Performing Univariate Analysis in SPSS

When you’re staring at a fresh dataset in SPSS, the urge to jump straight into regressions, t-tests, and fancy inferential statistics is real. But before you chase those complex outputs, there’s a humbler step you absolutely cannot skip: looking at each variable on its own. This is what researchers call univariate analysis, and it’s the foundation on which every meaningful statistical conclusion rests. Let’s walk through how to perform it in SPSS, what each statistic actually tells you, and how to interpret the output like a trained researcher.

What univariate analysis really means 🔗

Univariate analysis is exactly what it sounds like: the analysis of one variable at a time. There is no cause-and-effect relationship being tested here, because only a single variable is involved. The goal is simple but powerful – describe the data. You want to know what the typical value looks like, how spread out the observations are, and whether the distribution has any unusual shape.

Before running any inferential test, researchers use univariate analysis to check each variable for accuracy, identify typos or errors, count how many cases exist, and examine the values the variable contains. Skipping this step is one of the fastest ways to end up with a result that looks impressive but is quietly built on flawed data.

Why it matters for your research 🔗

Think of univariate analysis as the diagnostic X-ray of your dataset. Examining frequency distributions can reveal unexpected categories, miscoded values, or implausible outliers that need to be addressed before the analysis moves forward. If someone’s age is recorded as 998 because that was the code for a missing value, your mean age could be wildly off. Catching such issues early saves you from reporting nonsense later.

Knowing your variable’s level of measurement 🔗

Before clicking a single menu in SPSS, identify what type of variable you’re dealing with. This decision determines which statistics are meaningful and which are garbage. The four standard levels are nominal (categories with no order, like gender or religion), ordinal (ordered categories, like education level), interval, and ratio (both of which are continuous/scale variables, like age or income).

Here’s the catch: SPSS will calculate statistics even if the measure of central tendency and dispersion are not appropriate for the variable type. If you ask SPSS to calculate the mean of a nominal variable coded as 1 for male and 2 for female, it will dutifully return 1.4 – a number that means absolutely nothing. The software doesn’t know better. You have to.

The command path in SPSS 🔗

The primary route for univariate analysis in SPSS is through the Frequencies command. Navigate to Analyze → Descriptive Statistics → Frequencies. Once the dialogue box opens, you’ll see your list of variables on the left and an empty “Variable(s)” box on the right. Move the variables you want to analyse into this box using the blue arrow.

For categorical variables (nominal or ordinal), keep the Display frequency tables checkbox ticked. For continuous variables, you may want to uncheck it to avoid producing a giant table with hundreds of rows – one for every unique value. Next, click the Statistics button to choose which measures you want SPSS to compute.

Choosing the right statistics for the job 🔗

A concise workflow, adapted from standard guidance, looks like this: select Mode for nominal variables; Mode and Median for ordinal variables; and Mean, Median, Mode, Standard Deviation, Range, Skewness, and Kurtosis for continuous variables. You can also request percentiles and quartiles if needed. Then click Continue, and back in the main dialogue, click Charts to add a bar chart or pie chart for categorical data, or a histogram for continuous data. Click OK to run it.

Understanding the frequency table 🔗

Once SPSS produces the output, the first thing you’ll see for categorical variables is the frequency table. It contains four important columns: Frequency (the raw count of respondents in each category), Percent (the percentage of the total sample), Valid Percent (percentage after excluding missing cases), and Cumulative Percent (a running total).

The Valid Percent column is usually the one you report, because it accounts for respondents who skipped the question. A large gap between Percent and Valid Percent tells you that missing data is substantial – itself a finding worth flagging in your write-up.

Measures of central tendency 🔗

Central tendency statistics give you a single value that represents the “middle” of your data. The three classic measures are the mean, median, and mode – each appropriate in different situations.

Mean 🔗

The mean is the arithmetic average – the sum of all values divided by the number of observations. It includes every value in the dataset as part of its calculation, and it is the only measure where the sum of deviations of each value from the mean is always zero. That’s mathematically elegant, but it comes with a cost: the mean is highly sensitive to outliers. A single billionaire in a neighbourhood income dataset will drag the mean upward and paint a misleading picture of the “typical” resident.

Median 🔗

The median is the middle value when data is arranged in order. It’s the go-to measure when your distribution is skewed or contains outliers. The median is appropriate to use with ordinal variables, and with interval variables that have a skewed distribution. Household income, wealth, housing prices – these are all classic examples where the median tells a truer story than the mean.

