SPSS Data Exploration: 6 Key Characteristics to Examine Before Analysis

Before you run a single t-test or correlation in SPSS, there’s a step most students are tempted to skip: actually looking at your data. A preliminary analysis is where you get to know your dataset before asking it to answer research questions. Skip it, and you risk building an entire analysis on a foundation of typos, skewed distributions, or a handful of extreme scores quietly dragging your results in the wrong direction. This is why IBM’s own SPSS documentation lists data screening, outlier identification, and assumption checking among the core reasons researchers use exploratory tools before moving to inferential statistics.

Why preliminary analysis deserves your time 🔗

Every statistical test you’ll ever run in SPSS, from a simple ANOVA to a regression model, rests on assumptions about your data. Many tests assume your variable is roughly normally distributed, that variances across groups are similar, or that missing data won’t distort your sample size. You can’t verify any of this by staring at raw numbers in the Data View. You need to summarise, visualise, and screen the data first.

Think of preliminary analysis as a diagnostic check. It tells you whether your data behaves the way your chosen statistical test expects it to. If it doesn’t, you can decide whether to transform a variable, choose a non-parametric alternative, or investigate why a few responses look so different from the rest. Researchers who skip this stage often end up explaining away odd results after the fact, when catching the issue early would have saved the trouble. As one clinical research methods resource puts it, exploratory techniques help determine whether outliers exist, data are missing, and statistical assumptions will hold, well before conclusions are drawn.

Six characteristics of your dataset worth examining 🔗

Preliminary analysis usually revolves around a handful of dataset properties. Once you get comfortable checking these, they become second nature every time you open a new dataset in SPSS.

1. Shape: is your data normal or skewed? 🔗

The shape of a distribution tells you whether your scores cluster symmetrically around the middle or lean toward one end. A distribution can be normal (bell-shaped), positively skewed (a long tail toward higher values), negatively skewed (a long tail toward lower values), or even bimodal, with two distinct peaks. This matters because many statistical tests assume normality, and violating that assumption can distort your p-values and confidence intervals.

Skewness and kurtosis are the two statistics used to quantify shape. Skewness measures asymmetry, while kurtosis describes how peaked or flat the distribution is compared to a normal curve. SPSS calculates both automatically when you request descriptive statistics, giving you a quick numerical check alongside your histogram.

2. Location: where does the data centre? 🔗

Location refers to the typical or central value in your dataset, usually captured through the mean, median, or mode. Choosing the right measure matters. The mean works well for symmetric data, but the median is a more honest representation when your distribution is skewed, since it isn’t pulled around by extreme values the way the mean is.

For example, if you’re analysing exam scores for a class where most students scored between 60 and 80 but two students scored 15, the mean will drop noticeably while the median stays closer to where most students actually performed.

3. Spread: how much do the scores vary? 🔗

Spread, or dispersion, tells you how tightly or loosely your data points cluster around the centre. Range, variance, standard deviation, and interquartile range (IQR) are the common measures. A Sheffield Hallam University SPSS guide notes that standard deviation accounts for every data point but can be swayed by extreme values, while the IQR, based only on the middle 50 percent of scores, is more resistant to distortion from outliers.

Understanding spread helps you compare groups meaningfully. Two classes could have identical average scores but very different spreads, one tightly bunched, the other widely scattered, and that difference has real implications for how you interpret group performance.

4. Outliers: which values don’t belong? 🔗

Outliers are data points that sit far outside the expected range. They might reflect genuine variation, data entry errors, or unusual cases worth investigating separately. Left unchecked, a single extreme outlier can distort your mean, inflate your variance, and mislead an entire analysis. Research on outlier detection methods highlights how outliers can even be diagnostically useful, helping identify unusually strong or unusually weak performers within a dataset rather than simply being noise to discard.

SPSS’s boxplot output flags outliers automatically, marking them as individual points beyond the plot’s whiskers. This gives you a quick visual cue for cases that need a closer look before you decide whether to keep, transform, or exclude them.

5. Clustering: are there hidden subgroups? 🔗

Clustering refers to natural groupings within your data that aren’t always obvious from summary statistics alone. Students from different schools, teaching methods, or geographic regions often produce data that clusters by group rather than behaving as one uniform sample. This matters because standard statistical tests generally assume independent observations, and ignoring clustering can produce misleadingly precise results.

A study examining correlation analysis in international teacher education data found that when data has a hierarchical structure, such as teachers nested within training programs nested within institutions, failing to account for that clustering can produce misleading correlation estimates. Recognising this kind of structure early lets you choose analysis methods, such as multilevel modelling, that respect how your data was actually collected.

6. Associations: how do variables relate to each other? 🔗

Before running a formal correlation or regression, it helps to visually inspect how two variables move together. Scatterplots reveal whether the relationship is linear, curved, or absent altogether, and whether it’s positive (both variables increase together) or negative (one rises as the other falls). According to a statistics resource on interpreting scatterplots, stronger relationships between variables produce a tighter clustering of points around a trend line, while weaker relationships appear more scattered.

This preliminary check matters because correlation coefficients like Pearson’s r assume a linear relationship. If your scatterplot shows a curved pattern instead, a standard correlation analysis will understate the true strength of the relationship.

Using SPSS to run preliminary analysis 🔗

SPSS offers dedicated tools for exactly this stage of research, so you rarely need to calculate any of this by hand.

Descriptive statistics and the Explore procedure 🔗

The Descriptives procedure, found under Analyze > Descriptive Statistics, is useful when you want a quick side-by-side comparison of several variables’ means, standard deviations, and ranges. But for a deeper preliminary check, the Explore procedure is the better tool. As a university SPSS tutorial explains, Explore produces boxplots and stem-and-leaf plots by default, along with options for percentiles, outlier listings, and formal normality tests such as Shapiro-Wilk and Kolmogorov-Smirnov.

Running Explore on a new dataset typically gives you, in one output, a sense of shape, location, spread, and outliers together, which is precisely the combination you need before deciding on your main analysis.

Visual tools: histograms, boxplots, and scatterplots 🔗

Numbers alone rarely tell the full story, which is why visualisation plays such a central role in preliminary analysis. Histograms reveal shape at a glance, letting you spot skewness or multiple peaks that summary statistics might understate. Boxplots summarise the five-number distribution and flag outliers directly on the chart, based on the interquartile range. Scatterplots, meanwhile, are your go-to tool for spotting associations and clustering between two continuous variables.

Used together, these three visual tools cover most of what a preliminary analysis needs to establish before you move on to hypothesis testing. A quick habit worth building: run Explore, check the histogram and boxplot for every key variable, and glance at a scatterplot matrix if you plan to examine relationships between variables. That small investment of time before analysis often saves hours of confusion afterward.

What do you think? Have you ever run a full statistical test only to discover afterward that an outlier or a skewed distribution changed your results? What would you check first the next time you open a new dataset in SPSS?

References

1. https://www.ibm.com/docs/en/SSLVMB_26.0.0/pdf/en/IBM_SPSS_Statistics_Base.pdf
2. https://www.ncbi.nlm.nih.gov/books/NBK557570/
3. https://students.shu.ac.uk/lits/it/documents/pdf/analysing_data_using_spss.pdf
4. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4711812/
5. https://largescaleassessmentsineducation.springeropen.com/articles/10.1186/2196-0739-1-7
6. https://statisticsbyjim.com/graphs/scatterplots/
7. https://libguides.library.kent.edu/SPSS/Explore