Methods

The overall picture: means, spread, missing cells.

• Descriptive statistics

  Reduces every column in the dataset to one summary row: centre and spread for numeric columns, valid answers and category counts for text columns.

  ParametricNo group comparison

• Correlation matrix

  Gathers the pairwise coefficients of every numeric column in the dataset into one square table, without any test of significance.

  ParametricNo group comparison

• Missing value analysis

  Lists how many values are absent from each column in the dataset, what share of the rows that represents, and which columns carry no gaps at all.

  Non-parametricNo group comparison

• Weighting (rim weighting)

  Produces weights that pull the sample's distribution on variables such as gender, age and region towards known population shares, and reports how much information the correction costs.

  Non-parametricNo group comparison

• Net Promoter Score (NPS)

  Splits a single zero-to-ten recommendation question into three segments and subtracts the share of detractors from the share of promoters to produce one score.

  Non-parametricNo group comparison

Test preconditions: is the data normal, are there outliers.

• Normality tests

  Assesses whether the numeric columns you select depart from a normal distribution, alongside descriptive measures of their shape.

  Non-parametricNo group comparison

• Outlier detection

  Flags extreme values in numeric columns under three separate rules and reports how many were flagged and where they sit.

  Non-parametricNo group comparison

Do two variables move together, in which direction and how strongly.

• Pearson correlation

  Summarises the direction and strength of the linear co-variation between two continuous variables in a single coefficient between -1 and +1.

  ParametricNo group comparison

• Spearman rank correlation

  Relates two variables through their rank positions rather than their raw values, and reports the direction and strength of the monotonic association between them.

  Non-parametricNo group comparison

• Kendall's tau-b

  Measures association between two ordinal variables by counting how many pairs of observations line up in the same direction.

  Non-parametricNo group comparison

• Partial correlation

  Measures the linear relationship between two continuous variables once the variability they share with one or more other variables has been taken out.

  ParametricNo group comparison

Non-numeric variables: frequency, cross-tabs, independence.

• Chi-square test of independence

  Examines the joint distribution of two categorical variables and tests whether the observed cell counts depart meaningfully from the counts expected if the variables were unrelated.

  Non-parametricNo group comparison

• Cross-tabulation

  Brings two categorical variables into one table and describes the count in every cell alongside row, column and overall percentages.

  Non-parametricNo group comparison

• Frequency distribution

  Describes how many people fall in each category of one or several categorical columns, together with the percentage and the cumulative percentage.

  Non-parametricNo group comparison

• Chi-square goodness of fit

  Tests how closely the category counts of a single categorical variable match the counts expected from theory or from a known population distribution.

  Non-parametricNo group comparison

• Fisher's exact test

  Finds the probability of an association in a 2x2 table by enumeration rather than approximation, which keeps it trustworthy in small samples and sparse cells.

  Non-parametricNo group comparison

• McNemar test

  Tests whether the proportion giving a yes or no answer changed between two measurements of the same participants, using only the people who switched sides.

  Non-parametricTwo groups

• Cochran's Q test

  Evaluates in a single test whether the acceptance rates of three or more yes and no items measured in the same participants differ meaningfully from one another.

  Non-parametricThree or more groups

• Proportion test

  Tests whether an observed proportion on a binary response departs from a hypothesised value, or whether the proportions of two independent groups differ from each other.

  ParametricTwo groups

• Multiple response analysis

  Reads questions where a respondent may tick several options from the option columns that hold them, and reports a count per option alongside two percentages built on two different bases.

  Non-parametricNo group comparison

• Banner table (with significance letters)

  Places the answer options of one question down the rows and the categories of several breakdown variables across lettered column blocks, testing the pairs of columns inside each block row by row.

  Non-parametricNo group comparison

Is the difference between two or more groups significant.

• Independent samples t-test

  Compares the means of two independent groups and tests whether the observed difference is small enough to be explained by sampling variation.

  ParametricTwo groups

• Paired samples t-test

  Compares two measurements taken from the same people, or from matched pairs, by testing whether the average of the paired differences departs from zero.

  ParametricTwo groups

• One-way ANOVA

  Compares the means of three or more independent groups in a single test and reports whether at least one group stands apart from the others.

  ParametricThree or more groups

• Two-way ANOVA

  Tests the separate effects of two categorical factors on a continuous outcome and the interaction between them within a single model.

  ParametricThree or more groups

• ANCOVA

  Compares group means after statistically holding constant one or more continuous variables that also shape the outcome.

  ParametricThree or more groups

• MANOVA

  Combines several continuous outcomes into a single test of whether groups differ on that set of outcomes taken together.

  ParametricThree or more groups

• MANCOVA

  Takes several continuous outcomes together and tests group separation once the share of continuous variables that also explain those outcomes has been removed.

  ParametricThree or more groups

• Repeated measures ANOVA

  Compares the means of three or more continuous measurements taken from the same participants in a single test.

  ParametricThree or more groups

• Mann-Whitney U

  Compares two independent groups through their combined rank order rather than their means, without assuming a normal distribution.

  Non-parametricTwo groups

• Wilcoxon signed-rank

  Compares two measurements taken from the same participants using the signed ranks of their differences, without assuming the differences are normally distributed.

