Statistics guides
Statistics guides
Long-form guides written and reviewed by named statisticians. Each one covers the method properly — worked examples, assumptions, common mistakes and how to report it — rather than summarising a textbook.
Each guide is written and reviewed by one of five named statisticians holding UK doctorates, and the reviewer is credited on the page along with the date it was last checked. Guides include worked examples calculated step by step rather than summaries of textbook definitions.
Latest guide
Generalised estimating equations
What GEE actually estimates, how to choose a working correlation structure, why the robust standard errors matter more than the structure you pick, and when a mixed effects model answers your question better.
2,200 words · Sidra KhanQualitative methods
Analysing interviews, focus groups and open text.
Thematic analysis
Braun and Clarke's six phases, a worked coding example from extract to theme, the three variants compared, and how to write it up.
4,000 words · Elaine HalliburtonPublishedContent analysis
Manifest and latent content, building and testing a coding frame, a worked example with counts, and inter-coder reliability.
3,000 words · Elaine HalliburtonPublishedGrounded theory
The three schools, open, axial and selective coding, theoretical sampling and saturation, memos, and what examiners check.
3,000 words · Elaine HalliburtonPublishedDiscourse analysis
The approaches compared, what to look for in a transcript, a worked extract analysed line by line, and rigour without reliability coefficients.
3,000 words · Elaine HalliburtonPublishedReliability and validity
Every type of each, which statistic to report, thresholds, establishing them in your own study, and the qualitative equivalents.
3,000 words · Elaine HalliburtonStatistical tests
Choosing, running and reporting the right test.
T-tests
The three types, assumptions and what to do when they fail, three worked examples calculated by hand, effect sizes and APA reporting.
3,000 words · Hafiz Ahmad TariqPublishedChi-square test
Goodness of fit and independence, the expected-count rule, a worked contingency table, Cramer's V and adjusted residuals.
3,000 words · Hafiz Ahmad TariqPublishedANOVA
How variance is partitioned, a one-way ANOVA by hand, Welch's alternative, post-hoc tests and eta squared.
3,000 words · Hafiz Ahmad TariqPublishedMann-Whitney U test
When to use it over a t-test, what it really tests, U calculated by hand, ties and effect sizes.
3,000 words · Hafiz Ahmad TariqPublishedCorrelation
Pearson, Spearman and Kendall compared, r calculated by hand, partial correlation, r squared, and why correlation is not causation.
3,000 words · Hafiz Ahmad TariqPublishedLogistic regression
Why linear regression fails on binary outcomes, odds and log odds, interpreting odds ratios, separation and sample size.
3,000 words · Hafiz Ahmad TariqPublishedRegression analysis
Simple regression worked by hand, interpreting coefficients, R squared, what 'controlling for' means, and residual diagnostics.
3,000 words · Hafiz Ahmad TariqPublishedKruskal-Wallis
The rank-based alternative to ANOVA, H calculated by hand, Dunn's post-hoc test and epsilon squared.
3,000 words · Hafiz Ahmad TariqDesign and power
Getting the study right before data collection.
Self-selection bias
Why a bigger sample makes it worse, and exactly what weighting can and cannot fix.
2,600 words · Sidra KhanPublishedThe null hypothesis
Writing H0 and H1 correctly across designs, one- versus two-tailed, Type I and II errors, and why you never accept the null.
3,000 words · Hafiz Ahmad TariqPublishedExperimental design
Between and within designs, randomisation and concealment, control conditions, counterbalancing, blinding, threats to validity and sample size.
3,000 words · Hafiz Ahmad TariqPublishedHypothesis testing
The five steps in order, stating hypotheses, choosing the test, a worked example start to finish, and the errors you trade off.
3,000 words · Hafiz Ahmad TariqClustered and longitudinal data
When your observations are not independent.
Clustered data and the unit of analysis
Why repeated measures shrink your sample, the design effect, and the code for all four approaches.
3,000 words · Sidra KhanPublishedGeneralised estimating equations
What robust standard errors buy you, how far marginal and conditional diverge, and the code.
3,300 words · Sidra KhanPublishedMixed effects models
Variance components, the ICC and shrinkage calculated, plus the code for R, SPSS and Stata.
2,900 words · Sidra KhanEvidence synthesis
Combining findings across studies.
Meta-analysis
Inverse-variance weighting worked by hand, fixed versus random effects, heterogeneity and publication bias.
3,000 words · Hafiz Ahmad TariqPublishedSystematic reviews
PICO, protocol registration, search strategy, two-stage screening, risk of bias, synthesis and PRISMA 2020.
3,000 words · Hafiz Ahmad TariqPublishedForest plots
Every element explained, why the squares differ in size, judging heterogeneity by eye, and the common misreadings.
3,000 words · Hafiz Ahmad TariqEvaluation and audit
Measuring services and programmes in practice.
