One word. Four different machines.
This lab started by accident. Someone typed "new discoveries in survey" into a search engine, expecting land surveys, and got back a fistful of headlines that all said "survey" — and not one of them meant the same thing. That confusion is the first lesson. Before you can compare surveys, you have to notice they aren't the same animal.
The Opinion Poll
Measures how people feel. A sample stands in for a population; the result is a snapshot of mood at a moment.
The Census / Count
Not opinion at all. Counts how many things exist — people, students, dollars. No "feelings," just totals.
The Instrument Itself
The measuring stick is born, retired, or replaced. When the ruler changes, every trend line built on it quietly breaks.
A survey looks like a measurement — a clean number handed to you. It's really a stack of decisions: who to ask, how to word it, when to run it, what to count, and what to leave off entirely. Line six surveys up side by side and those decisions stop hiding. That's the whole skill: surveying the surveys.
The most important question in any survey is the one it didn't ask.
Everything that follows is just learning to see that blank.
→ Next: the actual six we pulled, and the five things to look for in any set.
Here's what one search actually returned.
Five real surveys, pulled from a single search for "new discoveries in survey" in June 2026 — nothing curated, nothing invented. Read across them. Hit "show the hidden choices" on each to see the decisions the headline didn't print.
- Who paid steers the frame. A growth collaborative ran it — so "do you want growth?" is the axis, and growth is the assumed good.
- Self-selected, not random. People who opt into a civic survey skew engaged and invested. The disengaged never show up.
- Residents only. Tourists, seasonal hospitality workers, and recent arrivals — the people growth affects most — aren't in the frame.
- The tension is manufactured by juxtaposition. "Proud" and "democracy is unhealthy" answer two different questions; pairing them creates the "uneasy" story.
- "Americans" = adults. Noncitizens and everyone under 18 aren't in the number, though they live here too.
- Who tells you matters too. The same 1,000-person poll gets a different spin depending on which outlet foregrounds which finding.
- "When" can poison "what." A poll is a snapshot at a point in time. If reality changes mid-fieldwork, early respondents and late respondents are answering different worlds.
- Honesty about it is rare. Most polls report one number and never tell you the days behind it. This one is unusual for showing its own blur.
- Slicing by week shrinks the sample. Splitting 1,000 people across weeks makes each slice small and noisy — a real cost of looking closely.
- "Science & engineering" only. Humanities and arts grad students simply don't exist in this dataset — a whole population defined out of the count.
- Institutions answer, not people. It's the registrar's-eye view: who's enrolled, not how they're doing or why they leave.
- The categories drift. NSF itself warns the field taxonomy has changed over the years, so old and new counts aren't cleanly comparable.
- The frame is a choice with consequences. Widen who's surveyed and you don't just get "more data" — you get harder-to-compare data.
- Single-site findings get over-generalized. Results from one campus often travel as if they were universal. This piece is about the seams.
- It's a survey whose subject is surveys. Easy to mistake for a results paper; it's really a methods confession.
□ The lens you carry to your own pull
- Who got asked? And who couldn't be in the frame at all?
- Who's asking / who paid? Funders rarely commission questions whose answers they'd hate.
- How was it worded? Small changes swing the result hugely.
- Sample & margin? Almost never in the headline — go find it.
- When did it run? A snapshot is only true for its moment.
- What got left out? The hardest one. That's Tab 03.
→ Now the centerpiece: across all five, what did each one decline to ask?
What did each one decline to ask?
Here's the move that turns you from a reader into an analyst. For each survey, take a second — in your head or in the box — and guess the question it carefully avoided. Then lift the bar. There's no wrong answer here; the point is only that you looked.
□ How to find the blank in your own set
- List who answered. Then say out loud who couldn't. The missing group usually holds the missing question.
- Follow the money. Ask what answer would embarrass whoever paid. That answer is often the one not asked.
- Separate count from experience. If it measured "how many," ask "how well." If it measured "how they feel," ask "why."
- Read across, not down. When five surveys on one topic all skip the same thing, that silence is the finding.
A survey is never just what it measured. It's also the shape of everything it chose not to. Read the set, find the silence, and you're reading like an analyst.