of adoptive and foster care children and their parents, there's information about the child themselves, there's, you know just a whole host of information that is available only within AFCARS. And there's really no comparable dataset. And I tell people the only way you're going to get better data is truly if you work directly with the state. That's the only way I could see you get even better data than the AFCARS. But that's not really feasible on a large scale level. Also, because AFCARS has been around for so long a lot of people have been using it and there's a lot of data support that has been built up and documentation and resources. Not only through NDACAN but through others particularly Children's Bureau. They put out what's called the AFCARS report every year and so those are sort of high-level statistics to give an idea of changing trends and sort of just basic counts of children in foster care and I think they break it down by demographics like race and sex too. But this is all to say there's a lot of stuff out there and NDACAN is here to help and we love to help people. So please you know if you're using AFCARS or want to start using AFCARS and need help please reach out. With the data itself, so besides being you know the best data on the subject, it's even better because we can link it with our additional great data holdings of NCANDS which is child maltreatment reports and NYTD. So NYTD is a the  gist of NYTD is a it's a survey given to youth aging out of foster care and kind of follows their outcomes over three waves of time. And so not only you know does AFCARS hold a lot of richness itself with the foster care system, it's also linkable to our other datasets which really provides the broad scope of you know a child experiencing the child welfare system. Maybe they start with a maltreatment report in NCANDS and they're removed from their home, they show up in AFCARS and unfortunately maybe they age out of foster care and end up in the NYTD survey and you can track outcomes thereof. So it's a hugely rich not just AFCARS but our whole you know admin holdings are hugely rich and so much research potential. As I've mentioned there are many years of data for long-term tracking changes over time or longitudinal analysis. All states have reported data since 2001. So you really have a complete look of all states over 20 over 20 years of data. And we'll talk about linking children you can link children between years. So like I said sort of creating a timeline of their experience with a child welfare system. There are some caveats which I'll get into in the limitations section though. But with all these years of data you can detect effects of impact or impact of policy or statutes. So any new policies that have been enacted at a state level, you could observe you know the effect in the data. You could observe the impact of historical events such as Covid-19. I also added the opioid epidemic which I have visuals for in a second but, yeah. So it's I very interesting just at you know a basic level it's really interesting to see you can make graphs which I'll show you and you can directly see the impact of these large scale events on the child welfare system. And not only do we have over time but we have complete data for all states since basically the creation of AFCARS so you can compare information between states. And not only states there are some identified counties I'll get into that I feel like that's more of a limitation if you're familiar with AFCARS. But there are some county level identifiers in AFCARS so not even just comparing between states but comparing some counties also. There are some limitations when you compare between states but I'll save that for our limitations section. Further as I mentioned it's not just information about the child itself, so like the child's experience, but there's information about the caregiver the you know the original caregiver like the parents from the home from which the child was removed and also foster care caregivers and adoptive parents. So there's demographic such as age, sex, race of them and services received by the family. Yes so that's my next bullet there's also information not just about the children and the family itself but also more broadly about the services or assistance received. And as I mentioned some of those things were expanded to include children you know 18, 19, 20, or 21 who still benefit from the foster care assistance. There's also information about risk factors for child and caretakers such as substance abuse or disability. So yeah that's really great to understand sort of the play between different risks and removal from the home. Yeah so let me take a brief hopefully fun slide from my rambling. I wanted to just visually show how you can observe historical trends over time just using the Foster Care File. This is this shows a graph of all states and this is entries into foster care. So this is number of children who are entering foster care whether it's a re-entry or a their first entry. This is just number of unique individuals entering Foster Care by state and year. You can see a lot of bumps and humps and ups and downs but when I put these slides together I will say I was reading the book "Dopesick" so I was very much trying trying to look at patterns related to opioid the opioid crisis. And so one thing I'll highlight