Dichotomous Questions

By Gojko Adzic · Published 11 July 2026

Dichotomous questions are questions that offer exactly two possible answers, mutually exclusive (with an unambiguous split). This article explains the key things you should know when using dichotomous questions in research.

• Dichotomous Format
• Types of Dichotomous Research Questions
  • Dichotomous Question Examples
• Dichotomous Surveys
  • Dichotomous Questionnaire Example
• Problems with Dichotomous Questions
  • Dichotomization reduces statistical power and shows false statistical significance
  • Acquiescence bias skews results towards positive answers 10-50%
  • Forced choices create artificial agreement
  • Binary scales have lower reliability than multi-point scales
• Applicability and alternatives

Dichotomous Format

The name “dichotomous” comes from the Greek word διχοτομία (dichotomía), meaning “dividing in two”. In the context of research, dichotomous format has several distinct meanings:

1. Question format: In this context, the name implies a question in a survey structured to offer only two possible answers (for example Yes/No). See the section on Dichotomous research questions for more information on this context.
2. Survey/Questionnaire type: In this context, the name implies a research instrument (such as a survey), mostly composed of questions in the dichotomous format. See the section on Dichotomous questionnaires for more information.
3. Binary analysis: Researchers sometimes also call a question “dichotomous” when a continuous construct (such as age or income) is treated as binary for analysis. Researchers pick a cutoff point (for example the median value in their sample) and turn the continuous variable into a binary one. See the section on Dichotomization for more information.
4. Research with a binary dependent variable: In this context, the name that the outcome of the research is binary (for example, “Does changing the heading title on the homepage positively impact customer conversion rates?” or “Does high level of social media usage correlate to higher voter turnout at local elections?”). The study itself may involve more complex research mechanisms, but the result can effectively have only two outcomes.

A dichotomous question is structured so that there are only two possible answers, such as True/False. Such question formats, particularly in the form of Yes/No, predate formalized survey research, so there is no particular known original work introducing this format.

Jon A. Krosnick, in Survey Research claims that dichotomous questions are very popular, appearing in “numerous batteries developed for attitude and personality measurement”, because they are “appealing from a practical standpoint”, and “easy to write and administer”. He also warns that “these formats are also seriously problematic”. See the section on Problems with Dichotomous Questions.

When creating a survey, it’s important to distinguish between two types of dichotomous survey questions:

1. True dichotomy, where the answers are genuinely binary and mutually exclusive, such as alive/dead.
2. Forced dichotomy, where a continuous variable or a multi-point scale is presented as a binary choice. For example, an attitudinal question that would usually imply nuance but only offers two response options such as Like/Dislike.

It’s important to evaluate if offering a third choice (such as “I don’t know” or “Not Applicable”) would be more appropriate than forcing the respondents to choose between only two options. (See the section on forced choice for more information.)

Dichotomous Question Examples

Dichotomous questions appear in many forms in surveys. Some of the most common formats are the following:

• Yes / No: for example, “Do you own a car?”. This is the most common format, appropriate for for factual, behavioral, and qualifying questions.
• True / False or Correct / Incorrect: for example, “Is the answer 3.14?”. This is a format common for knowledge assessment tests and in educational contexts.
• Agree / Disagree, for example, “Do you agree that the Republican Party represents your interests best?”. This is a two-point version of the Likert Scale, without any without the intermediate options.
• Forced-choice (this-or-that), for example “Which do you prefer: A or B?”, requiring the respondent to choose between two options.
• Checkbox / binary indicator, presented as “select if applicable”. This is effectively a dichotomous question because checking the box or selecting the option implies a “Yes” answer, and not selecting implies “No”.
• Pass / Fail used in quality control, compliance, and screening.

A dichotomous questionnaire (or dichotomous-type questionnaire) is a survey composed primarily or entirely of dichotomous questions. Dichotomous questionnaires are common in compliance checklists, health risk assessments, and quick customer-feedback forms.

Dichotomous-type questionnaires are appropriate when speed and simplicity matter more than nuance of responses, such as fast triage, screening or eligibility filtering before more in-depth research. They are also appropriate when all questions genuinely create a true dichotomy, such as compliance checklists where every item has a correct binary state or knowledge tests where answers are objectively right or wrong.

Surveys mostly composed from dichotomous format questions should be avoided when measuring attitudes, opinions, or satisfaction where more nuance is needed, or any construct where a meaningful middle ground exists (to avoid forcing an artificial choice).

Here are two dichotomous survey questions examples:

Dichotomous questions might be simple to write and easy to research in a survey, but they suffer from several systematic effects that are important to consider.

Dichotomization reduces statistical power and shows false statistical significance

Dichotomization is the process of survey result analysis when researchers treat a multi-point or continuous variable as binary. For example, the survey question might have a sliding-scale input for age, and the researchers simplify the sample by calculating the median of the responses, then assigning the value “young” to everyone below the median, and “old” to everyone above the median. The original numbers get thrown away, and research deals with effectively two buckets of responses.

The result behaves like a dichotomous question even though nobody actually answered “old” or “young” when filling in the survey, because the researcher imposed the binary split after the survey collection, during analysis. In On the Practice of Dichotomization of Quantitative Variables, MacCallum, Zhang, Preacher and Rucker argue that such “dichotomization” is “rarely defensible and often will yield misleading results.”

Maxwell and Delaney, in Bivariate Median Splits and Spurious Statistical Significance also argue that dichotomization “underestimates the strength of relationships and reduces statistical power”, and that the effects are even worse if the question posed as dichotomous is impacted by multiple continuous variables, which are then dichotomized together.

