When your child has a fever, you reach for a thermometer and trust the number. That trust is newer than it seems. Galileo’s early instrument, built around 1600, had no scale at all. It only showed that something had become warmer or colder. Later makers filled their tubes with alcohol, mercury, or air and marked them against different reference points, so two thermometers in the same room could disagree. As the historian Hasok Chang describes in Inventing Temperature, it took well over a century of comparing instruments against one another before readings from different devices meant the same thing.
Even then, errors lasted. In the 1860s the German physician Carl Wunderlich compiled more than a million armpit readings from about 25,000 patients and settled on 37°C as normal body temperature. Medicine used that figure for more than a century. When the physician Philip Mackowiak later tested one of Wunderlich’s thermometers, it read almost two degrees higher than modern instruments. The number had outlived the device that produced it.
When the instrument is wrong for some people
Some measurement errors cost more than a decimal point. Pulse oximeters, the clips that read oxygen through a fingertip, were calibrated largely on people with lighter skin. In 2020, Michael Sjoding and colleagues reported that low oxygen the device failed to detect was about three times as common in Black patients as in White patients. During the COVID-19 pandemic, oxygen readings decided who qualified for some treatments, and later work found that the bias delayed that recognition for Black and Hispanic patients. Hospitals kept using oximeters. Doctors learned where the readings mislead, and regulators began asking for testing across skin tones.
Loneliness measurement is still young
Loneliness is now a public health priority. The WHO Commission on Social Connection estimates that about one in six people worldwide are lonely, and many governments track loneliness in national surveys. Most of those surveys ask a single question, some version of “How often do you feel lonely?” It is cheap and quick, and it is the reason we have population data at all.
We have barely begun comparing our instruments against one another, though. With colleagues at the European Commission’s Joint Research Centre, we tested three loneliness measures in all 27 EU member states, with more than 25,000 respondents. A three-item scale produced scores that could be compared across every country. The single question met our minimum standard for validity in only 19 of the 27, and because it is one item, we could not test whether it means the same thing from one country to the next. The same paper notes that single-item and multi-item measures sometimes correlate as weakly as .27, and that two surveys run in the same year gave country estimates about four percentage points apart on average.
What happens when we get it wrong
Measurement errors don’t stay on paper. A country can look lonelier or less lonely than it is, and funding follows the numbers. A program can look like a failure when the measure simply missed what it changed. In our EU data, the single question tracked emotional distress more closely than people’s social lives, so a program that rebuilds someone’s network might barely move it.
I’ve written before about what this looks like in practice. A widely cited 2021 study of more than a million adolescents in 37 countries reported that school loneliness had nearly doubled between 2012 and 2018, and it pointed to smartphones. The study has been cited in U.S. Senate testimony and in the debate over Australia’s limits on adolescent social media use. When Sharanya Mosalakanti and I checked its measurement, not a single country met the standard for any level of measurement invariance. The study could not show that its loneliness score meant the same thing across countries or years, and policies on phones and young people were built partly on those readings. (For the record, we take no money from Meta, and I have criticized how Meta measures connection on this page before.)
Loneliness is not one thing
A thermometer measures one quantity. Loneliness does not work that way. In interviews with 354 people in eight countries, from Brazil to Zimbabwe, my colleagues and I found that people judge their relationships on many of the same elements, such as trust, reciprocity, and feeling safe to be themselves, yet weigh those elements very differently. A woman in rural Zimbabwe, surrounded by family and neighbors, described severe loneliness rooted in betrayal and lost trust. A university student in the United States who felt well connected described loneliness as the moments when nobody picked up the phone.
A single question asks each person to fold all of this into one word, and the word does different work for different people. In our interviews, some people experienced loneliness mainly as anger or self-doubt, and some described it as welcome time alone. The question also names the feeling outright, and in places where being social is a moral duty, people who are ashamed of feeling lonely can simply answer no. Two people with the same score can be missing very different things, and someone missing a great deal can score low.
Loneliness Essential Reads
That is why a measure that tells us only how lonely someone is cannot tell us what would help. Better measurement asks which parts of a person’s relationships are falling short, which is the case Peter Molenaar made two decades ago for studying people one at a time.
What to do with this
None of this is a reason to stop measuring loneliness, any more than doctors should have stopped taking temperatures. If you design surveys or set policy, use multi-item scales that have been tested for comparability where you can, and treat country rankings built on a single question with caution. If you’re trying to help someone who is lonely, ask what, specifically, is missing before deciding what to offer.