is you know West Virginia in particularly for those who are not so aware the opioid crisis kind of peaked leading up to 2010 and then there was some legislation and I think some acts to try to curtail the availability of oxycontin and I think that's sort of the effect we see of these bumps and then valleys. But then you know as we've been seeing in the news we've the drug opioid epidemic has kind of switched from pills to fentanyl. So I feel like some of the peaks we're seeing are from those original opioid epidemic and then maybe this fentanyl crisis. But you know there's so many other factors again I made this with "Dopesick" in mind so I was very much looking at a sort of proxy to the opioid crisis. One other thing really quickly while we're here. If you look at 2020 when Covid hit you see a steep decline in every single state. And that is strictly because as you saw in if you came to the NCANDS presentation I showed a graph of maltreatment reports and how Covid affected that and there was a steep decline of maltreatment reports and that's because children during Covid a large majority you know were at home and just not being observed by others so much. So it's really just a function of you know unobservable events and events that are were just not captured. So maltreatment reports therefore less children entering out-of-home care. So I just really interesting stuff. Continuing on in my opioid sort of purview, I wanted to see the proportion of removals attributed to reasons proxy to the opioid epidemic. So a child when they're removed from the home in the data we can see the removal reason and a child could have multiple removal reasons I will say so these do not all add up to one. So child could have multiple removal reasons but I've taken I think we have up to 15 in the data so I've taken a subset here of things that I thought would be related to the opioid epidemic. AA child is alcohol abuse of the child, DA is drug abuse so it's of the child versus parent. So again looking at West Virginia this pink line so this is children removed attributable to drug abuse by the parent and you can see in all of these sort of Appalachia states that this just goes up. So yeah I know this was just a sort of fun side project I thought to really highlight the benefits and what you can get out of AFCARS, I mean, I don't know about you guys I thought this was super interesting so yeah I just thought this was fun. This was again just a sort of higher level look so this is at the national level by the removal reasons I'll go quickly through this. But again you can see this drug abuse by parent kind of peaks then there's a little kind of slow decline and then it goes back up. But yeah so after all that fun stuff fun good stuff let's talk about the limitations because there are some real ones and you know with any dataset there's limitations and I think it's just when you're using AFCARS you need to understand what you have and understand what conclusions you can make and sometimes just be upfront with where you know you can't make you know direct conclusions or direct inference. So one of the things is the variation over states and time. So within the data itself I briefly mentioned you can link the years of AFCARS within themselves. So you can link you know 2019 AFCARS data with 2010 AFCARS data by the Child ID. So you're you're trying to see if a child was in 2009 to 2010 so you try to link them or whatever I just said. But the AFCARS ID which makes the data linkable between itself is an ID that's created at the state level and they create these IDs and then the IDs are encrypted before they're sent to the Children's Bureau and that's for privacy and protection and it definitely should be encrypted. The problem is is that states sometimes change their encryption algorithm in a year, between years. And when they change their encryption algorithm we just can't track that they've just basically created a whole new ID for the child. And so in some years we call we have what's called this sort of breakage in linkage where from one year to the next the state has changed their ID encryption which ultimately means we don't have a linkable ID from one year to the next of a child. We do have a table that tracks this. I did not include it here I've provided it in the NCANDS slides as well you see this in NCANDS as well but we have this available by request. I can elaborate on this and we have a table tracking which states are linkable. I will say I don't think the problem is as widespread in AFCARS as it is in NCANDS. And it's not even that widespread this is a problem that's more so in the earlier years and less so in the more recent years and really just a small handful of states. But very much something to keep in mind. The other huge thing is the state-to-state variation and child maltreatment laws, statutes, definitions and information systems. So as I said each state is like its own data collection entity. And not only like a data collection entity where you know the methods of data collection, their data collection computers and information systems are different, but the state laws and statutes and legal definitions are different between states. So what does that mean? Sometimes you're comparing apples with oranges because if a state let's say a state categorizes something one way and another state categorizes