How I Helped a Louisiana Tech Finance PhD Candidate Fix a Mediation Analysis in SPSS Using PROCESS

Mediation Analysis in SPSS is where most finance and management PhD candidates hit a wall, and it is usually not because the theory is weak. It is because the method taught in most coursework, three separate regressions, stopped being acceptable to committees years ago. Nobody tells students this until a committee member says it in a rejection email.

This happened to a Finance PhD candidate at Louisiana Tech University. Her topic: how corporate social responsibility (CSR) affects firm value, with risk-taking as the mediator. Good theory, clean data, wrong statistical method.

I’ll walk through what went wrong in her SPSS output, why her committee rejected it, and how we fixed it using the PROCESS macro. If you’re working through mediation, moderation, or any regression-heavy chapter in your own dissertation, organising your analysis chapter the right way from the start will save you exactly this kind of resubmission cycle.

The Problem: A Rejected SPSS Mediation Analysis

What Her Committee Actually Objected To

She sent me her syntax, her data file, and her output before we spoke. The theory was solid. The statistics were about fifteen years out of date.

One committee member said separate regressions do not test the indirect effect at all. Another told her she needed bootstrapped confidence intervals, and she had never even heard the term before. This is a fair rejection, and I’ve seen the exact same objection raised for students in India, the US and the UK alike. A poorly structured results chapter can genuinely stall a defence, and an outdated statistical method is really a structural problem wearing a statistics costume.

The Baron and Kenny Method She’d Used

She had run the classic Baron and Kenny causal steps approach from 1986: three regressions.

1. Firm value on CSR (the total effect)
2. Risk-taking on CSR (the a path)
3. Firm value on CSR and risk-taking together (the direct effect and b path)

She compared the CSR coefficient in model 1 against model 3, saw it shrink but stay significant, and called it partial mediation. That is the textbook definition of the method. Her mistake was not carelessness. The method itself never tests whether that shrinkage is statistically meaningful.

Why Comparing Coefficients Isn’t a Statistical Test

Here’s my honest opinion after 12 years of consulting on this exact problem: Baron and Kenny’s approach still gets taught as if it is the gold standard, and it isn’t, not anymore.

David Kenny, one of the original authors, has since written that the method’s own first step was a flawed gate. Requiring a significant total effect before testing mediation sounds reasonable. But a real, meaningful indirect effect can sit on top of a small or non-significant total effect. Baron and Kenny’s rule would tell you to stop testing before you even reach the mediator.

Baron and Kenny vs Bootstrapping

Baron and Kenny asks one question: did the CSR coefficient get smaller after adding risk-taking? Bootstrapping asks a sharper one: is the indirect effect (CSR through risk-taking to firm value) statistically different from zero, with an actual confidence interval around it?

That is the entire difference, and it isn’t a small one. A shrinking coefficient can happen by chance. A confidence interval that excludes zero cannot.

This distinction also matters because a mediation path is a different claim from a simple relationship between two variables. If you’re still deciding whether your two variables need a Pearson or a Spearman correlation in the first place, that’s a separate, earlier question from mediation. Conflating the two is a mistake I see often in early drafts.

Why the Sobel Test Wasn’t Enough Either

Her earlier draft had also reported a Sobel test, a fix some professors still recommend for Baron and Kenny’s weaknesses. I’d push back on that advice too.

The Sobel test assumes the indirect effect is normally distributed, which it usually isn’t. It also has noticeably lower statistical power than resampling methods. It’s an improvement on Baron and Kenny, but most finance and management journals now expect more.

What the PROCESS Macro (Andrew F. Hayes) Actually Does

Definition: The PROCESS macro is a free add-on for SPSS, SAS and R, written by Andrew F. Hayes, that runs mediation, moderation and conditional process models directly through regression. Instead of comparing coefficients across separate models, it reports the indirect effect with a bootstrapped confidence interval built in.

Hayes is a Distinguished Research Professor at the University of Calgary’s Haskayne School of Business, and PROCESS is documented in full in his mediation and moderation textbook published by Guilford Press, now in its third edition. You can download the tool directly from PROCESS macro’s official site. This isn’t a random SPSS plugin. It’s the standard reference tool for exactly the model her committee was asking for.

