Tipsheet - Question Wording

What would the “perfect survey” look like?

  • All questions would measure the concept they intend to measure.
  • No question would erroneously measure unintended concepts.
  •  All respondents would interpret questions in the same way.
  • All respondents would be able to respond accurately.
  • All respondents would be willing to answer

Survey questions almost never achieve all of these goals. The job of the researcher, then, is to reduce error, realizing that no survey and no question will be perfect. This tip sheet provides a number of “best practice” tips to reduce measurement and comprehension error as a result of the question wording process.

Avoid double-barreled questions.

Do you favor or oppose cutting taxes and government spending?

This question really asks two questions: what does the respondent think about cutting taxes and what does the respondent think about cutting government spending? Break questions like this into two items:

Do you favor or oppose cutting taxes?
Do you favor or oppose cutting government spending?

Sometimes there are phrases that seem double barreled because they include the word “and,” but they are common terms that many people use to refer to a single concept. The common phrases “free and fair elections” or “safe and legal abortion” are examples of these. Researchers should be careful when using such terms to make sure they are not really asking two distinct questions.

Avoid erroneous assumptions.

In the past twelve months, who has treated you for headaches – a general doctor, a specialist, a physician’s assistant, or a nurse?

The question assumes that a respondent had a headache in the last year and that the respondent saw a medical professional for treatment. Researchers should avoid asking questions that make assumptions that may not be true. This question can be broken down into two or three separate items. Though the respondent will have to answer more questions, this approach will yield better data quality and avoid confusing the respondent. Consider these alternatives:

Did you have a headache in the last twelve months?
(if yes) Did you make an appointment with a medical professional to get your headache treated?
(if yes) Who treated you for your headache - a general doctor, a specialist, a physician’s assistant, or a nurse?

In the last twelve months, did you have a headache that was serious enough that you sought treatment from a medical professional for it?
(if yes) Who treated you for your headache - a general doctor, a specialist, a physician’s assistant, or a nurse?

Clarify ambiguous and imprecise terms or break them down into several questions.

Do you agree or disagree that moral values are an important issue facing the nation?

The term “moral values” is vague and abstract. This term can mean different things to different people. Vague phrases like this are fine to use, but the researcher should define them to avoid differing comprehensions among respondents. Consider this alterative:

Do you agree or disagree that moral values are an important issue facing the nation? By moral values, we mean how much government should be involved in regulating citizen behavior when questions of right and wrong are involved.

Alternatively, if the researcher is not satisfied with the clarity or quality of a definition, it is common to ask a series of questions that collectively measure how respondents feel about a topic. Rather than asking about moral values per se, the researcher might ask about a number of specific issues that fall within the “moral values” umbrella:

Many issues that involve moral values are prominent in politics today. Below we have a list of these issues. For each issue, please tell us how important you think it is: 

  •  Abortion
  • Gay marriage
  • Death penalty
  • No fault divorce
  • The religiosity of public officials

Define terms very specifically when necessary.

What was your income last year?

Income is vague. Does this mean personal or household income? Does this mean income before or after taxes? Income needs a more specific definition. This will lengthen the question, but including the definition makes the question more precise. Consider:

What was your total household income before taxes in 2009?

Avoid loaded, leading, emotional, or evocative language as it can bias responses.

Do you believe the US should immediately withdraw troops from the failed war in Iraq?
Do you support or oppose the death tax?
Did the US make a mistake in deciding to defend Kuwait?
Do you approve or disapprove of how our president, George W. Bush, is doing his job?

The questions above all contain biasing language – failed war, death tax, defend, and our president. All of these terms can bias respondents toward a certain point of view. Sometimes the bias is subtle, even to respondents, but in other cases the respondent may actually feel pressured to respond in a certain way. These terms should be replaced with more balanced language.

Sometimes, though, there are terms that some people believe are slanted or offensive, but that are part of common language that most people understand. For example the terms “illegal immigrant,” “partial birth abortion,” and “death penalty” are likely more understandable than some of the alternative phrasings. Researchers will want to carefully choose their language in these instances to weigh ease of understanding against potential biasing language.

Avoid confusing technical or academic terms.

Has anyone in your family had a myocardial infarction in the last decade?

Technical terms may be appropriate when surveying experts, but they are often inappropriate for general population surveys. The question above might work on a survey of medical doctors, but for a general population a better phrasing might be:

Has anyone in your family had a heart attack in the last decade?

Balance questions to make positive and negative responses “ok.”

Do you feel that a woman should be allowed to have an abortion in the early months of a pregnancy if she wants to? 

Do you support increasing defense spending?

These questions place statements before respondents, but emphasize one side of the issue in the question wording (allowing abortion and supporting spending respectively). This type of construction can bias responses toward agreement with the policy statement. It is better to “allow” respondents to disagree with the statement by explicitly including disagreement language in the question itself. Consider instead:

Do you feel that a woman should be allowed to have an abortion in the early months of a pregnancy if she wants to, or do you not feel that this should be allowed?

Do you support or oppose increasing defense spending?

Consider providing counterarguments in the question itself.

Would you favor or oppose a law which would require a person to obtain a police permit before buying a gun?

Sometimes respondents may be ambivalent about a topic in the sense that they have both positive and negative considerations about it. With the gun example, people may have pro and con arguments for particular gun control policies. Ambivalent respondents are especially susceptible to question phrasing, and their responses may be biased in one direction by how the researcher frames the question. In instances like this, consider providing a counterargument that might balance the question. For example:

Would you favor a law which would require a person to obtain a police permit before buying a gun, or do you think such a law would interfere too much with the right to own guns?

Avoid complex sentences.

Do you agree or disagree that, controlling for inflation, your income has grown in the last year, where income means your gross household income calculated as the total financial receipts of all adults living in your household?

The question above is far too complex for a survey. Questions should be as simple as possible. Avoid excessive clauses and awkward phrasings that may confuse respondents. It is fine to use more conversational language in the survey setting if it will yield higher quality data than questions employing more technical and academic language.

General Advice

  • For self-administered modes, researchers do not need to include the responses as part of the question itself. The respondents will see them if using a web or mail survey. In faceto-face or phone formats, the responses must be read out loud, so they are necessarily part of the question in those instances.
  • Strike a balance between too many and too few words. Say what needs to be said in order to articulate the question clearly and precisely, but no more.
  • Avoid asking respondents to make any calculations beyond counting simple events in their heads, unless the survey is using a self-administered format in which respondents can leave the questionnaire to gather the information necessary to report a precise figure.
  • Give clear and precise instructions throughout the survey.
  • Grammar matters, especially for more educated respondents.
  • Don’t reinvent the wheel. If there is a good question from another survey that suits the purposes of your survey, then use it. Replicating questions also allows direct comparisons between surveys. However, if an existing question is faulty or does not suit your needs, feel free to modify it. With any change in the question wording or responses, though, the ability to directly compare a new question to an old one is lost. 

Author: Patrick R. Miller, DISM Survey Research Associate


Back to Top