Preliminary Data Analysis Techniques in SPSS

Before running any advanced statistical test, there is one step that researchers often underestimate: getting to know their data first. In educational research, jumping straight into regression or ANOVA without understanding what your dataset actually looks like is a recipe for misleading results. Preliminary data analysis (PDA) is the structured process of inspecting, exploring, and understanding the fundamental characteristics of your data before any in-depth analysis begins. SPSS (Statistical Package for the Social Sciences) provides a robust set of tools that make this process systematic and manageable – from checking data shape and spread to detecting outliers and exploring relationships between variables.

Why preliminary analysis is non-negotiable 🔗

Many researchers treat preliminary analysis as optional housekeeping. It is not. Preliminary analyses on any dataset include checking the reliability of measures, examining the distributions of individual variables, and identifying outliers – all before you begin testing hypotheses. Skipping this step and going straight to inferential statistics can produce results that are technically computed but fundamentally unreliable.

As the Springer text on SPSS for survey-based research highlights, preliminary data analysis is a critical stage that many existing guides overlook entirely. The decisions you make here – how to handle missing values, whether to transform skewed variables, which outliers to exclude – shape the integrity of everything that follows.

Understanding data shape and distribution 🔗

The first thing you want to know about your data is its shape. Is it normally distributed? Skewed to one side? Does it have multiple peaks? The shape of your data directly affects which statistical tests are appropriate.

SPSS provides several tools for this. Histograms are the most intuitive starting point – they give you an immediate visual of how your data is distributed and whether it approximates a bell curve. Frequency distributions work alongside histograms to quantify what the graph shows. For a more detailed examination, the Explore procedure (found under Analyze > Descriptive Statistics > Explore) goes deeper, producing stem-and-leaf plots, boxplots, and formal normality tests in a single run.

Two statistics are especially important for evaluating shape: skewness and kurtosis. Skewness measures the symmetrical nature of the data – a positive skew means data is concentrated on the left with a tail stretching right, while negative skew is the reverse. Kurtosis describes whether the distribution is peaked or flat relative to a normal distribution. In SPSS, you can access both through Analyze > Descriptive Statistics > Descriptives, then selecting skewness and kurtosis from the Options menu. Values for skewness and kurtosis between -2 and +2 are generally considered acceptable for assuming a normal univariate distribution.

Measures of location and spread 🔗

Once you know the shape of your data, you need to understand where it is centered and how widely it is dispersed. These are the measures of location (central tendency) and spread (variability).

Measures of location 🔗

SPSS computes the mean, median, and mode through the Descriptives and Frequencies procedures. In a perfectly normal distribution, all three are equal. When they diverge, it is a signal that your data may be skewed or that outliers are influencing the mean. The 5% trimmed mean, available through the Explore procedure, removes the top and bottom 5% of scores and recalculates the average. If the trimmed mean differs significantly from the regular mean, extreme scores are likely distorting your results.

Measures of spread 🔗

Standard deviation, variance, and range tell you how dispersed your data points are around the center. A large standard deviation in student test scores, for example, signals that performance varied widely across the sample. SPSS reports these automatically through the Descriptives procedure. The interquartile range (IQR) – the difference between the 75th and 25th percentiles – is especially useful because it is resistant to the influence of extreme values, making it a more stable measure of spread when outliers are present.

Detecting outliers in SPSS 🔗

Outliers are data points that fall far outside the normal range of a variable. They can have a significant impact on the results of statistical analyses and may result from measurement errors, data entry errors, or simply represent genuine but extreme observations. Identifying them early is essential.

SPSS offers several methods for outlier detection:

Boxplots are the most commonly used visual method. Generated through the Explore procedure, they display the median, IQR, and the range of non-outlying values. SPSS marks potential outliers with a circle and extreme outliers with an asterisk (*) on the boxplot, using Tukey’s method, which flags values that fall beyond 1.5 times the IQR from the quartiles.

Z-scores offer a statistical approach. A Z-score of 3 or more is considered an outlier, because it falls outside the range of approximately 99.7% of the data in a normal distribution. In SPSS, you can compute Z-scores using Analyze > Descriptive Statistics > Descriptives, then checking the “Save standardized values as variables” box.

Mahalanobis distance is used for detecting multivariate outliers – cases that appear normal on individual variables but are unusual when multiple variables are considered together. After running the analysis, multivariate outliers appear wherever the values of the probability variable are less than .001.

Once identified, you have several options: verify whether the outlier is a data entry error, retain it if it represents a legitimate observation, or handle it through winsorization, transformation, or exclusion – always documenting your decision.

Checking assumptions: normality tests in SPSS 🔗

Many inferential tests – t-tests, ANOVA, Pearson correlation – assume that data is approximately normally distributed. Preliminary analysis is where you verify this assumption before it is too late.

