SPSS Political Science Dissertation Help: A Working Guide to Voter Behaviour, Ordinal Data, and Cross-Tabs

If you are searching for SPSS political science dissertation help, you have probably already tried a generic tutorial and hit a wall. I have spent twelve years helping political science and international relations students get from “I have 150 survey responses and no idea what to run” to a statistics chapter their examiner actually signs off on. This guide is written from that experience, not from a textbook.

Most generic SPSS tutorials assume your dependent variable is a continuous number like income or test score. Political science data rarely works that way. Your dependent variable is usually an attitude: trust in government measured on a five-point scale, or likelihood to vote measured as unlikely, somewhat likely, very likely. That single difference changes almost every test you will run.

SPSS Political Science Dissertation Help: Why Generic Tutorials Fall Short

Generic SPSS tutorials don’t cover ordinal attitude data

I have read a lot of SPSS guidance written for business students and psychology students, and most of it quietly skips ordinal dependent variables. That is a problem because ordinal attitude scales (strongly disagree to strongly agree, unlikely to very likely) are the backbone of political science dissertations.

Run a plain linear regression on a five-point Likert DV and your examiner will ask why you treated unordered-looking category gaps as equal intervals. I have seen this exact comment on viva feedback sheets more times than I can count.

What examiners in the US and UK actually look for in your stats chapter

In my experience, examiners on both sides do not mainly care how advanced your statistics are. They care whether you picked the right test for your variable type and can defend why.

A clean chi-square test, correctly justified, beats an impressive-looking regression that ignores its own assumptions. How to know if you need a dissertation expert is a good place to check if you are already past the point where self-teaching is costing you more time than it saves.

Getting Your Data Dissertation-Ready in SPSS

Coding Likert-scale and categorical political variables correctly

Before any SPSS help for political science research means anything, your data has to be coded sensibly. Give every category a clear numeric code (1 = strongly disagree through 5 = strongly agree), label it properly in Variable View, and set the measurement level to Ordinal, not Scale.

I still find students who leave attitude variables set to “Scale” by default. SPSS will happily let you run the wrong test on badly labelled data and give you no warning at all.

Handling missing data and listwise deletion

SPSS deletes a whole case the moment one variable in your model is missing, unless you tell it otherwise. On a dissertation sample of 120 to 200 responses, that can quietly wipe out 15 to 20 percent of your data.

Check your missing values table before you run anything. If you are deciding between primary data collection and an existing dataset at this stage, primary versus secondary data for your dissertation walks through that trade-off properly.

Practising on free data: the ANES dataset

If you want to test your model before touching your real dissertation data, use the American National Election Studies dataset, hosted through ICPSR. It is free, well documented, and built around exactly the kind of variables political science dissertations use: party identification, policy attitudes, voting behaviour.

I tell every student I work with to run their planned analysis on an ANES subset first. If your model breaks on well-cleaned data, you know the problem is your approach, not your dataset.

Choosing the Right Statistical Test: A Decision Framework

Nominal vs ordinal vs continuous dependent variables

This is the single most important decision in your entire statistics chapter, and I say that after twelve years of watching dissertations go wrong here specifically.

1. If your DV is unordered categories (which party someone voted for) use multinomial logistic regression.
2. If your DV is ordered categories (unlikely, somewhat likely, very likely) use ordinal regression.
3. If your DV is two categories (voted or did not vote) use binary logistic regression.
4. If your DV is genuinely continuous (a 0 to 100 feeling thermometer score) linear regression is defensible.

Comparison table: which test fits your data

Test	Dependent variable type	SPSS menu path	Typical political science example
Chi-square and cross-tab	Two categorical variables	Analyze > Descriptive Statistics > Crosstabs	Party ID by stance on an issue
Binary logistic regression	Two categories	Analyze > Regression > Binary Logistic	Voted vs did not vote
Ordinal regression (PLUM)	Three or more ordered categories	Analyze > Regression > Ordinal	Trust in government, low to high
Multinomial logistic regression	Three or more unordered categories	Analyze > Regression > Multinomial Logistic	Party voted for

A quick note on the sources I see cited most often for this table. UCLA’s OARC statistical consulting pages are genuinely the best free walkthrough of the mechanics of ordinal logistic regression you will find anywhere online. What they do not tell you is what to do when the model fails its own assumption on a 150-respondent convenience sample, which is most of you reading this. That gap is exactly where a lot of dissertations stall.

Common dependent variables in political science dissertations

Party identification, trust in government, political efficacy, voting intention, and policy attitude scales cover a large share of dissertation DVs I have worked with across both US and UK students. Notice almost none of them are naturally continuous.

Cross-Tabulation Analysis for Political Attitudes

Running and reading chi-square and Cramer’s V in SPSS

Cross-tabulation is where almost every political science dissertation statistics chapter should start, even if a regression comes later. In SPSS: Analyze > Descriptive Statistics > Crosstabs, add your two categorical variables, tick Chi-square and Cramer’s V under Statistics.

Cross-tabulation, put simply, is a table that shows how two categorical variables relate to each other by counting how often each combination of categories occurs. If your variables are stored differently, what is tabular data is worth a quick read before you build the table. Chi-square tells you whether the relationship is statistically significant, and Cramer’s V tells you how strong it actually is, which the chi-square p-value alone does not.

