Hierarchical Multiple Regression: How to Run, Interpret and Report It (With a Worked Example)

Hierarchical multiple regression is a technique where you add predictors to a regression model in planned steps (blocks) to see how much additional variance each new set of variables explains beyond the ones already entered. If you're staring at your thesis data wondering whether your "interesting" predictor adds anything once you've controlled for age, gender and baseline scores — this is the test you need, and the R² change value is the number you're after.

Key Takeaways

• Hierarchical multiple regression tests whether a new block of predictors significantly improves a model after controlling for variables entered earlier, measured by the change in R² (ΔR²).
• You enter predictors in theory-driven steps: control variables in Block 1, your variables of interest in Block 2 (and so on), and compare models at each step.
• The key output is the ΔR² and its associated F-change test — a significant F-change means the new block explains meaningful additional variance.
• It is the same mathematics as standard multiple regression; "hierarchical" refers only to how you enter predictors, not to multilevel or mixed-effects models.
• APA 7 requires you to report R², ΔR², the F-change statistic, and the unstandardised (B) and standardised (β) coefficients for the final model.

What Is Hierarchical Multiple Regression (In Plain English)?

Hierarchical multiple regression answers one question: "Does my predictor still matter once I've accounted for the obvious things?" Imagine you want to show that job satisfaction predicts employee wellbeing. A sceptic says, "Of course — older, higher-paid people are just happier." Hierarchical regression lets you enter age and salary first (Block 1), then add job satisfaction (Block 2), and prove that satisfaction explains wellbeing over and above those controls.

The word "hierarchical" trips people up. It does not mean multilevel modelling or nested random effects. It simply means you decide, in advance and for theoretical reasons, the order in which blocks of predictors enter the model. Each step builds on the last.

When Should I Use Hierarchical Regression?

Use hierarchical multiple regression when you have:

• A continuous outcome variable (e.g. wellbeing score, exam mark, blood pressure).
• Two or more predictors you can group into theory-driven blocks.
• A clear reason to control for some variables before testing others (demographics, baseline measures, established predictors).

If you have no theoretical ordering and just want to find which variables predict the outcome, stepwise or standard (simultaneous) regression may suit you better. Hierarchical regression is for hypothesis testing with controls, not data dredging.

How to Run Hierarchical Multiple Regression Step by Step

The logic is the same whether you use StatRyx, SPSS or R:

1. Define your blocks. Decide which variables go in Block 1 (usually controls) and Block 2 (your predictors of interest). This must be theory-led, not data-led.
2. Run Model 1 with only the Block 1 predictors. Note its R².
3. Run Model 2 adding the Block 2 predictors. Note its R².
4. Calculate the change: ΔR² = R²(Model 2) − R²(Model 1).
5. Test the change with the F-change statistic. A significant result means your new block adds explanatory power.
6. Check assumptions across all models: linearity, independence of residuals, homoscedasticity, normality of residuals, and no severe multicollinearity (VIF < 10, ideally < 5).

In StatRyx, you select your outcome, drag predictors into named blocks, and the platform runs every model, computes ΔR² and F-change, checks the assumptions automatically, and writes the APA 7 result for you — no syntax, no clicking through five menus.

A Worked Example With Real Numbers

Suppose we study 120 university students and predict their wellbeing score (0–100). We want to know whether perceived social support adds anything beyond demographics.

• Block 1 (controls): age and gender.
• Block 2 (predictor of interest): social support.

Here is what the output looks like:

Reading this: Block 1 (age and gender) explained 8% of the variance in wellbeing. Adding social support in Block 2 raised the explained variance to 34% — an increase of 26 percentage points (ΔR² = .26), and that increase is highly significant, F(1, 116) = 45.70, p < .001.

Now look at the final model's coefficients:

Interpretation: Once social support is in the model, age and gender no longer predict wellbeing significantly. Social support is the dominant predictor: for every one-unit increase in support, wellbeing rises by 0.62 points (B = 0.62), and its standardised coefficient (β = .52) is by far the largest. This is exactly the pattern that justifies the hierarchical approach — the effect of interest survives the controls.

How Do I Interpret R² Change?

ΔR² tells you the proportion of additional outcome variance explained by the newly added block of predictors. An ΔR² of .26 means the new block accounts for 26% more of the variance than the earlier model did. Cohen's benchmarks for f² (derived from ΔR²) suggest roughly .02 = small, .15 = medium, and .35 = large effect. Always interpret the F-change test alongside it: a tiny ΔR² can reach significance in a large sample but mean little practically, while a large ΔR² in a small sample may not be significant at all.

How to Report Hierarchical Multiple Regression in APA 7

APA 7 expects you to report the model-building sequence, the variance explained at each step, the change statistics, and the final coefficients. A clean write-up for our example:

A hierarchical multiple regression was run to predict wellbeing. Age and gender were entered in Step 1, explaining 8% of the variance, R² = .08, F(2, 117) = 5.09, p = .008. Adding social support in Step 2 explained an additional 26% of the variance, ΔR² = .26, F-change(1, 116) = 45.70, p < .001. In the final model, social support was the only significant predictor, B = 0.62, 95% CI [0.44, 0.80], β = .52, p < .001.

Note the APA conventions: statistics italicised, leading zeros dropped on p and R² values (which cannot exceed 1), and a confidence interval reported for the key coefficient. If you're also deciding between predictors or testing interactions, see our related guide on moderation and interaction effects in regression.

Hierarchical vs Standard vs Stepwise Regression