How I Fixed a Political Science Candidate’s Analysis Before Her Defence

Written by Siddharth Gupta, dissertation and statistical consultant with 12 years of research and analysis guidance experience. Connect on LinkedIn

SPSS statistical consulting for PhD students becomes urgent the moment a dissertation committee sends back your analysis chapter with more red ink than approval marks. That is exactly what happened to a Political Science PhD candidate at a public university in the Midwest who wrote to me three weeks before her scheduled defence.

She was not confused about her theory. Her problem was sitting inside her SPSS output file, and she did not know it yet.

The Problem: Three Weeks to Defence and Stuck in SPSS

Her committee had already reviewed her draft and pushed it back. One reviewer said the analysis did not match her hypotheses. Another asked why her key independent variable came out insignificant when the existing literature said it should matter. She had three weeks left and was panicking, and she had good reason to be.

She sent me her codebook, her SPSS syntax file and her output. Within an hour of opening it, I could tell her argument was fine. Her data analysis was not.

What the Committee Pushed Back On

Committees rarely reject a chapter because the writing is weak. They reject it because the numbers do not support the claims being made in the text.

This mismatch shows up often in political science and other social science dissertations. Most students only get a single semester of SPSS training in their programme. That is not enough time to build real research-level statistical judgement.

What Was Actually Wrong With Her Analysis

Four separate problems were compounding each other in her dataset. None of them were visible unless you went through the syntax line by line.

What Is Likert Scale Reverse Coding in SPSS?

Reverse coding flips the scoring direction of a negatively worded survey item, so every item in a scale points the same way before you combine them into one score. In SPSS this is done through Transform > Recode into Different Variables.

Her survey measured political trust and institutional confidence using Likert-type items. Some items were worded positively, some negatively, which is standard practice to reduce acquiescence bias. She had entered every item in the same direction without applying Likert scale reverse coding in SPSS to the negative ones.

One item read something close to “People like me have no say in what government does.” Left unreversed, someone who agreed strongly with that statement, meaning they felt low political efficacy, was pulling the composite score in the same direction as people who actually felt high efficacy. Her “trust scale” was not measuring trust. It was measuring noise.

What Is Missing Data Imputation in SPSS?

Missing data imputation fills in missing values using statistically informed estimates, usually multiple times. You then run the analysis separately on each completed version of the dataset and pool the results into one final estimate. It is generally preferred over listwise deletion when data is not missing completely at random.

About 18 percent of her household income variable was missing. She had not deliberately chosen a missing data method. SPSS was quietly defaulting to listwise deletion, which drops any respondent missing even one variable used in the model.

That default was shrinking her effective sample size. Worse, it was biasing her results, since the missingness was not random. This is one of the most common invisible errors in dissertation data analysis. Nobody tells students that SPSS is making this choice on their behalf unless they change the setting themselves.

A Weak Item Was Dragging Down Cronbach’s Alpha

Her trust scale had a Cronbach’s alpha of only 0.54. Most dissertation committees expect at least 0.70 before accepting a scale as reliable, so this alone was enough to trigger the rejection.

One item in the scale was correlating poorly with the rest. It was not measuring the same underlying concept as the other items, and it was dragging the whole scale down with it.

Regression Assumption Checks in SPSS Were Skipped Entirely

When she ran her OLS regression, she had not checked a single assumption. The residual plots showed clear heteroscedasticity. This means the spread of errors was not constant across values of the predictors. Two of her state-level variables also had a variance inflation factor above 10, which signals serious multicollinearity.

Regression assumption checks in SPSS are not optional homework. They are the difference between a chapter that survives a defence and one that gets sent back. This includes checking normality with the Shapiro-Wilk test and, if your data has any time or sequence structure, reviewing the Durbin-Watson statistic as well.

How We Fixed It, Step by Step

I did not rewrite her dissertation for her. I walked her through every fix so she understood it well enough to defend it in front of her committee.

