Data Analysis Recommender
The recommender reads your project, including variables and how they are measured, hypotheses, research design, sampling and sample size, and sets out an analysis plan in the order a thesis reports it. Every method says why it suits, what it rests on, and what you'd need to justify.
Project details
Enter what you already know. Anything left blank is simply not used, and the plan says what's missing.
One per line.
One variable per line. Put its indicators after a colon, separated by semicolons, such as “wellbeing: life satisfaction; positive mood”.
A plan from the Sample Size Calculator, using Cochran's formula at 95% confidence.
Your analysis plan
In the order a thesis reports it. Open a method to see why it suits your project and what it rests on.
What the plan is based on
- Methodological choice
- Not recorded
- Research design
- Not recorded
- Same people measured more than once
- No
- Planned sample
- Not recorded
Methods at a glance
The families of methods your plan draws on, with the strongest verdict among each family's methods.
- Descriptive statistics: Strong recommendation
1. Describe your data
Describe the participants and each variable
Descriptive statistics Strong recommendation
Describing the sample and every variable comes before any test, and lets readers judge who the findings apply to.
Descriptive statistics: why, and what to check
Why it suits
- Describing the sample and every variable comes before any test, and lets readers judge who the findings apply to.
Assumptions to check
- The summaries chosen suit each variable's measurement level.
Frequency Possible recommendation
Counts describe participant characteristics, such as the number in each group, even when no study variable is categorical.
Frequency: why, and what to check
Why it suits
- Counts describe participant characteristics, such as the number in each group, even when no study variable is categorical.
Assumptions to check
- Categories don't overlap and every answer fits one.
Percentage Possible recommendation
Percentages describe the make-up of the sample alongside the counts.
Percentage: why, and what to check
Why it suits
- Percentages describe the make-up of the sample alongside the counts.
Assumptions to check
- The base (the number the percentage is of) is stated.
2. Check your measures
No variable is measured with several rating items, so scale reliability and factor analysis don't apply. If you use an established measure, report its published reliability and validity.
3. Answer your research questions
Record your dependent and independent variables, or hypotheses, to see which tests answer your research questions.
4. Test the whole model
Structural equation modelling suits frameworks linking three or more variables, at least two of them measured by several items, with several hypotheses. Your project doesn't have that yet, so the tests above answer each question separately.
Notes
- No variables are recorded, so no analysis can be matched to them.
- No research design is chosen, so the design can't inform the recommendations.
- The methodological choice (quantitative, qualitative or mixed) isn't recorded.
- No planned sample size is available, so sample size can't inform the recommendations.
Analysis methods explained
Every method the recommender covers, grouped by what it is for. References will be added after academic review.
Describing the data
Descriptive statistics
Summarises who took part and what each variable looks like before any test.
Learn more about Descriptive statistics
When it usually suits
Almost every quantitative study reports them first.
Assumptions to check
- The summaries chosen suit each variable's measurement level.
Limitations
- They describe the sample only; they don't test hypotheses or generalise to the population.
References awaiting academic review.
Frequency
Counts how many participants gave each answer or fall in each category.
Learn more about Frequency
When it usually suits
For categorical, binary and ordinal variables, and for participant characteristics.
Assumptions to check
- Categories don't overlap and every answer fits one.
Limitations
- Counts alone are hard to compare between groups of different sizes.
References awaiting academic review.
Percentage
Shows each category's share of the sample, so groups of different sizes can be compared.
Learn more about Percentage
When it usually suits
Alongside frequencies, for categorical, binary and ordinal variables.
Assumptions to check
- The base (the number the percentage is of) is stated.
Limitations
- Percentages of small samples can look more precise than they are.
References awaiting academic review.
Mean
Gives the average value of a numeric variable.
Learn more about Mean
When it usually suits
For numeric variables, and for scale scores made from several items.
Assumptions to check
- The variable is numeric, with equal intervals between values.
Limitations
- Extreme values pull it away from the typical case.
- It can mislead for ordinal answers such as a single rating.
References awaiting academic review.