Specifically, dichotomizing 2 continuous independent variables can lead to false statistical significance. As a result, the typical justification for using a median split as long as results continue to be statistically significant is invalid, because such results may in fact be spurious. Thus, researchers who dichotomize multiple continuous predictor variables not only may lose power to detect true predictor-criterion relationships in some situations but also may dramatically increase the probability of Type 1 errors in other situations.

– Maxwell and Delaney, Bivariate Median Splits and Spurious Statistical Significance

Even with large samples, that would normally reduce the effects of loss of statistical power through dichotomization, Maxwell and Delaney argue that if two variables are dichotomized together the resilts might not be reliable due to “spurious interactions”.

MacCallum and colleagues present two cases when dichotomization might be appropriate:

1. Taxometric analysis where “clear support for the existence of two types or taxons within the observed sample” exists, along with a “clear scale point that differentiated the classes.” In the paper, they warn that such support must not be assumed, but actually proven separately, and that using a median value for dichotomization is probably wrong.
2. Where the distribution of a count variable is extremely skewed, and a large proportion of responses is at the most extreme score on the distribution. As an example, the authors provide the question of “How many cigarettes do you smoke per day”, with the majority of respondents answering “zero”. In that case, it may be justified to dichotomize the results into Smokers/Non-Smokers.

Acquiescence bias skews results towards positive answers 10-50%

Jon A. Krosnick, in Survey Research, argues that people tend to agree with binary choice questions more than disagree, in a way that is systematic and statistically significant. Questions that require respondents to choose between True/False or Agree/Disagree suffer from the “acquiescence effect”, “the tendency to endorse any assertion made in a question, regardless of its content.”

Krosnick provides several possible causes for the acquiescence effect, including confirmatory bias (“most people typically begin by seeking reasons to agree rather than disagree”), psychological predisposition to be agreeable, low motivation or lack of expertise about the topic, or even just trying to be polite.

Other researchers have proven that a similar effect is present even for multi-point Likert Scale surveys (for example John J. Ray in 1990 in Acquiescence and Problems With Forced-Choice Scales and Bill Toner in 1987 in The Impact of Agreement Bias on the Ranking of Questionnaire Response), but Krosnick claims that the effect is particularly significant for dichotomous questions. Comparing data from several research experiments, Krosnick claims that there is “an average acquiescence effect of about 10%”.

Hill and Roberts studied the effects of acquiescence for political misconceptions, particularly conspiratorial beliefs, and concluded that the effects in those situations can be much larger, up to “50 percentage points” (the raw survey number is roughly double the real figure).

Jaak B. Billiet and McKee J. McClendon, in Modeling Acquiescence in Measurement Models for Two Balanced Sets of Items propose combining the “balanced scales” approach (where the same question is asked from two opposite perspectives to detect bias) with a separate variable unrelated to the original question, also tested using balanced scales, to get a kind of a signature for the bias of the respondents. Their paper is an excellent reference for further study of acquiescence detection and methods to compensate for it.

Forced choices create artificial agreement

Another potentially problematic aspect of dichotomous question format is that it forces a choice between two options, even when the respondents do not hold a particular opinion on the subject, or lack expertise to answer the question.

Howard Schuman and Stanley Presser compared surveys with Yes/No answers and adding a “Don’t Know” (DK) option, concluding that “the addition of a DK filter to an opinion question will typically induce more than a fifth of the sample to shift from substantive positions into the DK category”.

Kenneth A. Rasinski, David Mingay and Norman M. Bradburn in Do Respondents Really “Mark All That Apply” on Self-Administered Questions? compared a sequence of Yes/No forced choice questions with an alternative format in which the participants were asked to select all options that apply to them from a set. They found that significantly fewer response options were selected for with forced choices than when marking all that apply. Jolene D. Smyth, Don A. Dillman, Leah Melani Christian and Matthew J. Stern in Comparing Check-All and Forced-Choice Question Formats in Web Surveys repeated this research for online surveys, and while they agree with the previous conclusions, they present an alternative conclusion. Their argument is that forced choice questions, particularly in online surveys, cause respondents to think harder and spend more time on individual questions than in check-all-that-apply, so the forced choice options might be preferable. Mario Callegaro, Michael H. Murakami, Ziv Tepman and Vani Henderson in Yes–no Answers versus Check-All in Self-Administered Modes: A Systematic Review and Analyses measured that “endorsement levels increase by a factor of 1.42 when questions are posed in a forced-choice rather than check-all format”, pointing to significant systematic bias.

Binary scales have lower reliability than multi-point scales

Cicchetti, Showalter and Tyrer performed Monte Carlo simulations testing inter-rater reliability from binary up to multi-point scales, arguing that reliability increases up to 7 points, and then has no substantial gain beyond 7 categories. They concluded that “The level of interrater agreement was always lowest for 2 categories of classification “ and that “the most dramatic increase being between 2 and 3 categories of classification.”

Preston and Colman empirically studied reliability of binary and multi-point scales, asking respondents to rate the same items on scales of 2 to 11 items. Their conclusion is that the binary scale had significantly lower test-retest reliability than scales with 6–10 categories (notably, their research also claims that 3 and 4 point scales also performed poorly).

Dichotomous survey questions are best used for truly dichotomous variables, or for screening and filtering before more in-depth research. They are also applicable for knowledge tests and compliance checklists, where forced choice is critical for target metrics.

When reliability or statistical relevance is critical, consider the following alternatives:

• using “check all that apply” with multiple options instead of forcing a Yes/No choice for each item
• adding a “don’t know” option to get directional signals about the strength of conviction for opinion studies
• expanding the answers to a Likert Scale of 7 or more options for [Attitudinal measurement](Attitudinal Metrics).