The Explore procedure in SPSS can assess the normality of a numeric variable with special inferential statistics and detailed diagnostic plots. The two most commonly used tests are:

Shapiro-Wilk test – generally recommended for smaller samples (n < 50). A non-significant result (p > .05) suggests the data does not significantly depart from normality. Kolmogorov-Smirnov test – more appropriate for larger samples. Both tests are accessible via Explore > Plots > Normality plots with tests.

Beyond formal tests, Q-Q plots (quantile-quantile plots) offer a visual check. The Q-Q and detrended Q-Q plots show systematic deviations from normality – if data points fall along the diagonal reference line, the distribution is approximately normal. Deviations indicate non-normality that may require data transformation or the use of non-parametric alternatives.

Handling missing data 🔗

Missing data is one of the most common problems in educational research datasets. How you handle it in the preliminary stage has downstream consequences. In SPSS, missing values can be handled through listwise deletion, pairwise deletion, or imputation methods – and the best approach depends on how much data is missing and whether the pattern of missingness is random or systematic.

Random missing data is generally less problematic than systematic gaps. If certain groups of participants consistently have missing values on specific variables, that pattern itself becomes a finding worth reporting. SPSS flags missing values in the output of the Frequencies and Descriptives procedures, giving you an immediate count of how many cases have incomplete data per variable.

Exploring clustering patterns 🔗

In some educational studies, you may want to know whether your participants naturally fall into distinct groups – based on performance, learning styles, or demographic characteristics – before conducting group comparison tests. SPSS supports cluster analysis for this purpose.

Two primary algorithms are available: K-means clustering, which requires you to specify the number of clusters in advance, and hierarchical clustering, which builds a dendrogram that shows how cases merge into clusters at different levels of similarity. Running a cluster analysis as part of preliminary exploration can reveal subgroup structures that should inform how you set up your main analysis – for example, whether it makes sense to analyze groups separately or whether a covariate needs to be included.

Exploring relationships between variables 🔗

The final piece of preliminary analysis is to look at how variables relate to each other before running formal tests. This gives you an early sense of which relationships are worth investigating and helps avoid analytical dead ends.

Scatterplots 🔗

For two continuous variables, a scatterplot is the most direct way to visualize potential relationships. In SPSS, go to Graphs > Chart Builder > Scatter/Dot. A scatterplot quickly shows whether the relationship appears linear or curvilinear, and whether there are any outlier pairs that could distort a correlation coefficient.

Correlation analysis 🔗

Pearson’s r (for continuous, normally distributed variables) and Spearman’s rho (for ordinal or non-normal data) both quantify the strength and direction of a relationship. At the preliminary stage, correlation analysis can suggest potential relationships worth investigating further, though establishing true cause-and-effect requires more rigorous methods like regression or experimental design. In SPSS, run correlations through Analyze > Correlate > Bivariate.

Cross-tabulations for categorical variables 🔗

When both variables are categorical – for example, gender and pass/fail outcome – a cross-tabulation (crosstab) table shows the joint frequency distribution. This is accessed through Analyze > Descriptive Statistics > Crosstabs, and is often combined with a chi-square statistic to test whether the association is statistically meaningful.

A practical preliminary analysis sequence in SPSS 🔗

Bringing all these techniques together, a recommended workflow for preliminary analysis in SPSS follows a logical progression. Start with the Descriptives procedure to get quick summaries of multiple variables, then use the Explore procedure for in-depth examination of individual variables including normality tests and plots, create frequency tables for categorical variables, generate scatterplots to visualize relationships between continuous variables, and run correlation analyses to quantify associations. Throughout this process, document every decision – which outliers were excluded, how missing data was handled, and whether any transformations were applied. This documentation is essential for research transparency and replication.

The goal of preliminary analysis is not to answer your research questions – it is to make sure you are asking them the right way, with data you can trust. Descriptive statistics tell the story of what happened in a study, and while inferential statistics are important, it is essential to understand the descriptive statistics first. Rushing past this phase to reach inferential tests is one of the most common – and most avoidable – mistakes in research.

What do you think? When you review research studies in your field, how often do authors clearly describe the preliminary analysis steps they took before running their main tests? And if a dataset shows significant skewness in a key variable, what do you think is the most defensible next step – transformation, a non-parametric test, or something else?

References

1. https://opentext.wsu.edu/carriecuttler/chapter/12-4-conducting-your-analyses/
2. https://link.springer.com/book/10.1007/978-981-16-0193-4
3. https://resourcefulscholarshub.com/descriptive-statistics-in-spss/
4. https://www.researchgate.net/publication/334067633_SPSS_TRAINING_Data_Screening_Identification_of_Potential_Outliers
5. https://www.statisticssolutions.com/identifying-outliers-in-spss/
6. https://www.statisticshowto.com/outliers-spss/
7. https://www.statisticssolutions.com/identifying-multivariate-outliers-in-spss/
8. https://libguides.library.kent.edu/SPSS/Explore
9. https://distancelearning.institute/research/exploring-dataset-preliminary-analysis-spss/