In announcing this spring that homelessness was down in Los Angeles for the first time in four years, officials drew little attention to another statistic cryptically suggesting an opposite trend.
It was in a bullet point on page 13 of the 2018 homeless count report: the number of tents, vehicles and makeshift shelters had increased.
While the report put the increase at 5% over the previous year, a Los Angeles Times analysis of data obtained from the Los Angeles Homeless Services Authority shows a sharper increase in vehicles. There were 18% more cars and 11% more vans being used as homes. Tents were up 4%.
What could explain this apparent paradox?
Statisticians working for the homeless authority say the homeless population changed over the previous year — that fewer people are living in each car, tent or structure made of wood or tarps.
But they also acknowledge that the estimates for the size of homeless households, which are revised each year, are imperfect. It wouldn’t take much of an error, either this year or last, to turn the decrease into a small increase.
Critics have broadly questioned the accuracy of the count. In an analysis published last fall, the Economic Roundtable, a nonprofit research group that focuses on social issues in Southern California, said the methodology was not accurate enough to identify year-over-year changes.
If this year’s formula “had been used last year, it would have reduced the size of the count,” said Dan Flaming, president of the Economic Roundtable. “This could well produce annual counts showing a steady upward trend rather than a trend going up last year and down this year.”
While that wouldn’t significantly change the overall picture of homelessness, it would take some of the luster off the notion that the 2017 Measure H homeless sales tax had already turned the tide.
JuHyun Sakota, manager of data and research for the homeless authority, said the agency has confidence in the statistical model that produces the estimates.
The estimates change, Sakota said, “because that’s what the data tells us for that given year. We believe that’s the most accurate description.”
But she also conceded that the small change reported this year — a 3% decrease across Los Angeles County — would increase the impact of small statistical errors compared with last year, when the count increased 23%.
“We have been seeing so much fluctuation in our population that [household size] determines a small portion of our count,” Sakota said. “But now that the change is getting narrower and narrower, I think that this is going to become a bigger question.”
The numbers come from a survey conducted separately from the January count, when thousands of volunteers fan out to every census tract in the county. The volunteers count the number of tents, makeshift shelters and vehicles, but for safety and privacy are instructed not to engage people or attempt to count how many people are in them.
For the survey, paid workers interview about 5,000 homeless people to gauge the demographics of the population. In addition to recording age, race, geographic origin and health status, they determine how many live in each tent, car, van, recreational vehicle and makeshift shelter.
Applying a statistical formula to the survey information leads to a multiplier for each type of shelter. Those multipliers changed in 2018.
The average number of individuals living in each tent, for example, was fixed at 1.691 in the 2017 survey. This year it was reduced to 1.516, a difference of about one person for every five tents.
Even though that difference was small, it was more than enough, when multiplied over more than 2,000 tents, to more than compensate for the increase in tents counted. Thus, the estimated population living in tents was reduced by more than 300.
The multipliers for individuals living in makeshift shelters, vans and recreational vehicles were also reduced, as were the factors for families in most categories.
The challenge of conducting the survey is to select census tracts that reflect the whole population.
“From a methodological standpoint, we didn’t make a drastic change from 2017 to 2018,” Sakota said.
The process begins each year with a gathering of service providers who compile lists of census tracts where there are known homeless hot spots.
The tracts to be sampled are selected randomly and then weighted. Survey takers then attempt to interview every homeless person in the sampled tracts.
Though the basic methodology was unchanged, the survey was considerably more extensive this year. A Times analysis of data provided by the homeless authority shows that 483 tracts were sampled in 2018, about 16% more than the year before.
Sakota said the survey could have also improved this year because of improved training.
But a bigger change occurred when the homeless authority contracted with the USC Schaeffer Center in 2017 to take over the survey from the University of North Carolina.
“They really improved the reliability of these estimates by collecting larger samples from this population,” Sakota said.