Are You My Cousin or Half-Sibling?

Sometimes 23andMe unearths family secrets. Sometimes it just makes a mistake.

DNA strands
Photo illustration by Slate. Photo by Getty Images Plus.

A few years ago, Jenny participated in a study that gave her free 23andMe test results. She’d all but forgotten about it until a few weeks ago, when her cousin messaged her on Facebook: “lol why does 23andMe think we’re half sisters?”

When she logged in to her 23andMe account, sure enough, the site’s “Relatives” feature listed her cousin as her half-sister. Earlier this year, a shocking discovery had rocked Jenny’s family, says Jenny, leaving her questioning whether 23andMe might have revealed another one. “I played it off as a funny mistake to my cousin, but in my head I was thinking, ‘Well, I already know one family secret I never saw coming, so what if this is real?’ ” She thought through what it would mean for her cousin to actually be her half-sister—basically, one of her parents sleeping with one of her aunts or uncles—and the possible indiscretions “were all extremely upsetting,” she says.

As consumer DNA tests have become mainstream, there have been countless stories about the decades-old family secrets they’ve revealed. There’s the man who donated sperm in college and discovered, years later, that he had 17 biological children, the husband who discovered his daughter is actually the biological child of his ex-wife and the man she had an affair with, and a woman who was adopted at birth and reunited with her siblings. It’s so common that when I reached out to 23andMe about this piece, a representative directed me to an entire resource page the company launched earlier this year for users who have discovered unexpected results through their tests.

The narrative is almost old hat at this point: Of course the DNA test uncovered some skeleton in the closet. Genetic evidence, of all things, feels irrefutable; it not only reveals affairs, but also illuminates risk for diseases, and even solves crimes. It’s science! Many consumers assume that what a DNA testing company tells them must be true, but in some cases, like Jenny’s, those results might be the result of an algorithm jumping to conclusions. Those mislabelings can lead to familial strife—or at the very least, to the results’ recipient experiencing a few tense days of wondering about the truth.

It’s unclear how often DNA companies might miscategorize people’s relatives in test results, but judging from the consistent stream of freakouts in semi-anonymous Facebook groups and forums, it’s fairly common. After receiving results saying her aunt is actually her grandmother, a Reddit user named Melissa took to a 23andMe forum to ask how accurate the “Relatives” feature really is. “Should I be worried my whole life is a lie or is it just that it’s not completely accurate?” she posted. Others have had false reports of cousins listed as half-siblings (including one case in which cousins, related through the test-taker’s mother, were listed as half-siblings but were born seven months apart—a biological impossibility). In other cases, actual half-siblings were listed as cousins.

After doing some more research, Melissa decided she was “90 percent” sure that the results were a mistake. She’d tried other DNA test services, like Ancestry, and compared her results to other family members’, which led her to conclude that it was still most likely that her aunt is actually her aunt. “Maybe she is my grandmother, but I don’t think it’s enough to start a family feud.” To reflect the results she believes to be true, Melissa logged in to her 23andMe account and edited the relationship label in the service’s “Relatives” feature.

That feature is opt-in only, and, as Jenny told me, the company is “careful to use qualifiers like ‘probably’ and ‘likely’ with results, but beyond that, there’s nothing saying, for example, ‘Oh, here’s what else this could mean.’ ” After some additional research, Jenny, like Melissa, is pretty sure the algorithm was just wrong. An acquaintance who works as a genetic counselor reviewed the results, which showed that Jenny and her cousin shared 19 percent of their DNA. On average, half-siblings share about 25 percent, whereas cousins tend to share about 12.5 percent. The results also showed matches only through Jenny’s mother’s side of the family, with no matching segments on the X chromosome. So, Jenny says, unless her mother is actually also her cousin’s mother, it’s likely her cousin is not actually her half-sibling.

In an effort to be transparent with users, 23andMe has an entire page dedicated to explaining how it predicts relationships. By comparing your DNA from specific chromosomes to other users’, the company detects what percentage of your genome is shared. For instance, it’s a common belief that you have about 50 percent of your biological mother’s DNA and 50 percent of your father’s, but even that is not quite true. Kelley Harris, a genome scientist at the University of Washington, explains that, yes, each embryo is made by taking each chromosome from the mother and the matching chromosome from the father. “Then, the cell roughly cuts the chromosomes in half and pastes one half from each parent together, but they don’t have a super precise ruler, so sometimes it’s a little more than half from Mom, and sometimes a little less than half.” It gets more complicated as relationships become more distant: “It’s pretty hard just looking at a genetic profile to tell apart a sibling and a parent, because they both share 50 percent, and the same is true of an aunt and grandparent,” says Harris.

Even geneticists with the latest available tools can’t always figure out the exact relationship between two people based only on their DNA. Harris told me about a study of Neanderthal DNA from an individual whom scientists believed to be the result of inbreeding. The researchers had four hypotheses for the relationships between that individual’s parents—uncle and niece, double first cousins (e.g., their dads were brothers and their moms were sisters), grandfather and granddaughter, half-siblings—but there’s currently no way to reach a definitive conclusion. DNA tests are “good at distinguishing, say, first cousins from people who aren’t related at all, or first cousins from second cousins,” says Harris, but specific relationships are hard to glean. “Even if you had whole sequence data or did a lot to analyze the distributions of segments, it’s just a really hard thing to say with certainty because of all the randomness that comes with kids being made.”

In addition to percentage of shared DNA, 23andMe relies on age to make predictions about relationships. “When relationships have the same pattern of DNA sharing, we use age to try to tell them apart,” Katie Watson, 23andMe’s vice president of communications, told me in an email. “We compare the self-reported ages of the users against an average calculated generation time of 10 years. If the two users are within 10 years of age, we predict that they are half-siblings. If the age gap is great[er] than 10 years, we predict they are an aunt/uncle—nephew/niece pair.” That may explain why Melissa’s aunt—who’s about nine years older than her mother—was predicted to be her grandma.

I asked Melissa why she took to Reddit to ask about her aunt. Was she hoping to find others who shared that experience, or alternate explanations? “I wanted to understand why 23andMe was wrong since it’s one of the most accurate companies,” she said. Looking at the company’s many resources explaining its “Relatives” feature, there’s a clear effort to couch the results in nuance—like Jenny told me, they list results as “predicted relationships” and allow users to edit the relationships. But still, people put a lot of stock in those predictions and may not realize that their “discovery” might be a fluke—and that they’re not alone in their paranoia that they’ve uncovered a family secret.

To preempt users from jumping to conclusions, perhaps the “Relatives” feature could more prominently display uncertainty, especially in fringe cases like Jenny’s, where her percent match with her cousin just happened to fall in the middle of the range between the average relatedness of cousins and half-siblings. Or, perhaps instead of displaying a single relationship, the results could give the possibilities: Based on the percent of relatedness, this person could be your grandparent, grandchild, aunt, uncle, niece, nephew, or half-sibling. That kind of uncertainty might also drive home just how difficult it can be to make relationship predictions, and how wide the range of possibilities really is. “In a case like this, when science can’t tell us the answer, the only thing for scientists to do is admit that we can’t tell,” says Harris.

Future Tense is a partnership of Slate, New America, and Arizona State University that examines emerging technologies, public policy, and society.