Data Analysis – Definition, Types, Methods, Steps and Examples

Data analysis is the systematic process of preparing, examining, transforming, and interpreting data to answer questions, identify patterns, test explanations, and support conclusions. It can involve statistical calculations, qualitative coding, visualization, or computational models. A sound analysis matches the research question, study design, type of data, and assumptions of the chosen method.

Introduction

Collecting data does not automatically produce knowledge. Survey responses, interview recordings, laboratory measurements, documents, observations, and administrative records must be organized and examined before they can answer a research question.

Data analysis provides the connection between the evidence a researcher collects and the conclusions presented in a dissertation, journal article, policy report, or professional decision. It includes more than running a statistical test or generating a chart. Researchers must decide what the data represent, evaluate their quality, select appropriate methods, interpret uncertainty, consider alternative explanations, and communicate the findings transparently.

This guide explains:

• what data analysis means;
• the main types and methods of analysis;
• how qualitative, quantitative, and mixed-methods analysis differ;
• the steps in a defensible analytical workflow;
• how to select an appropriate method;
• how researchers handle missing values, outliers, assumptions, and bias;
• which digital tools are used;
• how artificial intelligence can assist without replacing human judgment;
• how results should be interpreted, reported, and preserved.

Key takeaways

• Data analysis begins with a research question, not with a software package.
• The correct method depends on the study design, data type, measurement level, sampling process, and intended conclusion.
• Qualitative analysis interprets meanings and experiences; quantitative analysis examines numerical patterns; mixed-methods analysis intentionally integrates both.
• Statistical significance alone is not enough. Effect sizes, confidence intervals, assumptions, uncertainty, and practical importance also matter.
• Reproducible analysis requires documentation, versioned code or decision records, transparent data preparation, and clearly reported limitations.
• AI can assist with code, summaries, transcription, and exploratory work, but its output must be verified and confidential data must be protected.

What Is Data Analysis?

Data analysis is the organized examination of data for the purpose of describing what was observed, answering a research question, evaluating a hypothesis, developing an interpretation, making a prediction, or informing a decision.

A complete analysis may include:

1. defining the question;
2. documenting the data and variables;
3. checking data quality;
4. cleaning or coding the data;
5. exploring patterns;
6. selecting an analytical method;
7. calculating, categorizing, or modeling;
8. evaluating assumptions and uncertainty;
9. interpreting the findings;
10. reporting and preserving the analytical record.

The process is usually iterative rather than perfectly linear. A graph may reveal an error that requires additional cleaning. An early qualitative code may require the codebook to be revised. A model diagnostic may show that a different specification is needed. Researchers should document these decisions rather than hide them.

A simple example

Suppose a university surveys students about online learning.

The raw dataset contains:

• age and programme of study;
• hours spent in online classes;
• a satisfaction score from 1 to 5;
• examination results;
• an open-ended question about learning difficulties.

The researcher might:

• calculate the distribution of satisfaction scores;
• compare average examination results across different levels of online participation;
• use regression to examine whether participation predicts results after accounting for programme and prior achievement;
• code the open-ended answers to identify recurring difficulties;
• integrate the statistical and qualitative findings.

The calculations, coding, interpretation, and integration are all parts of data analysis.

What Is the Purpose of Data Analysis?

The purpose of data analysis is to turn collected evidence into defensible answers. It helps researchers describe observations, identify patterns, compare groups, examine relationships, test explanations, estimate uncertainty, develop theory, make predictions, and communicate what the evidence does and does not support.

More specifically, analysis can help a researcher:

• summarize a dataset;
• assess the distribution of a variable;
• compare people, groups, locations, treatments, or time periods;
• identify associations among variables;
• examine whether an intervention produced a change;
• interpret participants’ experiences or language;
• discover unexpected patterns;
• evaluate a theoretical model;
• predict an outcome;
• combine evidence from different methods;
• identify uncertainty and limitations;
• support a practical or policy decision.

Analysis does not make poor data reliable. A sophisticated model cannot repair a severely biased sample, an invalid measure, an uncontrolled study, or inaccurate records. Analytical quality therefore depends on the complete research design.

Data Analysis, Data Analytics, Statistics, Data Science, and Data Interpretation

These terms overlap, but they are not identical.

Term	Main focus	Typical activities
Data analysis	Examining data to answer questions and develop conclusions	Cleaning, coding, summarizing, modeling, visualizing, interpreting
Data analytics	Applying analytical processes, often repeatedly, to support organizational decisions	Dashboards, performance monitoring, forecasting, optimization
Statistics	Principles and methods for learning from variable data under uncertainty	Estimation, hypothesis testing, sampling, probability models
Data science	A broader computational field combining data, programming, statistics, and domain knowledge	Data engineering, machine learning, modeling, deployment
Data interpretation	Explaining what analytical findings mean in context	Connecting results to questions, theory, prior evidence, and limitations

A thesis may use statistics as part of its data analysis and then interpret the statistical results in relation to its research question. An organization may automate similar analyses as part of a data-analytics system. A data-science project may additionally build and deploy a prediction model.

