SPSS Data Analysis Guide for Research Students

Introduction

For most Master’s and PhD scholars, the data analysis chapter is where confidence collapses. You collect the data, open SPSS for the first time, and stare at a grid of cells that looks suspiciously like Excel but behaves nothing like it. Where do you click? What is the difference between a variable and a value? Why did your t-test come back with a “cannot compute” error? A surprising number of dissertations stall at this exact point — not because the analysis is conceptually hard, but because no one ever taught the scholar a systematic SPSS data analysis workflow.

SPSS (Statistical Package for the Social Sciences), now sold by IBM as IBM SPSS Statistics, has been the workhorse of social-science research since 1968. It is point-and-click friendly, produces publication-ready tables, and handles everything from a 30-row descriptive summary to a 10,000-row multilevel regression. The catch is that the menu system lets you run tests whose assumptions you have not checked — which is how scholars end up reporting an ANOVA on data that violates normality, or a Pearson correlation on ordinal Likert items.

This guide walks you through a complete SPSS workflow, from importing data to interpreting output, with the assumption checks that examiners actually look for. If you would rather hand the analysis to a specialist, our data analysis service pairs you with a PhD-qualified statistician who runs the tests, writes the results chapter, and explains every table. For more on the broader context, see our data analysis chapter guide and our companion piece on interpreting SPSS output.

The SPSS Workspace: Data View, Variable View, Output

SPSS opens with two alternating spreadsheets. Data View shows one row per participant and one column per variable — this is the screen you actually look at when entering or inspecting data. Variable View, accessed by clicking the tab at the bottom, is where each variable’s metadata is defined. Most analysis errors originate in Variable View, so learn it early.

In Variable View, set the following for every column:

• Name: short, no spaces, no special characters (e.g., age_years, not Age in Years).
• Type: Numeric for numbers, String for text. Most analysis requires Numeric.
• Width and Decimals: set sensibly; do not let SPSS display 8 decimal places for age.
• Values: define value labels for categorical variables (1 = Male, 2 = Female). Without this, output shows raw numbers.
• Measure: declare the level of measurement — Nominal, Ordinal, or Scale. SPSS uses this to enable or disable certain procedures.
• Missing: declare any missing-data codes (e.g., 99 = did not respond) so SPSS excludes them rather than treating 99 as a real score.

The third window, Output Viewer, appears the moment you run your first procedure. It collects tables, charts, and notes in a navigable tree. Save the .spv file alongside your .sav data file so you can reopen both later.

Levels of Measurement: The Foundation

Every analysis in SPSS assumes you know the level of measurement of each variable. Get this wrong and the rest of the workflow collapses. The four levels, from weakest to strongest, are:

1. Nominal: categories with no order — gender, religion, marital status. Use frequencies and chi-square.
2. Ordinal: ordered categories with unequal spacing — Likert scales (Strongly Disagree to Strongly Agree), education level. Use medians, Mann-Whitney, Spearman.
3. Interval: ordered, equal spacing, no true zero — temperature in Celsius, IQ scores. Mean, t-test, Pearson are appropriate.
4. Ratio: interval with a true zero — age, height, reaction time, income. All statistics are permissible.

A common debate is whether a 5-point Likert item is ordinal or interval. Strictly, it is ordinal; pragmatically, many researchers treat summed multi-item scales as interval if the scale has at least 5 points and approximately normal distribution. Document your choice and justify it in your methodology chapter.

Step 1: Data Screening and Cleaning

Before any analysis, screen your data. A clean dataset is the difference between a defensible thesis and one sent back for major revisions. The screening checklist:

• Run Analyze → Descriptive Statistics → Frequencies on every variable. Look for out-of-range values (age = 999, gender = 3) and impossible combinations.
• Use Analyze → Descriptive Statistics → Explore to inspect boxplots for outliers. Decide whether to keep, transform, or remove outliers — and document the decision.
• Check missingness with Analyze → Missing Value Analysis. If more than 5% of cases are missing on a key variable, consider multiple imputation (Multiple Imputation menu under Analyze).
• Verify internal consistency of multi-item scales with Analyze → Scale → Reliability Analysis. A Cronbach’s alpha of 0.70 or above is the conventional threshold; below 0.60 means your scale needs revision.

Step 2: Descriptive Statistics

Descriptive statistics summarise your sample and are mandatory in every results chapter. For each continuous variable, report the mean, standard deviation, minimum, maximum, and sample size. For categorical variables, report frequencies and percentages.

In SPSS, run Analyze → Descriptive Statistics → Descriptives for the mean/SD block, and Frequencies for the count/percentage block. Always report the n on which each statistic is based — examiners check whether the n drops across analyses, which signals missing data.

Build at least one summary table in APA format: variable, M, SD, n, possible range, observed range. SPSS produces the raw numbers; format them in Word using the APA Table guidelines from the Publication Manual (7th edition). For a worked example, see our SPSS output interpretation guide.

