Machine learning applied to the Morningness-Eveningness Questionnaire revealed that a single metacognitive self-assessment item (Item 19) demonstrated 'threefold greater importance than any other item,' and a six-item combination achieved robust overall classification, potentially reducing assessment burden by 70%.
Key Findings
Results
Item 19 ('Which chronotype do you think you are?') demonstrated exceptional predictive utility, showing threefold greater importance than any other item in the MEQ.
Item 19 is a direct metacognitive self-assessment of chronotype
Its importance was approximately three times greater than the next most predictive item
The authors interpret this as reflecting 'relatively accurate chronotype identification rooted in lived experience'
This finding was derived from machine learning feature importance analysis applied to the German MEQ
Results
Chronotype-specific analyses revealed distinct predictive item patterns across morning, neutral, and evening types, with each type relying on different item combinations for optimal classification.
Three chronotype groups were examined: morning type, neutral type, and evening type
The study used the Dortmund Vital Study cohort (ClinicalTrials.gov NCT05155397), a prospective cohort study on healthy cognitive aging
Classification patterns were heterogeneous, meaning no single set of items was universally optimal across all chronotypes
Chronotype-specific analyses were conducted separately to uncover these differential patterns
Results
Partial dependence analysis identified non-linear response patterns in MEQ items, including sigmoid curves, threshold effects, and plateau regions.
Non-linear patterns challenge the assumption underlying composite scoring that item responses contribute linearly and equally to chronotype classification
Three specific pattern types were identified: sigmoid curves, threshold effects, and plateau regions
These findings support methodological concerns about multidimensionality and unequal item contributions in the MEQ
The analysis was conducted using machine learning techniques applied to item-level data
Results
A six-item combination achieved robust overall chronotype classification, potentially reducing MEQ assessment burden by approximately 70%.
The full German MEQ contains 19 items; a six-item subset was identified as sufficient for robust classification
This represents a reduction of approximately 70% in the number of items required
The abbreviated version was identified through machine learning-derived item importance hierarchies
The authors propose this as a foundation for 'abbreviated screening tools for primary care consultations, large-scale epidemiological studies, and clinical contexts'
Methods
The study applied machine learning techniques to the German MEQ within the Dortmund Vital Study, a prospective cohort study focused on healthy cognitive aging.
The dataset was drawn from the Dortmund Vital Study, registered at ClinicalTrials.gov under NCT05155397
The study population consisted of healthy aging adults
Machine learning was used to identify item-level predictive hierarchies and response patterns
An open-access R tutorial was produced to ensure reproducibility and facilitate adaptation across diverse populations and instruments
Background
The MEQ's reliance on composite scores is methodologically problematic due to concerns about multidimensionality and unequal item contributions.
Widely used chronotype tools like the MEQ aggregate items into a single composite score
The authors identify this as a limitation that obscures item-level predictive differences
Machine learning approaches were applied specifically to address this limitation
The study aimed to advance 'theoretical understanding and empirical precision in chronotype assessment' beyond composite scoring
What This Means
This research suggests that the most commonly used questionnaire for assessing whether someone is a 'morning person' or 'evening person' — the Morningness-Eveningness Questionnaire (MEQ) — contains a single standout question that is far more informative than all the others. That question simply asks people what chronotype they think they are. According to the study, this self-assessment was about three times more predictive of a person's actual chronotype than any other question on the 19-item questionnaire, suggesting that people's own sense of their daily rhythm patterns is surprisingly accurate.
The researchers also found that the standard practice of adding up all questionnaire responses into one total score may be overly simplistic. Using machine learning on data from a large German cohort study on healthy aging, they discovered that different chronotype groups (morning, neutral, and evening types) are best identified by different combinations of questions, and that the relationship between individual question responses and chronotype classification is often non-linear — meaning the pattern is not a simple straight line but can involve tipping points or leveling-off effects. These findings challenge the assumption that all questions contribute equally and in a straightforward way to chronotype measurement.
Practically, this research suggests it may be possible to assess chronotype reliably using just six questions instead of nineteen — a 70% reduction in assessment length — which could make chronotype screening faster and more feasible in clinical settings, large population studies, and primary care. The authors also released an open-access tutorial so that other researchers can apply these methods to different populations and questionnaires, potentially improving how circadian preferences are measured across a wide range of health and research contexts.
Check Your Own Numbers
Upload your bloodwork. We'll cross-reference your results against this study and 4,700 others.
Metzler Y, Schade H, Nitsche M, Wascher E, Getzmann S, Gajewski P, et al.. (2026). Beyond composite scores in chronotype assessment: item-level predictive patterns in the Morningness-Eveningness Questionnaire.. Scientific reports. https://doi.org/10.1038/s41598-026-54301-w