Effects of truncation on reaction time analysis

Area of Science:

• Cognitive Psychology
• Psychometrics
• Behavioral Neuroscience

Background:

• Reaction time (RT) data analysis often involves excluding outliers.
• Data truncation, removing RTs outside a set range, is a common but potentially biasing practice.
• Extreme RTs, though infrequent, can be valid measures of cognitive processes.

Purpose of the Study:

• To examine the biasing effects of data truncation on RT measures.
• To investigate how truncation impacts various statistical properties of RT distributions.
• To introduce a statistical method for mitigating truncation bias.

Main Methods:

• Simulations under different distributions of valid and spurious RTs.
• Analysis of truncation effects on mean, median, standard deviation, and skewness.
• Evaluation of bias on RT-independent variable relationships, factorial designs, and hazard functions.
• Development and presentation of a maximum likelihood estimation procedure.

Main Results:

• Truncation bias is substantial, often exceeding known experimental effects.
• Significant distortion of linear RT relationships, factorial patterns, and hazard functions observed.
• Statistical power remains largely unaffected by truncation.
• The proposed maximum likelihood method shows promise for estimating untruncated RT properties.

Conclusions:

• Data truncation in RT research introduces significant bias, distorting key statistical measures and relationships.
• Researchers should be cautious about excluding extreme RTs without considering potential biases.
• A maximum likelihood procedure offers a viable approach to estimate unbiased RT distributions and control for truncation effects.