A Method for Removing Outliers to Improve Factor Analytic Results.

Area of Science:

• Statistics
• Psychometrics
• Data Analysis

Background:

• Inaccurate data, including faked responses and errors, can significantly distort correlation matrices.
• Distorted correlation matrices lead to unreliable results in factor analysis.
• Identifying and addressing problematic data cases is crucial for valid statistical outcomes.

Purpose of the Study:

• To present a novel method for detecting potentially problematic data cases (outliers) within correlation matrices.
• To compare the effectiveness of this new outlier detection method with an existing program (BMD 10M).
• To evaluate the utility of outlier detection programs in data cleaning for factor analysis.

Main Methods:

• A new outlier detection technique was developed based on the average squared deviation of a subject's cross-product of standard scores.
• This method was applied to data matrices and compared against the BMD 10M outlier program.
• The overlap in outlier identification between the two methods was analyzed.

Main Results:

• The novel method identified outliers based on deviations in cross-product standard scores.
• Approximately 40-60% of outliers detected by the new method and the BMD 10M program were consistent.
• A significant portion of identified outliers did not represent 'bad data' but required further investigation.

Conclusions:

• The developed method is a viable tool for identifying potential outliers in data for factor analysis.
• Outlier detection programs are valuable for flagging cases for scrutiny, not for automatic data removal.
• Careful examination of flagged cases is essential to determine the nature of the deviation and appropriate action.