Main Types of Data Analysis

There is no single universally correct list of data-analysis types. Different classifications answer different questions.

One classification focuses on the form of data and methodological tradition:

• qualitative analysis;
• quantitative analysis;
• mixed-methods analysis.

Another classification focuses on the purpose of the analysis:

• descriptive;
• exploratory;
• inferential;
• diagnostic;
• predictive;
• causal;
• prescriptive.

These categories can overlap. A study may conduct exploratory quantitative analysis, descriptive qualitative analysis, or predictive analysis using both text and numerical variables.

Qualitative, Quantitative, and Mixed-Methods Data Analysis

Qualitative data analysis

Qualitative data analysis is the systematic interpretation of non-numerical material, such as interview transcripts, documents, images, observations, field notes, audio, or video, to understand meanings, experiences, processes, language, or social contexts.

Common qualitative approaches include:

• thematic analysis;
• qualitative content analysis;
• framework analysis;
• grounded theory analysis;
• narrative analysis;
• discourse analysis;
• conversation analysis;
• interpretative phenomenological analysis;
• ethnographic analysis;
• case-study analysis.

Qualitative data are not scientifically inferior to numerical data. Their appropriateness depends on the research question. An interview study may be better than a questionnaire for understanding how patients experience a new treatment, how teachers interpret a policy, or how employees negotiate professional identity.

Typical qualitative workflow

1. Prepare and securely store the material.
2. Transcribe or organize it where necessary.
3. Read, watch, or listen repeatedly to become familiar with it.
4. Record initial observations and reflexive notes.
5. Create codes that identify relevant features.
6. Compare codes across cases and contexts.
7. Develop categories, themes, narratives, or theoretical concepts.
8. review patterns against the complete dataset.
9. examine contradictions and negative cases.
10. interpret the findings in relation to the question and methodological position.
11. support claims with appropriate extracts or examples.
12. preserve an audit trail of analytical decisions.

In reflexive thematic analysis, themes do not simply appear independently of the researcher. The analyst actively interprets patterns in relation to the research question, theoretical position, and context (Braun & Clarke, 2006).

Quantitative data analysis

Quantitative data analysis uses numerical information and statistical or mathematical methods to describe variables, estimate population characteristics, compare groups, examine relationships, test models, or predict outcomes.

Common techniques include:

• frequencies and percentages;
• mean, median, and mode;
• range, variance, and standard deviation;
• confidence intervals;
• cross-tabulation;
• correlation;
• t tests;
• analysis of variance;
• chi-square tests;
• linear and logistic regression;
• nonparametric tests;
• time-series analysis;
• survival analysis;
• multilevel modeling;
• factor analysis;
• structural equation modeling;
• machine-learning models.

Quantitative analysis is not defined solely by the use of software or a p-value. It requires attention to measurement, sampling, assumptions, uncertainty, effect magnitude, model fit, and the design that produced the data.

Mixed-methods data analysis

Mixed-methods analysis combines and intentionally integrates qualitative and quantitative evidence so that the combined interpretation answers a question more fully than either form of evidence could answer alone.

Merely placing a survey beside a set of interviews does not produce meaningful integration. Researchers must explain how the two components relate.

Common integration strategies include:

• merging: comparing qualitative and quantitative results;
• connecting: using results from one phase to select participants or questions for another;
• building: using one form of evidence to develop an instrument, intervention, or model;
• embedding: placing a secondary method within a larger primary design;
• joint displays: presenting both forms of evidence in one table, figure, or analytical framework;
• data transformation: converting one form of evidence into another where methodologically defensible.

Mixed-methods researchers should identify agreements, contradictions, expansions, and silences across the components. The final product is an integrated conclusion or meta-inference, not two unrelated results sections (Fetters et al., 2013).

Types of Data Analysis by Analytical Purpose

Descriptive analysis

Descriptive analysis summarizes what is present in the data without extending the conclusion beyond the observed dataset.

Examples include:

• percentages of participants in each category;
• mean or median examination scores;
• the range of temperatures recorded;
• a table of interview-participant characteristics;
• a qualitative summary of the subjects discussed in policy documents.

For a numerical variable (x), the arithmetic mean is:

x̄ = Σxᵢ / n

where (xᵢ) represents each observed value and (n) is the number of observations.

Descriptive results are normally the first analytical layer, but they should not be treated as trivial. An inappropriate average or misleading graph can distort all later interpretation.

Exploratory data analysis

Exploratory data analysis examines data openly to identify distributions, unusual observations, possible relationships, patterns, and questions that deserve further investigation.

Typical techniques include:

• histograms;
• box plots;
• scatterplots;
• cross-tabulations;
• grouped summaries;
• correlation matrices;
• cluster exploration;
• close reading and preliminary coding;
• memo writing;
• visual inspection of time trends.

