Logo site TopDissertations
+
Order Now
Request Callback
Call Back
Logo site TopDissertations

What This Guide Covers (And What It Doesn’t)

If you’re here, you probably already know how to run a t-test or basic ANOVA in SPSS. You’ve covered the basics and now it’s time to level up.

  • This guide covers advanced SPSS methods beyond basic t-tests and ANOVA — exactly what master’s and PhD students need for dissertation analysis.
  • Every method includes the SPSS menu path, assumption-checking workflow, interpretation tips, and an APA-ready reporting template.
  • The comparison table at the end is your quick reference for choosing the right test and knowing what to report.
  • Effect sizes (Cohen’s d, η², odds ratios) are mandatory for dissertation approval — don’t skip them.

This article covers the advanced methods that show up in real dissertations:

What’s covered in this article Why it matters for your dissertation
Multiple linear regression Predict outcomes with 2+ predictors — the single most-requested advanced method
Logistic regression (binary, multinomial, ordinal) Handle categorical outcomes (pass/fail, brand preference, treatment choice)
Factor analysis / PCA Validate questionnaires and reduce dozens of survey items into meaningful constructs
ANOVA with post-hoc tests Know exactly which groups differ, not just that "something differs"
Non-parametric tests When your data violates assumptions — Mann-Whitney, Kruskal-Wallis, Spearman
ANCOVA Control for baseline differences between groups
MANOVA Test multiple outcomes simultaneously instead of running separate ANOVAs
Mixed-effects models Analyze longitudinal and nested data structures

Let’s break each one down with menu paths, assumption checks, interpretation tips, and APA reporting templates.

1. Multiple Linear Regression

When to use: You have one continuous outcome (e.g., GPA, satisfaction score) and multiple predictor variables (e.g., study hours, sleep, stress level) and want to know how each predictor independently affects the outcome.

SPSS menu path: Analyze > Regression > Linear

Assumption-checking workflow:

  1. Linearity: Scatterplot of each predictor vs. the outcome. If the plot curves, try transforming the predictor.
  2. Normality: Histogram of standardized residuals or Q-Q plot. Shapiro-Wilk test if n < 50.
  3. Homoscedasticity: Scatterplot of standardized residuals vs. predicted values. Look for a funnel shape (bad) or a random cloud (good).
  4. Multicollinearity: Check the "Collinearity Diagnostics" output. If VIF > 10 for any predictor, you have a problem. If tolerance < 0.1, drop one predictor.
  5. Outliers: Cook’s Distance > 1 suggests influential outliers. Check the box.

Interpreting the output:

  • Look at the "Model Summary" table for R² (how much variance your predictors explain).
  • Check the "ANOVA" table for the overall model significance (the F value and p value).
  • The "Coefficients" table shows each predictor’s β, SE, t, and p value.
  • Don’t forget: Report effect sizes. For multiple regression, report the overall F test and each predictor’s β with its 95% CI.

APA reporting template:

A multiple linear regression was conducted to predict [outcome variable] from [predictor variables]. The model was significant, F(5, 113) = 4.23, p < .05, R² = .15. All predictors were significantly associated with the outcome: [predictor 1], β = .34, p < .001; [predictor 2], β = .21, p = .012; [predictor 3], β = −.18, p = .028.

Common mistake: Running a regression without checking multicollinearity first. If two predictors are highly correlated (r > .7), your model results will be unreliable. Always check VIF before trusting coefficients.

2. Logistic Regression (Binary, Multinomial, Ordinal)

When to use: Your outcome variable is categorical, not continuous. The type of logistic regression depends on the number of categories and whether they have a natural order.

Binary Logistic Regression

Use when your outcome has exactly 2 categories (e.g., pass/fail, yes/no, disease/no disease).

SPSS menu path: Analyze > Regression > Binary Logistic

Key steps:

  1. Put your binary outcome in the "Dependent" box.
  2. Put your predictors in the "Covariates" box.
  3. Click "Categorical" if you have any categorical predictors — SPSS will automatically set them up with reference categories.
  4. Click "Statistics" and check: "Hosmer-Lemeshow goodness of fit", "Exp (E)" for exponentiated coefficients (odds ratios), "Classification plots", "CI for 95% confidence intervals".

Interpreting the output:

  • The "Omnibus Tests of Model Fit" table tells you if the model improved over the intercept-only model.
  • The "Hosmer-Lemeshow" test checks model calibration (p > .05 = good fit).
  • The "Hypothesis Test" table shows each predictor’s Wald chi-square, β, SE, and odds ratio (Exp(B)).
  • Odds ratio > 1: The predictor increases the odds of the outcome.
  • Odds ratio < 1: The predictor decreases the odds.

