Statistical Analysis for a Master’s Thesis: The Complete Roadmap

By the Mastermind PhD team · Last updated: 3 November 2026 · Editorial policy

Why do two students with the same questionnaire and the same kind of sample end up with two very different results chapters, one accepted at the first defense and the other sent back three times? The difference is rarely the software or the student’s ability. It is whether a clear path connects each statistical decision to the one before and the one after: the research question determines the test, the test determines the measurement scale and the sample size, and correct data entry determines whether every later table can be trusted.

This article is part of our guide How to Write a Master’s Thesis by the Mastermind PhD team and relates to our Statistical Analysis service. Sources last checked: November 2026.

This guide draws the path of statistical analysis for a thesis in seven stages, from the moment you write your research questions to the last table in the results chapter, and points to a detailed article at each stage where one exists. Read it once before you collect data, and return to it at every stage.

Stage 1: Before data collection, where most statistical decisions are made

In statistical analysis for a thesis, the costliest errors happen before SPSS is opened. Three decisions taken in the proposal determine everything that follows.

1. Write questions and hypotheses that imply a test. The verb in the question chooses the test:

If a question has no obvious test (“What is the role of X in Y?”), rewrite it now; the committee will ask which table answers it.

2. Justify the sample size. Use the Krejcie and Morgan (1970) table for a population of known size (380 for a population of 40,000, for example), a power analysis in G*Power from the expected effect size and the desired power (usually .80), or both. Multiple regression needs at least 50 + 8 times the number of predictors (Green, 1991), and each group in a comparison should ideally have 30 or more cases. State the method and the reference in the methodology chapter.

3. Design an instrument that yields analyzable data. Each construct measured by at least four items, a symmetric Likert format, a small number of reverse-worded items whose numbers you record in advance, and demographic variables with categories balanced enough to compare: a category of “more than 20 years” that will hold six people is a problem you can prevent at this stage.

Stage 2: Data entry, coding and cleaning

Data that go in wrong come out wrong whatever the test. The order of work in statistical analysis for a thesis:

1. A codebook linking every item to a variable, a short name, its permitted values and a missing-value code.
2. Define the variables in Variable View before entering any number: numeric type, the item text as the label, value labels, a missing code (9 or 99) and the correct measurement level (nominal, ordinal, scale).
3. Entry or import. Give each questionnaire an ID that matches the paper copy or the online response number. Import from Excel or Google Forms through File, then Import Data, after cleaning the column names.
4. Cleaning. Run frequencies on every variable to find out-of-range values (a 55 among values from 1 to 5), check incomplete questionnaires, and exclude those missing more than a stated threshold (commonly 10% to 20% of items), documenting how many were removed.
5. Reverse-code negative items into new variables with an R suffix, keeping the originals.
6. Compute scale scores with the MEAN function rather than SUM, so that scores stay on the original 1-to-5 metric and partially missing cases are handled consistently.

📱 Ask a statistician about your data — Send a short description of your instrument and sample size on WhatsApp and get a suitable analysis plan. Every analysis comes with the raw SPSS file and a results report.

Stage 3: Reliability and validity

This stage of statistical analysis for a thesis is reported in the methodology chapter. It is computed first on the pilot sample (30 or more cases) and repeated on the final sample.

• Face and content validity. Expert review by five to ten judges, an agreement index or a content validity ratio, and a table of changes made.
• Internal consistency validity. Pearson correlations between each item and its scale score, and between each scale and the total; non-significant items are reviewed.
• Construct validity. Exploratory factor analysis for new instruments (KMO above .60, a significant Bartlett’s test, loadings above .40), or confirmatory factor analysis in AMOS for instruments built on an established theory.
• Reliability. Cronbach’s alpha for every scale and for the whole instrument: .70 or higher is the conventional threshold (Nunnally, 1978), and a value between .60 and .69 on a short scale needs a stated justification. Test-retest reliability where the design allows.

Compute alpha only after reverse-coding; a negative item left unreversed can drag alpha below .50 and send you hunting for a problem that does not exist.

Stage 4: Descriptive statistics

The results chapter opens with two tables: a sample profile (frequencies and percentages of the demographic variables) and a description of the study variables (mean, standard deviation and, where used, rank and verbal level for each scale, ordered from highest to lowest, with items inside each scale if required). Where an interpretation key is expected, derive it from the class interval, (5 − 1) ÷ 5 = 0.80, state it in the methodology and apply it without rounding toward a nicer label. UK and most EU programs do not expect a key and are content with means and standard deviations against the scale range.

In the descriptive stage of statistical analysis for a thesis, the text that accompanies a table does not repeat its numbers. It connects the result to the research question, highlights the highest and lowest items, and comments on a large standard deviation (above about 1.20 on a 5-point scale) as a sign of divided responses worth discussing.