Exploration is valuable, but exploratory findings should be labelled honestly. A hypothesis developed after inspecting a dataset should not be presented as though it had been specified in advance. Confirmation normally requires new data, a held-out sample, or an explicitly separated validation process.

Inferential analysis

Inferential analysis uses sample data and probability-based reasoning to estimate or test conclusions about a wider population or data-generating process.

It includes:

• point estimates;
• confidence intervals;
• hypothesis tests;
• regression estimates;
• Bayesian posterior estimates;
• population comparisons;
• generalization from probability samples.

A simplified confidence interval has the form:

estimate ± critical value × standard error

The interval expresses uncertainty under the assumptions of the method. It is usually more informative than reporting a p-value alone because it shows the range of effect sizes compatible with the data and model.

Diagnostic analysis

Diagnostic analysis investigates why a pattern or event may have occurred.

It may involve:

• subgroup analysis;
• process tracing;
• residual analysis;
• root-cause diagrams;
• comparison of competing explanations;
• follow-up interviews;
• mediation or pathway analysis;
• examination of changes before and after an event.

Diagnostic analysis should not automatically be called causal. Observational data may identify plausible explanations without proving that one factor produced the outcome.

Predictive analysis

Predictive analysis estimates an unknown or future outcome for new observations.

Examples include:

• predicting student dropout;
• forecasting energy demand;
• classifying fraudulent transactions;
• estimating hospital readmission risk;
• predicting future sales.

Prediction requires evaluation on data not used to fit the model. A model that performs well on its training data may fail on new cases because of overfitting or data leakage.

Useful predictive metrics include:

• mean absolute error;
• root mean squared error;
• sensitivity and specificity;
• precision and recall;
• area under the receiver operating characteristic curve;
• calibration;
• out-of-sample accuracy.

High predictive accuracy does not explain why an outcome occurs and does not prove causation.

Causal analysis

Causal analysis estimates how an outcome would differ under one condition or intervention compared with a relevant alternative condition.

Randomized experiments can support causal conclusions when they are properly designed, implemented, and analyzed. Observational studies require additional assumptions and methods, such as:

• matching;
• weighting;
• interrupted time series;
• difference-in-differences;
• instrumental variables;
• regression discontinuity;
• sensitivity analysis for unmeasured confounding.

A correlation, regression coefficient, or predictive association is not automatically a causal effect. Causal language must be justified by the design and assumptions.

Prescriptive analysis

Prescriptive analysis recommends an action by combining evidence, predictions, objectives, constraints, costs, risks, and possible consequences.

Examples include:

• selecting an allocation strategy for limited medical resources;
• choosing an inventory level;
• optimizing a transport schedule;
• recommending an intervention under budget constraints.

Prescriptive results depend on value judgments and operational constraints as well as data. Researchers should make those assumptions visible.

The Data-Analysis Process

Step 1: Define the research question

State exactly what the analysis must answer.

A vague question such as “Does technology affect students?” should be refined:

Among first-year university students, is weekly use of a structured online practice platform associated with final examination performance after accounting for prior achievement?

The refined question identifies:

• the population;
• exposure or predictor;
• outcome;
• possible adjustment variable;
• intended relationship.

Step 2: Connect the question to the research design

Determine whether the study is:

• experimental or observational;
• cross-sectional or longitudinal;
• qualitative, quantitative, or mixed methods;
• based on primary or secondary data;
• exploratory or confirmatory;
• intended to describe, compare, explain, predict, or estimate a causal effect.

The same dataset may support some conclusions but not others. A cross-sectional survey can estimate associations at one time point, but it rarely establishes temporal order or causation.

Step 3: Prepare an analysis plan

A good analysis plan is written before the main analysis whenever possible. It should specify:

• research questions and hypotheses;
• variables or qualitative domains;
• inclusion and exclusion rules;
• coding decisions;
• primary and secondary outcomes;
• statistical models or qualitative procedures;
• handling of missing data;
• treatment of outliers;
• subgroup analyses;
• robustness checks;
• integration procedures for mixed methods;
• software;
• reporting outputs.

Confirmatory studies may preregister these decisions. Exploratory studies should still document them, even when the plan evolves.

Step 4: Inspect and document the data

Create or review:

• a data dictionary;
• variable names and labels;
• coding schemes;
• units of measurement;
• permitted values;
• source information;
• dates and versions;
• consent or access restrictions;
• transformations already performed.

For qualitative research, document transcription conventions, anonymization decisions, file organization, researcher roles, and the relationship between cases and data sources.

Step 5: Evaluate data quality and ethics

Check whether the data are:

• relevant to the question;
• sufficiently complete;
• internally consistent;
• accurately recorded;
• representative enough for the intended conclusion;
• collected using valid and reliable measures;
• legally and ethically usable;
• appropriately de-identified;
• protected against unauthorized access.

Data quality is not just a technical problem. A dataset can be clean but systematically biased because important people or events were excluded.