1. Reverse-coded the negatively worded items using Transform > Recode into Different Variables, then checked the new distributions to confirm the direction was correct.
2. Ran multiple imputation for the missing income data and pooled the results across the imputed datasets instead of deleting cases.
3. Removed the weak item from the trust scale, which raised Cronbach’s alpha from 0.54 to 0.78.
4. Mean-centred the collinear variables, built the interaction term properly, and re-ran the model using bootstrapped confidence intervals.

The Result: A Defence That Went Well

When the corrected output came back, the interaction term was significant. Her original conclusion, that state context did not moderate the relationship she was studying, was wrong. It did moderate the relationship. But only once the variables were centred correctly and the model was specified properly.

She updated her results chapter, went back to her committee with the corrected analysis, and defended successfully a few months later. What she said afterward has stayed with me since: “I thought I was bad at statistics. I just didn’t know what I didn’t know.”

That line sums up almost every dissertation SPSS problem I come across. SPSS will run a regression on badly prepared data without complaint. It does not check whether your reverse coding is correct, whether your missing data is handled properly, or whether your assumptions are met. That is the researcher’s job, and very few doctoral programmes teach it properly.

Is It Ethical to Hire a Statistician for Your Dissertation?

Yes, provided the statistician helps you understand and defend the analysis rather than producing findings you cannot explain yourself.

Michigan State University’s Center for Statistical Training and Consulting states this directly in its own guidance for graduate students: a statistician can help you understand results and check your methodology, but should not perform the methodological research for your dissertation. You must still be able to defend the content, explain your method choices, and interpret the results yourself.

That is the exact line I hold myself to. Where I think a large part of this market gets it wrong is with services that treat your dataset like a factory order.

Send data, receive output, no explanation attached. That approach will not survive a committee member asking one follow-up question about why you centred a variable or chose a particular imputation method. If you cannot answer that in your own words, the correctness of the underlying statistics will not save you.

What a Political Science PhD Statistical Consultant Should (and Shouldn’t) Do For You

I will be direct about this, since a lot of services in this space blur the line on purpose because “we do it all for you” is an easier sell.

A consultant should:

• Explain why an assumption was violated and what fixing it changes about your conclusion
• Walk you through the corrected syntax so you can run it yourself afterward
• Prepare you for the exact questions a committee is likely to ask

A consultant should not:

• Hand you a finished chapter with no explanation of the choices made
• Change your research question or hypotheses to fit convenient results
• Leave you unable to answer basic questions about your own analysis

Signs Your Own Dissertation Analysis Might Have This Problem

You do not need to wait for a committee rejection to check for these. If two or more of these apply to you, it is worth reading how to know if you need a dissertation expert before your next committee meeting.

• A key variable is insignificant even though the literature strongly suggests it should matter
• Your Cronbach’s alpha is below 0.70 and you have not checked which item is weak
• You used the default missing data handling in SPSS without deciding on it yourself
• You have not looked at a residual plot or checked VIF values before running your regression
• You can produce the output but struggle to explain in plain language why a result came out the way it did

SPSS Statistical Consulting for PhD Students: What I Can Help You With

I work as a statistical consultant and analysis coach for PhD and master’s students across disciplines, using SPSS, R and Stata. If you are searching for SPSS dissertation data analysis help, this is exactly what that looks like in practice. You can also start with a shorter session through my SPSS data analysis tutor option if you only need help understanding one part of your output, or see the full range on my statistical consulting services page.

• Clean and prepare your dataset properly before any modelling begins
• Apply correct Likert scale reverse coding and build reliable, defensible scales
• Handle missing data using the correct method for your situation, not the SPSS default
• Complete regression assumption checks in SPSS before you present a single result, reported in line with APA standards
• Run and interpret OLS, logistic, mediation, moderation and panel models
• Make sure you understand your own output well enough to defend it
• Troubleshoot SPSS syntax and error messages
• Turn confusing output into a results chapter your committee can actually follow

Every dataset you send me is used only for your own project and is not shared, reused or stored beyond what is needed to complete your work. You stay in control of your research. My job as a statistical consultant is to make sure you get the numbers right and understand why they are right, which is the part most services skip.

If you are stuck at the analysis stage right now, send me your dataset and your defence date. I will tell you honestly what needs fixing and how long it will realistically take, before you commit to anything.

Frequently Asked Questions