Python for Statistical Analysis Training Course

Participants can organise analytical data using core Python and pandas data structures.

• Lists, tuples, dictionaries, and sets selected for analytical tasks
• NumPy arrays created for efficient numerical calculation and indexing
• pandas Series and DataFrames structured for tabular statistical analysis
• Vectorised operations replace slow row-by-row processing patterns
• Boolean masks filter observations using explicit analytical conditions
• Functions accept parameters, return results, and support reusable workflows

Participants can import, profile and reshape source data for statistical analysis.

• CSV, Excel, JSON, and SQL extracts imported with pandas
• Encoding, delimiters, headers, and date parsing diagnosed before analysis
• Wide and long data reshaped using melt, pivot, and stack
• Column types standardised with categorical, datetime, integer, and string dtypes
• Data dictionaries documented with source, field definition, unit, and owner
• Raw data retained separately from processed analytical datasets

Participants can identify data quality problems and apply documented cleaning rules.

• Missingness patterns profiled using counts, proportions, and visual matrices
• Duplicate records identified through keys, composite fields, and business rules
• Outliers screened with IQR fences, z-scores, and domain thresholds
• Invalid categories reconciled through mapping tables and controlled vocabularies
• Cleaning decisions recorded in an auditable data quality log
• Assertions and validation checks embedded in repeatable pandas pipelines

Participants can describe distributions, groups and relationships before formal modelling.

• Distribution shape assessed with histograms, density plots, and quantile summaries
• Central tendency and dispersion calculated using mean, median, variance, and IQR
• Grouped summaries produced with pandas groupby, aggregation, and crosstabs
• Relationships inspected through scatterplots, correlation matrices, and stratification
• Business questions translated into measurable variables, populations, and time windows
• Analytical findings recorded with assumptions, exclusions, and source caveats

Participants can create accurate charts that communicate statistical evidence clearly.

• Matplotlib and seaborn charts selected according to variable type and audience
• Bar, line, box, violin, and scatter charts constructed correctly
• Axes, scales, labels, uncertainty bands, and annotations checked for accuracy
• Colour palettes designed for contrast, accessibility, and categorical distinction
• Small multiples and faceting used to compare groups without distortion
• Publication-ready figures exported with reproducible code and consistent formatting

Participants can apply probability concepts and simulations to quantify analytical uncertainty.

• Sample spaces, events, conditional probability, and Bayes' theorem applied correctly
• Random variables distinguished as discrete, continuous, or mixed measurements
• Expected value, variance, covariance, and correlation interpreted from business data
• Binomial, Poisson, normal, and lognormal distributions matched to processes
• Sampling distributions demonstrated through simulation and repeated resampling
• Monte Carlo methods used to quantify uncertainty in operational scenarios

Participants can design sampling approaches and estimate population values with stated precision.

• Target populations, sampling frames, units, and eligibility criteria specified precisely
• Simple random, stratified, cluster, and systematic sampling compared
• Selection bias, nonresponse bias, coverage error, and measurement error diagnosed
• Sample size estimated from precision, confidence level, variability, and constraints
• Bootstrap confidence intervals generated for medians, proportions, and percentiles
• Survey weights and representativeness limitations documented in reporting

Participants can test operational hypotheses and interpret statistical and practical significance.

• Null and alternative hypotheses framed from operational decision questions
• Type I, Type II, power, and effect size interpreted
• Confidence intervals calculated and reported alongside point estimates
• One-sample and two-sample t-tests selected from data structures
• Paired tests applied when observations share matched subjects or periods
• Practical significance separated from statistical significance in recommendations

Participants can select and interpret tests for categorical, ordinal and non-normal data.

• Chi-square goodness-of-fit and independence tests applied to count data
• Fisher's exact test selected for sparse contingency tables
• Mann-Whitney and Wilcoxon tests used for ordinal or nonnormal outcomes
• Kruskal-Wallis testing followed by appropriate adjusted pairwise group comparisons
• Assumption checks guide parametric versus nonparametric test selection
• Effect sizes reported using odds ratios, rank measures, and Cramér's V

Participants can compare multiple groups using ANOVA and appropriate post-hoc methods.

