While language and anonymous societies give humans our superpowers, they can also limit our thinking. Those limitations can be hard to see, because social discourse—“the conversation,” as many people have taken to calling it—defines many of the building blocks of thought itself.
First, by identifying with abstract, discrete symbols like “Palestinian” or “Israeli,” “straight” or “gay,” we can impose arbitrary thresholds on properties that usually come in many shades of in-between. We may struggle to notice when people don’t fall neatly into categories. The result can be uncomfortable for those in the ambiguous middle. When a continuous human landscape coalesces into opposing camps—tribalism—the consequences can be severe, even deadly, especially when those camps are unequal in size or privilege.
When I started to think about how to run online surveys about identity in 2016, I puzzled over the technical problem of how to map both the human landscape and the tribal encampments on it. Ultimately the surveys would have to rely on language to ask questions and get answers. It seemed a bit like trying to lift yourself up by your own bootstraps. Language is all about categories. How could I explore the continuum underneath language using language? Would I need to ask respondents to use sliders or knobs?
I tried to simplify. First, I decided to go in the opposite direction, and stick to plain, yes/no questions. Second, I hit on the idea of trying out my method by exploring handedness before tackling anything trickier. Both of these decisions need some explaining.
Why handedness? I knew that I wanted to figure out how to ask and reason about identity categories like gender and sexuality, which involve majorities and minorities, blur the lines between the biological and the cultural, and involve both visible signs and behaviors. But gender and sexuality are complicated and politically fraught. Debates rage today about the definitions of words and identities in these domains, the harms that language can or can’t cause, who has the right to use words that may offend, and even about grammar (capitalization of races, the singular “they”). Before wading into such landmine-strewn territory, I needed what physicists call a “toy problem” with lower stakes.
Handedness seemed like the perfect solution. Something in human neurophysiology makes a majority of us right-handed, but some of us are left-handed. Handedness strongly influences everyday behavior, such as which hand you write with, and how you hold tools.
It’s a trait, but it’s also an identity—something people say that they “are.” With a survey on handedness, I could begin to explore how identity works in a simplified setting—one that seemed less likely to land me in hot water during the learning process. There’s no talk radio about deporting left-handed people, no grave offense taken for guessing someone’s handedness incorrectly, no epic debate over terminology, no moral panic about whether left-handed couples should be allowed to reproduce—or adopt, lest their left-handedness rub off on an innocent child.
I sensed, on running the first handedness surveys, that the experiment had pretty much worked as planned. There was no blowback. Dozens of respondents wrote some variation on “I liked this survey, thanks!” 1 Some were also bemused. “What are you testing for? I’m curious!” 2
Partly, what made the survey fun and quick to take was the other simplification—its format. It consisted almost entirely of short yes/no questions, with no assumptions made and no definitions given. Finding out people’s handedness and which hand they write with required asking not one or two, but four yes/no questions:
Are you left-handed?
Are you right-handed?
Do you write with your left hand?
Do you write with your right hand? 3
At first glance, this simpleminded approach might seem like a head-scratcher if the goal is to get at nuance. As one respondent put it, “A very binary test for nonbinary concepts.” 4 Answering seemingly unnecessary questions could be annoying for some people, too—in the words of a man from Morrison, Illinois, “This could really be condensed into a short set of multiple choice questions.” Yes, and no.
Normally, all four yes/no questions would be presumed to follow from a single, two-choice question: “Are you (a) left-handed, or (b) right-handed?” This would give us a single “bit” of information for every person. (“Bit” is short for “binary digit,” meaning a zero or a one, or equivalently an (a) or (b).) Since each yes/no question also gives us a single bit of information, using the “yes/no method” with the four questions above produces four bits per respondent instead of one, which works out to 16 different possibilities (2×2×2×2) instead of just two. The man from Morrison might argue that all of these extra possibilities are redundant, because if someone answers “yes” to being left-handed, they will also answer “no” to being right-handed, “yes” to writing with their left hand, and “no” to writing with their right hand. Isn’t that what being left-handed means?
