Do human or natural variables better predict wolf predation on American livestock?
HOW THE WOLF PROJECT BEGAN
My sophomore year of college, I joined Dr. Todd Michael Anderson's lab, mostly out of curiosity. His lab focuses on African savanna ecology, specifically species densities and wildfires. He also leads a decades-long camera trap project based in the Serengeti that generates thousands and thousands of images. I helped out by classifying hundreds and hundreds of images by species. Originally, I wanted to pursue research with cheetahs (everyone in the lab makes fun of me for being a cheetah nerd), using Dr. Anderson's camera trap data. But after filing through all of those images, and considering cheetahs aren't captured *that* often, the idea became less appealing. I then had a project lined up having to with wild feline fear behavior - I was going to be working with lions, leopards, caracals, and more. Unfortunately, there was an accident at the conservation center where my project was to be based, and I was back at the drawing board.
I first started reading extensively about wolves the beginning of my junior year, in 2018. An article struck my interest about how wolves are ravaging livestock. This seemed a bit odd to me, so I did more digging. One evening, I spent upwards 6 hours looking at numbers and USDA data until I had an entire powerpoint presentation of notes and graphs that I had made out of complete curiosity- notes and graphs that ended up pointing to the (accurate) conclusion that wolves account for only 0.02% of livestock losses in America. I had the idea to do some work with data analysis, but wolf data is notoriously hard to come by. I started combing through 50 page reports from Fish and Wildlife or DNR for the different states, extracting variables and starting to create a giant data set for American gray wolves.
Some research had been done on the influence of predator killing to protect livestock, but the literature for wolves in America was sparse and not very scientifically sound. I wanted to improve it. As I had been researching, I saw the hatred people had for wolves, and I couldn't understand why. I knew that nobody would listen to an emotional young woman who just said "don't kill wolves". I needed actual data and sound science. This way, perhaps I could find a solution to ranchers depending on their animals and to the heavy persecution of wolves. I discussed the idea with Dr. Anderson, and he was immediately on board. Soon enough, my project idea was approved by the university for independent undergraduate research, with funding (thankfully, statistical analysis doesn't require much funding, but it's still nice to have). It was time to get to work.
I first started reading extensively about wolves the beginning of my junior year, in 2018. An article struck my interest about how wolves are ravaging livestock. This seemed a bit odd to me, so I did more digging. One evening, I spent upwards 6 hours looking at numbers and USDA data until I had an entire powerpoint presentation of notes and graphs that I had made out of complete curiosity- notes and graphs that ended up pointing to the (accurate) conclusion that wolves account for only 0.02% of livestock losses in America. I had the idea to do some work with data analysis, but wolf data is notoriously hard to come by. I started combing through 50 page reports from Fish and Wildlife or DNR for the different states, extracting variables and starting to create a giant data set for American gray wolves.
Some research had been done on the influence of predator killing to protect livestock, but the literature for wolves in America was sparse and not very scientifically sound. I wanted to improve it. As I had been researching, I saw the hatred people had for wolves, and I couldn't understand why. I knew that nobody would listen to an emotional young woman who just said "don't kill wolves". I needed actual data and sound science. This way, perhaps I could find a solution to ranchers depending on their animals and to the heavy persecution of wolves. I discussed the idea with Dr. Anderson, and he was immediately on board. Soon enough, my project idea was approved by the university for independent undergraduate research, with funding (thankfully, statistical analysis doesn't require much funding, but it's still nice to have). It was time to get to work.
The Goal
Livestock predation is perhaps the biggest instigator in carnivore-human conflict, and has gone on ever since the domestication of livestock. It does not just apply to the United States, either. In Africa, the Maasai have had a centuries-long war against lions. The goal of my project is to find the variables that most influence livestock predations- cattle and sheep- by wolves, so that solutions can be found to improve prevention.
HOW IT WORKS
1. Collect data. I started collecting data at the end of 2018, from Fish and Wildlife and DNR reports, in addition to USDA, BLM, and several other government agencies. The data runs from 2009-2018, from eight different states that wolves occupy: Minnesota, Wisconsin, Michigan, Wyoming, Montana, Idaho, Oregon, and Washington. The original list of variables topped 40, and included things like number of breeding pairs, number of cattle and sheep predations, average pack size, wolf harvest, wolf control, ungulate harvest, precipitation, etc. That's 40 variables from 8 different states for 10 years. It isn't as easy as it sounds. It's been more than 6 months and I'm still not finished yet, although I just have to finish collecting ungulate and BLM data. Fun fact- my data set is the biggest one in existence on gray wolves in the lower 48 states.