Stage 5: Checking the assumptions

Before any inferential test in statistical analysis for a thesis, three things are checked and reported in a short paragraph or table:

• Normality of the scale scores within each group (or of the residuals in regression): Shapiro-Wilk and Kolmogorov-Smirnov together with skewness, kurtosis and plots. With samples above about 100, give more weight to the indices and plots than to the test’s p value, because the tests reject normality for trivial departures.
• Homogeneity of variance: Levene’s test. If it fails, read the Welch row in the t-test output, or use Welch’s ANOVA with Games-Howell post hoc comparisons.
• Outliers: boxplots and standardized scores beyond ±3.29. Correct data-entry errors first; then document any exclusion.

In statistical analysis for a thesis, the two-out-of-three rule works well: treat the distribution as acceptable if two of the following hold: indices within the threshold, plots without marked asymmetry, and a non-significant Shapiro-Wilk. If normality genuinely fails, the option examiners accept most readily is the matching non-parametric test; the second is to rely on the robustness of the parametric test when every group exceeds 30; the third is bootstrapping.

Stage 6: Hypothesis tests and effect sizes

Answer every question with one appropriate table, following the selection table in Stage 1 of statistical analysis for a thesis. The tests that appear most often in master’s theses:

Rules that apply to every test in statistical analysis for a thesis: set the alpha level before the analysis (usually .05); write p to two or three decimals and p < .001 rather than .000; report an effect size with every significant result and interpret it in practical terms; run post hoc tests only after a significant F; and never run the parametric and non-parametric versions and keep the “better” one.

For complex models in statistical analysis for a thesis (latent variables, mediation, moderation) use structural equation modeling: covariance-based SEM in AMOS for confirmatory models with large samples, or PLS-SEM in SmartPLS for exploratory models and smaller samples, reporting fit indices and path coefficients instead of the tables above.

Stage 7: Writing the results chapter

The structure examiners expect when statistical analysis for a thesis is written up:

1. Introduction: the purpose of the chapter, the software and version, the tests used and the alpha level.
2. Sample profile in one table.
3. Reliability and validity on the final sample, or a cross-reference to the methodology chapter.
4. Assumption checks in a paragraph or compact table.
5. Results for each question in turn: the question restated, the APA table with its number and title above it, an interpretation paragraph that opens with the main finding and the effect size, then a brief comment on patterns inside the table.
6. A summary table linking each question and hypothesis to the decision (supported or not supported) and the key statistic.

Templates speed up writing up the statistical analysis for a thesis. For descriptives: “Job satisfaction was high (M = 3.62, SD = 0.71); the highest-rated item was …” For differences: “An independent samples t-test showed that women (M = 3.49, SD = 0.64) reported higher burnout than men (M = 3.21, SD = 0.78), t(274.8) = 3.35, p = .001, d = 0.39.” For relationships: “Emotional intelligence was moderately and positively correlated with performance, r(298) = .42, p < .001.” APA 7 tables carry the number in bold above the italic title, horizontal rules only, two decimals, and a note below for symbols and probability markers.

What belongs in the discussion chapter after statistical analysis for a thesis?

The discussion is not a second results chapter. For each main finding it makes three moves: what you found, in a sentence; which earlier studies it agrees or disagrees with and why (sample characteristics, context, instrument); and what it means theoretically and practically. Then the limitations, stated honestly: a cross-sectional design that cannot establish causation, a sample from one region, self-report measures, a small effect despite statistical significance. Recommendations follow from significant results and meaningful effect sizes, not from general opinion, and future research addresses the limitations.

Worked example: one thesis through the seven stages

To see how the stages of statistical analysis for a thesis connect, follow one titled “Transformational leadership and organizational commitment among secondary-school teachers”:

• Before data collection. Four questions: two descriptive (level of leadership, level of commitment), one relationship (Pearson), one effect (multiple regression with four leadership dimensions), plus a differences question by gender and experience (t-test and ANOVA). The population is 4,200 teachers, so the Krejcie and Morgan sample is 351, raised to 400 distributed questionnaires to allow for a 12% non-return. The instrument has a four-dimension leadership scale (20 items) and a commitment scale (12 items) on a 5-point format, with two reverse-worded items recorded in advance.
• Data entry. 362 questionnaires returned; 14 excluded for more than 15% missing items; 348 retained. The codebook is ready, the two negative items are reversed into new variables, and dimension scores are computed with MEAN.
• Reliability and validity. On a pilot of 40 teachers, item-to-scale correlations range from .46 to .81; Cronbach’s alpha ranges from .78 to .90 across dimensions and is .94 for the whole instrument. The values are recomputed on the final sample with similar results.
• Descriptives. Transformational leadership M = 3.71 (high) and commitment M = 3.58 (high), with dimensions and items presented in rank order.
• Assumptions. Skewness and kurtosis of the scale scores within ±1; Q-Q plots acceptable although Shapiro-Wilk is significant because of the sample size; Levene’s test non-significant for gender; VIF between 1.9 and 3.6 in the regression.
• Tests. Pearson’s r between leadership and commitment = .55 (large by Cohen’s benchmarks); the regression explains 34% of the variance in commitment, with inspirational motivation the strongest predictor; no significant gender difference (d = 0.11); a significant difference by experience with a small effect (η² = .03), favoring the most experienced group after Tukey’s test.
• Results chapter. Six APA tables matching the questions in order, a summary table linking each hypothesis to its decision, and a discussion that explains the absence of a gender difference against two earlier studies and comments on the small size of the experience effect despite its significance.