Step 3: Test Assumptions Before Choosing a Test

The biggest single mistake in SPSS data analysis is choosing a statistical test without checking its assumptions. Every parametric test in SPSS rests on three assumptions: normality, homogeneity of variance, and independence of observations. Violate them silently and your p-values are untrustworthy.

For normality, run Explore with Plots → Normality plots with tests. SPSS reports both Kolmogorov-Smirnov (with Lilliefors correction) and Shapiro-Wilk. Use Shapiro-Wilk for samples below 50; K-S for larger samples. A non-significant result (p > 0.05) indicates normality. Also inspect the Q-Q plot visually — if points deviate substantially from the diagonal line, treat the variable as non-normal.

For homogeneity of variance, use Levene’s test (automatically produced by the independent-samples t-test and ANOVA procedures). A significant Levene’s (p < 0.05) means variances are unequal — read the “equal variances not assumed” row in t-test output, or use Welch’s ANOVA instead of standard one-way ANOVA.

If assumptions are violated, switch to the non-parametric equivalent:

• Independent-samples t-test → Mann-Whitney U test
• Paired-samples t-test → Wilcoxon signed-rank test
• One-way ANOVA → Kruskal-Wallis test
• Pearson correlation → Spearman’s rho

Step 4: Running the Core Tests

With assumptions checked, the analysis itself is fast. Below are the five most common procedures in graduate research and the menu path to each:

1. Independent-samples t-test: Analyze → Compare Means → Independent-Samples T Test. Compares two groups on a continuous outcome (e.g., male vs female exam scores).
2. Paired-samples t-test: Analyze → Compare Means → Paired-Samples T Test. Compares the same participants at two time points (pre-test vs post-test).
3. One-way ANOVA: Analyze → Compare Means → One-Way ANOVA. Compares three or more groups. Always run Post Hoc (Tukey or Bonferroni) to identify which pairs differ.
4. Chi-square test of independence: Analyze → Descriptive Statistics → Crosstabs → Statistics → Chi-square. Tests association between two categorical variables.
5. Linear regression: Analyze → Regression → Linear. Predicts a continuous outcome from one or more predictors. Check R², the F-test, and each predictor’s beta and p-value.

Report effect sizes alongside p-values. Cohen’s d for t-tests (0.2 = small, 0.5 = medium, 0.8 = large), eta-squared or partial eta-squared for ANOVA (0.01, 0.06, 0.14), and R² for regression. Many journals now require effect sizes; reporting only p-values is no longer acceptable. For a fuller catalogue of which test to run when, see our statistical tests overview.

SPSS output is a starting point, not a destination. The numbers it produces are only as trustworthy as the assumptions you checked before clicking OK — and only as meaningful as the effect sizes you report alongside the p-values.

Step 5: Interpreting and Reporting Output

SPSS output is dense; learn to read the four pieces that matter for every test:

• The test statistic (t, F, χ², r) and its degrees of freedom.
• The p-value (labeled “Sig.” in SPSS). A value below 0.05 is conventionally significant, but report the exact value (p = 0.032, not p < 0.05).
• The effect size, which tells you whether a statistically significant finding is practically meaningful.
• The confidence interval, which gives the plausible range of the true effect in the population.

Write results in APA style: “An independent-samples t-test showed that students in the flipped-classroom condition (M = 78.4, SD = 6.2) scored significantly higher than those in the traditional condition (M = 72.1, SD = 7.5), t(118) = 4.82, p < .001, d = 0.88.” Notice how the sentence carries the mean, SD, t, df, p, and effect size in one line — this is what reviewers expect.

Common SPSS Mistakes to Avoid

Over hundreds of student manuscripts, the same errors recur. Audit your own work against this list before submission:

• Treating ordinal Likert items as interval without justification. Either justify the treatment with a normality check and a citation (e.g., Norman 2010), or use non-parametric tests.
• Running parametric tests without checking normality. A Shapiro-Wilk test takes 10 seconds and saves your chapter from a reviewer’s demand for re-analysis.
• Reporting only p-values. Always include effect sizes and confidence intervals.
• Ignoring missing data patterns. Listwise deletion can shrink your sample by 30% if every variable has 5% missing. Investigate and impute if needed.
• Failing to weight or cluster. If your sample used stratified or cluster sampling, you must weight the data; otherwise standard errors are wrong.
• Not saving syntax. Without syntax, you cannot reproduce the analysis — and reproducibility is now a publication requirement at many journals.

Conclusion

SPSS data analysis is a workflow, not a single skill. Define your variables correctly in Variable View, screen and clean the data, run descriptives, check assumptions, choose the right test, and report results in APA format with effect sizes and confidence intervals. Each step has a menu path and a small set of conventions; once internalised, the software fades into the background and your attention shifts to the actual findings. If you would like expert eyes on your dataset, our data analysis service offers end-to-end support from cleaning through to the written results chapter. Pair it with our guides on R for statistical analysis, survey design, and sampling methods for a complete data-skills toolkit, and contact us on our contact page for a free consultation.

Frequently Asked Questions