🧪 This platform is in early beta. Features may change and you might encounter bugs. We appreciate your patience!
central_tendency_bias
Central tendency bias occurs when observers or respondents avoid the extreme ends of a rating scale, clustering their responses near the middle. This compression of variance reduces the ability to distinguish between truly different subjects, weakens statistical power, and can mask real patterns in the data.
A manager rates all 20 team members between 3 and 4 on a 5-point performance scale, despite clear differences in actual performance. The compressed ratings make it impossible to identify top performers or those needing improvement.
Customers asked to rate their satisfaction with a new government service on a 1-to-7 scale predominantly choose 3, 4, or 5, regardless of their actual experience. Policymakers interpret the middling scores as moderate satisfaction, when in reality many respondents were either very happy or very frustrated but felt uncomfortable selecting extreme options on an official form.
Medical students evaluating each other's clinical communication skills in a peer assessment exercise cluster nearly all scores between 6 and 8 out of 10, even when some peers clearly struggled and others excelled. The compressed scores prevent faculty from identifying students who need additional support or those ready for advanced responsibilities.
Binary (yes/no) questions an LLM must answer to identify this aspect:
Does the data collection rely on subjective rating scales?
Type: binaryIs there an unusually high concentration of responses near the midpoint of the scale?
Type: binaryCould raters be avoiding extreme categories due to uncertainty or social pressure?
Type: binaryDoes the compressed range of responses reduce the ability to detect true differences?
Type: binaryCentral tendency bias occurs when observers or respondents avoid the extreme ends of a rating scale, clustering their responses near the middle. This compression of variance reduces the ability to distinguish between truly different subjects, weakens statistical power, and can mask real patterns in the data.
Extreme ratings feel risky — they require stronger justification and may attract scrutiny. Raters feel safer in the middle, especially when uncertain about their judgment. This tendency is amplified in cultures that value modesty and consensus.
Use behaviorally anchored rating scales (BARS) that define each point with specific examples. Force ranking or forced distribution methods can counteract central clustering. Train raters to use the full scale and provide clear criteria for extreme ratings.
Employee performance reviews are notorious for central tendency bias. Most organizations find that 80-90% of employees are rated 'meets expectations' or equivalent, despite wide variation in actual performance. This undermines merit-based decisions on promotions, raises, and development.
Respondents agree with statements regardless of content, inflating affirmative responses.
Researcher expectations systematically influence how observations are recorded.
Equal measurement error across groups that typically biases estimates toward the null.
Tendency to round measurements to preferred digits, distorting data distributions.
Use these tools to detect, analyze, or train this aspect.