Length of stay is critical for animal shelters. It affects how many animals the shelter can care for, how many staff are needed, and how stressed the animals become. Shorter stays mean more animals can move through the shelter and find homes. Longer stays mean animals spend more time in kennels, which increases stress and risk of illness.

Currently, animal shelters only count animals who leave during a specific time period, like a month or a quarter, in their length-of-stay calculations. This seems logical, but as the author of this study argues, it creates a distorted picture.

Imagine a shelter in May has many dogs who’ve been there for months. These long-term residents don’t show up in May’s length-of-stay calculation because they haven’t left yet. The shelter only counts the dogs who leave, and these tend to be shorter-term residents. This makes May’s numbers look better than they are in reality. Then in June, some of the long-term residents finally get adopted or transferred to another organization. Now June’s length-of-stay calculation suddenly looks alarming because it includes all the dogs who were there for months. In other words, it appears as though the shelter got worse at moving animals out of care from one month to the next when actually it got better.

This study proposes a better way to calculate length of stay that accounts for all animals present during any part of a time period. The researcher analyzed six years of dog data from July 2018 to June 2024 from Orange County Animal Care, an animal shelter in California, to demonstrate how the traditional method fails and how the new, corrected method works.

The New Method

The researcher used two well-established statistical tools: the Kaplan-Meier method and Cox regression.

The Kaplan-Meier method tracks how long it takes for something to happen across a group — in this case, how long before animals leave the shelter. It produces a curve that shows, at any given point in time, what fraction of animals are still in care. The key innovation here is that this curve can include animals who were already in the shelter when the measurement period began, and animals who were still there when it ended — not just those whose entire stay fell neatly within the window.

Cox regression then allows the researcher to test whether the curve shifted meaningfully from one time period to another, and to account for the influence of animal characteristics such as age or size on length of stay.

With this approach, the researcher can ask: did length of stay actually change, and if so, could that change be explained by the kinds of animals coming in, or by something the shelter was doing differently?

Testing The New Method

The researcher put these tools to work in three ways: a focused comparison of two consecutive months, a sweep across 24 quarterly periods over six years, and a longer before-and-after look at how the shelter’s operations changed following COVID-19.

The clearest example of the difference between the two methods comes from October and November 2023. A fire near the shelter led to emergency transfers of many dogs in November. Using the traditional method, it looked as though length of stay got worse in November, increasing from about 16 days to 32 days. Using the researcher’s corrected method showed the opposite: length of stay actually improved dramatically as the shelter successfully moved long-term residents out, decreasing from roughly 28 days to 10 days.

The new method is also able to catch meaningful changes over time. Over the six years studied, length of stay was relatively stable at around 15 days until mid-2021, when it increased and stayed elevated until the fall of 2023, when the fire and large transfer of dogs occurred. Just over a third (35%) of quarter-to-quarter comparisons showed significant changes, far more than random chance would explain. Detecting these shifts is important because they can impact shelter resources like housing and staffing.

The before-and-after COVID-19 comparison revealed something important. After COVID-19, the shelter kept visitor restrictions in place, limiting public access and adoption hours. Length of stay increased significantly during this period from about 15 days to 21 days. The researcher tested whether this was just because different types of dogs were arriving (e.g., more puppies, fewer medium-sized dogs), but the analysis showed the increase happened regardless of dog characteristics. The operational restrictions appear to have caused the longer stays. Thus, the new method potentially has explanatory power as well, as it can show the impact of shelter decisions.

Why The New Method Matters

Length of stay directly determines how many animals are in the shelter at any time. More animals mean more kennels, more staff, more food, more veterinary care, and more supplies. When this study’s shelter had a longer length of stay, it needed space for dozens more dogs, which required several additional full-time staff positions. The traditional calculation method hides problems as they build up, then makes things look worse when shelters try to solve them. This discourages exactly the kind of problem-solving shelters need to do.

The new, corrected method shows what’s actually happening in real time. Because it’s rather sophisticated, the author recommends that shelter databases incorporate it directly in their length-of-stay reports for staff ease of use. With it, shelters can track changes month by month, see whether operational decisions work, predict resource needs, and spot problems before they become crises. Better measurement leads to better decisions, which means animals spend less unnecessary time in shelters.