Median
Gives the middle value, which extreme values don't distort.
Learn more about Median
When it usually suits
For ordinal variables, and for numeric variables that are skewed.
Assumptions to check
- Values can be put in order.
Limitations
- It ignores how far values lie from the middle.
References awaiting academic review.
Mode
Gives the most common value or category.
Learn more about Mode
When it usually suits
For nominal variables, where it is the only average that makes sense, and alongside the median for ordinal variables.
Assumptions to check
- Values or categories can be counted.
Limitations
- A variable can have several modes, or none that stands out.
- It ignores every value except the most common one.
References awaiting academic review.
Standard deviation
Shows how spread out a numeric variable's values are around the mean.
Learn more about Standard deviation
When it usually suits
Reported with every mean.
Assumptions to check
- The variable is numeric.
Limitations
- Like the mean, it is sensitive to extreme values.
References awaiting academic review.
Checking the measures
Reliability
Checks that a measure gives consistent results, such as items in a scale agreeing with each other.
Learn more about Reliability
When it usually suits
Whenever a variable is measured with several items combined into one score.
Assumptions to check
- The items are meant to measure the same thing.
Limitations
- A reliable measure isn't necessarily valid: it can consistently measure the wrong thing.
References awaiting academic review.
Cronbach's alpha
Estimates the internal consistency of a set of items that form one scale.
Learn more about Cronbach's alpha
When it usually suits
For multi-item scales, such as several Likert items measuring one variable.
Assumptions to check
- The items measure one underlying construct.
- Reverse-worded items are reverse-scored first.
Limitations
- It rises with the number of items, so long scales can look consistent when they aren't.
- It assumes every item contributes equally, which often isn't true.
References awaiting academic review.
Validity
Checks that a measure captures the construct it is meant to, such as items grouping as the theory expects.
Learn more about Validity
When it usually suits
When scales are new, adapted or used in a new population.
Assumptions to check
- There is a theory of which items belong to which construct.
Limitations
- Statistical checks cover construct validity only; content and face validity need expert review.
References awaiting academic review.
Kaiser–Meyer–Olkin (KMO) measure
Checks whether the items share enough variance for factor analysis to be worthwhile.
Learn more about Kaiser–Meyer–Olkin (KMO) measure
When it usually suits
As a check before factor analysis.
Assumptions to check
- The items are numeric or treated as numeric.
Limitations
- It says whether factor analysis is sensible, not what the factors are.
References awaiting academic review.
Bartlett's test of sphericity
Checks that the items are correlated at all, which factor analysis needs.
Learn more about Bartlett's test of sphericity
When it usually suits
As a check before factor analysis.
Assumptions to check
- The items are roughly normally distributed.
Limitations
- With large samples it is almost always significant, so it is a minimal check.
References awaiting academic review.
Factor analysis
Finds or confirms groups of items that measure the same underlying construct.
Learn more about Factor analysis
When it usually suits
For multi-item scales: exploratory when the structure is unknown, confirmatory when theory specifies it.
Assumptions to check
- Enough participants for stable results.
- Items are related to each other in straight-line ways.
Limitations
- Small samples give unstable factors.
- Choices such as how many factors to keep need justifying.
References awaiting academic review.
Relationships between variables
Correlation
Measures how strongly two variables move together.
Learn more about Correlation
When it usually suits
When a hypothesis or objective concerns a relationship between two measured variables.
Assumptions to check
- The relationship is monotonic (it keeps going one way).
Limitations
- Correlation doesn't show cause and effect.
References awaiting academic review.
Pearson correlation
Measures the strength and direction of a straight-line relationship between two numeric variables.
Learn more about Pearson correlation
When it usually suits
When both variables are numeric and roughly normally distributed.
Assumptions to check
- Both variables are numeric.
- The relationship is linear.
- No extreme outliers.
- The variables are roughly normally distributed.
Limitations
- It misses curved relationships.
- Outliers can distort it strongly.
References awaiting academic review.
Spearman's rank correlation
Measures how consistently two variables rise or fall together, using ranks.