Mode 🔗

The mode is simply the most frequently occurring value. It’s the only measure of central tendency you can meaningfully use with nominal data. If you ask respondents about their preferred mode of transport and “metro” shows up most often, that’s your mode. Note that a dataset can be bimodal (two modes) or even multimodal, which itself reveals something about the underlying population.

Measures of dispersion 🔗

Central tendency alone is never enough. Two datasets can have the same mean but look completely different – one tightly clustered, the other wildly spread. That’s why you also need measures of dispersion to describe how far individual values fall from the centre.

Range 🔗

The range is the simplest measure: the difference between the maximum and minimum values. It’s easy to calculate but highly sensitive to extreme values. Still, it gives you a quick sense of how wide the spread is.

Variance and standard deviation 🔗

Variance quantifies the average squared deviation from the mean. Because it’s in squared units, interpreting it directly is awkward – if your variable is measured in rupees, variance is in “rupees squared.” That’s why we usually report the standard deviation, which is simply the square root of variance and is expressed in the same units as the original variable. A small standard deviation means observations cluster tightly around the mean; a large one means they’re widely scattered.

When the distribution is roughly normal, the mean and standard deviation are the preferred summary. If the distribution is skewed or has notable outliers, report the median and interquartile range (IQR) instead, and briefly mention the skew or outliers.

Shape of the distribution: skewness and kurtosis 🔗

Beyond central tendency and dispersion, SPSS lets you examine the shape of a distribution through two statistics: skewness and kurtosis. These are often overlooked by beginners but are vital when you plan to run parametric tests that assume normality.

Skewness 🔗

Skewness measures the symmetry of a distribution. A value of zero indicates perfect symmetry. A positive value means the distribution has a long tail on the right (right-skewed – think of income data), while a negative value means a long tail on the left (left-skewed – think of exam scores when most students do well). As a rule of thumb, skewness values between -2 and +2 are generally considered acceptable for assuming a normal univariate distribution.

Kurtosis 🔗

Kurtosis describes the “tailedness” of a distribution – whether the data produces more or fewer extreme outliers compared to a normal distribution. If a distribution has kurtosis greater than zero, it tends to produce more outliers than the normal distribution. A negative kurtosis indicates lighter tails and a flatter shape. SPSS reports excess kurtosis, meaning zero represents a perfectly normal distribution.

When both skewness and kurtosis fall within acceptable ranges, you can generally proceed with parametric tests. When they don’t, consider data transformations (like taking a log) or switching to non-parametric alternatives.

Interpreting the output – an example 🔗

Suppose you’re analysing the age variable in a public policy survey of 500 respondents. SPSS returns: mean = 38.4, median = 37, mode = 35, standard deviation = 11.2, range = 62, skewness = 0.42, kurtosis = -0.18. What does this tell you?

The mean and median are close, suggesting a reasonably symmetric distribution. Skewness at 0.42 confirms a mild rightward lean – a small group of older respondents pulls the mean slightly above the median. Kurtosis near zero suggests tails close to normal. The standard deviation of 11.2 indicates respondents typically fall within about 11 years above or below the average age. Based on this, age is likely suitable for parametric analyses without transformation.

Common pitfalls to avoid 🔗

Beginners often fall into a few traps. First, calculating a mean for a categorical variable just because SPSS allowed it. Second, reporting raw frequencies without percentages, which makes comparison across studies difficult. Third, forgetting to define missing value codes – if “99” represents “refused to answer” and you don’t tell SPSS, it treats 99 as a real age. Finally, treating univariate output as the end of the analysis rather than the beginning. These statistics set up everything that comes next, including your choice of inferential tests.

What do you think? When analysing a dataset from a citizen satisfaction survey with hundreds of variables, how would you prioritise which variables to examine first through univariate analysis? And when skewness and kurtosis suggest non-normality, would you prefer transforming the data or switching to non-parametric tests – and why?

References

1. https://manifold.open.umn.edu/read/chapter-5-data-management-and-cleaning-in-spss
2. https://sociology.institute/research-methodologies-methods/univariate-analysis-spss-guide/
3. https://subjectguides.sunyempire.edu/c.php?g=659059&p=4626896
4. https://statistics.laerd.com/statistical-guides/measures-central-tendency-mean-mode-median.php
5. https://www.betterevaluation.org/methods-approaches/methods/measures-central-tendency
6. https://spssservices.com/univariate-analysis-spss-guide/
7. https://imaging.mrc-cbu.cam.ac.uk/statswiki/FAQ/Simon
8. https://www.statology.org/skewness-kurtosis-in-spss/