Every decision in this thesis can be traced to a specific question and a specific reference. That is what sound statistical analysis for a thesis means in practice.

Ten reasons examiners send a results chapter back

1. A research question with no table, or a table with no research question.
2. Single Likert items compared with t-tests and ANOVA instead of scale scores.
3. Cronbach’s alpha computed before reverse-coding the negative items.
4. No assumption checks, or normality tested for every item instead of the scale scores.
5. p = 0.000 and no effect sizes.
6. The first row of the t-test output read regardless of Levene’s test.
7. Post hoc tests after a non-significant F, or three post hoc tests with the favorable one reported.
8. Regression without a multicollinearity check, or with a sample smaller than 50 + 8 times the number of predictors.
9. An interpretation key copied from a thesis that used a different scale.
10. Interpretation that repeats the table in words, does not answer the question, or turns a correlation into a cause.

Which software suits statistical analysis for a thesis?

In statistical analysis for a thesis, the software is a tool. What the committee evaluates is whether the test fits the question, whether the assumptions were checked, and whether the interpretation is sound.

If you work with a statistician on statistical analysis for a thesis: what to ask for

Most universities allow statistical support on the condition that you understand every table and can defend it, and the analytical decisions remain yours. Before work starts, ask to be told the specialist’s field and degree, and ask for a meeting to go through your faculty’s requirements: table format, alpha level, mandated software.

On delivery, ask for the raw data file in SPSS format (.sav) together with the syntax file, a report that links each research question to its table and interprets the effect size, and a session that walks you through the tables before the defense. The Mastermind PhD team provides the raw SPSS file with every analysis, tells you the specialist’s field and degree before work starts and schedules a meeting, works on two-stage payment (80/20), and revises free of charge until supervisor approval within 30% of the chapter.

📱 Ask on WhatsApp now — 5 years of experience across all disciplines and 14 academic services. Two-stage payment (80/20); revisions are free until supervisor approval, within 30% of the chapter.

Frequently asked questions

How many chapters of the thesis depend on the statistical analysis?

At least three: the methodology (sample, instrument, reliability, validity and planned tests), the results (tables and interpretation) and the discussion (meaning in the light of earlier studies). The key figures also appear in the abstract and the recommendations.

Should I start the analysis after all the data are collected or during collection?

Prepare the SPSS file and the codebook before collection, and enter the first 30 to 40 questionnaires as a pilot to compute reliability and validity and adjust the instrument if needed. The final inferential analysis is run once the full sample is in and cleaned.

How long does the statistical analysis of a master’s thesis take?

Once entry and cleaning are complete, a questionnaire with three to five scales and the usual difference and relationship questions typically takes three to seven working days of analysis and reporting. Structural equation modeling, incomplete data or repeated rounds of supervisor changes can extend this to two weeks or more.

Do examiners accept non-parametric tests in a master’s thesis?

Yes, and many prefer them when the assumptions of parametric tests are not met, provided the choice is justified by the assumption checks and the results are reported with medians and an appropriate effect size.

What is the difference between reliability and validity, and why are both required?

Reliability is consistency: the instrument gives the same results when repeated. Validity is whether the instrument measures what it claims to measure. A reliable instrument can consistently measure the wrong thing, so both are required and each is established in a different way.

Sources

• Field, A. (2018). Discovering Statistics Using IBM SPSS Statistics (5th ed.). SAGE.
• Pallant, J. (2020). SPSS Survival Manual (7th ed.). Open University Press.
• Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences (2nd ed.). Lawrence Erlbaum.
• Nunnally, J. C. (1978). Psychometric Theory (2nd ed.). McGraw-Hill.
• Green, S. B. (1991). How many subjects does it take to do a regression analysis? Multivariate Behavioral Research, 26(3), 499–510. https://doi.org/10.1207/s15327906mbr2603_7
• Krejcie, R. V., & Morgan, D. W. (1970). Determining sample size for research activities. Educational and Psychological Measurement, 30(3), 607–610.
• American Psychological Association. (2020). Publication Manual of the American Psychological Association (7th ed.); Journal Article Reporting Standards: Quantitative. https://apastyle.apa.org/jars/quantitative
• IBM. IBM SPSS Statistics documentation. https://www.ibm.com/docs/en/spss-statistics

Sources last checked: November 2026. This article is general educational guidance and does not replace your university regulations or your supervisor’s instructions. How we prepare our content.