Data Science for Beginners

1. 1Data science begins with a decision, population and clock
2. 2Data science core concepts: grain, distributions, uncertainty and action

Work through one boundary at a time and verify the example against real data.

1. 1Dataset grain and join cardinality: protect the unit of analysis
2. 2Missing data policy: distinguish absence from a measured zero
3. 3Exploratory analysis without peeking: inspect the data and preserve the test
4. 4Metric denominators and cohorts: make a rate reproducible
5. 5Reproducible analysis snapshots: pin data, code and cutoff together
6. 6Bootstrap intervals: estimate uncertainty at the right sampling unit

Audit the frame, design and response process before claiming a population rate.

1. 1Sampling frames and coverage error: who could enter the analysis?
2. 2Stratified sampling and design weights for operational estimates
3. 3Nonresponse bias: diagnose the missing outcomes before adjusting
4. 4Precision budgets and effective sample size for weighted analyses

Validate fields, investigate tails and distinguish composition from process change.

1. 1Column profiles and domain constraints before analysis
2. 2Outlier investigation: impossible values, rare events and quarantines
3. 3Resistant summaries: median, trimmed mean and median absolute deviation
4. 4Subgroup distributions and aggregation reversal

Define entry, preserve open outcomes, audit proxy fields and test assumptions.

1. 1Cohort entry and survivorship bias: count the cases that could fail
2. 2Right censoring and time-to-event analysis for open cases
3. 3Proxy measurements and label error in operational datasets
4. 4Sensitivity analysis: find which assumptions can reverse a decision

Keep unknowns, denominators, event versions and row counts explicit.

1. 1SQL null semantics: count records without inventing outcomes
2. 2SQL conditional aggregation: rates with an auditable denominator
3. 3SQL window functions: select one event with a deterministic rule
4. 4SQL reconciliation: prove the cohort survived each transformation

Fix event identity, sequence paths, mature cohorts and reconcile independent outcomes.

1. 1Product event contracts: identity, deduplication and eligibility
2. 2Ordered funnels: step order, entry rules and conversion windows
3. 3Retention cohorts: return behavior only after a full observation window
4. 4Instrumentation changes and metric guardrails for product analysis

Preserve units and resolution, check device bias and carry uncertainty into decisions.

1. 1Quantity units and conversion contracts for mixed datasets
2. 2Measurement resolution and rounding: avoid invented precision
3. 3Calibration drift: compare sensor readings with a reference
4. 4Derived metrics: carry input bounds and shared error into the result

Preserve namespaces, evaluate candidate coverage and review merge effects.

1. 1Identifier normalization and exact linkage without false merges
2. 2Blocking for record linkage: reduce comparisons without hiding true matches
3. 3Match scores and review bands: separate similarity from identity
4. 4Entity merge lineage: show how identity decisions change metrics

Translate estimates into feasible actions, whole-number plans and stress-tested choices.

1. 1Decision objectives: define the action before optimizing a score
2. 2Constraint feasibility and slack: reject impossible plans early
3. 3Integer allocation and relaxation gaps for indivisible work
4. 4Scenario stress tests and regret for uncertain allocation benefits

Diagnose missingness, selection, safe imputation and sensitivity to unseen outcomes.

1. 1Missingness mechanisms: model why a value is absent
2. 2Complete-case selection: know whose outcome remains
3. 3Train-only imputation and missingness indicators
4. 4Missing-not-at-random sensitivity: test what unseen outcomes could change

Check priors, update event rates, quantify decisions and inspect group pooling.

1. 1Prior predictive checks for a binary service rate
2. 2Beta-binomial updating with an auditable case count
3. 3Posterior intervals and threshold decisions
4. 4Shared priors, shrinkage and the limits of fixed pooling

Define the survey construct, code response states, test the form and preserve comparable versions.

1. 1Survey constructs, response units and recall windows
2. 2Response options and ordinal coding without false precision
3. 3Cognitive pilots and survey branch-logic audits
4. 4Question versions, order effects and comparable trends

Criticize model fit, stress priors, score future cases and replay mature batches.

1. 1Posterior predictive checks for a hidden group gap
2. 2Prior sensitivity for an operational action threshold
3. 3Held-out predictive log scores for a binary rate model
4. 4Sequential posterior monitoring with mature outcomes

Track selection, adjust response, calibrate to controls and inspect precision.

1. 1Survey inclusion probabilities and base weights
2. 2Nonresponse adjustment cells and the positivity check
3. 3Survey calibration to known population margins
4. 4Weight extremes, effective sample size and trimming tradeoffs

Build risk sets, estimate unresolved time, separate terminal outcomes and report a supported horizon.

1. 1Event-time risk sets and tied outcomes
2. 2Kaplan–Meier curves for support resolution
3. 3Competing risks and resolution incidence
4. 4Restricted mean unresolved time at a business horizon

Handle delayed entry, quantify uncertainty, avoid immortal time and compare queues at one horizon.

1. 1Delayed entry and left-truncated risk sets
2. 2Survival-curve uncertainty and thin-tail support
3. 3Landmark analysis and immortal-time traps
4. 4Compare queues by restricted mean open time

Specify a modeled decision, control simulation error, preserve shared shocks and compare policies on paired shifts.