Clinical audit
The six-stage cycle, setting standards, sample size, and why most audits never close the loop.
2,400 words · Sidra KhanService evaluation
What it is, how it differs from audit and research, what to measure and how to report it.
2,000 words · Sidra KhanTheory of change
The five levels, outputs vs outcomes, and how to build one backwards from impact.
2,300 words · Elaine HalliburtonClinical and epidemiological
Effect measures for binary and time-to-event outcomes.
Incidence and prevalence
Person-time calculated by hand, direct age standardisation, and why crude rates deceive.
2,900 words · Sidra KhanPublishedSurvival analysis and Kaplan-Meier
The curve and the log-rank test calculated by hand, competing risks, and the code.
3,000 words · Sidra KhanPublishedCox regression and the hazard ratio
Reading the output line by line, what a proportional hazards failure hides, and the code.
2,800 words · Sidra KhanPublishedROC curves and AUC
Building the curve by hand, two ways to the area, and when precision-recall tells you more.
3,000 words · Sidra KhanPublishedSensitivity, specificity, PPV and NPV
All four measures worked through by hand, why predictive values move with prevalence, and the code for R, SPSS and Stata.
2,900 words · Sidra KhanPublishedAbsolute and relative risk
ARR, RRR, risk ratio, NNT and NNH worked through step by step, with intervals and the code.
2,800 words · Sidra KhanPublishedOdds ratios
Odds versus probability, a 2x2 worked by hand, the log-scale interval, and when a risk ratio is the better measure.
3,000 words · Hafiz Ahmad TariqReporting and data quality
Turning output into defensible write-up.
The STROBE statement
Weak against strong answers for the four hard items, and the RECORD extension.
2,500 words · Sidra KhanPublishedCONSORT checklist
A reconciling flow diagram, the sample size calculation, and concealment versus blinding.
2,600 words · Sidra KhanPublishedMissing data: MCAR, MAR and MNAR
The three mechanisms, how much complete-case analysis costs, and the imputation code.
2,600 words · Sidra KhanPublishedP-values
What a p-value measures, five wrong interpretations, why .05 is arbitrary, p-hacking and what to report instead.
3,000 words · Hafiz Ahmad TariqPublishedConfidence intervals
What 95% confidence actually means, intervals for means and proportions worked by hand, intervals for differences, and why they beat a p-value.
3,200 words · Hafiz Ahmad TariqPublishedNormality and Shapiro-Wilk
Why the Shapiro-Wilk p-value misleads at both ends, reading a Q-Q plot, which tests need normality, and what to do when it fails.
3,000 words · Hafiz Ahmad TariqPublishedData cleaning
A defensible workflow, validity checks, the missingness mechanisms, outlier decisions, reverse scoring and reproducible documentation.
3,000 words · Hafiz Ahmad TariqPublishedLikert scales
Item versus scale, how many points, the ordinal-interval argument settled, Cronbach's alpha, response bias and honest charting.
3,000 words · Hafiz Ahmad TariqMore guides are published regularly. Each is reviewed by a named statistician and dated, so you can see when it was last checked.
Send the data. Get a fixed quote.
Attach your dataset or just describe the project. A named statistician replies with a price and a deadline, usually within one working day.
Software walkthroughs
Doing it in SPSS, NVivo and R.
How to test for normality in SPSS
The Explore click path, reading Shapiro-Wilk, Q-Q plots, and what to do when the assumption fails.
1,800 words · Hafiz Ahmad TariqPublishedHow to run a chi-square test in SPSS
Crosstabs step by step, the expected-count check, Cramer's V and adjusted residuals.
1,900 words · Hafiz Ahmad TariqPublishedCronbach's alpha in SPSS
Reliability Analysis, the item-total table, fixing reverse-scored items and what alpha misses.
1,500 words · Elaine HalliburtonPublishedMultiple regression in SPSS
The diagnostics that are off by default, VIF, dummy coding and APA reporting.
1,300 words · Hafiz Ahmad TariqPublishedDescriptive statistics in SPSS
Which of the three dialogs to use, where the median hides, and mean versus median.
1,200 words · Hafiz Ahmad TariqPublishedHow to use NVivo
Projects, sources, nodes and coding — what NVivo does and does not do for your analysis.
1,900 words · Elaine HalliburtonPublishedCoding in NVivo
Naming codes that survive to transcript thirty, when to build a hierarchy, coding stripes.
1,400 words · Elaine HalliburtonPublishedNVivo for thematic analysis
Braun and Clarke's six phases mapped onto what you actually click, and where NVivo misleads.
1,500 words · Elaine HalliburtonPublishedHow to run ANOVA in R
One-way, two-way and repeated measures with working code, Type III sums of squares and post-hoc tests.
1,300 words · Hafiz Ahmad Tariq