Statistics for Research - L11 - What are Outliers and How to Solve the Issue using SPSS?

Outliers—unusually high or low data points—can quietly distort research results by pulling summary statistics like the mean away from what most participants actually report. A common example given is a survey about daily coffee intake: if most respondents fall between 0 and 4 cups, but one person reports 17 cups, that extreme value is far from the typical pattern and can inflate the sample mean even though it doesn’t represent the average coffee drinker. The practical takeaway is that outliers aren’t just a statistical curiosity; they can change conclusions if they’re handled incorrectly.

The session lays out several ways to identify outliers in a dataset, then connects those findings to decision rules and remedies. On the identification side, it recommends starting with descriptive checks such as minimum and maximum values, then using visual tools like histograms and box plots to spot points that sit far outside the expected range. For a more standardized approach, the transcript describes converting responses into Z scores. Z scores standardize each value relative to the sample (mean 0, standard deviation 1), making it easier to compare across cases. A Z score beyond about ±3.3 is treated as a strong indicator of an outlier.

Once outliers are found, the next step is deciding what to do with them—an issue framed as context-dependent rather than automatic. If an extreme value is clearly spurious (for example, inconsistent with the questionnaire context or without a plausible justification), removing it can be appropriate. If the value seems legitimate but is only slightly beyond the Z-score threshold, keeping it may be safer. Several handling options are presented: ignore outliers when they don’t materially affect results; trim them by removing extreme cases (especially when they stem from data entry or measurement errors); transform the data (such as applying a log transformation) to reduce outlier impact; or replace outliers using percentile-based rules—specifically substituting with the 5th or 95th percentile value.

The transcript then demonstrates how to carry out these steps in SPSS using a variable labeled “Vision one.” The first method uses Analyze → Descriptive Statistics → Frequencies, where minimum and maximum values reveal extremes (for instance, a maximum of 16 with only a couple of observations at the high end). Those specific cases can be located via Edit → Find, then traced back to the original dataset or questionnaire to verify whether the value is a genuine response or an entry mistake. A second method uses Analyze → Descriptive Statistics → Explore with outlier statistics and a box plot, where “extreme outliers” appear as marked points (stars) and can be navigated directly to the corresponding cases. Finally, the session shows how to compute standardized values in SPSS (saving standardized values) and then sort them to quickly spot cases with Z scores well above ±3.3, after which the analyst can choose to delete or replace them using the percentile approach or other replacement strategies like interpolation. The overarching message: outlier handling must be deliberate, because removing or ignoring them can shift results and lead to incorrect conclusions.