Let's talk about Chronic Homelessness

Let me ask a question: how do you calculate chronic homelessness? We know that the federal definition is 6 months in the past year or 18 months in the past 3 years, but we have to deal with a lack of information - there's lots of holes in our data. So how do we address these gaps? Let's go through a few scenarios and discuss.

Scenario 1

Our client, Aladdin, stays in our shelter every night for six months. He books in January 1st and then books out June 30. That's 181 days, so we can calculate for a complete certainty that Aladdin has been homeless for at least 6 months in the past year. Easy peasy, lemon squeezy.

Scenario 2

Aladdin likes to switch back and forth between two different shelters. He stays for January at shelter A, February at shelter B, March at shelter A, April at shelter B, May at shelter A, and June at shelter B. (Maybe they have a 30-day limit, who knows?) It's still the same total amount of time, split between two shelters, so we can just add up the days spent at each shelter for a total of 181 days. Again, we know for sure that Aladdin is chronically homeless. (But we wouldn't know that if the shelters weren't sharing their data - in different clusters, for example. If that was the case, both shelters would think that Aladdin is not chronically homeless because he'd only spent 3 months there.)

Scenario 3

Aladdin spends some time sleeping in shelters and some other time sleeping in a tent. He was very forthcoming about the whens and wheres of it, so we know that Aladdin stayed in shelters from January 1 to March 31, and then in tents from April 1 to June 30. It's still the same total amount of time, but the tent stays would be recorded in the client's Housing History, and the shelter stays would be recorded in their Admissions history. We would need to combine these two sources of information, but once you can do that, the answer remains the same; definitely chronically homeless.

Those first three scenarios were easy, comparatively speaking. They all had 6 months of complete data. So let's move on to some harder questions:

Scenario 4

Aladdin is in shelter from January 1 to March 31, then disappears for a few days, then comes back April 7th and stays until June 30. Uh-oh. There's a gap, and so we only know for sure about 173 days in the last year. We don't have enough information to conclusively decide if Aladdin is or is not chronically homeless.

Let's say that today is June 30th and there's a housing unit that's available right now and would be a good match for Aladdin... if he was chronically homeless. If he wasn't, the unit would go to someone else (maybe with lower acuity but who is definitely chronically homeless). And he's so close! Let's further assume that you have a report that simply magically spits out a "yes" or a "no," whether Aladdin is chronically homeless or not. Because the report just gives you a yes or no, close doesn't help. As they say, close only counts in horseshoes and hand grenades. So how do we handle this gap?

One approach, which is fairly common among communities, is to only count the days that we know about for sure. Using that method, Aladdin is officially not chronically homeless (at least for a few more days). In fact, most of our reports use this method.

However, it's pretty reasonable to make the assumption that even if we don't know where Aladdin was for the missing week, he probably wasn't housed. He was homeless before the gap and homeless again after the gap and the gap was only a week long. It's pretty hard to call yourself housed if you're only in that situation for a week (that's hardly permanent). So we could reasonably assume that Aladdin was homeless during this gap, even though we don't know where he was - he could have been in a shelter in a different city, or in a motel, or crashing with a buddy, or camping, or in the hospital.

The benefit of this approach is it becomes much, much easier for staff; they don't need to hound clients for all the details about where they were staying every single night. That said, embracing this approach could result in your staff becoming much, much lazier about data entry, at the expense of data quality. If staff learn that they don't need to update a client's Housing History after a shelter re-admission, then you could be losing valuable data about what homelessness actually looks like in your community.

So let's assume for a moment that we like this idea: if there is a small gap between known periods of homelessness, then we assume it should be filled in. But now we need to ask: what constitutes a "small" gap? 1 day? 3 days? 7 days? 14 days? 30 days? Where do we draw the line between assuming that someone is still homeless and deciding that you really don't know?

In order to answer that question, we really need to ask what we're using this information for. Usually, the answer is prioritization for housing. Now if we think about this from one point of view, we might find out that how long the gap is doesn't matter... much. Why's that? Consider the following:

Scenario 5

Now let's move forward in time, and follow a few different paths.

In scenario 5A, we don't hear from Aladdin again. July 31st Aladdin now has 210 days of homelessness (and is still chronic) and we haven't heard from him in 60 days. Assuming your inactivity policy is something like 90 days, Aladdin is still on your prioritization list. Fast forward some more; it's August 31st. Aladdin has 240ish days of homelessness and is still considered chronic, but we haven't seen him in 90 days, so Aladdin becomes inactive. So it really doesn't matter if he was homeless from June 1 to August 31st, he is no longer on the prioritization list anymore, which includes people who are chronically homeless but only those who are active.

In scenario 5B, we eventually find out that Aladdin has become housed. Let's say the same thing happened; on July 31st we assume he's had 210 days of homelessness and we haven't heard from him in 60 days. But then he gets in touch with an outreach worker who finds out that he's been living in his own apartment since June 1. The outreach worker updates Aladdin's Housing History (retroactively) and we now know for sure that Aladdin is not chronically homeless as of July 31st. He's taken off the prioritization list.

In scenario 5C, we eventually meet Aladdin, and find out that he has been homeless all along. On July 31st we are still assuming that he's had 210 days of homelessness, and he gets in touch with an outreach worker who finds out he's been living in a tent since June 1. The outreach worker updates Aladdin's Housing History and we know for sure that Aladdin is chronically homeless. He's kept on the list, and additionally is still considered to be active.

In scenario 5D, we eventually meet Aladdin again, and he's still homeless, or homeless again. Maybe he was incarcerated or hospitalized for a few months, but he's not very forthcoming about it. All we know is that on July 31st, he enters shelter again. This is the trickiest scenario, because he could have been housed for two months - but you could make the argument that even if he was housed, he wasn't able to maintain it so it couldn't have been very stable, so it could go either way. This is the only scenario in which the duration of the gap makes a big difference.

Conclusion

Considering these scenarios above, I propose that it makes sense to count gaps up to a certain limit - say, 30 days - as being homeless, but if there is a gap for longer than 30 days, call it "unknown" housing status. Days spent in >30 day gaps don't count towards chronic homelessness; days spent in <30 gaps do.

As an aside, if you start using this definition today, your number of chronically homeless people could possibly skyrocket if you have lower data completeness (i.e. you have lots of gaps), so don't make this decision lightly! Consider this to be more of a thought experiment.

What do you think? Am I missing anything? Do you think this method would improve your accuracy in identifying who is chronically homeless?