Data Analysis in Research: Methods, Steps and Examples

In a research context, data analysis means applying defined, transparent procedures to a dataset in order to describe it, identify patterns or relationships, and draw conclusions that relate back to your research questions or hypotheses. A useful working definition: data analysis is the step that converts observations into evidence.

When people ask what it means to analyze data, they usually mean three linked activities:

• Data processing: entering, coding, transforming and cleaning raw data so it is in a usable form (for example, turning survey responses into numeric codes or transcribing interviews).
• Analysis of data: applying statistical tests, models or coding frameworks to that prepared data.
• Data interpretation: explaining what the results mean for the research question, how they compare with prior studies, and what their limits are.

The distinction between data processing and data analysis matters because many errors happen before any test is run. A perfectly chosen regression model will still mislead if missing values were coded as zero or if duplicate cases were left in the file.

Data analysis is also different from simply reporting data. A table of raw survey responses is reporting. Stating that respondents in group A scored on average three points higher than group B, that the difference is unlikely to be due to chance, and that this supports your hypothesis is analysis and interpretation.

There are several kinds of data analysis, and most research projects use more than one. The table below summarizes the different types of data analysis you will meet in academic research.

Descriptive and inferential analysis are the two you will use most often in a thesis. If the difference between them is unclear, see our guide to descriptive vs inferential statistics.

Quantitative data analysis, also called statistical data analysis, works with numeric data: test scores, survey scales, counts, measurements and coded categories. The core data analysis methods are:

• Descriptive statistics: measures of central tendency (mean, median, mode) and spread (standard deviation, range, interquartile range).
• Comparison of groups: independent and paired t-tests for two groups, ANOVA for three or more, and nonparametric alternatives such as Mann-Whitney U and Kruskal-Wallis when assumptions fail.
• Association between categorical variables: chi-square test of independence.
• Relationships between continuous variables: Pearson or Spearman correlation.
• Prediction and explanation: linear regression for continuous outcomes, logistic regression for binary outcomes, multilevel models for nested data.
• Latent constructs: factor analysis and structural equation modeling for questionnaires that measure concepts such as motivation or trust.

A worked example

Suppose a small pilot study compares quiz scores (out of 20) for students taught with two methods. Method A: 12, 14, 15, 16, 18. Method B: 10, 11, 13, 14, 12.

1. Describe. Mean A = 75 / 5 = 15.0; mean B = 60 / 5 = 12.0. Squared deviations sum to 20 for A and 10 for B, so the sample variances are 20 / 4 = 5.0 and 10 / 4 = 2.5, giving standard deviations of 2.24 and 1.58.
2. Choose a test. Two independent groups and a continuous outcome point to an independent-samples t-test.
3. Compute. Pooled variance =(20 + 10) / (5 + 5 − 2) = 3.75. Standard error = √(3.75 × (1/5 + 1/5)) = √1.5 = 1.22. t = (15.0 − 12.0) / 1.22 = 2.45, with df = 8.
4. Decide. The two-tailed p value is about .040, below the conventional .05 threshold, so the difference is statistically significant.
5. Quantify the effect. Cohen's d = 3.0 / √3.75 = 1.55, a large effect, though with only five students per group the estimate is very imprecise.

Note step 5: a good analysis does not stop at the p value. Effect sizes and confidence intervals tell readers how large and how certain a result is. For more on this test, read the t-test explained.

Qualitative analysis works with non-numeric data such as interview transcripts, focus group recordings, open-ended survey answers, field notes, documents and images. Instead of testing hypotheses, it aims to interpret meaning, experience and process. Common approaches include:

• Thematic analysis: identifying, reviewing and defining patterns of meaning (themes) across a dataset. It is flexible and the most common starting point for students. See our thematic analysis guide.
• Qualitative content analysis: systematically categorizing text, sometimes counting how often categories occur.
• Grounded theory: building a theory from the data through iterative coding and constant comparison.
• Narrative analysis: examining how people structure and tell stories about their experiences.
• Discourse analysis: studying how language constructs social reality, power or identity.
• Framework analysis: charting data into a matrix of predefined and emerging themes, common in applied health and policy research.

How coding works in practice

Imagine an interview excerpt: "I only check my email after the kids are asleep, and by then I'm too tired to reply properly." A first-cycle code might be work after caregiving hours. A second code might be fatigue affecting quality. After coding all transcripts, you might group these and related codes into a theme such as "Work squeezed into leftover time." The analysis then explains this theme, shows representative quotes, and relates it to the research question.

Rigor in qualitative analysis comes from transparency: a clear codebook, an audit trail of decisions, reflexive notes on your own position, and, where appropriate, a second coder or member checking.

Whatever your field, analyzing data follows a predictable sequence. Skipping a step usually means returning to it later at greater cost.

1. Restate the research questions and plan the analysis. Before touching data, write down each question or hypothesis, the variables or data sources that address it, and the technique you intend to use. This analysis plan protects you from testing everything and reporting only what is significant.
2. Prepare the data. Import files, assign variable names and labels, code categories consistently, transcribe recordings and anonymize identifying details.
3. Clean the data. Check for impossible values (an age of 210), duplicates, inconsistent spellings and missing data. Record every change in a log or script so it can be reproduced.
4. Explore the data. Run descriptive statistics and plots, or read transcripts in full and note first impressions. Exploration reveals outliers, skewed distributions and unexpected patterns.
5. Analyze the data. Apply the planned tests, models or coding procedures. Check assumptions (normality, equal variances, independence, linearity) and use alternatives if they are violated.
6. Interpret the results. Ask what each result means for the research question, how large and precise the effects are, whether alternative explanations exist, and how the findings fit existing literature.
7. Report the results. Present findings in tables, figures and text following your discipline's style guide, and keep interpretation in the discussion where your field expects it.