Step 6: Clean, code, and transform the data

Possible quantitative tasks include:

• removing exact duplicate records;
• correcting impossible values;
• standardizing dates and units;
• recoding categories;
• checking skip patterns;
• deriving variables according to documented rules;
• identifying missing values;
• detecting data-entry errors.

Possible qualitative tasks include:

• correcting transcripts against recordings;
• removing identifying information;
• defining units of analysis;
• organizing cases;
• developing initial codes;
• documenting changes to the codebook.

Never delete a value simply because it is inconvenient. Preserve the raw data separately and record every alteration.

Step 7: Conduct exploratory analysis

Before fitting a complex model:

• inspect distributions;
• calculate appropriate summaries;
• visualize relationships;
• examine missingness;
• look for impossible combinations;
• identify influential observations;
• review category sizes;
• compare groups descriptively;
• read qualitative material in context.

Exploration frequently reveals that an initially planned model is inappropriate or that additional data-quality work is required.

Step 8: Select and perform the main analysis

Choose a method based on:

• the research question;
• study design;
• outcome type;
• predictor type;
• independence or dependence of observations;
• sample size;
• distribution and model assumptions;
• measurement quality;
• intended population;
• whether the analysis is exploratory or confirmatory.

Software should implement the analytical decision, not make it on behalf of the researcher.

Step 9: Check assumptions and analytical quality

Depending on the method, assess:

• independence;
• distribution of residuals;
• linearity;
• homogeneity of variance;
• proportional hazards;
• multicollinearity;
• influential observations;
• model calibration;
• convergence;
• coding consistency;
• coherence of qualitative themes;
• adequacy of evidence for each interpretation;
• contradictory or negative cases.

An assumption check is not a ritual. Its purpose is to determine whether the chosen method produces meaningful estimates for the data and question.

Step 10: Assess robustness

Robustness or sensitivity analysis asks whether a conclusion changes under reasonable alternative decisions.

Examples include:

• using mean and median summaries;
• comparing complete-case and imputed analyses;
• fitting models with and without influential observations;
• using alternative variable definitions;
• applying parametric and nonparametric approaches;
• comparing coding interpretations;
• checking whether a theme appears across different cases or sources;
• evaluating alternative model specifications.

A finding that disappears under minor reasonable changes should be presented cautiously.

Step 11: Interpret the results

Interpretation should address:

• the direction of the finding;
• its magnitude;
• uncertainty;
• practical or theoretical importance;
• compatibility with the research question;
• alternative explanations;
• the role of design limitations;
• agreement or disagreement with prior evidence;
• implications for theory, practice, or future research.

A statistically significant result may be too small to matter. A nonsignificant result may be too imprecise to rule out an important effect. A common qualitative theme may not represent every participant’s experience.

Step 12: Communicate and preserve the analysis

A complete report should state:

• what was planned;
• what was changed and why;
• how the data were prepared;
• which methods were used;
• which assumptions were evaluated;
• how uncertainty was expressed;
• which software and relevant packages were used;
• how qualitative interpretations were developed;
• how mixed-methods findings were integrated;
• what limitations remain;
• where permitted data, code, or materials can be accessed.

Preserve the analysis scripts, output, codebook, decision log, and data documentation according to ethical, legal, institutional, and funder requirements.

How to Choose a Data-Analysis Method

Choose a method by starting with the research question and design, then identify the outcome type, predictor structure, dependence among observations, assumptions, sampling process, and intended inference. Variable type alone is not enough to determine the correct method.

Quantitative method-selection guide

Research aim	Typical data situation	Possible starting method	Important considerations
Describe one categorical variable	Nominal or ordinal categories	Frequencies, percentages, bar chart	Report denominator and missing values
Describe one continuous variable	Numerical measurements	Mean and SD or median and IQR; histogram or box plot	Choose summaries according to distribution and purpose
Compare two independent groups	Continuous outcome, two groups	Independent-samples t test or suitable robust/nonparametric alternative	Independence, distribution, variance, effect size, confidence interval
Compare two paired measurements	Same participants measured twice	Paired t test or Wilcoxon signed-rank test	Pairing, change distribution, time effects
Compare three or more groups	Continuous outcome	ANOVA, regression, or suitable alternatives	Planned contrasts, multiplicity, assumptions
Compare categorical variables	Contingency table	Chi-square test or Fisher’s exact test	Expected cell counts, sampling design
Examine two continuous variables	Paired numerical measurements	Scatterplot, Pearson or Spearman correlation	Nonlinearity, outliers, range restriction
Predict a continuous outcome	One or more predictors	Linear regression	Functional form, residuals, collinearity, validation
Predict a binary outcome	Two-category outcome	Logistic regression	Events per parameter, calibration, discrimination
Analyze repeated observations	Longitudinal or clustered data	Mixed-effects model, generalized estimating equation	Dependence, time structure, missingness
Analyze time until an event	Censored event-time data	Kaplan–Meier and survival regression	Censoring, proportional-hazards assumption
Identify latent dimensions	Multiple related items	Exploratory or confirmatory factor analysis	Sample adequacy, theory, estimator, model fit
Forecast a sequence over time	Ordered time-series data	Time-series model	Trend, seasonality, autocorrelation, external shocks
Estimate an intervention effect	Experimental or defensible quasi-experimental design	Design-appropriate causal model	Randomization, confounding, attrition, compliance
Develop a prediction model	High-dimensional predictors	Penalized regression or machine learning	Data leakage, tuning, validation, calibration

This table provides starting points, not automatic prescriptions. Complex survey designs, small samples, measurement error, hierarchical data, and missingness may require specialist advice.