Learn more about Spearman's rank correlation
When it usually suits
When a variable is ordinal, or numeric data aren't normally distributed.
Assumptions to check
- Both variables can be ranked.
- The relationship is monotonic.
Limitations
- It has less power than Pearson when Pearson's assumptions hold.
References awaiting academic review.
Predicting an outcome
Regression
Predicts an outcome from one or more variables and estimates each one's contribution.
Learn more about Regression
When it usually suits
When a hypothesis or objective predicts an outcome from other variables.
Assumptions to check
- The outcome's type matches the kind of regression.
Limitations
- Prediction isn't causation unless the design supports it.
References awaiting academic review.
Simple linear regression
Predicts a numeric outcome from one variable.
Learn more about Simple linear regression
When it usually suits
One numeric predictor and one numeric outcome.
Assumptions to check
- Linear relationship.
- Independent observations.
- Residuals are normally distributed with equal spread.
Limitations
- Leaves out other variables that may matter.
References awaiting academic review.
Multiple regression
Predicts a numeric outcome from several variables at once, each adjusted for the others.
Learn more about Multiple regression
When it usually suits
Two or more predictors of one numeric outcome.
Assumptions to check
- Linear relationships.
- Independent observations.
- Residuals are normally distributed with equal spread.
- Predictors aren't too highly correlated with each other (no multicollinearity).
Limitations
- Needs more participants as predictors are added.
- Highly related predictors make individual estimates unstable.
References awaiting academic review.
Hierarchical regression
Adds predictors in stages, showing what each set adds beyond the ones before, such as control variables first.
Learn more about Hierarchical regression
When it usually suits
When control variables should be accounted for before the variables of interest.
Assumptions to check
- As for multiple regression.
- The order of entry is set by theory before analysis.
Limitations
- The order of entry must be justified; changing it changes the results.
References awaiting academic review.
Logistic regression
Predicts the chance of a two-category outcome, such as yes or no.
Learn more about Logistic regression
When it usually suits
When the outcome is binary.
Assumptions to check
- Independent observations.
- A straight-line relationship between numeric predictors and the log odds of the outcome.
- Enough cases in the smaller outcome category.
Limitations
- Odds ratios are easy to misread as risks.
- Needs larger samples than linear regression.
References awaiting academic review.
Differences between groups or times
One-sample t-test
Compares the mean of a numeric outcome in one sample with a known or hypothesised value, such as a target or a published average.
Learn more about One-sample t-test
When it usually suits
When one group's mean is compared with a fixed value rather than with another group.
Assumptions to check
- The outcome is numeric.
- Observations are independent.
- The outcome is roughly normally distributed, or the sample is large enough for the mean to be.
Limitations
- The comparison value must come from outside the sample, and its source should be stated.
- Sensitive to outliers in small samples.
References awaiting academic review.
Independent-samples t-test
Compares the means of a numeric outcome between two separate groups.
Learn more about Independent-samples t-test
When it usually suits
When a difference hypothesis compares two independent groups.
Assumptions to check
- The outcome is numeric.
- The groups are independent.
- The outcome is roughly normally distributed in each group.
- The groups have similar variances, or Welch's version is used.
Limitations
- Only two groups.
- Sensitive to outliers in small samples.
References awaiting academic review.
Paired-samples t-test
Compares the means of a numeric outcome measured twice on the same participants, such as before and after.
Learn more about Paired-samples t-test
When it usually suits
When the same people are measured at two times or in two conditions.
Assumptions to check
- The outcome is numeric.
- Measurements are paired.
- The differences between pairs are roughly normally distributed.
Limitations
- Only two measurements.
References awaiting academic review.
One-way ANOVA
Compares the means of a numeric outcome across three or more independent groups.
Learn more about One-way ANOVA
When it usually suits
When a difference hypothesis compares three or more groups on one factor.
Assumptions to check
- The outcome is numeric.
- Groups are independent.
- Roughly normal within each group.
- Similar variances across groups.
Limitations
- Shows that groups differ, not which ones: post hoc tests are needed.
References awaiting academic review.