2. Double, triple, and quadruple check the data. Possibly even more time consuming than collecting data is reviewing it over and over to make sure it is accurate and also standardized. For example, when retrieving data for cattle and sheep predations, I had to decide 1) do I combine cattle and sheep for one variable- which would be easier- or compare separately? 2) do I take confirmed predations, suspected predations, probable predations, or both? I ended up deciding to split cattle and sheep - which meant more data collection and a lot of hand counting USDA reports- due to their very different behaviors in regards to wolf depredation. I also decided to take only confirmed predations- no doubt, more predations by wolves occurred, but also no doubt that the hatred for wolves also led to a lot of false blame. Confirmed predations was the most accurate way to go. I had to do this for all the variables, think through them and make sure they're standardized between all eight states, and no doubt this step will continue until I finish the project.
3. Run preliminary tests. In order to assess which variables best predict livestock predation, I make statistical models. I use a coding software called R Studio. Before I actually make the models, however, I need to go through a lot of statistical tests involving each of the variables. This includes things like checking for outliers and normality, and either fixing or getting rid of n/as (An n/a is a data point that is missing; either I haven't filled it in yet, or it isn't available.). Then, I run each variable in a preliminary model individually, just to start getting an idea for relationships, and which variables initially may influence predations. So, for the variable "percent breeding pairs" (the percent of wolf packs in a state's population that contain a breeding female and male) , I'll run a model that uses pack size as the only variable to predict cattle predations, then sheep predations. This 1) gives me visuals for how breeding pairs affect predations, 2) tells me if the variable is statistically significant, and 3) gives me an estimate for how many cattle/sheep predations are predicted for each additional increase in percent of breeding pairs. That last one sounds hard, but it's basically just like a slope.
2. Double, triple, and quadruple check the data. Possibly even more time consuming than collecting data is reviewing it over and over to make sure it is accurate and also standardized. For example, when retrieving data for cattle and sheep predations, I had to decide 1) do I combine cattle and sheep for one variable- which would be easier- or compare separately? 2) do I take confirmed predations, suspected predations, probable predations, or both? I ended up deciding to split cattle and sheep - which meant more data collection and a lot of hand counting USDA reports- due to their very different behaviors in regards to wolf depredation. I also decided to take only confirmed predations- no doubt, more predations by wolves occurred, but also no doubt that the hatred for wolves also led to a lot of false blame. Confirmed predations was the most accurate way to go. I had to do this for all the variables, think through them and make sure they're standardized between all eight states, and no doubt this step will continue until I finish the project.
3. Run preliminary tests. In order to assess which variables best predict livestock predation, I make statistical models. I use a coding software called R Studio. Before I actually make the models, however, I need to go through a lot of statistical tests involving each of the variables. This includes things like checking for outliers and normality, and either fixing or getting rid of n/as (An n/a is a data point that is missing; either I haven't filled it in yet, or it isn't available.). Then, I run each variable in a preliminary model individually, just to start getting an idea for relationships, and which variables initially may influence predations. So, for the variable "percent breeding pairs" (the percent of wolf packs in a state's population that contain a breeding female and male) , I'll run a model that uses pack size as the only variable to predict cattle predations, then sheep predations. This 1) gives me visuals for how breeding pairs affect predations, 2) tells me if the variable is statistically significant, and 3) gives me an estimate for how many cattle/sheep predations are predicted for each additional increase in percent of breeding pairs. That last one sounds hard, but it's basically just like a slope.
Resources
Here are some previously published studies that relate to my research. Click here for FAQs about my research, or contact me with questions.
- Adaptive use of nonlethal strategies for minimizing wolf–sheep conflict in Idaho
- Trends and management of wolf-livestock conflicts in Minnesota
- Effectiveness of lethal, directed wolf-depredation control in Minnesota
- Effects of wolf removal on livestock depredation recurrence and wolf recovery in Montana, Idaho, and Wyoming
- Public attitudes towards wolves and wolf management in Wisconsin
- Aversive and disruptive stimulus applications for managing predation
- Non lethal radio activated guard for deterring wolf depredation in Idaho: summary and call for research
- Social and cognitive correlates of Utah residents’ acceptance of the lethal control of wolves.
- Limits to plasticity in gray wolf, Canis lupus, pack structure: conservation implications for recovering populations
Header photo credit: Steven Miley