Qualitative method-selection guide

Research aim	Possible approach	Typical output
Identify patterned meanings across a dataset	Thematic analysis	Themes and an interpretive account
Systematically categorize text or media	Qualitative content analysis	Categories, frequencies where appropriate, contextual interpretation
Answer applied questions using a structured matrix	Framework analysis	Case-by-theme matrix and policy or practice findings
Develop an explanatory theory from data	Grounded theory	Concepts, categories, relationships, substantive theory
Understand lived experience	Phenomenological or interpretative phenomenological analysis	Detailed experiential themes
Examine how stories are organized and given meaning	Narrative analysis	Narrative structures, identities, turning points
Analyze language, power, and social construction	Discourse analysis	Interpretations of language practices and social effects
Study interaction turn by turn	Conversation analysis	Detailed account of conversational organization
Understand cultural practices in context	Ethnographic analysis	Thick description and cultural interpretation
Investigate a bounded case using multiple sources	Case-study analysis	Integrated account of the case and its context

The researcher’s philosophical and theoretical position should be compatible with the chosen approach. For example, different forms of thematic analysis make different assumptions about coding, researcher subjectivity, and what constitutes a theme.

Worked Example: Analyzing a University Learning Intervention

Research question

Does a structured online practice programme improve first-year statistics performance, and how do students experience the programme?

Study design

A mixed-methods study includes:

• an intervention and comparison group;
• a pre-course assessment;
• a final examination score;
• attendance records;
• a satisfaction scale;
• interviews with selected students.

Quantitative analysis

The researcher could:

1. describe both groups at baseline;
2. inspect missing scores and attrition;
3. visualize pre-course and final scores;
4. estimate the adjusted difference in final scores while accounting for baseline performance;
5. report the effect estimate and confidence interval;
6. conduct a sensitivity analysis for missing outcomes;
7. avoid describing the difference as causal unless allocation and implementation justify that claim.

Qualitative analysis

The researcher could:

1. transcribe and anonymize the interviews;
2. read the transcripts closely;
3. code statements about usability, motivation, workload, feedback, and access;
4. develop themes such as “immediate feedback supports persistence” or “fixed deadlines create unequal pressure”;
5. examine cases that contradict the dominant interpretation;
6. support each theme with contextualized extracts.

Integration

A joint display could compare quantitative and qualitative findings:

Quantitative finding	Qualitative finding	Integrated interpretation
Frequent users had higher adjusted scores	Students described immediate feedback as helping them correct misunderstandings	Feedback may be one mechanism connecting engagement with performance, although self-selection remains possible
Satisfaction was lower among working students	Working students described difficulty meeting fixed deadlines	Scheduling requirements may create unequal usability despite general academic value
Some low-frequency users still improved	Several students used the programme selectively before assessments	Total usage time may not capture strategic use

The integrated interpretation is more informative than either component alone.

Data Cleaning, Missing Values, and Outliers

Data cleaning

Data cleaning should correct identifiable errors and standardize the analytical dataset while preserving the original evidence.

A cleaning log should record:

• the issue found;
• how it was detected;
• the decision made;
• the code or rule used;
• the number of records affected;
• the person and date responsible.

Missing data

Missing values should not automatically be replaced with zero or deleted.

Researchers should ask:

• Why is the value missing?
• Is missingness related to participant characteristics or outcomes?
• How much information is missing?
• Is the pattern concentrated in particular variables or groups?
• Would complete-case analysis change the target population?
• Is imputation appropriate?
• Which assumptions does the chosen method require?

Possible approaches include:

• transparent descriptive reporting;
• complete-case analysis when defensible;
• available-case methods;
• model-based estimation;
• multiple imputation;
• weighting;
• sensitivity analysis under different missingness assumptions.

The correct approach depends on why data are missing, the design, the proportion and pattern of missingness, and the intended analysis.

Outliers

An outlier is an observation that is unusual relative to other data or to a fitted model. It may represent:

• a data-entry error;
• a measurement problem;
• a legitimate rare case;
• a different subpopulation;
• an important event;
• model misspecification.

Researchers should verify unusual values, investigate their context, and assess their influence. Deletion is justified only by a documented rule or defensible error determination—not by whether removing the case creates a preferred result.