APA template:

A binary logistic regression was conducted to predict [outcome] from [predictors]. The model was statistically significant, χ²(5) = 12.45, p = .014, R² = .23, Cox & Snell R² = .21. The Hosmer-Lemeshow test was non-significant, χ²(8) = 6.12, p = .421, indicating adequate model fit. [Predictor 1] significantly predicted the outcome, OR = 1.85, p = .008, 95% CI [1.21, 2.83].

Multinomial Logistic Regression

Use when your outcome has 3+ categories with NO natural order (e.g., brand preference: A, B, C; learning mode: visual, auditory, kinesthetic).

SPSS menu path: Analyze > Regression > Multinomial

Key difference from binary: You must select a "Reference Category" (the baseline group to compare against). This is usually the most common group or your control condition.

Ordinal Logistic Regression

Use when your outcome has ordered categories (e.g., Likert scale: strongly disagree → strongly agree).

SPSS menu path: Analyze > Regression > Non-Linear (for proportional odds models) or use specialized procedures.

Why it matters: Running separate t-tests or ANOVAs on ordered categories throws away the ordinal structure. Ordinal logistic regression is the correct approach.

3. Factor Analysis and Principal Component Analysis

When to use: You have many correlated variables (like survey questionnaire items) and want to reduce them into a smaller number of underlying "factors" or "components." This is essential for validating multi-item scales and constructs.

SPSS menu path: Analyze > Dimension Reduction > Factor

The workflow:

  1. Check sample size: You need at least 10 participants per item (rule of thumb). With 50 items, you need 500 participants minimum.
  2. KMO and Bartlett’s test: Click "Descriptive" and check "KMO and Kunique Bartlett’s test." KMO > .6 is acceptable; Bartlett’s p < .05.
  3. Extract factors: Choose "Eigenover, then scree" — this lets the data decide how many factors to extract. Look at the "Total Variance Explained" table.
  4. Rotate: Use "Varimax" (orthogonal rotation) for a cleaner, more interpretable solution. If you suspect your factors are correlated, use "Promax" instead.
  5. Interpret: Look at the "Rotated Component Matrix." Each item loads on the factor where its loading is highest.

Critical tip: Don’t use SPSS’s built-in factor score extraction (which gives z-scores with mean = 0, SD = 1). Instead, compute factor scores as the mean of your variables that load on each factor. This yields interpretable, positively-meaningful scores.

APA template:

A principal component analysis with varimax rotation was conducted on [N] questionnaire items. Kaiser’s KMO was .82 and Bartlett’s test of sphericity was significant, χ²(120) = 542.3, p < .001. The scree plot suggested three components, which explained 68% of the variance. Items loaded cleanly on three factors, with no cross-loadings exceeding .35. Component 1 (n = 5 items) represented [construct name], Component 2 (n = 4 items) represented [construct], and Component 3 (n = 3 items) represented [construct].

Common mistake: Extracting too many factors (overfitting) or too few (losing information). Use both the eigenvalue > 1 rule AND the scree plot to decide.

4. ANOVA with Post-hoc Tests

When to use: You already ran ANOVA and found "something is different." Now you need to know which specific groups differ from each other.

SPSS menu path: Analyze > Compare Means > One-Way ANOVA

After ANOVA, run post-hoc tests:

  1. In the One-Way ANOVA dialog, click "Post Hoc."
  2. Choose your test:
  3. Tukey HSD: Use when groups are equal size and you want all pairwise comparisons (most common choice).
  4. Bonferroni: Use when groups are unequal size or you have a small number of planned comparisons.
  5. Scheffe: Most conservative; use when group sizes are very unequal and you want the most conservative test.
  6. Click "Continue" and run.

Interpreting post-hoc output:

  • The "Post Hoc Tests" table shows pairwise comparisons and their p values.
  • Look for pairs where p < .05 (often marked with asterisks in SPSS output).
  • Report the mean difference and confidence interval for each significant pair.

APA template:

A one-way ANOVA revealed a significant effect of [independent variable] on [dependent variable], F(2, 57) = 8.42, p = .001, η² = .22. Tukey HSD post-hoc tests indicated that Group A (M = 45.2, SD = 8.1) differed significantly from Group C (M = 31.5, SD = 7.6), p = .002, 95% CI [3.4, 24.1]. Groups A and B (M = 42.1, SD = 7.9) and Groups B and C did not differ significantly, ps > .05.

Why effect size matters: Your thesis examiners will ask for η² (eta-squared). SPSS doesn’t calculate this automatically — you need to compute it as SSbetween / SStotal from the ANOVA output.

5. Non-Parametric Tests (When Assumptions Fail)

When to use: Your data violates the assumptions of parametric tests (not normally distributed, unequal variances, ordinal data). Non-parametric tests don’t assume a specific distribution.