In real life, we make assumptions like these all the time. They are, however, just that—assumptions. Put another way, they’re testable hypotheses.
To test them, we just have to ask thousands of people the same four questions. Then, we can see how well our assumptions match the data. Here’s the breakdown of answers to the first two questions (2×2=4 possibilities), from a total of 5,590 respondents:
| Not left-handed | Left-handed | |
|---|---|---|
| Not right-handed | 1.14% | 11.93% |
| Right-handed | 85.89% | 1.04% |
Unsurprisingly, a large majority of people—about 86%—answer “yes” to right-handed, and “no” to left-handed. A minority, just under 12%, answer “yes” to left-handed, and “no” to right-handed. This accounts for about 97.8% of the population, which tells us that the assumption that a person is either left- or right-handed, but not both or neither, is generally correct. However, it doesn’t cover everyone. About 1 in 46 people answer either “yes” to both or “no” to both.
These people belong to an “excluded middle.” By answering as they have, they’ve opted out of our usual multiple-choice, either/or assumption about left- and right-handedness. One in 46 isn’t such a small number, either. If the sample is representative of the US, this works out to about 7.2 million people. That’s more than the combined population of LA and Chicago.
That we can see this excluded middle in the response statistics shows us that yes/no questions are not as binary as they seem. Or rather, each individual question is binary, and forces each respondent to “round up or down” with every answer. However, looking at the aggregated answers of multiple yes/no questions over many subjects reveals the excluded middle in the pattern of yeses and nos.
This is analogous to the old printing and graphics technique known as “halftoning” or “dithering,” which allows a black and white printer or display to render images in shades of gray—even if each pixel can only be black or white. I’ll explain shortly how one can produce a “halftoned” portrait of some nuanced human characteristic using only the unpromising ingredients of binary (and redundant-seeming) yes/no questions.
But first, you may be wondering: what if these 1 in 46 people are wrong—misguided, delusional about their handedness, or just sloppy at answering the questions? The last, at a minimum, is definitely a fair concern. I automatically paid all who completed the survey, even if they clicked at random, because I didn’t want to get into a petty dispute with a Mechanical Turk worker over a dollar and change. I figured that I was paying for their time, and hoping that by paying fairly they’d reciprocate by answering carefully and honestly.
As a social experiment, this was interesting in its own right. The results were, on the whole, affirming. The great majority of respondents took obvious care with their work—pride, even—clarifying any apparent contradictions in their answers using the free response section, or sometimes even by sending me explanations by email. As you’ve seen, they didn’t hesitate to let me know if they felt I might have overlooked something in the survey design (this feedback was often valuable), or when they worried they might have taken the same survey twice (this happened when I revised my methods and ran an updated version). 5 In an era that sometimes seems characterized by abusive behavior online, it was reassuring to see that decency, reciprocity, and a work ethic seemed the norm.
A cynic might argue that this is a function of the power dynamic between Amazon and its gig workers on Mechanical Turk—that they’re kept on a short leash, subject to a harsh reputation economy. I’m not a romantic about either people or corporations, but I think this view does both a disservice. Reputation probably does matter, online just as in real life, and the anonymity of burner accounts almost certainly does undermine social accountability on the internet, but my experiences with Mechanical Turk have left me with the impression that most people, regardless of their identities or beliefs, are wired for decent social behavior by default.
Nonetheless, I did set some “traps” in the surveys, to filter out respondents who weren’t answering carefully. More details, along with a careful look at the demographics of Mechanical Turk workers, are in the Appendix for data nerds. As will soon become clear, there’s good reason to believe that for the 90% or so of survey respondents who made it through the filter, the excluded middle is unlikely to be the result of random clicking. Most of them meant to answer as they did.
