Data Screening, Cleaning and How to Replace Missing Values in SPSS

Before any model testing—such as checking whether one construct influences another—data must be screened for basic quality problems: entry errors, outliers, and signs that respondents didn’t answer thoughtfully. The workflow starts with setting up an ID column so each case can be traced back after sorting or filtering. From there, the first practical check is respondent abandonment: if someone stops answering near the end of a questionnaire, the record may be deleted only when the missing portion is large (for example, 40–50% of items), while small gaps (like missing one or two final questions) can be handled by keeping the rest of the responses.

Next comes respondent misconduct, especially in Likert-style items where answers should vary. A common red flag is marking the same option across nearly all questions on a 1–7 scale, which suggests the respondent may not be reading items. Attention checks—such as reverse-coded questions or items that require selecting a specific number—can help detect this behavior. Another quantitative approach is to compute the standard deviation of each respondent’s Likert responses and flag cases with extremely low variance. A rule of thumb used here is that standard deviation below 0.25 indicates too little variation and should trigger deletion, though the threshold should ultimately be judged against survey size and what level of agreement/disagreement is plausible.

After handling misconduct and abandonment, the screening process shifts to impermissible values and missing data. Impermissible values are those outside an expected range, often caused by keying errors. In SPSS, this is checked through descriptive statistics by requesting minimum and maximum values for selected variables; if the observed extremes match expected bounds, the dataset passes this particular test.

Missing data is assessed using SPSS frequencies and descriptive statistics to confirm whether blanks or system-missing values exist. When missingness is present, the transcript contrasts two broad strategies: deletion versus imputation. Listwise or pairwise deletion is discouraged because it discards substantial data—dropping an entire survey response if even one item is missing. Instead, imputation is presented as preferable when missingness is not excessive; prior research cited suggests that imputation can recover parameter estimates even when 20–30% of values are missing.

Two imputation methods are emphasized. The most common is series mean imputation, which replaces each missing value with the mean of that indicator. It’s easy to implement but can reduce variance and ignore individual differences. The second method is linear interpolation, which estimates missing values by using the last valid value before the gap and the next value after it, assuming a linear progression. In SPSS, both methods are implemented via Transform → Replace Missing Values, selecting the variables with missing entries and choosing either the default series mean method or linear interpolation (using the Change option to switch methods). The overall message is that careful screening and thoughtful imputation choices protect the validity of later analyses.