One face, two meanings
In 2014 I ran sessions in a psychology lab, gelling EEG electrodes and watching people decide when a smiling face turned angry. Everyone drew that line somewhere, and it landed in a different place for different people. A reading only earned trust when it repeated and the other measures pointed the same way. This one is about holding an employee survey to that same rule, and what a study of one health system and a fired teacher show about what one number can miss.
Sponsored by
How 2M+ Professionals Stay Ahead on AI
AI is moving fast and most people are falling behind.
The Rundown AI keeps you ahead of the curve.
It's a free AI newsletter that keeps you up-to-date on the latest AI news, and teaches you how to apply it in just 5 minutes a day.
Plus, complete the quiz after signing up and they’ll recommend the best AI tools, guides, and courses — tailored to your needs.
Where the face turns angry
In 2014 I worked in Lisa Feldman Barrett's lab at Northeastern, on a study led by Spencer Lynn. We used eleven images that blended one person's smile into their scowl, and the software flashed them one at a time, half a second each, with a single question: was that anger? Somewhere along that range, each person drew a line between angry and not angry, and that line sat in a different place for different people.
My job covered each session from start to finish. I welcomed the participant, walked them through what the session involved, hooked up the EEG and gelled each electrode, ran the research software, and then gathered and compiled the data. Once a week the team met to walk through different parts of the research and the technology, so we kept learning how all of it worked.
The study measured each person more than one way: the face judgments, a working memory test, a set of questionnaires, and, in the sessions I ran, an EEG recording. The EEG had one hard rule: each electrode needed the gel to hold a strong connection to the scalp, and when one didn't, we counted that reading as bad.
What I took from the lab is a simple rule: a reading earns trust when it repeats and when separate measures point the same way. You look at the outliers, then build the finding on the pattern.
That rule is the lens I bring to work now. A survey is one measure, and it needs other data next to it before anyone should trust what it says.
One face, two meanings
The paper's abstract makes the point in passing: "a scowling facial expression can indicate anger or concentration." One face can carry two meanings, and nothing on the face itself tells you which. You need a second reading.
Redrawn from Figure 1 of Lynn et al. (2016).
An employee survey works the same way. It's one measure, taken on one day, in people's own words, and on its own it can't tell you which meaning it carries. That makes it the first reading, the one you check against something else before you act on it.
A 2025 study by Rhodes and colleagues in the Journal of Healthcare Management shows what that second reading can catch. The researchers took the well-being survey one health system ran across its OB/GYN clinics and set it next to a second record: who left in the months that followed. Advanced practitioners who skipped the survey were nearly 10 times as likely to leave in the quarter after it as the ones who answered. Among physicians below retirement age, the ones who skipped it were about 5 times as likely to leave over the next two quarters. The survey on its own had no way to show who it was missing, and the second record did.
When a system skips that check, a person pays for it. In 2012, Education Week covered the case of a fifth-grade teacher in the D.C. public schools. She had excellent observation ratings and praise from her colleagues and her students' parents, and she was fired anyway because a low score built from her students' test results outweighed all of it. Three readings pointed one way, one pointed the other, and the system trusted the one.
Here's the strongest case against all this. People's own words carry information nothing else can, and Barrett's research argues that the brain builds each feeling out of a person's past experience. If that's right, nothing outside a person can fully stand in for asking them. So the claim narrows: a survey is necessary, and it's never enough on its own.
The Vitals
Employee willingness to support enterprise change fell from 74 percent in 2016 to 43 percent in 2022 (Gartner research published in Harvard Business Review, May 2023). The published figure gives two readings, six years apart.
A 2003 review of research methods recommends measuring the cause and the outcome from different sources whenever the same people would otherwise report both (Podsakoff, MacKenzie, Lee, and Podsakoff, 2003).
From the build
Companies run engagement surveys because they want to know whether people are on board with where things are going, what's in their way, and whether they feel heard. SteadMark goes after those same questions every day, without asking anyone to fill anything out. It observes how work is already moving across a team and turns that into leading indicators a manager can act on the same week, while a change is still taking shape. The people doing the work see what their input changed, so they have a real say in where their work goes next, and everyone stays on the same page as the company grows through the change.
This is my fourth newsletter. Reply and tell me one number your company trusts about its people that nothing else backs up. I'll read every response.
With love,
Jim Sullivan
