New UMBC research is helping dismantle gender and publication biases in science.
A team of researchers working across disciplines has developed a new statistical technique to understand similarity, rather than difference, in the natural world. With this new technique, they’ve determined that among Eastern Bluebirds the structure of songs female birds sing is statistically indistinguishable from songs males sing.
Awareness of female birdsong is growing worldwide, thanks in part to a breakthrough paper by Karan Odom, Ph.D. ’16, biological sciences, but it’s still understood as a trait found primarily in tropical birds. Evangeline Rose, a current Ph.D. student in the same lab and first author on a new paper in Animal Behavior, wanted to look at song in a temperate species.
During Rose’s fieldwork, “I was finding that the females were singing, to me, what sounded just like male songs,” she says. “So we started thinking about equality, and equivalence, and how to test for it.” On the advice of her advisor, Kevin Omland, professor of biological sciences, she reached out to Thomas Mathew, professor of statistics, who has expertise in statistical equivalence.
Challenging a paradigm
Working together, the team modified a statistical method used in generic drug testing to meet their needs for ecology and animal behavior studies. The existing test helps determine whether generic and brand name drugs are “statistically equivalent,” meaning they are similar enough to be prescribed safely for the same purpose. The new modification will allow scientists in other fields to test for equivalence. Before, researchers could only report they did not find a significant difference—a very different statement than saying two things are conclusively equivalent.
“We’re really hoping this new method is going to address some issues with what kinds of data get published,” Rose says. “The most important thing about being a good scientist is to be unbiased. And the whole tradition of testing for difference really leads to incredible biases in scientists,” Omland says. He adds, “There’s a whole realm of things in nature that we find interesting and important because of their similarity.”
For example, in addition to similarities in songs between the sexes in birds, researchers could use the new test to ask if two species use the same type of habitat, respond the same way to predators, or consume the same food sources. Answers to those questions could fill long-standing knowledge gaps, or even inform conservation efforts.
“This test is really broadly applicable,” says Rose, “and we’re hoping to introduce it more to the ecology and evolution field.”
A new approach
One advantage of the new method is it accounts for unequal sample sizes. In a medical study, researchers can carefully control the size of treatment and control groups. In other fields, from ecology, to engineering, to agriculture, that’s often not possible. The new test also allows researchers to determine the equivalence of several traits simultaneously, Mathew explains. For example, in this study, the authors found that the male and female birds’ songs were statistically equivalent across five different characteristics, such as duration of each song and the range of pitches the birds produced.
Rather than testing whether two things are exactly equal, the team was looking for a way to determine if two things were “close enough,” given a defined allowable margin of difference. Because of that added layer, “There are additional challenges here,” Mathew says.
“Even though this methodology is out there, it hasn’t been applied—even in statistics—with this kind of data. That’s why I was very excited when they brought this project to me,” Mathew says. Rose adds, “It ended up being a really great partnership to look at these questions that hadn’t been asked before for female song, and we also ended up modifying this test in a really cool, new way.”
As research on similarities grows, there is also a growing drive to remove the bias against publishing studies that do not find a significant difference, often termed a “negative result.” This paper “is part of an amazing drumbeat that’s building up in the scientific community,” Omland says. “There’s a broader problem with the scientific method that’s being increasingly acknowledged, and the test we’ve developed can at least play a small role, and I hope a big role, in addressing it.”
Rose, who plans to next investigate the function of female bluebird songs, says she will carry these new techniques with her as she moves through her research career. “I think in the future, I’ll be thinking about how equivalence can change the questions we’re asking, and I’ll always keep in mind that we have extra tools in the toolkit.”
The Latest on: Biases in science
via Google News
The Latest on: Biases in science
- Republicans Can't Count On Electoral College Edge in 2024on November 22, 2020 at 6:22 am
To understand why, think about the electoral college arithmetic. Biden won Wisconsin by just 0.6 percentage points. That means that if there had been a uniform nationwide shift to Trump of 0.6 ...
- Practical strategies to minimize bias in machine learningon November 21, 2020 at 2:16 pm
The question isn't whether a machine learning model will systematically discriminate against people -- it's who, when, and how.
- New York City wants to restrict artificial intelligence in hiringon November 20, 2020 at 1:08 pm
New York City is trying to rein in the use of algorithms used to screen job applicants. It's one of the first cities in the U.S. to try to regulate what is an increasingly common — and opaque — hiring ...
- Parler’s explosive growth is rooted in half a century of US partisanshipon November 19, 2020 at 7:30 am
The political social media apps have seen a sudden surge in downloads, but the forces fueling their rise go way back.
- Wrongful arrest exposes racial bias in facial recognition technologyon November 19, 2020 at 4:00 am
Detroit police defend the system, even though the police chief once said technology alone may misidentify people 96% of the time.
- “Male Science Fiction Movies are About Men Having a Romance with Their A.I. Women”: Shalini Kantayya on Coded Biason November 16, 2020 at 10:46 am
A.I. and “machine learning models” can decide who is accepted into college, who gets housing, who gets approved for loans, who gets a job, what advertisements appear on our social media and when. The ...
- WTIA creates ‘Anti-racism in Tech Pact’ to guide organizations in eliminating bias across industryon November 16, 2020 at 6:00 am
In hopes of addressing what it considers widespread racial bias across the tech industry, the Washington Technology Industry Association (WTIA) has created an "Anti-racism in Tech Pact" ...
- This Film Examines the Biases in the Code That Runs Our Liveson November 15, 2020 at 4:03 am
Coded Bias follows MIT researcher Joy Buolamwini as she investigates ... As someone who doesn't have advanced degrees in data science, I had this fear of improperly explaining ideas like algorithms or ...
- You can’t eliminate bias from machine learning, but you can pick your biason November 14, 2020 at 4:45 pm
Removing one source of bias from your model will frequently introduce another. It's a matter of deciding what you can live with.
via Bing News