🧪 This platform is in early beta. Features may change and you might encounter bugs. We appreciate your patience!
range_restriction
Range restriction occurs when the variability in one or more variables is artificially reduced, typically through sample selection, truncation, or censoring. When a variable's range is restricted, correlations with other variables are attenuated — they appear weaker than they truly are in the full population. This can lead to incorrect conclusions about the strength or even existence of relationships between variables.
A company studies whether SAT scores predict job performance but only examines current employees, all of whom had high SAT scores sufficient to be hired. The restricted range of SAT scores makes the correlation with performance appear near zero, leading the company to conclude SAT scores are useless — when in the full applicant pool, the relationship is substantial.
A sports psychologist studies whether mental toughness scores predict athletic performance among Olympic sprinters. Because all athletes at that level have already been filtered for exceptional mental toughness, the scores cluster tightly together, and the correlation with race times appears negligible — even though mental toughness strongly differentiates performance across the broader population of athletes.
A dating app analyzes whether profile attractiveness ratings predict match success, but only examines profiles that received at least 50 swipes — effectively excluding the least attractive profiles. The restricted sample shows almost no correlation between attractiveness and matches, because the variability in attractiveness within the included group is too narrow to detect the true relationship.
Binary (yes/no) questions an LLM must answer to identify this aspect:
Has the sample been selected or filtered in a way that reduces variability in a key variable?
Type: binaryIs the range of values in the sample narrower than the range in the full population?
Type: binaryDoes the restricted range weaken observed correlations compared to what would be found in the full population?
Type: binaryAre conclusions about relationships drawn without correcting for the restricted range?
Type: binaryRange restriction occurs when the variability in one or more variables is artificially reduced, typically through sample selection, truncation, or censoring. When a variable's range is restricted, correlations with other variables are attenuated — they appear weaker than they truly are in the full population. This can lead to incorrect conclusions about the strength or even existence of relationships between variables.
Selection processes that filter out extreme values are common and often unnoticed. Researchers may not realize their sample has been pre-selected on the variable of interest, and the resulting weak correlations are taken at face value rather than recognized as artifacts of restriction.
Identify whether sample selection has reduced the range of key variables. Apply statistical corrections for range restriction (e.g., Thorndike's correction formulas). When possible, obtain data from the full, unrestricted population. Report the range of observed values and compare it to the expected population range.
Common in personnel selection research, university admissions studies, and clinical populations where only patients above a diagnostic threshold are included.
Measurement error in predictor variables biases effect estimates toward zero.
A measurement instrument cannot distinguish differences at the upper extreme of the scale.
A measurement instrument cannot distinguish differences at the lower extreme of the scale.
The statistical error of drawing conclusions from a dataset that has been filtered by a survival or success criterion, without accounting for the filtered-out cases. The surviving sample is systematically different from the full population, and conclusions drawn from it are biased.
Use these tools to detect, analyze, or train this aspect.