Lecture 4 - handling rt data


Introduction

Course: PSYC204 - Using Response Time to Study Human Behaviour

Lecture Focus: Response Time (RT) Data Distribution - Dr Andrew Hodrien

Data Preparation before Analysis

Proper preparation of RT data is crucial for robust data analysis. Just like preparing a carrot, it involves careful selection and cleaning of data to ensure accuracy in outcomes.

Participants Inclusion:

  • Conduct thorough attention checks to maintain data integrity.

  • Ensure participants are performing above chance levels to confirm they are engaging with the task meaningfully.

Data Points Inclusion:

  • Include only the responses to 'Yes' trials which are indicative of the critical experimental tasks.

  • Include only trials that were responded to correctly, discarding any incorrect responses to ensure reliability of findings.

  • Exclude responses that are excessively fast (e.g., below a certain threshold) to eliminate responses that may indicate misunderstandings of the task or random error.

Trial-Level Data Processing

Focus should be directed at the RT data gathered from test trials resulting in correct "Yes" responses. This approach minimizes the influence of erroneous or misleading data points, enabling clearer insights into participant behavior.

Distribution Analysis:

  • Examine the normality of RT data by evaluating skewness and kurtosis values. This involves using statistical tests and visual inspections to characterize data distribution.

  • If the data is normally distributed, parametric analysis techniques can be applied; if not, extreme values may need to be trimmed or the data transformed to meet analysis requirements.

  • Consider treating "No" responses as incorrect data points, impacting the interpretation of overall performance statistics.

Addressing Slow Responses

In-depth analysis of the slow responses is necessary:

  • Investigate whether the distribution of slow responses exhibits significant differences from faster responses; patterns may indicate different cognitive strategies or difficulties.

  • Identify which slow response data points should be viewed as informative signals, potentially revealing underlying cognitive processes.

Understanding RT Data Distribution

Key Characteristics:

  • Positive Skewness: This is a common phenomenon in RT data where most responses cluster at the lower end of the scale and few are stretched toward higher values. It's essential to understand its implications for mean reporting.

  • Mean vs. Median Reporting: Due to positive skewness, the mean may be disproportionately influenced by extreme values, making medians a preferred measure of central tendency in such cases.

  • Excel Functions: Utilize the following functions for calculations:

    • =MEDIAN((cell 1):(cell 30)) for median calculation.

    • =AVERAGE((cell 1):(cell 30)) for mean calculation.

Normal Distribution Requirements

For statistical analyses that rely on normal distribution, it is crucial that:

  • Data should be normally distributed or close to it to validate the reporting of means or conduct parametric analyses (e.g., ANOVA, t-tests).

  • While ANOVA can tolerate some deviations from perfect normality, extreme non-normality may compromise results.

Key Indices for Normal Distribution

  • Skewness: Refers to the degree of asymmetry of the distribution; a skewness of 0 indicates a normal distribution.

  • Kurtosis: Measures the sharpness or flatness of the distribution; a kurtosis value of 0 represents a normal distribution's peak.

Checking Normality

Assess normality through a combination of statistical tables and histograms. Tools such as SPSS, JASP, or R can facilitate this process by providing visual and quantitative assessments.

Handling Non-Normal Data

Strategies:

  1. Data Trimming:

    • Exclude trials with slow responses using conservative criteria (e.g., mean +/- 3.5 standard deviations). The goal is to maximize the retention of valid data points while reducing noise from outliers.

  2. Data Transformation:

    • Implement transformations such as logarithmic or square root to normalize data distributions. However, it's important to acknowledge that transformations may result in decreased transparency in results.

Examination of Data Trimming

Excessive trimming of data can lead to misrepresentations; therefore, caution is essential to ensure that the data represents genuine cognitive processes rather than skewed interpretations. Remaining mindful of biases during data manipulation is crucial.

Data Description

Clearly articulate the percentage of data lost due to trimming. A typical expectation is a 2-7% loss, whereas a loss greater than 10% may suggest potential over-trimming risks that could undermine the study’s conclusions.

Data Transformation Overview

If the data does not conform to normal distribution, alternative modeling of the complete dataset or cautious application of transformations is advised. Recognize that no single method is flawless; thus, the rationale behind the chosen methods should be transparently justified.

Knowledge Check

  • Identifying Skewness: Positive skewness reveals a longer tail to the right of the distribution, affecting the interpretation of RT data.

  • Mean and Median Relations: In cases of positive skewness, the mean will be greater than the median, whereas in negative skewness, the opposite is true. This knowledge is vital for accurate data interpretation.

Summarizing Key Terms

  • Key Terms: Response Time (RT), Error Rate, Signal to Noise Ratio (S/N), Speed-Accuracy Tradeoff (SAT).

Key Knowledge for Research Design

An understanding of experimental structure is essential for designing robust RT studies. Considerations include:

  • Maximizing the signal-to-noise ratio is critical for enhancing data quality.

  • Acknowledging different experimental effects, such as potential congruency effects, which have been established by prior studies.

Key Skills for Data Analysis

  • Psychopy: A tool to design, adjust, and execute experiments effectively.

  • Excel: Utilized for data organization (e.g., pivot tables), calculations (e.g., data transformations), and filtering.

  • Collaborative Skills: Essential for group work and effective scientific communication to enhance research output.

Closing Remarks

Continue to explore available resources for ongoing learning and prepare thoroughly for upcoming presentations related to this course's topics.