Two-way ANOVA
Compares means across two grouping factors at once, including whether they interact.
Learn more about Two-way ANOVA
When it usually suits
When two categorical factors may each, and together, affect a numeric outcome.
Assumptions to check
- As for one-way ANOVA, for every combination of groups.
Limitations
- Every combination of groups needs enough participants.
References awaiting academic review.
MANOVA
Compares groups on several related numeric outcomes at once.
Learn more about MANOVA
When it usually suits
When one grouping factor may affect two or more related numeric outcomes.
Assumptions to check
- Outcomes are numeric and moderately related.
- Multivariate normality.
- Similar covariance matrices across groups.
Limitations
- Harder to interpret, and needs follow-up tests on each outcome.
References awaiting academic review.
ANCOVA
Compares group means on a numeric outcome after adjusting for a numeric covariate, such as a control variable or a pretest score.
Learn more about ANCOVA
When it usually suits
When groups are compared and a control variable should be adjusted for.
Assumptions to check
- As for ANOVA.
- The covariate relates to the outcome in a straight line.
- The covariate's relationship with the outcome is the same in every group.
Limitations
- Adjustment can't make non-equivalent groups truly comparable.
References awaiting academic review.
Repeated measures ANOVA
Compares the means of a numeric outcome measured three or more times on the same participants.
Learn more about Repeated measures ANOVA
When it usually suits
When the same people are measured repeatedly over time or conditions.
Assumptions to check
- The outcome is numeric.
- Roughly normal at each time.
- Sphericity: similar variances of the differences between times.
Limitations
- Participants who miss one time point drop out of the analysis.
References awaiting academic review.
Wilcoxon signed-rank test
Compares two related measurements, such as before and after, using ranks.
Learn more about Wilcoxon signed-rank test
When it usually suits
The non-parametric counterpart of the paired t-test, for ordinal or non-normal data.
Assumptions to check
- Measurements are paired.
- The differences can be ranked.
Limitations
- Less power than the paired t-test when its assumptions hold.
References awaiting academic review.
Mann–Whitney U test
Compares two independent groups using ranks.
Learn more about Mann–Whitney U test
When it usually suits
The non-parametric counterpart of the independent t-test, for ordinal or non-normal data.
Assumptions to check
- Independent groups.
- The outcome can be ranked.
Limitations
- Tests whether values tend to be higher in one group, not whether means differ.
References awaiting academic review.
Kruskal–Wallis test
Compares three or more independent groups using ranks.
Learn more about Kruskal–Wallis test
When it usually suits
The non-parametric counterpart of one-way ANOVA.
Assumptions to check
- Independent groups.
- The outcome can be ranked.
Limitations
- Needs follow-up tests to find which groups differ.
References awaiting academic review.
Associations between categories
Chi-square test of independence
Tests whether two categorical variables are associated.
Learn more about Chi-square test of independence
When it usually suits
When both variables are categorical, such as group and a yes/no answer.
Assumptions to check
- Independent observations.
- Expected counts of at least 5 in most cells.
Limitations
- Shows association, not its strength or cause; report an effect size too.
References awaiting academic review.
Chi-square goodness-of-fit test
Tests whether the counts in one categorical variable's categories match expected proportions, such as equal shares or known population shares.
Learn more about Chi-square goodness-of-fit test
When it usually suits
When one categorical variable is compared with a stated expected distribution.
Assumptions to check
- Independent observations.
- Each participant falls in exactly one category.
- Expected counts of at least 5 in most categories.
Limitations
- The expected proportions must be stated before looking at the data.
- Shows that counts differ from expectation, not why.
References awaiting academic review.
Fisher's exact test
Tests an association between two categorical variables exactly, without large-sample approximations.
Learn more about Fisher's exact test
When it usually suits
When samples are small and some expected counts fall below 5.
Assumptions to check
- Independent observations.
Limitations
- Mainly used for two-by-two tables.
References awaiting academic review.
Testing a whole model
Moderation analysis
Tests whether a third variable changes the strength or direction of a relationship, using an interaction term.
