🧪 This platform is in early beta. Features may change and you might encounter bugs. We appreciate your patience!
digit_preference_bias
Digit preference bias occurs when observers systematically round measurements to preferred numbers, typically those ending in 0 or 5. This seemingly minor habit can have significant consequences when rounding pushes values across diagnostic or treatment thresholds, distorting prevalence estimates and clinical decisions.
A nurse recording blood pressures rounds readings to the nearest 10 mmHg. A true reading of 138/88 is recorded as 140/90, pushing the patient above the hypertension threshold. Across a population study, this inflates the apparent prevalence of hypertension.
A field epidemiologist recording ages of patients during a disease outbreak in a region without reliable birth records finds a suspicious spike in reported ages of 30, 35, 40, and 45, with almost no one reporting ages like 31, 33, 37, or 42. The apparent age-disease relationship is distorted because local interviewers are rounding to the nearest five years.
A researcher analyzing self-reported body weights from a large health survey notices that weights cluster heavily at 140, 150, 160, and 170 pounds with far fewer entries at 143, 157, or 163 pounds. This rounding compresses the true weight distribution and skews calculations of average BMI and obesity prevalence across the sample.
Binary (yes/no) questions an LLM must answer to identify this aspect:
Does the data show an unusual concentration of values ending in certain digits (0, 5)?
Type: binaryWere measurements recorded by human observers rather than automated instruments?
Type: binaryCould rounding have shifted measurements across a clinically meaningful threshold?
Type: binaryWas the digit distribution analyzed to detect rounding patterns?
Type: binaryDigit preference bias occurs when observers systematically round measurements to preferred numbers, typically those ending in 0 or 5. This seemingly minor habit can have significant consequences when rounding pushes values across diagnostic or treatment thresholds, distorting prevalence estimates and clinical decisions.
Human observers have a natural tendency to simplify numerical values. Rounding feels insignificant in any single measurement, but when it occurs systematically across thousands of observations, it creates measurable distortions in data distributions.
Use automated digital measurement devices that record exact values. Train data collectors to record precise readings. Check data distributions for heaping at round numbers as a quality control step. Flag and investigate suspicious digit patterns.
Blood pressure measurement studies consistently show excess readings at 0 and 5, even with mercury sphygmomanometers that allow precise readings. This has influenced clinical guidelines and prevalence estimates of hypertension worldwide, leading to the push for automated measurement devices.
Systematic error arising from faulty or poorly calibrated measurement instruments.
Raters avoid extreme values, compressing variability in subjective assessments.
Equal measurement error across groups that typically biases estimates toward the null.
Researcher expectations systematically influence how observations are recorded.
Use these tools to detect, analyze, or train this aspect.