Effect Sizes, Confidence Intervals, and P-Values

A complete quantitative interpretation should normally report the estimated effect, its uncertainty, and its practical meaning. A p-value alone cannot show how large or important a finding is.

A p-value indicates how incompatible the observed data are with a specified statistical model under its assumptions. It does not provide:

• the probability that the null hypothesis is true;
• the probability that the result occurred “by chance”;
• the size of an effect;
• the importance of an effect;
• proof that a finding will replicate.

The American Statistical Association recommends interpreting statistical evidence in context rather than treating a threshold such as p < .05 as an automatic boundary between truth and falsehood (Wasserstein & Lazar, 2016).

Depending on the method, report measures such as:

• mean difference;
• risk difference;
• risk ratio;
• odds ratio;
• correlation;
• standardized mean difference;
• regression coefficient;
• model-based predicted probability;
• confidence or credible interval.

Practical significance must be judged using disciplinary knowledge, costs, benefits, risks, and the measurement scale.

How to Interpret and Report Data Analysis

Separate results from interpretation where appropriate

The results section usually reports what the analysis produced. The discussion explains what the findings mean. Some qualitative traditions combine analysis and interpretation more closely, so the appropriate structure depends on the methodology and publication convention.

Report enough detail for evaluation

For quantitative analysis, report:

• sample sizes and exclusions;
• missing-data handling;
• descriptive statistics;
• model specification;
• assumption checks;
• effect estimates and uncertainty;
• test statistics where relevant;
• sensitivity or robustness analyses;
• software and relevant package versions.

For qualitative analysis, report:

• the methodological approach;
• researcher role and reflexivity;
• data preparation;
• coding and theme-development process;
• whether coding was inductive, deductive, or combined;
• how disagreements or alternative interpretations were handled;
• evidence supporting themes;
• contradictory cases;
• context and limits of transferability.

SRQR and COREQ can help researchers report qualitative work transparently, although a checklist should support rather than replace methodological reasoning (O’Brien et al., 2014; Tong et al., 2007).

Avoid overstating conclusions

Prefer:

After adjustment for baseline score, the intervention group had a higher estimated mean final score.

Avoid:

The intervention proved that online learning is better.

The first statement describes the model and scope. The second overgeneralizes beyond the evidence.

Use visualizations honestly

A good visualization should:

• match the data type;
• label axes and units;
• show denominators;
• represent uncertainty where relevant;
• avoid distorted scales;
• distinguish raw data from model estimates;
• remain readable without decorative clutter;
• use accessible colors and alternative text;
• explain transformations or exclusions.

Data-Analysis Tools

No single tool is best for every project.

Tool category	Examples	Appropriate uses	Main caution
Spreadsheets	Excel, Google Sheets	Small datasets, cleaning, formulas, pivot tables, basic graphs	Manual actions are difficult to audit and reproduce
Menu-based statistical software	SPSS, Stata, SAS, JMP	Established statistical procedures and institutional workflows	Default output can encourage analysis without understanding
Open statistical software	R, JASP, jamovi	Statistics, graphics, reproducible reports, teaching	Complex work still requires methodological knowledge
General programming	Python	Data cleaning, automation, modeling, machine learning	Code and packages must be validated and documented
Database tools	SQL	Extracting, joining, filtering, and aggregating structured records	Incorrect joins or filters can silently alter the sample
Qualitative software	NVivo, MAXQDA, ATLAS.ti	Coding, retrieval, memos, multimedia organization	Software organizes analysis but does not perform interpretation independently
Visualization and BI	Tableau, Power BI	Dashboards, interactive graphics, stakeholder reporting	Dashboards may conceal analytical assumptions
Reproducible notebooks	Jupyter, Quarto, R Markdown	Combining code, output, explanation, and citations	Sensitive information can be exposed if files are shared carelessly
Large-scale processing	Spark and cloud platforms	Distributed analysis of large datasets	Cost, governance, access control, and complexity

Tool selection should consider:

• the analysis required;
• dataset size and format;
• team skills;
• institutional support;
• cost and licensing;
• reproducibility;
• security;
• collaboration;
• long-term access to files and code.

Artificial Intelligence in Data Analysis

AI can assist with data preparation, programming, pattern detection, transcription, coding suggestions, visualization, and explanation. It should not be treated as an independent authority. Researchers remain responsible for the validity, confidentiality, interpretation, and reporting of every output.

Appropriate supporting uses

AI tools may assist researchers by:

• explaining unfamiliar code;
• drafting code that is then tested;
• translating an analytical plan between software languages;
• proposing data-quality checks;
• detecting possible anomalies for human review;
• suggesting visualization options;
• summarizing non-sensitive documentation;
• assisting transcription where consent and security permit;
• proposing preliminary qualitative codes;
• comparing a codebook against selected text;
• generating simulated data for teaching or software testing;
• improving plain-language explanations.