Learn more about Moderation analysis
When it usually suits
When a hypothesis names a moderator.
Assumptions to check
- As for regression.
- The moderator is measured before or independently of the outcome.
Limitations
- Interaction effects need larger samples to detect.
- Continuous moderators are best mean-centred to make results interpretable.
References awaiting academic review.
Mediation analysis
Tests whether one variable affects an outcome through another, estimating the indirect effect.
Learn more about Mediation analysis
When it usually suits
When a hypothesis names a mediator.
Assumptions to check
- As for regression.
- The causal order of predictor, mediator and outcome is justified by theory or design.
- The indirect effect is usually tested with bootstrapped confidence intervals.
Limitations
- With data from one time point, mediation shows a pattern consistent with a process, not proof of it.
References awaiting academic review.
Structural equation modelling
Tests a whole model of relationships at once, often between constructs measured by several items each.
Learn more about Structural equation modelling
When it usually suits
When a conceptual framework links several constructs, each measured by multiple items.
Assumptions to check
- The model is specified from theory before analysis.
- Enough participants for the model's size.
Limitations
- Needs large samples and careful model specification.
- Good fit doesn't prove the model is the right one.
References awaiting academic review.
PLS-SEM
Estimates a structural model by maximising the variance explained in the outcomes.
Learn more about PLS-SEM
When it usually suits
Often chosen for prediction-focused or exploratory models, complex models, or smaller samples.
Assumptions to check
- The measurement and structural models are specified from theory.
- Indicators relate to their constructs as specified.
Limitations
- It doesn't give the same overall model-fit tests as CB-SEM.
- Its estimates can be biased when constructs are truly latent factors.
References awaiting academic review.
CB-SEM
Tests how well a theorised model reproduces the observed relationships, with overall fit statistics.
Learn more about CB-SEM
When it usually suits
Usually chosen for confirming an established theory with a large sample.
Assumptions to check
- Multivariate normality, or a robust estimator.
- A large sample.
- A model identified from theory.
Limitations
- Needs larger samples than PLS-SEM.
- Complex models may fail to converge.
References awaiting academic review.
Next: Statistical assumptions
What to check before running each analysis in your plan.
How the recommender works
- The recommender reads your project draft: variables and their measurement levels, indicators and questionnaire items, hypotheses, research onion choices, research design, sampling technique and sample size. Enter them in the first step; later versions will bring them across from the other research tools.
- Each rule looks at how variables are measured, what each hypothesis claims, whether the same people are measured more than once, and how large the sample will be.
- Every method gets one of three verdicts: Strong recommendation, Possible recommendation or Needs justification. None is ever called wrong.
- Normality can't be known until data are collected, so every test that assumes it comes with its non-parametric alternative.
- The recommender only reads the project draft. It writes nothing back.
What this tool can't do
- Recommendations follow from the project as recorded: its variables, measurement levels, hypotheses, design, sample and approach. If those are missing or wrong, so are the recommendations.
- Normality and other assumptions can only be checked once data are collected, so every parametric test comes with its non-parametric alternative.
- Rules of thumb for sample size, such as for factor analysis and SEM, are common guidance rather than fixed requirements; the right size depends on the model.
- Qualitative analysis, such as thematic analysis, is outside this tool, as are methods it doesn't list, such as multilevel models, multinomial logistic regression and the Friedman test's follow-up comparisons.
- The tool doesn't run any analysis or look at data. Use statistical software, and seek statistical advice for complex designs.
- It never says a method is wrong: “Needs justification” means the project, as recorded, doesn't obviously support it.
Awaiting review
- Statistical review: every decision rule and the wording of each method's assumptions, by a statistician.
- Sample size guidance: the thresholds used for small samples, factor analysis and SEM.
- References: sources for each method, after academic review.
- Worked examples: complete projects checked against the recommendations.
Privacy
Everything stays in your browser. Nothing is sent anywhere or saved, and it is cleared when you leave the page. If you start a project in your research workspace, your work is also saved in this browser, and only this browser, until you delete the project.