  Non-parametricTwo groups

• Kruskal-Wallis

  Compares three or more independent groups using rank positions rather than raw values, and tests whether the groups occupy different positions in the pooled distribution.

  Non-parametricThree or more groups

• Friedman

  Compares ordinal measurements taken from the same participants under three or more conditions by ranking each row within itself.

  Non-parametricThree or more groups

• One-sample t-test

  Compares the mean of one continuous variable against a fixed reference value supplied from outside the data, and reports a confidence interval for the difference.

  ParametricOne group

• Welch's ANOVA (unequal variances)

  Compares the means of three or more independent groups without assuming that the group variances are equal.

  ParametricThree or more groups

• Dunn's test (post-hoc)

  Compares the ranks of three or more independent groups pair by pair, ranking the sample once and adjusting the p values for multiplicity.

  Non-parametricThree or more groups

Which variables explain an outcome, and by how much.

• Linear regression (OLS)

  Explains a continuous outcome as a linear combination of one or more numeric predictors, estimating each slope by the least squares criterion.

  ParametricNo group comparison

• Logistic regression

  Models the probability of a two-outcome event from numeric predictors and expresses each predictor's contribution as an odds ratio.

  ParametricNo group comparison

• Ordinal logistic regression

  Models an ordered outcome from numeric predictors and reports, as an odds ratio, how strongly each predictor pushes the outcome towards higher categories.

  ParametricNo group comparison

• Multinomial logistic regression

  Models an unordered outcome with more than two categories from numeric predictors, contrasting every category against a chosen baseline.

  ParametricNo group comparison

• Hierarchical (block-wise) regression

  Enters predictors in blocks the researcher decides on and tests separately what each block adds to the explained variance.

  ParametricNo group comparison

Do the scale items measure the same thing.

• Cronbach's Alpha

  Summarises in a single coefficient how consistently the items of a scale measure the same construct, and shows which item is dragging that consistency down.

  ParametricNo group comparison

• Split-half reliability

  Divides a scale into two halves and estimates the reliability of the whole from how closely the half scores agree, reporting which split was used.

  ParametricNo group comparison

• McDonald's omega

  Estimates scale reliability from the common factor loadings of the items, without assuming that every item carries the construct with equal weight.

  ParametricNo group comparison

• Inter-rater agreement (kappa, ICC)

  Measures how far people coding the same material reach the same decision, using kappa coefficients corrected for chance and the intraclass correlation.

  Non-parametricNo group comparison

The latent dimensions behind many items.

• Exploratory factor analysis (EFA)

  Works out from the data how many latent dimensions sit behind a set of items, and which items move together on each of them.

  ParametricNo group comparison

• Principal component analysis (PCA)

  Turns many numeric columns into a handful of components that between them carry as much of the total spread in the data as possible.

  ParametricNo group comparison

• Parallel analysis (number of factors)

  Compares the eigenvalues from your data with eigenvalues generated from random datasets of the same size to show how many factors are worth retaining.

  ParametricNo group comparison

Segmenting respondents by how similar they are.

• K-means cluster analysis

  Divides respondents into groups that resemble one another on the continuous variables you choose, and profiles each group against the overall average.

  Non-parametricThree or more groups

• Hierarchical cluster analysis

  Merges respondents step by step starting from the most similar pairs, stops at the number of clusters you ask for, and profiles each one.

  Non-parametricThree or more groups

A model of direct and indirect effects between variables.

• Structural equation modelling (SEM)

  Estimates measurement relationships and directed paths between latent and observed variables within one model.

  ParametricNo group comparison

• Confirmatory factor analysis (CFA)

  Tests how well a factor structure specified in advance matches the collected data and reports the validity of the measurement model through fit indices.

  ParametricNo group comparison

• Path analysis

  Estimates directed relationships among observed variables as one system of equations and reports indirect routes alongside the direct paths.

  ParametricNo group comparison

• Mediation analysis

  Estimates how much of a predictor's effect on an outcome travels through an intervening third variable and tests that share with a Sobel test.

  ParametricNo group comparison

• Moderation analysis

  Tests whether the link between a predictor and an outcome strengthens or weakens across the levels of a third variable, and reports the link separately at those levels.

  ParametricNo group comparison

Preference data: how much each attribute is worth, which option wins.

• MaxDiff (best-worst scaling)

  Turns a design that asks respondents to mark the best and worst option in each small set into a preference score and a rank for every item.

  Non-parametricNo group comparison

• Conjoint analysis

  Turns a design in which respondents pick one of several product profiles per task into part-worth utilities for feature levels and relative importance for the features.

  Non-parametricNo group comparison

• TURF analysis (reach and frequency)

  Counts how many different people a set of options covers and finds the combination of a given size that reaches the largest number of them.

  Non-parametricNo group comparison

• Van Westendorp price sensitivity

  Builds cumulative curves from four price questions put to respondents and reads an acceptable price range off the points where those curves cross.

  Non-parametricNo group comparison

Is the sequence random, or is there a pattern.

• Runs test for randomness

  Tests whether the values in a column are ordered at random, by counting the unbroken stretches that form on either side of the median.

  Non-parametricNo group comparison