Important risks

Fabricated or incorrect output

An AI system can produce plausible but invalid code, nonexistent functions, incorrect statistical explanations, or fabricated references. Code must be run, inspected, tested, and compared with trusted documentation.

Confidentiality

Identifiable, proprietary, embargoed, or ethically restricted data should not be entered into a public AI service unless the institutional agreement, consent process, data-management plan, and legal requirements permit it.

Bias and missing context

AI systems may reproduce social or data biases and may not understand the field context, sampling process, measurement limitations, or research design.

Nondeterminism

The same prompt may produce different outputs. Researchers should preserve prompts, model or service information, dates, settings where available, and the verification process when AI materially contributes to analysis.

False objectivity in qualitative research

AI-generated codes do not remove interpretation. Model outputs are shaped by training data, prompts, system design, and the material supplied. Human researchers must evaluate context, latent meaning, contradiction, power, and methodological coherence.

Responsible-use checklist

Before using AI:

1. Check institutional, funder, ethics, and publication rules.
2. Classify the sensitivity of the data.
3. Determine whether the service retains or trains on submitted content.
4. Remove identifying information where appropriate.
5. Use AI for a defined supporting task.
6. Independently verify code, calculations, classifications, and references.
7. Compare outputs with a non-AI method or human review.
8. document material use.
9. disclose it where required.
10. Keep final analytical responsibility with the research team.

NIST’s AI Risk Management Framework, UNESCO’s guidance on generative AI in education and research, and research-integrity guidance from organizations such as UKRIO provide useful governance principles for responsible use (Autio et al., 2024; Miao & Holmes, 2023).

Data Analysis in Modern Research

Reproducible workflows

A reproducible analysis enables another qualified person to understand how the analytical dataset and results were produced.

Useful practices include:

• preserving raw data separately;
• using scripts rather than undocumented manual edits;
• assigning stable variable names;
• maintaining a data dictionary;
• recording package and software versions;
• using version control;
• setting random seeds where relevant;
• preserving analysis logs;
• generating tables and figures directly from code;
• documenting deviations from the original plan.

Preregistration and registered reports

Preregistration records hypotheses, outcomes, exclusions, and analytical decisions before results are known. It is most useful for confirmatory work. It does not prevent exploration, but exploratory analyses should be labelled as such.

A registered report undergoes peer review before the results are available. This can reduce publication decisions based mainly on whether findings are statistically significant or surprising.

Open and FAIR research

Where ethical and legal conditions permit, researchers can share:

• de-identified data;
• metadata;
• code;
• questionnaires;
• codebooks;
• analysis plans;
• computational environments;
• supporting documentation.

The FAIR principles encourage research objects to be findable, accessible under stated conditions, interoperable, and reusable (Wilkinson et al., 2016). FAIR does not mean that all data must be publicly open. Sensitive data may require controlled access.

The NIH Data Management and Sharing Policy, effective since January 25, 2023, also illustrates the growing expectation that researchers plan prospectively for data preservation and appropriate sharing.

Reporting guidelines

Reporting guidelines vary by design and discipline. Examples include:

• CONSORT for randomized trials;
• STROBE for observational studies;
• PRISMA for systematic reviews;
• SRQR for qualitative research;
• COREQ for interviews and focus groups.

Researchers should consult the current guideline relevant to their study rather than applying one checklist to every design.

Advantages of Data Analysis

Appropriate data analysis can:

• convert complex evidence into understandable findings;
• make uncertainty visible;
• identify patterns that are difficult to detect informally;
• test theoretical expectations;
• compare groups or conditions systematically;
• incorporate multiple sources of evidence;
• reveal contradictory and unexpected cases;
• improve transparency in decision-making;
• support prediction and planning;
• enable other researchers to evaluate or reproduce findings.

Limitations of Data Analysis

Analysis is constrained by:

• the quality and relevance of the data;
• the validity of measurements;
• sampling and nonresponse bias;
• uncontrolled confounding;
• missing information;
• imperfect models;
• researcher judgment;
• software and coding errors;
• limited statistical power;
• violations of assumptions;
• social and historical context;
• restrictions on data access;
• changing populations and environments.

No analytical method can make a conclusion stronger than the design and evidence allow.

Common Data-Analysis Mistakes

Starting with software instead of a question

Selecting a technique because it is available in a menu reverses the proper order. Begin with the question, design, data, and intended conclusion.

Confusing correlation with causation

An association may be produced by confounding, reverse causation, selection, measurement error, or chance. Causal language requires a defensible causal design and assumptions.

Treating p < .05 as proof

A threshold does not determine truth, importance, or reproducibility. Report magnitude, uncertainty, assumptions, and context.

Ignoring missing data

Deleting incomplete cases without investigation can change the sample and bias estimates.

Removing inconvenient outliers

Outlier rules should be justified independently of the desired result.

Testing many hypotheses without disclosure

Trying numerous outcomes, subgroups, and models increases the likelihood of apparently convincing chance findings. Distinguish planned and exploratory analyses and address multiplicity where appropriate.