Steps 2 to 4 often take longer than the analysis itself. Budget time for them in your project plan.

The right technique follows from three things: your research question, the type of data you have, and your study design. Use the questions below as a decision path.

• Is the question about description, difference, relationship or prediction? Description needs descriptive statistics; differences need group comparisons; relationships need correlation or chi-square; prediction needs regression.
• What is the measurement level of each variable? Nominal (categories without order), ordinal (ordered categories such as Likert items), interval or ratio (continuous numbers).
• How many groups or variables are involved, and are observations independent? Repeated measurements on the same people require paired or repeated-measures methods.
• Are assumptions met? Small samples and strongly skewed data often call for nonparametric tests.

For a fuller decision guide covering more designs, see choosing a statistical test.

Software does not choose your method, but it shapes how efficiently and reproducibly you work. Common programs for data analysis in academic research include:

Whichever tool you use, work from a script or syntax file rather than only clicking menus. A script documents every step, lets you rerun the analysis after a data correction, and makes your methods section easier to write. If you use Stata, our Stata commands cheat sheet covers the commands researchers use most.

In a thesis, dissertation or journal article, the data analysis section usually sits at the end of the methodology chapter. Its job is to explain what you did to the data and why, in enough detail that another researcher could repeat it. Results belong in the next chapter.

What to include

• The software and version used.
• How data were prepared and cleaned, including how missing data and outliers were handled.
• The descriptive statistics reported.
• Each inferential test or model, linked to the research question or hypothesis it addresses.
• How assumptions were checked and what you did if they were violated.
• The significance level and any correction for multiple comparisons.
• For qualitative work: the analytic approach, coding process, how themes were developed, and the steps taken to ensure trustworthiness.

A template paragraph

Data were analyzed in [software, version]. Cases with more than [X]% missing responses were excluded, leaving [n] participants; remaining missing values were handled using [method]. Descriptive statistics (means, standard deviations and frequencies) were calculated for all study variables. To address RQ1, [test] was used to compare [outcome] between [groups]. Assumptions of [normality / homogeneity of variance] were assessed using [method]; where violated, [alternative] was used. RQ2 was examined using [model] with [outcome] as the dependent variable and [predictors] as independent variables. Statistical significance was set at α = .05, and effect sizes are reported as [measure].

Adapt the template to your study rather than filling blanks mechanically. For the rest of the chapter, see writing a methodology chapter.

The three short examples below show how the same process plays out in different designs. They are invented for illustration.

Example 1: Quantitative survey study

Question: Does perceived workload predict burnout among graduate teaching assistants? Data: An online survey of 180 assistants with validated scales for workload and burnout, plus hours taught per week. Analysis: After removing incomplete responses and checking scale reliability with Cronbach's alpha, the researcher reports means and standard deviations, examines a correlation matrix, then fits a multiple linear regression with burnout as the outcome and workload and hours as predictors. Residual plots are used to check linearity and constant variance. Interpretation: The write-up reports the regression coefficients with confidence intervals and R², and discusses whether workload perception matters more than actual hours.

Example 2: Qualitative interview study

Question: How do first-generation students describe their sense of belonging in their first semester? Data: Fifteen semi-structured interviews, transcribed verbatim. Analysis: Reflexive thematic analysis: repeated reading, line-by-line coding, grouping codes into candidate themes, reviewing themes against the full dataset, and naming them. Interpretation: Three themes are presented with illustrative quotes and connected to belonging theory, with a reflexive note on the researcher's own background.

Example 3: Mixed methods evaluation

Question: Did a peer-mentoring program improve retention, and why? Data: Enrollment records for participants and non-participants, plus focus groups with mentees. Analysis: A chi-square test compares retention rates; logistic regression adjusts for prior grades; thematic analysis of focus groups explains mechanisms. Interpretation: The discussion integrates both strands, using the qualitative themes to explain the quantitative pattern.

Good data analysis skills are less about memorizing tests and more about judgment. The skills that matter most in research are:

• Statistical reasoning: understanding sampling, variability, effect size and uncertainty.
• Data management: organizing files, documenting variables and keeping a reproducible workflow.
• Software fluency: enough command of one program to run, check and rerun analyses.
• Critical interpretation: separating what the data show from what you hoped they would show.
• Clear reporting: presenting results accurately in tables, figures and prose.

Mistakes to avoid

• Choosing the test before checking measurement levels and assumptions.
• Running many tests and reporting only significant ones, which inflates false positives.
• Treating a non-significant result as proof that no effect exists.
• Inferring causation from correlational data.
• Reporting p values without effect sizes or confidence intervals.
• Presenting qualitative themes as lists of topics without analytic interpretation.
• Failing to describe how missing data were handled.

If you want a second pair of eyes on your analysis plan or output, our statistics help service reviews methods and explains results. We coach, check and explain; you stay the author of your work.