Mann-Whitney U Test

The non-parametric alternative to the independent-samples t-test.

SPSS menu path: Analyze > Nonparametric Tests > Legacy Dialogs > Two Independent Samples

When to use: Compare two independent groups on an ordinal or non-normal outcome.

Interpretation: Report the U statistic, the z value, and the p value. Also report the effect size (r = z / √n).

Kruskal-Wallis Test

The non-parametric alternative to one-way ANOVA.

SPSS menu path: Analyze > Nonparametric Tests > Legacy Dialogs > K Independent Samples

When to use: Compare 3+ independent groups on an ordinal or non-normal outcome.

Follow-up: Use Mann-Whitney with Bonferroni correction (divide p by the number of comparisons) to find which groups differ.

Spearman’s Rank Correlation

The non-parametric alternative to Pearson’s correlation.

SPSS menu path: Analyze > Correlators > Bivariate (choose Spearman instead of Pearson)

When to use: At least one variable is ordinal, or both are non-normal.

6. ANCOVA (Analysis of Covariance)

When to use: You want to compare groups but need to "adjust" for baseline differences. For example, comparing teaching methods while controlling for students’ prior knowledge.

SPSS menu path: Analyze > General Linear Model > Univariate

Key steps:

  1. Put the outcome in "Dependent Variable."
  2. Put the grouping variable (e.g., teaching method) in "Fixed Factors."
  3. Put the covariate (e.g., pretest score) in "Covariates."
  4. Run.

Interpretation: Look at the "Adjusted Means" table. ANCOVA adjusts the group means for the covariate, giving you a fairer comparison.

APA template:

An ANCOVA was conducted to compare [outcome] between [groups] while controlling for [covariate]. The model was significant, F(2, 47) = 6.84, p = .002. After adjusting for baseline scores, Group A (adjusted mean = 45.6) scored significantly higher than Group C (adjusted mean = 31.8), F(1, 47) = 9.42, p = .003.

7. MANOVA (Multivariate ANOVA)

When to use: You have multiple outcome variables and want to test them simultaneously instead of running separate ANOVAs. This reduces Type I error and captures intercorrelations between outcomes.

SPSS menu path: Analyze > General Linear Model > Multivariate

Key steps:

  1. Put all your outcomes in the "Dependent Variables" box.
  2. Put your grouping factor in "Fixed Factor."
  3. Click "Statistics" and request "Effect sizes," "Estimate of effect size," and "Observed power."

Why MANOVA beats separate ANOVAs: When you have correlated outcomes, separate ANOVAs inflate Type I error. MANOVA tests all outcomes together and accounts for their correlations.

Interpretation: Look at the "Multivariate Tests" table. The "Pillai’s Trace" statistic is the most robust (use it when group sizes are unequal). Report the statistic, df, and p value for each factor.

8. Mixed-Effects Models (Linear Mixed Models)

When to use: Your data has a nested or longitudinal structure — for example, students measured at multiple time points, or students nested within classrooms.

SPSS menu path: Analyze > Mixed Models > Linear

Why this matters for PhD candidates: Longitudinal data analysis is becoming the standard. Traditional repeated-measures ANOVA assumes sphericity (equal variances of all pairwise differences), which is almost never true in real data. Mixed models relax this assumption.

Key steps:

  1. Put the outcome in "Dependent Variable."
  2. Put time in "Fixed Factor" and the subject ID in "Subject."
  3. In the "Fixed Effects" dialog, add your time variable.
  4. In the "Random Effects" dialog, choose the correct covariance structure (Unstructured is most flexible; if it doesn’t fit, try AR1 or Diagonal).

Interpretation: Report the fixed-effects estimates, standard errors, and significance tests. For the random effects, report the variance components.

Decision Matrix: Choosing the Right Test

This table is your quick reference for selecting the correct statistical method based on your research question and data structure.

