How to Detect Outliers in IBM SPSS Statistics

This Tech Tip focuses on a vital step in data preparation: detecting outliers. Outliers are data points that fall significantly outside the expected range and can distort statistical results if not properly identified and addressed. Detecting them early ensures cleaner, more reliable analysis. 

To enhance your experience with IBM SPSS Statistics, the Version 1 SPSS experts have developed a series of Tech Tips designed to streamline your workflow and boost your confidence in data analysis. IBM SPSS Statistics provides a powerful, integrated interface for conducting descriptive statistics, regression analysis, advanced analytics, and machine learning – all within a single environment. Its seamless integration with open-source tools and intuitive help features ensures that analysts and researchers are well-equipped for any data-driven task. 

To begin, follow these steps: 

1. Open IBM SPSS Statistics and load your dataset. 

2. Navigate to Analyze > Descriptive Statistics > Explore. 

3. Select the variable(s) you want to analyse. 

4. Choose the statistics and plots you wish to generate. 

5. Click OK to run the procedure. 

The output includes several tools to help detect outliers: 

• Descriptives: Compare the mean and 5% trimmed mean to assess skewness and distribution. 

• Boxplots: Outliers and extreme values appear as individual points beyond the whiskers. 

• Stem-and-leaf plots: Help visualise data distribution and identify extreme cases. 

Data Validation can also be utilized to identify rule violations. For instance, if instances where “TV hours per day > 7” need to be highlighted and examined, SPSS can automatically flag these instances for review. This process does not remove cases from the dataset; rather, it facilitates easier identification of such cases.  

Best practices for managing outliers include visualising data, understanding the impact of outliers, and documenting your approach. Whether an outlier is valid or erroneous, recognising and addressing it is essential for accurate and trustworthy analysis.