Floor and Ceiling Effects in Anxiety Measures: Why Scores Can Be Misleading at the Extremes

A questionnaire score of zero does not mean the absence of anxiety. It means the questionnaire cannot detect any more. This distinction — between the lower bound of a scale and the lower bound of a symptom — is the floor effect, and it creates problems throughout anxiety research that are easy to overlook.

What floor and ceiling effects are

Floor effect: a large proportion of respondents score at or near the minimum of a scale, making it impossible to distinguish among them. The scale’s resolution runs out before the construct does.

Ceiling effect: a large proportion score at or near the maximum, with the same blinding at the other end.

Quality criteria for health status questionnaires recommend flagging a floor or ceiling effect when more than 15% of respondents score at the scale’s minimum or maximum [2]. In anxiety research, floor effects are more common and more consequential, because they emerge precisely where accurate measurement matters most: among people who are getting better.

The GAD-7 as an example

The GAD-7 is the most widely used brief measure of generalized anxiety. It has seven items, each scored 0 (not at all) to 3 (nearly every day), for a total range of 0 to 21 [1]. It is quick, validated and free to use, which explains its ubiquity.

But a 22-point range is narrow. After successful treatment, many patients score 0–3. At the floor, the scale cannot distinguish between someone who is genuinely well and someone who has residual worry that does not meet the threshold of “several days” on any item. Both score 0. The resolution is gone.

At the ceiling, a person with extreme anxiety that prevents them from leaving the house and a person with constant moderate worry may both score 20 or 21. The scale cannot capture the severity difference.

This is not a failure of the GAD-7 in particular. It is a structural consequence of using a short scale with few items and a narrow response range. The scale was designed for efficient screening, not for precise measurement across the full severity range.

How floor effects distort treatment research

Treatment comparison sensitivity. If two treatments both bring most patients to the scale floor, they look equally effective even if one produces deeper recovery. The scale cannot detect the difference. This biases comparison trials toward finding no difference between active treatments — a conclusion that may be an artifact of measurement insensitivity.

Relapse measurement. Patients who reach the floor after treatment cannot show deterioration until their symptoms cross a threshold large enough to register on the scale. Early relapse signals are invisible. By the time a change is detectable, the relapse may be well advanced.

Change scores. Pre-to-post change scores are compressed for patients who start near the floor or improve to it. A patient who starts at 5 and reaches 0 shows a change of 5 — but their true improvement might be much larger or smaller than that of another patient with the same numbers.

Exclusion effects. If researchers require a minimum score for trial entry (a common practice), they exclude people near the floor — but those people may still have clinically meaningful anxiety that the scale cannot capture.

Item response theory offers a different lens

Classical test theory treats every point on a scale as equally informative. Item response theory (IRT) does not: it maps where each item provides the most information across the severity continuum [4].

IRT analyses of anxiety scales typically show that most items cluster in the middle of the severity range. They are most informative for moderate symptoms and least informative at the extremes. This is the psychometric explanation for why floor and ceiling effects occur: the items were not designed to discriminate at those levels.

An IRT analysis of clinical measurement demonstrated that composite scores lose precision at the extremes of their range, and that this loss can be substantial enough to miss clinically significant changes in individual patients [3].

Computerized adaptive testing

One solution is computerized adaptive testing (CAT), which selects items from a large bank based on the person’s previous answers. If a respondent’s answers suggest low severity, the next item is chosen to discriminate well in that range. The result is a precise measurement at any severity level without requiring the person to answer dozens of items.

The PROMIS (Patient-Reported Outcomes Measurement Information System) anxiety item bank uses this approach, drawing from a large pool of calibrated items to provide precise measurement across a wide severity range [5]. PROMIS anxiety scores have been shown to maintain measurement precision at levels where the GAD-7 hits its floor.

The trade-off is complexity: CAT requires software, an item bank and validated calibrations. The GAD-7 requires a printed page and a pen. In routine clinical settings, convenience often wins.

How to read research with this in mind

1. Check the distribution of scores. If the paper reports means and standard deviations, note whether the mean is near a scale boundary. If it is, floor or ceiling effects are likely.
2. Ask what scale was used and what its range is. Brief screening tools have narrower ranges than comprehensive measures.
3. For treatment trials, look at whether participants clustered at the floor post-treatment. If most scored 0–2, the scale may not have captured meaningful differences.
4. For comparisons showing no difference between treatments, consider whether measurement insensitivity at the floor could explain the null finding.

For a broader discussion of what different measurement approaches can and cannot capture, see self-report versus clinician-rated measures. For what the GAD-7 score actually represents, see GAD-7 scores: what they mean.

When to look for more support. A low score on a screening questionnaire does not necessarily mean you are fine. If you still feel anxious despite a normal score, trust your experience and talk to a clinician. Questionnaires are screening tools, not diagnoses. If you are in crisis, call or text 988 (US) or Samaritans on 116 123 (UK and Ireland).