So what does it mean to be neither exclusively left-handed nor exclusively right-handed? As noted earlier, when their answers aren’t self-evident, people often elaborate on them in the free response field at the end of the survey. I left the wording of this final question deliberately vague, along the lines of “Is there anything else you’d like to add?” to encourage elaboration. Its open-endedness helps make up for the rigidity of the yes/no questions. The tradeoff is that while those yes/no questions are perfect for statistical analysis, free text isn’t so quantifiable. Stories do give us insight, though, like this one, from a 62-year-old woman living in a small town in rural Washington:
Due to an industrial accident ten years ago, I lost half of my right hand, so [some] of these questions were hard to answer because I’ve had to “adapt” by learning how to use my left hand for things like writing, etc. I wouldn’t necessarily call myself [ambidextrous], though, because if that hadn’t happened, I would still be strictly right-handed.
Injuries leading to outcomes like these aren’t as uncommon as one might think. A recent paper in a medical journal estimated that “One in 190 Americans is currently living with the loss of a limb.” 6 Many respondents described other, more commonplace injuries affecting handedness: “I badly burnt my left hand as a kid and was forced to [learn] to write with my right hand” 7 ; “When I was young I was becoming ambidextrous, but due to an injury my right hand became dominant” 8 ; “I was forced to switch when I got [juvenile rheumatoid] arthritis at 13” 9 ; or simply “I can write reasonably well with my right hand because I broke my left wrist for a while.” 10
Although some hand or arm problems are chronic or congenital, most injuries are acute, meaning that they happen at a particular moment, and may affect life from then onward. This makes it interesting to look beyond overall percentages, and start to break down people’s responses by age.
I’ll be using age breakdowns a lot, so some explanation is in order. These graphs are generated by dividing responses into age brackets or “bins,” which are shown in alternating shades. Here, the bins are ages 18–21, 21–25, 25–30, 30–40, 40–50, and 50–80. 11 For each labeled series, straight solid lines connect values from the center of one bin to the next, showing how these quantities vary with age (hence the left edge of the graph is at the midpoint of the first bin, 19.5, and the right edge is at the midpoint of the last, 65). The “Strictly left-handed” line shows the percentage of people who both answered “yes” to “Are you left-handed?” and “no” to “Are you right-handed?”; the “Strictly right-handed” line is the converse. Putting multiple quantities on the same graph this way can help us see patterns in the combined data.
If you look closely, you’ll spot shaded regions just around the lines. When you zoom out to look at majorities like right-handedness, they look small, but when you zoom in to look at minorities like left-handedness, they look much bigger. These shaded regions are important. They’re what data scientists call “error bars.” They represent the 90% confidence interval, meaning that if one assumes every bin contains a random sample from a much larger candidate population, then 90% of such random draws of the same size would produce an estimated percentage within the shaded range. 12
Nobody can say for certain that the real percentage of the candidate population falls within this (or any) range, because it’s only possible to sample a tiny fraction of, say, all of the 18- to 21-year-olds in the United States, and it’s impossible to guarantee that any given sample isn’t biased—though I have taken pains to make my sample as unbiased as possible, using “stratified sampling” methods as described in the Appendix for data nerds. Still, the error bars are useful in showing how seriously to take the estimate. When they’re tight around the solid lines, as they are here, it means those numbers are pretty reliable, statistically speaking.
So what are the patterns in the data? There’s more going on than tables of numbers like the one earlier in this chapter can reveal. For one, a large proportion of the youngest adults, probably over 90%, characterize themselves as strictly right-handed, while only a small proportion (a bit over 5%) say they’re strictly left-handed. However, this rapidly shifts to only about 80% strictly right-handed by the mid-20s, and from there, the number of strictly left- or right-handed people declines subtly but steadily with age.
Big meta-analyses of the scientific literature suggest that about 10% of people are left-handed on average, and that left-handedness is about 23% more common in men (so, about 9% of women and 11% of men). 13 It’s been theorized that left-handedness is due to some inherent sex-linked difference in brain development. There may be such a factor, but zooming in on the strictly left-handed curve and breaking it down by sex reveals something interesting.
On the whole, the survey data are consistent with big studies in the research literature—showing 9.6% of women are strictly left-handed, and 11.7% of men. 14 However, that difference may only be significant before middle age; by 45, men and women seem pretty much alike, with both below 10% and falling. What’s going on here?
Looking more closely at strict right-handedness broken down by sex adds some further color.