something a different way all the states have to conform to those data elements that I talked about like slide three each state submits 66 I think it was data elements. So while each state has these different variations they have to conform their data so that we can put it all together in this database. And so sometimes there are differences when you see a one or an indicator on one variable in one state versus an indicator in a different state. And so you should always keep that in mind sometimes you know what you're observing in the data or differences in the data may simply be just differences in state variation of laws or data collection systems. That's not often the case but it is something you know I with your research project and dissertations should be like a point that you investigate to kind of understand those variations. And investigating includes looking at our state footnotes associated with each state so. State footnotes come with the date data I'm hesitating because I can't remember if it's NCANDS or AFCARS or both. But there are state footnotes that provide a little context about you know what a state captures in a variable. I usually direct people to the the children the Child Maltreatment Report. Children's Bureau puts out a Child Maltreatment Report every year and in the appendix of these child maltreatment reports there are state summaries and there are kind of nuggets of information there. Like a state might say well "we group reports from these people all into neglect" or I don't know. Just like the state-level differences. There's also a dataset called SCAN that's capital s c a n. That's basically metadata about a state's statute. So it's indicators of like what a state includes in their definitions or like who's a mandated reporter for instance or like you know are they including X Y and Z in the definition of neglect or physical abuse or XYZ. So SCAN dataset is a hugely beneficial resource when trying to understand these variations over states. But yeah that's I feel like the biggest one you have to think about when analyzing any administrative dataset. Okay so another limitation is I mentioned you can link AFCARS and NCANDS and the way you can link AFCARS and NCANDS is there's a variable in both datasets it's called the AFCARS ID I don't know I think that V I think that's the variable name in both datasets but it's the AFCARS ID and it's found in NCANDS and AFCARS the AFCARS ID is always present and available for everyone in the AFCARS data but it's not always available in all states in NCANDS. And that's because in NCANDS, NCANDS is a separate data system so the addition of the AFCARS ID is kind of at the discretion of the states. And so some states just don't provide an AFCARS ID. And so this is what this table is trying to show. So what you have here is out of all of the submitting states of NCANDS which states do not provide an AFCARS ID. So these are basically states and years in which you could not link to AFCARS. So what you see is the early years, the early years are kind of rocky for NCANDS it's always a case so you can't really link the early years with AFCARS. This is saying all of the submitting states in NCANDS in this year did not have an AFCARS ID. So it's kind of maybe like weirdly worded but this is saying that no states could be linked. There was no states submitting AFCARS ID in 2000. And then you see over time this gets better and better there's some like you know common offenders Puerto Rico, Vermont. But it gets better over time. But still to this day Illinois, Pennsylvania, and Vermont are going to cause problems if you're trying to link between NCANDS and AFCARS. So yeah I'll just, I'll just leave it at that you can't always link with NCANDS but I mean you can see you can still link a lot so there's still a huge benefit in trying to do so and I think still worthwhile in trying. We have, like any data suppressed, and missing data well missing data is in any data but we also apply some suppression. There are some also sort of weird missing nuances. So there's some incomplete records you'll see in AFCARS and there's a really great description in the users guide that Michael Dineen has like draft has written up it's on our it's on the AFCARS website. But basically sometimes a child will enter foster care and they appear in the records but they don't have an entry date. Or you'll find a child was in 2015 and they don't have a discharge date so it seems like they never left foster care but then they don't show up ever again and I think Michael's called those we've called them lost cases. You know you don't know what happened to them. So it's very small proportion but you know these things exist. These are administrative datasets and there are you know just small fun hurdles that come up. Missing data and missing codes I noted here just again this comes up in basically any data any real dataset you have there's going to be missingness. I will say missingness varies between state and year. I generally tell people you should do a you know of all of your main responses or explanatory variables do a sensitivity check or understand your main variables' missingness by state and year. Some states just outright don't report stuff and some states you'll see have kind of weird missing patterns and I've seen cases where one variable in a state it's all ones or missing