48 Types of Question Bias: The NIH Research Every Researcher Should Know

Most surveys fail before a single response comes in. The problem is not sample size, not timing, not incentives. It is the questions themselves.

In 2005, researchers Bernard Choi and Anita Pak published a landmark study through the National Institutes of Health cataloging 48 distinct types of bias that can compromise questionnaire data. Two decades later, their taxonomy remains the most comprehensive classification of question bias ever assembled, and most research teams have never seen it.

We turned their research into a free, practical whitepaper: The Complete Guide to Question Bias: 48 Types Every Researcher Must Know.

Here is what it covers and why it matters for anyone designing surveys, interviews, or feedback forms.

The Three Sources of Bias

Choi and Pak organized all 48 bias types into three categories based on where the bias originates:

Question Design (24 types): How individual questions are worded, scaled, and structured. This is the largest category and the one most teams think about, but even experienced researchers routinely miss several of these.

Questionnaire Design (6 types): How questions are formatted, ordered, and organized as a whole. Even perfectly worded questions can produce bad data if the questionnaire structure introduces bias.

Administration (18 types): How the questionnaire is delivered and completed. These biases arise from interviewer behavior, respondent psychology, recall errors, and cultural differences. They are the hardest to detect because they happen during the interaction, not on paper.

Bias Types Most Teams Miss

Everyone knows about leading questions and double-barreled questions. But how many teams check for these?

Framing Bias

Patients prefer "an operation where 90% survive" over "an operation with 5% mortality," even though the first option is actually worse (10% mortality). How you frame choices changes the answer.

Telescope Bias

People remember distant events as more recent than they actually happened. An event from November gets reported as March. This distorts any time-based data collection.

Mind-Set Bias

When question format suddenly changes mid-survey (per week shifts to per month), respondents on autopilot give wrong answers without realizing it.

Hypothesis Guessing

When respondents figure out what your study is testing, they alter their answers. A sequence about headaches followed by questions about battery exposure makes the connection obvious.

Social Desirability (Faking Good)

Respondents give socially acceptable answers rather than truthful ones. Mothers tend to deny smoking during pregnancy even when they did. This affects any question touching identity, values, or behavior.

Rumination Bias

People who are sick think harder about possible causes and report higher exposure rates than healthy controls, even when actual exposure is the same.

Why This Matters for Voice Interviews

Voice interviews amplify some of these biases and reduce others. The social presence of an AI interviewer makes respondents more polite (social desirability increases), but the conversational format reduces response fatigue and gives richer answers than checkbox surveys.

Understanding these 48 bias types helps you design voice interview questions that are neutral, clear, and structured to capture authentic emotional responses rather than compliance or politeness.

The Bias Prevention Checklist

The whitepaper includes a 16-point checklist you can use before launching any study:

Each question asks exactly one thing
All questions use plain, everyday language
Questions are free of leading phrasing
No loaded adjectives or emotional words
No assumptions or presuppositions
Response scales have no gaps or overlaps
Sensitive questions placed late with normalizing preambles
Time references use fixed dates, not vague terms
The questionnaire has been pretested with a small sample

Download the full whitepaper (10 pages, no form required) from our Whitepapers page.

And if you want to check your questions right now, try our free AI-powered Question Bias Checker. It is trained on all 48 bias types and rewrites your questions using voice interview best practices.

Most surveys fail before a single response comes in. The problem is not sample size, not timing, not incentives. It is the questions themselves.

We turned their research into a free, practical whitepaper: The Complete Guide to Question Bias: 48 Types Every Researcher Must Know.

Here is what it covers and why it matters for anyone designing surveys, interviews, or feedback forms.

The Three Sources of Bias

Choi and Pak organized all 48 bias types into three categories based on where the bias originates:

Question Design (24 types): How individual questions are worded, scaled, and structured. This is the largest category and the one most teams think about, but even experienced researchers routinely miss several of these.

Questionnaire Design (6 types): How questions are formatted, ordered, and organized as a whole. Even perfectly worded questions can produce bad data if the questionnaire structure introduces bias.

Administration (18 types): How the questionnaire is delivered and completed. These biases arise from interviewer behavior, respondent psychology, recall errors, and cultural differences. They are the hardest to detect because they happen during the interaction, not on paper.