Worked example: party ID and a policy attitude

Say you cross-tabulate party identification against support for a specific policy in a sample of 180 respondents. A significant chi-square with a Cramer’s V of 0.35 tells a different story than the same significant chi-square with a Cramer’s V of 0.08. Pearson versus Spearman correlation is useful here if any of your variables are actually ordinal pairs rather than nominal, since correlation can sometimes say more than a cross-tab.

Ordinal Regression for Categorical Political Attitudes

Why ordinal regression beats linear regression for Likert DVs

Ordinal regression is a statistical technique used when your dependent variable has three or more categories with a natural order, and it avoids falsely assuming the gap between “somewhat likely” and “very likely” equals the gap between “unlikely” and “somewhat likely.” That single assumption is exactly where linear regression quietly fails on attitude data.

This is the core of ordinal regression political science work, and it is the test I recommend most often once cross-tabs have set up the groundwork.

Running PLUM in SPSS, step by step

1. Go to Analyze > Regression > Ordinal.
2. Move your ordered attitude DV into the Dependent box.
3. Add your predictors (party ID, demographics, policy attitudes) as Factors or Covariates depending on their type.
4. Under Output, request Test of Parallel Lines.
5. Run the model and check the Model Fitting Information table first for overall significance.

To run a regression in SPSS for political science data, in short: pick Ordinal from the Regression menu, slot in your ordered DV and categorical predictors, and request the parallel lines test before you trust a single coefficient.

The proportional odds assumption, and what to do if it fails

SPSS calls this the Test of Parallel Lines. If that test comes back significant, your ordinal model has technically failed its core assumption, and most students panic at this point.

In practice, I usually recommend either running a generalised ordinal model outside SPSS’s default output or falling back to multinomial logistic regression and reporting honestly why. Hiding a failed assumption is the fastest way to lose marks at viva.

Interpreting coefficients and odds ratios for a dissertation audience

SPSS gives you log-odds coefficients by default, which mean very little to anyone reading your dissertation. Exponentiate them to get odds ratios, and report those instead, with a plain-language sentence explaining direction and size for every significant predictor.

Case Study: Analysing Voter Behaviour with a Public Opinion Dataset

The dataset and the research question

I worked with an International Relations MA student in the UK, I will call her Meera, whose dissertation examined whether media trust predicted voting intention. She had a 220-respondent survey and a five-point voting likelihood scale as her DV.

Cross-tab first, then the ordinal model

We started with a cross-tab of media trust category against voting intention, which showed a significant but modest relationship, Cramer’s V around 0.21. That told us a relationship existed but was not the whole story.

Moving to ordinal regression (PLUM), we controlled for age, education, and party ID. Media trust remained a significant predictor even after that control, with an odds ratio suggesting each step up in trust roughly doubled the odds of higher voting likelihood.

What made the committee happy

The parallel lines test passed, so we reported the model as is, with the cross-tab kept as supporting evidence in the appendix rather than dropped entirely. That sequencing, cross-tab first, then regression, then honest assumption checking, is what got the chapter through her viva without a single statistics-related query.

Reporting SPSS Results in Your Dissertation (US and UK Conventions)

APA style vs Harvard style reporting of regression tables

US dissertations generally expect APA 7th formatting for statistical reporting, with italicised test statistics and specific decimal conventions. UK dissertations more often expect Harvard-style referencing alongside a looser table format, though this varies by university, so check your handbook rather than assume.

IRB approval vs ethics committee approval

US institutions require IRB (Institutional Review Board) approval before you collect survey data involving human respondents. UK institutions use a university or departmental research ethics committee, and the paperwork, while similar in spirit, is rarely interchangeable. Mention whichever one your institution required, by name, in your methodology chapter.

Common mistakes examiners flag at the viva (UK) or defense (US)

In my experience, three mistakes repeat more than any others:

1. Running linear regression on an ordinal dependent variable.
2. Reporting a significant chi-square without mentioning effect size.
3. Not disclosing a failed proportional odds test.

All three are avoidable with a five-minute assumption check before you write a single results paragraph.

When to Get Expert SPSS Help vs Doing It Yourself

Red flags you are in over your head

A few signs you’re past the point where self-teaching helps:

1. Your proportional odds test keeps failing and you don’t know why.
2. Your supervisor has asked you to justify your sample and you’re not sure how.
3. Your deadline is under three weeks and your results chapter is still blank.

Is convenience sampling defensible in a dissertation and the sample size calculator are both worth checking before that conversation with your supervisor. If your sampling approach is still undecided, stratified vs cluster sampling is also worth reading before you lock your survey design, since it affects which tests are defensible later.

What a 12-year dissertation statistics consultant actually does differently

I do not just run the test and hand back output. I check whether your research question matches your test, whether your sample supports the claims you want to make, and whether your write-up will survive a sceptical examiner’s questions. That last part is the one most SPSS tutorials, free or paid, never cover at all.

To be clear, this is guidance on choosing and running the right test and defending it, not someone writing your dissertation for you. That distinction matters for your academic integrity, and it’s one I’m explicit about with every student I work with.

Frequently Asked Questions

Author: Siddharth Gupta, 12-year dissertation statistics consultant. LinkedIn