---
name: survey-design
description: Use when asked to design a survey, questionnaire, or feedback form — writing questions and response scales that actually measure what they're meant to without leading or confusing respondents — as distinct from a [[user-interview-guide]], which is a qualitative, open-ended conversation rather than a standardized instrument.
---

# Survey Design

Survey design is the discipline of writing a survey instrument that
actually measures what it's meant to measure, rather than one that
produces answers that look like data but can't be trusted. A survey is
easy to write badly and hard to fix once it's been sent — every
ambiguous or leading question quietly corrupts the results it
produces.

## Key components

- **One clear construct per question** — each question measures a
  single thing. A question trying to measure two things at once
  produces answers that can't be interpreted, because a respondent
  might mean yes to one part and no to the other.
- **Response scales matched to what's being measured** — a rating
  scale for a quantifiable attitude, a genuinely open text field for
  something that doesn't reduce to a number, and never forcing a
  naturally open-ended answer into a rigid scale just because scales
  are easier to tabulate.
- **A pilot test before full deployment** — running the survey with a
  small sample first to catch confusing wording, check that questions
  are interpreted as intended, and confirm the survey isn't so long
  that respondents abandon it partway through.
- **Neutral, unbiased question wording** — phrasing that doesn't hint
  at a preferred or "correct" answer.
- **A logical question order** — general questions before specific
  ones, and sensitive or demographic questions placed late so they
  don't prime or discourage responses to what comes after.
- **Clear instructions and definitions** — any term that could be
  interpreted multiple ways (e.g. "recently," "the team") defined
  explicitly rather than left to the respondent to guess.

## Why one construct per question matters

A "double-barreled" question like "was the service fast and friendly?"
asks about two independent things at once. A respondent who found the
service fast but rude has no valid way to answer — picking "yes"
overstates the friendliness, picking "no" overstates the slowness, and
either way the resulting data point is uninterpretable. The fix is
always to split it into two questions: one about speed, one about
friendliness. This applies beyond the obvious "and" cases — any
question where two respondents could reasonably disagree about which
part it's actually asking about needs to be split.

## Common pitfalls

- **Leading questions** — wording that suggests the "right" answer,
  such as "how much did you enjoy our excellent new feature?" instead
  of a neutral "how would you rate the new feature?"
- **Double-barreled questions** — combining two distinct things into
  one question, making the answer impossible to interpret cleanly.
- **Skipped pilot test** — going straight to full deployment means
  confusing wording, unclear scales, or an overly long survey are only
  discovered after the real data collection is already compromised.
- **Forcing an open-ended answer into a scale** — asking respondents to
  rate something on 1-5 when what's actually needed is their own words,
  losing the nuance a scale can't capture.
- **Too many questions** — a long survey depresses completion rates and
  degrades the quality of answers toward the end as respondents fatigue.
- **Unbalanced or missing response options** — scales that skew
  positive, or that omit a "not applicable" option, forcing an answer
  that doesn't reflect the respondent's actual situation.

## Learn more

- [[user-interview-guide]] for open-ended qualitative conversations
  rather than a standardized instrument.
- [[research-proposal]] for the broader research plan a survey often
  serves as the method for.
- [[literature-review]] for checking whether a validated instrument
  already exists before designing a new one from scratch.