Statistical Analysis Does Not Match Research Question? Here’s How to Check and Fix It

If your statistical analysis does not match research question wording, the cause is almost always a mismatch between what your RQ is asking (a relationship, a difference or a prediction) and what your test is actually measuring, not a calculation error. Rewriting your RQ in plain words and checking it against your variable type fixes this in most cases, without recollecting any data.

I have reviewed this exact problem in hundreds of dissertations over the last 12 years, and it follows the same pattern in almost every case I review. In this article, I will walk you through the same diagnostic process I use when a student sends me their SPSS output along with their supervisor’s feedback in a panic.

Does a Significant P-Value Mean Your Analysis Is Correct?

Here is the trap almost every student falls into. You run a test, SPSS gives you a p-value under 0.05, and you assume the job is done.

A significant p-value only tells you that your result is unlikely to be due to chance. It does not tell you if you asked SPSS the right question in the first place. It also says nothing about effect size, how large the relationship or difference actually is, which most UK and US committees now expect alongside p-values. This is the single biggest confusion I see when students ask me how to know if statistical analysis is correct.

An analysis is correct when two conditions are met together. First, the test’s assumptions are satisfied (normality, sample size, data type). Second, the test actually answers what your research question is asking.

A lot of the statistics advice you will find online explains this second point in purely theoretical language, write your research question in “operational terms” before choosing a test. That advice is not wrong, but it is not very useful to a student who has already collected data and is three weeks from submission. What you need is a practical way to check your own work, which is what the rest of this article gives you.

What It Means When Your Statistical Analysis Does Not Match Your Research Question

When a supervisor says analysis is wrong, students usually assume they made an SPSS error, clicked the wrong box, or got a formula backwards. In my experience, that is rarely the issue.

What is actually happening is that your analysis does not answer research question as it is worded. Your RQ might ask about a relationship. Your test measured a difference instead. These are not the same thing statistically, even though they sound similar in everyday English.

Example of the mismatch: A research question reading “What is the relationship between remote work hours and employee productivity?” needs a correlation or regression. If you instead grouped employees into “high remote hours” and “low remote hours” and ran an independent samples t-test, you have answered a different question than the one you asked. This is a classic research question and statistical test mismatch, and it is exactly what a sharp supervisor or examiner catches within minutes.

How Do I Know If My Statistical Analysis Is Correct? (The 4-Step Check)

This is the diagnostic process I run with every student who comes to me with this problem. It takes less time than you think, and none of it requires your supervisor or you to be a statistics specialist. It is written in plain English on purpose.

Step 1: Rewrite Your RQ in Plain, Operational Words

Strip out academic phrasing. What are you actually asking: is there a relationship, a difference, or a prediction?

Step 2: Identify the Action Verb in Your RQ

“Relationship between,” “effect of,” “difference between,” and “predicts” each point to a different family of tests. This one word usually gives away the whole answer.

Step 3: Check Your Variable Types

Nominal, ordinal, interval, or ratio. This decides whether you can use a parametric test at all, and whether your DV is continuous or categorical.

Step 4: Match RQ Type and Variable Type to the Correct Test

Use the table below. This is how to match analysis to research question wording, every single time, without needing a statistics degree. For a broader academic reference alongside the SPSS-specific table here, Scribbr’s guide to choosing the right statistical test is a useful general companion.

RQ Wording to SPSS Test: A Quick Reference Table

This table covers most dissertation-level SPSS analysis for research questions I see in business, psychology, nursing and social science dissertations.

Save this before you go: this is the exact checklist I run through with every dissertation student who sends me their SPSS output for review before submission. Bookmark the page, you will need to come back to it when you write up your methodology section.

Common Mismatches I See Most Often

After 12 years of reviewing dissertation data, three mistakes show up again and again.