Now, it’s young men and women who are all but indistinguishable; but as they age, fewer and fewer men report being strictly right-handed. By age 65, only about 75% are, as compared to about 83% of women. Those 90% confidence intervals show us that this effect is probably quite significant.
It has been well-established that, on average, men are more accident prone than women. They use (and misuse) more power tools, fall off more roofs, lose more limbs, and are even (somewhat bafflingly) struck by lightning more often! 15 Perhaps they lack the good sense to stay indoors during a thunderstorm. As a cohort, young men are—it will surprise nobody to learn—especially unwise in their life choices, likely due to a complex stew of biological, cultural, and behavioral factors.
Regardless of initial handedness, with every passing year, men have on average a greater likelihood of needing to change their handedness due to injury. Even if women and men started off with equal (and very low) probabilities of left-handedness at birth, this would result not only in an excess of left-handed men by age 18 due to disproportionate childhood injuries, but also would result in more sharply declining numbers of both strictly left- and strictly right-handed men over time—especially since injury of the dominant hand is more common, as it’s in harm’s way more often. And this is exactly what the graphs show.
The evidence mounts further on considering the excluded middle. With declines in both strict left- and strict right-handedness, it follows that an increasing number of people are answering the handedness questions ambiguously. This is true; that curve is similar to the rate at which people answer “yes” to “Are you ambidextrous?” and consists largely of the same population (though “ambidextrous” seems to be a somewhat stronger statement, as it’s used a tad less often, at every age).
When people say they’re ambidextrous, they often include an account of injury, as in “I have severe deficiencies with my right arm I’ve had 9 surgeries so I became ambidextrous” 16 or “ambidextrous in some things due to having a severely broken right hand wrist, in a cast/pins for 6 months” 17 . So, injuries likely account for the increasing numbers with age.
None of this means that there’s no innate biological component to handedness. There is evidence for at least some heritability of handedness (based on studies of twins), and there may well be inherently sex-linked differences too. 18 Eventually, our understanding of neuroscience and genetics might pinpoint a sex-linked developmental mechanism. However, given the variations by age and sex in the survey data, the burden of proof would be on medical researchers to demonstrate such an effect—and show that differences in overall handedness averages between the sexes can’t be accounted for simply by different rates of injury over time. This is a good illustration of the old saying “correlation is not causation.”
And so, we’ve arrived at this book’s first run-in with medical authority. It won’t be the last.
We have much to thank physicians for. Many of us would die young were it not for everyday miracles like antibiotics, perinatal care, insulin, even modern dentistry. (In the last chapters of this book, I’ll return to the key role these innovations have played in the larger human story.)
However, as with any in-group, the medical community has its preconceptions, biases, and blind spots. Medical history is rife with long-held assumptions that turned out to be wrong, and with at least the usual share of expert overconfidence in those assumptions. We often entrust doctors with our bodies and our lives. Maybe that’s why we invest them with a kind of intellectual and even moral authority that we don’t tend to extend to other researchers or knowledge specialists.
I reflexively reached for that authority when, in wondering what ambidexterity “really means,” I consulted medical sources—many people would do the same to resolve a dinner table dispute as to the “official” definition. So many people claiming to be ambidextrous puzzled me; I had assumed it was less common. Indeed, according to much of the medical literature, ambidexterity or mixed-handedness is a fairly rare condition “afflicting” about 1% of infants and “associated with atypical cerebral laterality” resulting in “a greater likelihood of having language, scholastic, and mental health problems” later in life. 19
Older research ascribes similar woes to the left-handed. As an influential 1977 paper on measuring handedness by psychologists Curtis Hardyck (of UC Berkeley) and Lewis Petrinovich (of UC Riverside) put it,
Reaction to the problems of explanation posed by the left-handed has followed two courses. Perhaps the most common approach has been to assume that left-handedness is a signal that something is wrong—that the left-handed represent an aberrancy or abnormality and can thus be excluded from consideration in theories of normal cerebral functioning. Certainly the search for deficit associated with left-handedness has been both extensive and unceasing. […] A second approach has been to disregard the left-handed […]. 20
Despite a lack of evidence that left-handed people have “something wrong,” this attitude filters into people’s lives in a variety of ways. For instance, a 24-year-old survey respondent from Wewahitchka, Florida wrote, “I was left-handed at birth but the doctor encouraged my parents to train me right-handed due to me being on the spectrum, the doctor said I would have enough problems without adding left-handedness to them.” Speaking from the majority’s point of view, a 42-year-old from East China, Michigan wrote, “I seem to be right-handed dominant. I think I have less emotional problems than left-handed people.”