1. 1Monte Carlo estimands and reproducible draws
2. 2Simulation error and replication budgets
3. 3Joint inputs and correlated operational shocks
4. 4Paired policy simulations with common random numbers

Separate service targets from control limits, account for changing denominators and investigate signals against a frozen baseline.

1. 1Control limits and service targets answer different questions
2. 2P-charts with changing daily denominators
3. 3EWMA monitoring for small persistent rate shifts
4. 4Frozen baselines and alert investigation

Interpret relative hazards, test time-constant effects, preserve inspection intervals and validate fixed-horizon probabilities.

1. 1Cox partial likelihood and hazard-ratio interpretation
2. 2Diagnose the proportional-hazards assumption
3. 3Interval-censored support resolution times
4. 4Calibrate resolution predictions at a fixed horizon

Build time-updated case histories, examine censoring support and validate survival predictions on later arrivals.

1. 1Time-updated case histories and the observation clock
2. 2Counting-process intervals for a time-varying Cox model
3. 3Censoring survival weights and horizon support
4. 4IPCW Brier score for censored survival predictions
5. 5Temporal validation of a support-resolution model

Define competing exits, estimate incidence, validate event probabilities and quantify score variation.

1. 1Competing-outcome contracts at a fixed horizon
2. 2Aalen–Johansen cumulative incidence from case events
3. 3From cause-specific event rates to absolute risk
4. 4Calibrate competing-risk predictions at a fixed horizon
5. 5Paired bootstrap uncertainty for prediction-score differences

Build branch-month panels, compare adoption cohorts, inspect pre-trends and spillovers, then report branch-level uncertainty.

1. 1Branch-month panels and the rollout clock
2. 2Cohort-time difference-in-differences for staggered adoption
3. 3Event-time pre-trends and support for a staggered rollout
4. 4Spillover-aware comparison pools for branch rollouts
5. 5Branch-cluster bootstrap for a rollout contrast

Inspect assignment support, influential branches and indirect exposure before claiming a rollout effect.

1. 1Assignment-aware permutation tests for a few branches
2. 2Leave-one-branch-out stability for rollout effects
3. 3Direct and indirect exposure maps for branch rollouts
4. 4A spillover difference-in-differences contrast
5. 5Negative-control outcomes and placebo rollout checks

Audit donors, fitted counterfactuals and the population to which a local effect may apply.

1. 1Synthetic-control donor eligibility and pre-fit
2. 2Synthetic-control weight search and prediction
3. 3Synthetic-control placebos and fit ratios
4. 4Synthetic-control donor and window sensitivity
5. 5Target-population standardization for branch effects

Audit intervention timing, level and slope breaks, comparison paths, residuals and lags.

1. 1Interrupted-series outcome clock and data contract
2. 2Segmented regression for level and slope changes
3. 3Controlled interrupted time series with a comparison series
4. 4Seasonality and residual dependence in interrupted series
5. 5Transition windows and lag sensitivity for policy series

Design a reference audit, estimate group-specific label error and stress-test the decision.

1. 1Design a validation subsample for outcome labels
2. 2Estimate outcome-label sensitivity and specificity by group
3. 3Correct a binary outcome rate for label error
4. 4Differential outcome-label error in group contrasts
5. 5Misclassification assumption grids and decision ranges

Keep missing-outcome assumptions visible and test whether the policy decision survives them.

1. 1Partial identification and an assumption ledger
2. 2Worst-case bounds for missing binary outcomes
3. 3Treatment-effect bounds under differential attrition
4. 4Bounded missing-outcome risk scenarios
5. 5Decision thresholds and follow-up value under bounds

1. 1Project: audit a receipt-review rate from raw submissions
2. 2Project: estimate support-case defects from an auditable sample
3. 3Project: audit fulfillment delays without hiding the tail
4. 4Project: audit support resolution with open cases and imperfect labels
5. 5Project: build an auditable support-intake SQL report
6. 6Project: audit checkout conversion and repeat purchase
7. 7Project: audit cold-chain sensor measurements before classifying excursions
8. 8Project: resolve customer records and audit the metric restatement
9. 9Project: allocate support reviews under capacity and uncertainty
10. 10Project: audit missing delivery scans before reporting service rates
11. 11Project: Bayesian monitoring of support escalations
12. 12Project: measure support-handoff clarity with a versioned survey
13. 13Project: check a Bayesian support-rate model before acting
14. 14Project: audit weights for a support-handoff survey
15. 15Project: audit support resolution time and competing exits
16. 16Project: audit specialist-queue resolution comparisons
17. 17Project: simulate warehouse capacity under shared shocks
18. 18Project: release a support breach monitoring packet
19. 19Project: review a support-resolution survival model
20. 20Project: validate a time-updated resolution model
21. 21Project: review resolution predictions with competing exits
22. 22Project: audit a staggered support-branch rollout
23. 23Project: review a small branch rollout with spillovers
24. 24Project: review a single-branch policy and its transfer
25. 25Project: audit an interrupted policy series
26. 26Project: audit a deadline-breach outcome label
27. 27Project: decide a routing policy with missing outcomes