Shelters can now measure length of stay correctly, with free software they can run on their own data

What is it about?

How long animals stay in a shelter affects their welfare, crowding, staff workload, and costs. But the usual way of computing length of stay for a month or a quarter is misleading. It counts only the animals that leave during that period, credits each one's entire stay to that period (even days spent months earlier), and ignores animals still in care at the end. A slow period can look fast, and a fast one slow. An earlier study showed how to fix this using survival analysis, the statistics of time-to-event data. This paper presents mLOS, free open-source software that does this for any shelter. The shelter supplies a simple file of intake and outcome records and chooses its calendar periods. A companion tool, ShelterDataPrep, helps convert raw shelter exports. The shelter then gets plots, tables, and a workbook. mLOS shows: • the fraction of animals still in care each day after intake • how long current residents have already stayed • how many more days a resident can be expected to stay, and its likely outcome, given the time it has already spent • how adoption, transfer, and non-live outcomes compete over time mLOS compares groups such as dog size, intake type, and calendar period, adjusting each for the others. It also turns stay lengths into census and days of care. The code has been tested extensively against hand-computed and simulated cases. An example with dogs at Orange County Animal Care (2019–2022) shows large dogs were 38% of intakes but about 70% of the dogs in care. They also accounted for over 90% of the future care owed to the resident population. Between the pre-COVID period and mid-2021 through September 2022, the median stay did not change. The average stay rose by half, and the longest stays roughly doubled.

Featured Image

Photo by LUM3N on Unsplash

Photo by LUM3N on Unsplash

Why is it important?

Most shelters track only an average length of stay, and they compute it in a way that can mislead. mLOS lets a shelter measure its own numbers correctly and see the whole picture. That means not just the typical stay, but the long-stay animals who fill the kennels and take up most of the care. A shelter can check whether a policy change shortened stays. It can see which animals would free up the most kennel-days if they found homes faster, and how its numbers compare with other shelters or published studies. Research findings become easier to act on when a shelter can measure the same things in its own data. The software is free and documented for practitioners. It can even run in a web browser through Google Colab, with no local installation of R. The Orange County example tells a policy story. During the pandemic the shelter switched to adoption by appointment, and it kept that policy long after other shelters had reopened. Visitors could no longer walk through the kennels and come across a dog they hadn't planned to see. From July 2021 through September 2022, intakes were still about 30% below pre-pandemic levels, yet the number of dogs in care climbed back above its pre-pandemic level, because stays got longer. The typical stay didn't change: the median was 6 days both before and after. The increase was concentrated among the longest-staying dogs, the ones who benefit most when visitors can walk by and notice them. Before the pandemic, one dog in ten stayed longer than 25 days; by 2022 that threshold had risen to 54 days. My earlier studies found that dogs visitors could see were adopted at much higher rates, and that hard-to-place dogs were adopted 1.5 times as often when visitors could view the kennels. Had the shelter had mLOS in 2022, it could have watched this unfold in its own numbers. The Board of Supervisors eventually voted unanimously to restore kennel access.

Perspectives