I included questions on the survey about depression, bipolar disorder, alcoholism, and a number of other conditions that have been at one time or another associated with left-handedness. The results are underwhelming, with the exception of cerebral palsy—one case where a link with left-handedness has long been well established. 21 Left-handed people are also slightly overrepresented among the bipolar population, though the effect is modest.
On the whole, though, the survey doesn’t support the idea that either left-handedness or ambidexterity are “aberrant or abnormal” traits. Ambidexterity in childhood may indeed be rare, perhaps as low as 1% or even less, based on the youngest respondents (unfortunately I couldn’t survey people under 18). But as we’ve seen, by middle age it also rises, and to well above 10%—even more common than strict left-handedness. That’s a pretty big excluded middle. The survey data are also inconsistent with the idea that ambidexterity is a rare medical condition you’re either born with or not. By middle age, at least 10 times more people say they’re ambidextrous than were born that way. So whom should we believe, the doctors or the survey respondents?
A man from Charlotte, North Carolina.
A woman from Odessa, Florida.
In practice, these would turn up in random order.
This respondent was, in fact, a non-binary 23-year-old from Chicago.
After some initial experimentation to get survey designs right, I used a Mechanical Turk feature called “qualifications” to make sure that nobody could take a survey twice (unless they went to the trouble of creating a fake second account, which is against the site’s policies).
Ziegler-Graham et al., “Estimating the Prevalence of Limb Loss in the United States: 2005 to 2050,” 2008.
A 36-year-old man from Henderson, Kentucky.
A 35-year-old man from Springville, Utah.
A 67-year-old woman from Kearney, Nebraska.
A 34-year-old woman from Bellevue, Nebraska.
These bins or intervals are “half-open” and would be written in mathematical set notation as [18,21), [21,25) and so on, meaning that someone aged exactly 21 would fall in the [21,25) bin, not the [18,21) bin. Questions about birth month (“Were you born in January or February?”), together with age in years, allow the ages of respondents to be calculated to within two months. Bin sizes are chosen to ensure that each bin contains enough samples for the error bar to be reasonably small, while still capturing significant changes by age, as described in the Appendix.
I’ve chosen a generous 90% confidence interval rather than the more common 68% confidence interval (one “standard deviation”) both in order to emphasize uncertainty where it exists and, where the regions are tight, to make clear how many of the effects I’ll describe are so large that they’re highly unlikely to be statistical artifacts.
Papadatou-Pastou et al., “Sex Differences in Left-Handedness: A Meta-Analysis of 144 Studies,” 2008.
Later I’ll delve into the non-binariness of sex and gender. For purposes of the analysis here, “women” is a shorthand for those who answer “yes” to “Do you identify as female?” and “no” to “Do you identify as male?” and vice versa for “men.” There are a number of other possible definitions based on the survey questions, but none of them materially affect these results.
Jensenius, “A Detailed Analysis of Lightning Deaths in the United States from 2006 Through 2019,” 2020; Sorenson, “Gender Disparities in Injury Mortality: Consistent, Persistent, and Larger than You’d Think,” 2011.
A woman from Elkton, Maryland.
A woman from Kansas City, Kansas.
Porac, Laterality: Exploring the Enigma of Left-Handedness, 2016.
Rodriguez et al., “Mixed-Handedness Is Linked to Mental Health Problems in Children and Adolescents,” 2010.
Hardyck and Petrinovich, “Left-Handedness,” 1977.
Only 14 people surveyed have cerebral palsy, but 4 of those are left-handed, which at about 29% is much higher than the expected rate of left-handedness. This is consistent with the medical literature.