Research Question Type Variable Configuration Recommended Test Why
Comparing 2 groups on continuous outcome 1 IV (2 levels), 1 continuous DV Independent-samples t-test Basic comparison (covered in our basics guide)
Comparing 3+ groups on continuous outcome 1 IV (3+ levels), 1 continuous DV One-way ANOVA → post-hoc ANOVA tells you "something differs"; post-hoc tells you which
Predicting a continuous outcome from multiple predictors Multiple continuous IVs, 1 continuous DV Multiple linear regression Extends simple regression to multivariate prediction
Predicting a binary outcome from predictors Multiple IVs, 1 binary DV (pass/fail) Binary logistic regression Handles categorical outcomes; reports odds ratios
Predicting an unordered categorical outcome Multiple IVs, 1 nominal DV (3+ categories) Multinomial logistic regression Correct model for nominal outcomes
Predicting an ordered categorical outcome Multiple IVs, 1 ordinal DV (Likert scale) Ordinal logistic regression Preserves the ordinal structure
Validating a questionnaire scale Many correlated items, wanting fewer constructs Factor analysis / PCA Reduces items into interpretable factors
Comparing groups while controlling for baseline 1 IV (categorical), 1 DV, 1 covariate ANCOVA Adjusts for pre-existing differences
Testing multiple outcomes simultaneously 1 IV (categorical), 2+ DVs MANOVA Reduces Type I error; accounts for intercorrelations
Longitudinal or nested data Repeated measures or hierarchical structure Linear mixed models Handles non-sphericity; flexible covariance structures
Data violates normality assumptions Ordinal data or skewed distributions Non-parametric tests Distribution-free; robust to outliers

The Assumption-Checking Workflow You Can’t Skip

This is the single biggest gap between what students know and what thesis examiners expect. Running the test is not enough. You must check assumptions and report on them.

Step 1: Data preparation

  • Check for missing data (Analyze > Descriptive Statistics > Frequencies)
  • Code-categorical variables properly (Analyze > Classifications > Define Variable Classes)

Step 2: Assumption testing

  • Normality: Shapiro-Wilk test (n < 50) or histogram/Q-Q plot (n ≥ 50)
  • Homogeneity of variance: Levene’s test (in ANOVA output)
  • Multicollinearity: VIF or correlation matrix (for regression)
  • Linearity: Scatterplot matrix (for regression)

Step 3: Run the test

  • Follow the menu paths above
  • Request all relevant diagnostics (confidence intervals, effect sizes)

Step 4: Interpret and report

  • Report the test statistic, df, p value, and effect size
  • Include assumption-check results in the methods section

What You Should Report (APA Template Checklist)

Every advanced method needs specific reporting. Don’t just copy-paste SPSS output — here’s what examiners actually look for:

For regression models:

  • R² and adjusted R²
  • F test for the overall model
  • Each predictor’s β, SE, t, p, and 95% CI
  • Effect size (Cohen’s f² or η²)
  • Assumption-check results (normality, multicollarity, linearity, homoscedasticity)

For ANOVA/post-hoc:

  • F value, df, p value
  • η² (eta-squared) effect size
  • Post-hoc test results with mean differences and CIs
  • Levene’s test result

For logistic regression:

  • Model χ² and df
  • R² (Cox & Snell, Nagelkerke)
  • Hosmer-Lemeshow test result
  • Each predictor’s odds ratio, 95% CI, and p value

For factor analysis:

  • KMO and Bartlett’s test
  • Number of components extracted
  • Total variance explained (%)
  • Rotated component matrix with loadings
  • Scree plot (include in appendix)

Common Mistakes (and How to Avoid Them)

1. Running separate ANOVAs instead of MANOVA

The mistake: Testing multiple outcomes with separate ANOVAs inflates Type I error and ignores intercorrelations.

The fix: Use MANOVA. It’s one click in SPSS (Analyze > GLM > Multivariate) and gives you a cleaner, more accurate result.

2. Ignoring assumption checks

The mistake: Running a regression or ANOVA without checking assumptions.

The fix: Follow the workflow above. Even a one-paragraph description of assumption checks in your methods section will satisfy examiners.

3. Reporting only p values, not effect sizes

The mistake: Only reporting "p < .05" without effect sizes. Thesis examiners will flag this.

The fix: Always report effect sizes. Cohen’s d for t-tests, η² for ANOVA, odds ratios for logistic regression, η² partial for ANCOVA.

4. Using the wrong post-hoc test

The mistake: Using Tukey HSD when groups are unequal size.

The fix: Use Bonferroni for unequal group sizes; Scheffe for very unequal groups; Tukey for equal groups.

5. Interpreting factor loadings as correlations

The mistake: Treating factor loadings as if they were Pearson correlations.

The fix: Factor loadings are analogous to correlations, but they represent the correlation between an item and the factor, not the correlation between items.

Advanced SPSS methods can feel overwhelming at first — especially the difference between binary, multinomial, and ordinal logistic regression, or the choice between different covariance structures in mixed models. But every one of these methods follows a predictable pattern: choose the right test → check assumptions → run the procedure → interpret → report.

The most common thread across all advanced methods is assumption checking. That’s what separates a student who has "run the analysis" from one who has "validated the analysis." Thesis examiners notice this difference, and so does the statistical rigor your dissertation deserves.

You’ve covered the basics and now it’s time to level up — come back to this guide for the advanced methods. Together, they cover everything you need to handle dissertation data analysis confidently.