Cases and projects
Apply statistics to realistic questions
Each case gives you a brief, a dataset and a sequence of graded tasks — from checking the data to recommending a decision. You finish by writing up your analysis in the report workspace.
Applied cases
Focused investigations of one professional question, in three or four stages.
- Retail · 60 min · FreeDid the promotion work?Estimate the sales lift of a chain-wide discount promotion fairly — separating it from holidays and store mix — put an honest interval on it, and decide whether it pays for itself.4 stages · 10 graded tasks · data: Weekly store sales
- Operations · 60 min · MembersStaffing the contact centreUse hourly call data to model arrivals as Poisson, plan staffing to a demand quantile rather than the average, and connect waiting times to abandoned calls.4 stages · 9 graded tasks · data: Contact-centre hourly calls
- Banking · 75 min · MembersDefault rates in a loan bookEstimate default rates with honest intervals, compare segments, fit a logistic regression and interpret odds ratios — and keep risk prediction separate from causal claims.4 stages · 10 graded tasks · data: Personal loan applications
- HR · 60 min · MembersCan we trust the engagement survey?Check response rates and response quality, summarise Likert items properly, and correct an engagement headline for nonresponse bias with post-stratification weights.4 stages · 10 graded tasks · data: Employee engagement survey
- Logistics · 60 min · MembersDid the new routing cut delivery times?Choose between paired and independent analyses, estimate the saving with an interval and an effect size, check assumptions, and judge what a before–after study can and cannot show.4 stages · 10 graded tasks · data: Delivery times before and after a routing change (paired)
- Public health · 60 min · MembersEvaluating a screening programmeBuild the 2×2 table, estimate sensitivity and specificity with intervals, and show why predictive values depend on base rates — then decide who the programme should target.4 stages · 9 graded tasks · data: Population screening test
- Insurance · 75 min · MembersPricing for rare large claimsMeasure claim frequency per policy-year, recognise overdispersion, describe a heavy-tailed severity distribution, and price a premium that neither ignores nor over-reacts to one huge claim.4 stages · 10 graded tasks · data: Motor insurance claims
- Manufacturing · 60 min · MembersAre defects under control?Build a p-chart with varying sample sizes, find the out-of-control day, see why a c-chart misleads when volumes vary, and focus root-cause work with a Pareto of defect types.4 stages · 10 graded tasks · data: Daily inspection defects
- Marketing analytics · 60 min · MembersWhere does the funnel leak?Trace 26 weeks of funnel counts for five channels, find the stage where each one loses customers, put a price on it, and recommend where the next quarter's budget should go — without over-reading last-click data.4 stages · 10 graded tasks · data: Marketing funnel by channel
- Real estate and lending · 75 min · MembersWhat drives house prices?Build a valuation model for 800 home sales: choose the right scale for price, untangle floor area from bedrooms, interpret log-scale coefficients as percentages and give an honest prediction interval for a single home.4 stages · 9 graded tasks · data: Home sales in a fictional city
- Energy and facilities · 70 min · MembersEnergy use and the weatherA year of daily meter readings from an office tower: discover why a straight line says temperature barely matters, find the curve the residuals point to, add the working-week effect and check what the residuals still say.4 stages · 9 graded tasks · data: Office building daily energy use
- Sports analytics · 60 min · MembersDoes training volume pay off?Five hundred 10 km finishers and their weekly mileage: measure a curved association honestly, adjust for age and sex, and decide what a running club can — and cannot — promise its members.4 stages · 9 graded tasks · data: 10 km race results
- Healthcare and clinical research · 80 min · MembersEvaluating a (fictional) blood-pressure trialAnalyse a randomised, placebo-controlled trial end to end: check the randomisation and the dropouts, choose the comparison that actually estimates the drug effect, quantify it with an interval and effect size, and weigh it against adverse events.4 stages · 9 graded tasks · data: Blood-pressure drug trial (fictional)
- Transportation and operations · 65 min · MembersWhich carrier is more punctual?Compare departure delays for four carriers when most flights leave on time and a few are hours late: choose summaries and tests that respect heavy skew, measure the share of late flights, and check whether weather explains the gaps.4 stages · 10 graded tasks · data: Departure delays
- Education policy · 70 min · MembersDoes tutoring raise exam scores?Tutored students scored lower than untutored ones. Discover why the raw gap misleads, adjust for prior attainment with regression, see why gain scores overstate the effect, and give the board an honest, uncertain answer.4 stages · 9 graded tasks · data: Student exam scores
Capstone projects
Integrated projects that combine several tracks: data quality, exploration, method choice, inference and a written recommendation.
- Retail · 210 min · MembersRetail performance review: two years of Lumen Bay StoresAudit two years of store-week data, build the KPIs a board actually needs, compare store formats at the right unit of analysis, separate holidays from trend, estimate the promotion effect three ways with honest uncertainty, and write the executive recommendation.5 stages · 18 graded tasks · data: Weekly store sales
- Customer analytics and data quality · 220 min · MembersFull statistical investigation: from a messy export to a decisionTake a raw customer export full of duplicates, typos, placeholders, impossible values and a unit mix-up; audit it, clean it with documented rules, analyse the cleaned data, and show which conclusions survive — and which collapse — when the cleaning changes.5 stages · 19 graded tasks · data: Customer records (messy)
- E-commerce and product analytics · 180 min · MembersCustomer conversion experiment: ship the new checkout?Run a checkout A/B test the way a senior analyst would: pre-register the metric, effect size and sample size, check the randomisation, analyse conversion with an interval for the lift, handle revenue, guardrails and segments honestly, and make a ship / no-ship call that weighs uncertainty against money.5 stages · 20 graded tasks · data: Checkout redesign A/B test
- Human resources and reward · 200 min · MembersEmployee compensation and pay equity reviewReview pay at a 600-person company: describe a skewed salary distribution, separate the unadjusted gender gap from the adjusted one, handle job level as a possible mediator rather than a neutral control, test the sensitivity of the result, and give HR careful, costed recommendations.5 stages · 19 graded tasks · data: Employee salaries at a fictional company
- Manufacturing · 210 min · MembersManufacturing quality investigationAudit two production processes, build and read control charts, separate special from common causes, judge capability honestly, compare lathes and shifts without pseudo-replication, and recommend what the plant manager should fix first — with the money attached.6 stages · 24 graded tasks · data: Machined shaft diameters
- Finance · 210 min · MembersHow bad can a bad day get? A tail-risk reviewAudit three years of daily fund returns, show why a normal model understates extreme losses, estimate historical Value-at-Risk and expected shortfall with honest uncertainty, test diversification under stress, and recommend how an endowment should size its liquidity buffer.6 stages · 23 graded tasks · data: Daily returns of three funds
- HR · 180 min · MembersEngagement survey study: from fieldwork to a funded decisionRun a complete survey study: judge the design and response, audit response quality, summarise Likert items properly, weight for nonresponse with honest intervals, compare groups without fishing, and cost a retention programme.6 stages · 23 graded tasks · data: Employee engagement survey
- Operations · 210 min · MembersDemand forecasting for planningForecast monthly water-heater demand for a distributor whose market shifted three years ago: audit and decompose the series, handle a structural break, compare benchmarks with Holt–Winters on a holdout and a rolling origin, quantify forecast uncertainty and turn it into an order quantity that balances stock-out and holding costs.6 stages · 23 graded tasks · data: Monthly product demand