Mistake 1: Treating a “Relationship” RQ as a “Difference” RQ

Students run a t-test because it feels simpler than regression, even when the RQ word is “relationship,” not “difference.” That is a “statistical test does not match hypothesis” problem hiding inside a correctly written RQ. It gets more complicated when the RQ also implies a moderating or mediating variable, for example “How does training moderate the relationship between workload and burnout”, since that needs a proper moderation model, not a simple correlation or t-test.

Mistake 2: Treating Likert-Scale Data as Interval Without Checking

A five-point agreement scale is ordinal, strictly speaking, not interval. Running a parametric ANOVA on it without justification is a common reason reviewers flag the analysis section, and this is the one area where I genuinely disagree with how most online guides handle it. Most of them tell you Likert data is “usually fine” to treat as interval and leave it there, which is technically defensible but weak advice to hand a student facing an examiner who disagrees. Run and report a Shapiro-Wilk test for normality on your scale data first, then justify your choice either way in the methodology section, rather than assuming.

Mistake 3: Ignoring Repeated Measures Structure

If you measured the same group before and after an intervention, a regular t-test is the wrong family of test. You need a paired design test instead. The difference is explained properly in this comparison of a paired t-test versus repeated measures ANOVA for pre-post data.

A quick case from my own files: A UK-based MBA student I worked with last year had the RQ “What is the relationship between leadership style and employee engagement?” She had run an independent samples t-test by splitting leadership style into two arbitrary groups. Her supervisor flagged it as not answering the RQ. We re-ran it as a multiple regression with leadership style as a continuous predictor, same dataset, no recollection needed, and the results chapter was rewritten in two days.

Multiple regression brings its own checks too. Before trusting a model with more than one predictor, it is worth checking VIF for multicollinearity so you are not reporting an unstable model.

Do You Need to Recollect Data, or Just Re-Run the Analysis?

This is the question that causes the most panic, so let me be direct about it.

In most of what students bring me as a wrong analysis for dissertation problem, the existing data is fine. The fix is choosing the correct test for the data you already have, not starting over. This was true in the MBA case above and in the majority of corrections I handle.

The exception is when your study design itself cannot answer the RQ, for example if you only measured one group when the RQ genuinely needs a comparison group. This is rarer than students fear, but it does happen, and it usually traces back to the research design stage rather than the SPSS stage.

How to Fix and Report It Without Blowing Your Timeline

Once you know the correct test, re-running it in SPSS usually takes an afternoon, not weeks. Switching from a t-test to a correlation or regression is typically a matter of Analyze then Correlate, or Analyze then Regression, rather than starting a new file. The output format changes accordingly, a correlation coefficient and significance value instead of a t-statistic and group means.

The bigger task is rewriting your results and discussion sections to match the corrected output, since the numbers, tables and interpretation will all change. For the actual write-up, follow a recognised style guide rather than guessing the format. Scribbr’s guide to reporting statistics in APA style is a useful reference for how to present test statistics, p-values and effect sizes consistently once your corrected analysis is ready.

Go back to your supervisor with the corrected test already run, not just a promise to fix it. Supervisors and committees respond far better to “I found the mismatch and here is the corrected analysis” than to a vague apology.

When to Get a Second Opinion Before Your Viva or Defence

If you have gone through the checklist above and you are still unsure, that uncertainty itself is a signal. Having someone outside your own head run a dissertation statistical analysis review catches blind spots you cannot see after staring at the same dataset for months, whether your panel calls it a viva, a defence, or a committee review.

Proper SPSS analysis review help checks three things together, not just one test in isolation: whether your RQ and hypotheses are worded consistently, whether the test matches both, and whether your SPSS output is interpreted correctly for your discussion chapter. This is different from the generic “I will run your SPSS tests” gigs you find online, which usually just execute whatever test you ask for without checking if it is the right one in the first place. Closing that exact gap is what this whole article has been building toward. If you want hands-on support working through your own RQ and output, I run these checks directly in one-on-one SPSS tutoring sessions.

FAQ

Written by Siddharth Gupta, dissertation and statistics consultant with 12 years of experience reviewing SPSS analyses for dissertation students across the USA and UK. Connect on LinkedIn.