Using the same data for discovery and confirmation

A pattern discovered and confirmed in the same observations may reflect overfitting. Use new data, hold-out samples, cross-validation, or transparent exploratory labelling.

Reporting software output without interpretation

A table of coefficients is not an explanation. Connect estimates to the research question, scale, context, and uncertainty.

Treating qualitative coding as mechanical

Coding is an analytical act. Researchers should explain how interpretations were developed, how their position influenced the process, and how the data support the claims.

Listing qualitative and quantitative findings without integration

Mixed-methods research should explain how the components modify or deepen one another.

Uploading sensitive data to an unapproved AI tool

Convenience does not override consent, confidentiality, legal duties, or institutional policy.

Hiding analytical changes

Changes are sometimes necessary. Undisclosed changes damage trust; documented and justified changes improve transparency.

Data-Analysis Plan Template

Use the following template before beginning the main analysis.

1. Study information

• Study title:
• Research design:
• Population or cases:
• Data sources:
• Unit of analysis:
• Planned software:

2. Research questions

• Primary research question:
• Secondary questions:
• Confirmatory hypotheses:
• Exploratory questions:

3. Variables or qualitative domains

• Primary outcome:
• Predictors or exposures:
• Covariates:
• Grouping variables:
• Qualitative topics or sensitizing concepts:
• Measurement scales and coding:

4. Inclusion and exclusion rules

• Eligible records or participants:
• Exclusion criteria:
• Duplicate handling:
• Minimum data requirements:
• Treatment of protocol deviations:

5. Data preparation

• Cleaning rules:
• Derived variables:
• Category combinations:
• Transformations:
• Transcription procedure:
• Anonymization:
• Codebook development:

6. Missing data and unusual observations

• Missingness assessment:
• Primary handling strategy:
• Sensitivity analysis:
• Outlier definition:
• Influence assessment:
• Conditions for correction or exclusion:

7. Main analysis

• Descriptive outputs:
• Primary model or qualitative method:
• Assumptions:
• Effect-size or uncertainty measures:
• Coding and theme-development process:
• Mixed-methods integration strategy:

8. Additional analysis

• Planned subgroup analyses:
• Robustness checks:
• Alternative specifications:
• Validation approach:
• Treatment of multiple comparisons:

9. Reporting

• Tables:
• Figures:
• Qualitative extracts:
• Reporting guideline:
• Data and code availability:
• AI-use disclosure:
• Known limitations:

Conclusion

Data analysis is the disciplined process through which evidence is converted into an answer. Reliable analysis requires more than calculation: it requires a focused question, an appropriate design, documented preparation, suitable methods, assumption checks, transparent interpretation, and honest reporting.

The strongest analysis is not necessarily the most complicated. It is the analysis that best matches the question and data, expresses uncertainty, recognizes its limitations, and leaves a clear record of how the conclusion was reached.

References

• Autio, C., Schwartz, R., Dunietz, J., Jain, S., Stanley, M., Tabassi, E., Hall, P., & Roberts, K. (2024). Artificial intelligence risk management framework: Generative artificial intelligence profile (NIST AI 600-1). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.AI.600-1
• Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology, 3(2), 77–101. https://doi.org/10.1191/1478088706qp063oa
• Fetters, M. D., Curry, L. A., & Creswell, J. W. (2013). Achieving integration in mixed methods designs—Principles and practices. Health Services Research, 48(6, Part 2), 2134–2156. https://doi.org/10.1111/1475-6773.12117
• Miao, F., & Holmes, W. (2023). Guidance for generative AI in education and research. UNESCO.
• National Institute of Standards and Technology. (2002). NIST/SEMATECH e-handbook of statistical methods (NIST Handbook 151).
• National Institutes of Health. (2023). Data Management and Sharing Policy.
• O’Brien, B. C., Harris, I. B., Beckman, T. J., Reed, D. A., & Cook, D. A. (2014). Standards for reporting qualitative research: A synthesis of recommendations. Academic Medicine, 89(9), 1245–1251. https://doi.org/10.1097/ACM.0000000000000388
• Tong, A., Sainsbury, P., & Craig, J. (2007). Consolidated criteria for reporting qualitative research: A 32-item checklist for interviews and focus groups. International Journal for Quality in Health Care, 19(6), 349–357. https://doi.org/10.1093/intqhc/mzm042
• Tukey, J. W. (1977). Exploratory data analysis. Addison-Wesley.
• Wasserstein, R. L., & Lazar, N. A. (2016). The ASA’s statement on p-values: Context, process, and purpose. The American Statistician, 70(2), 129–133. https://doi.org/10.1080/00031305.2016.1154108
• Wilkinson, M. D., Dumontier, M., Aalbersberg, I. J., et al. (2016). The FAIR guiding principles for scientific data management and stewardship. Scientific Data, 3, Article 160018. https://doi.org/10.1038/sdata.2016.18