Why Bootstrapping Is the Modern Standard

Bootstrapping in SPSS mediation analysis works by resampling your dataset thousands of times, recalculating the indirect effect each time, and building a confidence interval from that distribution. There’s no assumption that the indirect effect is normally distributed, unlike some traditional regression diagnostics where you’d still run a check like the Shapiro-Wilk normality test. This is exactly why bootstrapping indirect effect in SPSS has become the expected standard in finance and management research, not an optional extra.

Step-by-Step: Installing and Running PROCESS in SPSS

We installed PROCESS together over a screen share. If you’re doing this for the first time, here’s the exact sequence:

1. Download the PROCESS macro from PROCESS macro’s official download page; it’s free.
2. In SPSS, go to Utilities > Custom Dialogs > Install Custom Dialog and select the downloaded .spd file.
3. Restart SPSS. You’ll now see Analyze > Regression > PROCESS by Andrew F. Hayes.
4. Set your variables:
   • Y (outcome): firm value, measured as Tobin’s Q (a market-to-book ratio commonly used as a proxy for firm value in finance research)
   • X (predictor): CSR score
   • M (mediator): risk-taking, measured as standard deviation of ROA (return on assets)
   • Model number: 4 (simple mediation)
   • Bootstrap samples: 5,000
   • Confidence interval: 95%
5. Under Options, tick “Show total effect model” and “Effect size”.
6. Run it.

Setting Up Model 4 for Simple Mediation

Model 4 is PROCESS’s number for the basic X to M to Y mediation path, which is what her theory needed. If your dissertation also involves an interaction term, that’s a different model number and a separate conversation about moderated mediation, which I come back to further down.

Bootstrap Samples and Confidence Level

5,000 resamples is a reasonable working minimum for a dissertation-level dataset. Many journals now expect 10,000. I’d rather a student run 10,000 once than get asked to rerun it during revisions, so use the higher number from the start if your processing time allows it.

Reading the Output: Total, Direct, Indirect Effects

PROCESS gave us three numbers that mattered:

• Total effect of CSR on firm value
• Direct effect of CSR on firm value, controlling for risk-taking
• Indirect effect of CSR on firm value through risk-taking, with its bootstrap confidence interval

What a Bootstrapped CI Tells You

The indirect effect came out to 0.08, with a 95% bootstrap confidence interval of [0.03, 0.14]. That interval doesn’t cross zero, so the indirect effect is statistically significant. That single sentence is what her committee had been asking for all along, and Baron and Kenny’s method simply cannot produce it.

Full vs Partial Mediation

The direct effect stayed significant even after adding risk-taking, so this is partial mediation, not full mediation. The difference matters for your discussion chapter. Full mediation means the mediator explains the entire relationship. Partial mediation means CSR still affects firm value directly, with risk-taking explaining only part of the story.

Before you get to this stage, check your model for a multicollinearity problem between CSR and risk-taking. I’d always run a VIF check before PROCESS, not after.

The Result: A Defensible Mediation Chapter

She rewrote her results section, dropped the Baron and Kenny language, and reported the bootstrapped indirect effect with proper APA phrasing. Her committee approved the revision on the next pass.

What she said afterward has stayed with me. She thought mediation was just three regressions and had no idea there was an entirely different, more accepted way to test it. That’s not her fault. It’s a gap in how mediation gets taught in most statistics courses, in India and abroad both.

How I Help Finance and Business PhD Candidates With SPSS

If you’re searching for Louisiana Tech dissertation SPSS help, or the same kind of help from any US or UK university, this is exactly the kind of problem I work on daily as a finance PhD statistical consultant offering PROCESS macro SPSS help. I don’t write dissertations. I get your numbers right and make sure you can defend them in your own words in front of a committee.

I’ve done the same fix for a marketing PhD candidate running a moderated mediation model (PROCESS model 7) on brand trust and purchase intention, where the client had never touched an interaction term before. The mechanics change; the underlying problem, outdated methods presented as current, does not.

If any of this sounds familiar, I offer SPSS data analysis tutoring and full statistical consulting services for exactly this stage of a dissertation. Any data, syntax or output you share with me stays confidential and is used only for your own analysis, never reused or passed on elsewhere. If you’re unsure whether your situation needs outside help at all, this guide on knowing when you need a dissertation expert is a fair starting point before you spend money on anyone, me included.

FAQ

Siddharth Gupta has 12+ years of experience in dissertation and statistics consulting across SPSS, R, Stata and Power BI, and runs Statssy, a statistics tutoring and dissertation consulting practice. Connect on LinkedIn.