• One-way ANOVA compares mean outcomes across independent groups
• Two-way ANOVA tests main effects and interaction effects
• Residual normality, variance homogeneity, and independence assumptions examined
• Levene's test and residual plots support assumption assessment
• Tukey HSD identifies group differences after significant omnibus tests
• Eta-squared and partial eta-squared quantify explained outcome variance

Participants can build and interpret linear regression models for continuous outcomes.

• Simple linear regression models continuous outcomes from one predictor
• Multiple regression estimates adjusted relationships among several predictors
• Coefficients, intercepts, confidence intervals, and p-values interpreted responsibly
• Categorical predictors encoded using reference categories and indicator variables
• Interaction terms model conditional relationships between explanatory variables
• R-squared, adjusted R-squared, and residual standard error reported

Participants can diagnose regression weaknesses and validate model performance.

• Residual plots reveal nonlinearity, unequal variance, and influential observations
• Variance inflation factors identify problematic multicollinearity among predictors
• Cook's distance and hat values flag influential cases for review
• Transformations and polynomial terms address specified nonlinear functional forms
• Heteroscedasticity-consistent robust standard errors support valid coefficient inference
• Cross-validation estimates out-of-sample error for competing regression specifications

Participants can model binary events and count outcomes using generalised linear models.

• Logistic regression estimates event probabilities from categorical and numeric predictors
• Odds ratios, marginal effects, and predicted probabilities communicated clearly
• Classification thresholds selected using costs, prevalence, and operational capacity
• ROC curves and AUC assess discrimination across threshold choices
• Poisson regression models counts with exposure offsets and rate interpretation
• Overdispersion checked before choosing Poisson, quasi-Poisson, or negative binomial models

Participants can prepare time-series data, fit ARIMA models and assess forecast accuracy.

• Datetime indexes, resampling, lags, and rolling windows prepared in pandas
• Trend, seasonality, cyclicality, and irregular variation separated conceptually
• Autocorrelation and partial autocorrelation plots guide model identification
• Stationarity assessed using plots, differencing, and augmented Dickey-Fuller tests
• ARIMA models fitted with documented orders and residual diagnostics
• Forecast accuracy compared using MAE, RMSE, and MAPE

Participants can analyse controlled experiments and make evidence-based rollout recommendations.

• Randomised controlled trial designs distinguish treatment effects from confounding
• Primary metrics, guardrail metrics, and stopping rules defined before analysis
• Randomisation balance checked across baseline customer or participant characteristics
• Conversion-rate tests calculated with confidence intervals and risk differences
• Multiple comparisons controlled using Bonferroni adjustment and false discovery rates
• Experiment results translated into rollout, holdout, or further-testing decisions

Participants can prepare reviewable statistical reports with documented methods and limitations.

• Parameterized notebooks separate inputs, transformations, outputs, and narrative findings
• Reusable functions reduce repeated code across recurring statistical reports
• pytest tests verify calculation logic, boundary cases, and expected outputs
• Versioned data, code, and environments support reproducibility and audit review
• Executive summaries state methods, uncertainty, limitations, and recommended actions
• Statistical reports distinguish association, prediction, and causal claims

Participants can build, review and present a complete reproducible statistical analysis addressing a defined workplace decision.

• Problem statement defines decision, stakeholders, success metric, and analytical scope
• Raw dataset imported, cleaned, documented, and transformed through reproducible code
• Exploratory analysis identifies distributions, data quality issues, and candidate relationships
• Appropriate inferential or predictive model justified through assumptions and diagnostics
• Results visualised with uncertainty, effect sizes, and decision-relevant comparisons
• Peer review tests code, interpretations, limitations, and recommendation traceability
• Final briefing presents methods, findings, and implementation actions to stakeholders