and so that seems to me a sort of more informative missing pattern than just a bunch of ones and zeros and missings. So I just sort of the standard things to think about with missing data I wrote missing codes here just to note that some variables we code missing as nine for instance and other variables might be a 99. You should always be working with your codebook up as you work with the data to make sure that you're capturing all that. We do suppress some information. The biggest one I'll say that comes up a lot I think is the county masking and so we do mask counties who have less than a thousand records in the year. And so inevitably you we almost always mask all small counties and rural populations just because of the sheer number of records just never reaches a thousand. So as of 2021 in the Foster Care File there was 101 identifiable counties and to put it in perspective there's over 3,000 counties in the United States. So this is not a huge proportion of identifiable counties. I will note we are changing our suppression threshold so there will be new there will be AFCARS re-released and we are changing the suppression threshold from 1,000 to 700. And what does that mean? There will be more identifiable counties. I think changing the threshold off the top of my head changes this from about 101 to about 150 or something thereabouts. So again not great but we do this for the sake of balancing data privacy, data disclosure risk, you know these are really at-risk populations and we'd rather be more cautious than not and so yeah the suppression is it's just an unfortunate fact that needs to happen when we're dealing with data like this. So that's the county masking. We also mask dates for instance we mask the date of birth to the 15th of the month. And so that's sort of our anchor date, everyone's datee of birth is masked to the 15th of the month and any other date that's provided in the data are then shifted accordingly to maintain the time span of the date of birth with the other dates and all the other dates between themselves. So for instance if the date of birth was June 13th and we rounded it to the 15th all the other dates in the dataset would be shifted 2 days forward to maintain that time span. So just to visualize the county suppression I made this fun figure so in red are the counties identifiable at our current threshold of a thousand. And so you can see this state which I'm is really embarrassing I don't what state is this this is Montana this is Iowa this is Idaho this is embarrassing but whatever state this is has no counties represented. This one is North Dakota I know that North Dakota has no counties. So Louisiana under the current threshold and even with the new threshold unfortunately we just don't have you know super rich county information. You have your metropolitan and populace counties and so you know there is there is information to be had here and it is something. But the pink are the additional counties that will be available under our new threshold. And so you can see they're sort of scattered all over but and Hawaii and Alaska were omitted by the limitations of my R package but. So other things just to keep in mind with AFCARS as you're using it there are problems when you want to link the Foster Care File and the Adoption File. As I kind of mentioned back way back when the Foster Care File is the main focus of this you know presentation and most of the limitations also apply to the Adoption File but when we're talking about utilizing them together you can have some problems with linkage and it's really state it's really like a state basis. Some states give a give children a new identifier when they're adopted. So because you know they're getting a new identifier they're just not able to to be linked with their original AFCARS ID. And I have a table actually in the next slide that will speak to that. The annual file uses the most recent information. So if a child has multiple foster care episodes in the same fiscal year you might lose some of that information. So this is what I was going on about with the six-month files being conformed into an annual file and how you have like the most recent information but not as much richness or granularity if a child is going through multiple episodes. Generally administrative data is developed as an ongoing data collection system and does not conform to rigorous criteria for scientific research design. I have this in our NCANDS slide and it's just the case for all administrative data what it means is it's a data collection that's mandated by Congress but it there's there's no research design or whatnot put into it. It's really like data for the sake of collecting data and getting you know the best research we can out of it. But as you can see like admin data just inherently have its flaws. I just mention this quickly here if you've ever used AFCARS the AFCARS IDs uses UTF-8 characters and what you need to know is it's more extensive than standard ASCII. So like standard ASCII is what you would see the characters on your keyboard that's like your standard ASCII but UTF-8 characters include what I call weird characters sort of like your Copyright symbol like that could be a UTF-8 character. And so the AFCARS ID uses these UTF-8 characters and some programming languages just can't read them and it just causes problems. And I think it's just most problematic if you're trying to link because then you're going to have a problem with your linking ID. But otherwise it's not so much of a problem if you're like if you're aggregating data or using one year at a time. So this was just to touch on linking Adoption and Foster Care File to give you a sense of quote unquote good states and bad states that shows the proportion of children in the 2021 Adoption File who could be linked with their child ID to the children in the 2021 oops Foster Care File who exited by adoption. So basically just trying to link the foster care children who exited by adoption with the children who are in the Adoption File. In theory they should be this overlap completely. But you'll see here a low proportion or zero means children in the Adoption File cannot be linked to the AFCARS file. So for instance Alabama 0% have been matched. In contrast to Alaska where you can basically match everyone. So again this is just to highlight linking foster care and the Adoption File the limitations but again broadly the very important case that each state is very different from each other and like if you could do you know understanding especially of your main variables at a state level you should definitely just make sure you understand all the differences between states. Some quick additional considerations as I mentioned always do state bye exploration of any variables for research. Understand missingnesso reporting differences. Just do frequencies, crosstabs, just basic visualizations. Just give yourself a sanity check that the differences and conclusions and any significance that you're observing in your model is a function of the data and not just simply because this state didn't submit anything or you know it it's not attributable to the state differences. Consider multi-level modeling. If you've ever heard a presentation by Frank, Frank's like our multi-level modeling guy I think we have a lot of resources on our site or past presentations by him. This is just a great way to account for those differences at the state level, state and years. Because you know even within a state, statutes and reporting and things change over time. So multi-level modeling is a great way to help adjust and account for just those basic variations. Refer to publish reports or state footnotes for additional information. Use context see how other people are using it, what other people adjusted for, or, you know, listed as their limitations to understand the scope of it. And seek assistance as I said we are always here we offer hands-on direct support to users we love help helping users, we want to help users, we want you to use the data appropriately and get the best information and research you can out of it. There's also CB (Children's Bureau) reports I mentioned and published literature if you just go Google Scholar or NDACAN maintains a database of citations related to our datasets. So we have it's a Zotero citation database and so we have tags for each of our datasets and you can find published lit related to AFCARS. And the last slide I have here I just wanted to circle back because if you're a current AFCARS user you're going to be in for like a huge change in the next few years because as I briefly mentioned this is all with respect to the existing AFCARS which was created under the 1993 rule but the AFCARS is going to be undergoing a huge change over the next year and two as we're like you know adjusting. And it's going to be a really great change. It's going to be great. There's going to be bumpiness as we adapt but it's just going to be so much more information there's going to be more granularity at each placement setting. So some of the limitations we've talked about will actually be ironed out with this new dataset. We'll have more granularity. There's going to be more granularity and information about Native American affiliation which is you know has become a big point of research for people. There's expanded children demographics such their education. Are they getting What grade are they in? Are they being homeschooled or not? There's information about teen pregnancy and placements with their baby or their siblings or other family. So it's going to be really really it's gonna be great. I think it's gonna be super great. The only thing is you know the or because of the expanded information the data files are just going to look different and I think that's the biggest thing. NDACAN will be here to help us all through this transition with the additional help of CB. I will say the first few years is going to be Rocky some of you might have already noticed we have not put out a 2022 AFCARS. And that's because everyone sort of in this state of transition, states were transitioning away from the old one and trying to collect new ones, some states were still collecting old ones and while adapting to the new one. So like these first few years as we adapt to this new data are going to be bumpy and the data quality might not be as great. But once we get this infrastructure and all of the wrinkles ironed out this new AFCARS is is going to be really really awesome. And I've included some resources here you can see this is from CB just more details about it. But felt like I couldn't talk about AFCARS without talking about the new AFCARS. And I've actually finished with six minutes.

[Clayton Covington]
All right what a record, thank you Sarah!

[Sarah Sernaker]
It is! 

[Clayton Covington]
we're gonna start with our first question and just so our presenter or our audience knows we're going to get to as many of these as we can but we might not get to all of them. The first asks, is the annual AFCARS file for foster care one file per child for the entire year? Just want to make sure I understand that correctly.

[Sarah Sernaker]
Yes so with the annual AFCARS file there is one observation per child for the entire year. Yes. 

[Clayton Covington]
All right the next question says if someone has more than two placements in a single year is there a way to tell this in the data? Also, does AFCARS have any information that can be helpful in identifying siblings in foster care? I'm interested in understanding what happens to siblings in foster care placement and having difficulty locating a dataset that speaks to that demographic.

[Sarah Sernaker]
So the first part of your question if someone has more than two placements in a single year is there a way to tell this in the data? So there are some cumulative measures. So for instance if a child was first removed, let's say in Winter of 2009, and then they enter they leave and they re-enter in 2010, so the second half. So we would have the 2010 their re-entry data but we'd have a cumulative measure of the number of times they were removed, and that would reflect that this child was removed twice that year. There's also a variable called number of placement settings, but when you're thinking about this foster care you have to be careful: removals from the home are not exactly the same as episodes, or placement settings. But so all you can do is use those cumulative measures and sometimes they're imperfect but that's really the best we have. And I think at the very least you should be able to identify like is this the child's first entry or re-entry? Sometimes you can't discern exactly which number re-entry it is but you should be able to at the very least say this is a re-entry and not a first entry. So for the second half does AFCARS have any information to identify siblings? Unfortunately not and as I just mentioned one of the you know features of the new AFCARS is we should have more information to identify the family unit and siblings in foster care. I think the best I would say you could maybe sort sort of get at is you'd have to take AFCARS and link it to NCANDS and if you could find the child in NCANDS they might have a maltreatment report and the report might list more than one child. But even then like there's no there's no explicit measure in any of our data to say this child is this child's sibling or like this child has a sibling in foster care. We just don't have that measure. So hopefully in the new AFCARS that will be able to help you identify such situations. But you have to wait a little bit for that.

[Clayton Covington]
All right the next question says if a child only has a record in the first reporting period, the first six months of that year, and not in the second reporting period, will the record from the first reporting period be retained in the annual file? 

[Sarah Sernaker]
No. So because I kind of mentioned in a few different ways the first report itself will not be retained. So all of the raw measured information is not retained but as I said there are some cumulative measures that would account for the first report or record in the first report such as like I said number of placements or number of removals, those are cumulative things that can be sort of used to piece together if it was an entry or a re-entry or what. But I mean it is imperfect and sometimes we can't exactly get at the measure we hope.

[Clayton Covington]
One quick followup to that: is the annual file is at the child level, what level is the six-month file at?

[Sarah Sernaker]
Six-month file is also at the child-level. Really think of the six-month file as sort of the same but like worse quality annual file. So it's kind of like all the same things you'd see in the annual file which is at the child level and child case-level things but it's just measured at more time points and it's not cleaned as well. I can't emphasize that enough like we don't usually recommend using the six-month file. So to to answer question it is at the child level.

[Clayton Covington]
So the last question we'll be able to get to today is is there a resource that provides specifications for the outcome measures in the child welfare Outcomes Reports to Congress I've struggled to figure out how to do this for measures that require information on multiple custody episodes for example children re-entering foster care.

[Sarah Sernaker]
So the child welfare outcomes reports to Congress I'll just say this is the first I'm a becoming aware of such reports. So I'll just say we're not involved with the reports it's probably on CB's end so I don't know how they're constructed. I'll I will say I mean reach out to me and if you're if you've tried to put it together and are not even coming close then I we can help you know try to figure it out. But I don't know how those reports are put together or have them I've not seen them. So I think that'd be a good one to reach out and I can help. 

[Clayton Covington]
Okay Sarah. 

[Sarah Sernaker]
Someone told me that was Wyoming thank you yeah!

[Clayton Covington]
Do you mind going to the last side Sarah?

[Sarah Sernaker]
Yes.

[Clayton Covington]
All right so again thank you everyone for joining us today for this session of the summer training series. Our next presentation is going to happen at the same time next week 12:00 p.m. Eastern time and we're going to be continuing with Sarah who will be talking about survey design using weights. Just as an FYI after that point I'm actually going to be finishing up my time here at NDACAN. So Paige Logan who's actually on the call right now is going to be assuming this role so she will host based on the sessions that begin August 7th and the subsequent one on August 14th. So you'll get to meet Paige soon but just wanted to flag that. But we'll see you all next week.

[Sarah Sernaker]
Yes thank you everyone feel free to email me with questions too.

[VOICEOVER]
The National Data Archive on Child Abuse and Neglect is a collaboration between Cornell University and Duke University funding for NDACAN is provided by the Children's Bureau an office of the administration for Children and Families

[MUSIC]