The whole curriculum, before you start it

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Installing a working environment offline, and why uv or conda beats a system Python
• Reading CSV, Excel and fixed-width exports without losing leading zeros in identifiers
• pandas selection, filtering and grouping, with the SettingWithCopyWarning explained rather than silenced
• Dates, ages and reporting periods — the arithmetic behind an age-in-months disaggregation
• Categorical codes, missing-value sentinels, and turning 99 into NA before it becomes a mean
• Writing a script someone else can run — arguments, paths and a reproducible output folder

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Projects, renv and the working-directory habits that make an analysis reopenable a year later
• readr, readxl and haven for CSV, Excel, Stata and SPSS exports, keeping value labels intact
• The dplyr verbs, and why group_by() + summarise() is where most indicator bugs are born
• Factors with forcats — ordering categories so a table reads the way the report needs it to
• tidyr pivots, and the pivot_longer() call that undoes a flattened repeat group
• Writing functions, so the same cleaning runs on next quarter's export unchanged

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Profiling a fresh export before touching it — row counts, key uniqueness, value ranges
• Missingness that is not random, and why a per-site missingness table beats a global percentage
• Duplicate detection when there is no clean identifier — fuzzy matching on name, age and village
• Plausibility rules from the sector — a MUAC of 250 mm on a six-month-old is a data error, not a finding
• Harmonising free-text site and village names against an administrative list
• The cleaning log — every rule, how many rows it touched, and who approved the deletion

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• The four join types, and how to prove a join did not silently drop or duplicate rows
• Household-to-member joins, and the many-to-many that turns 400 households into 9,000
• Long and wide, and choosing the shape the analysis needs rather than the shape it arrived in
• Joining programme data to an administrative population frame for a coverage denominator
• Reconciling a facility register against a DHIS2 aggregate, and reporting the gap honestly

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• The five quality dimensions — accuracy, completeness, timeliness, consistency, integrity
• Verification factor — recounting source documents against what was reported upward
• Trend and outlier checks that flag a site reporting a suspiciously round number every month
• Sampling facilities for verification when you cannot visit all of them
• Writing the DQA report — findings, root cause, corrective action, owner, deadline

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Theory of Change to LogFrame to indicator, and what breaks at each translation
• The indicator reference sheet — numerator, denominator, disaggregation, frequency, source, and the decision it informs
• Reach, coverage and cumulative counts, and why summing monthly reach double-counts people
• Standard indicator sets — UNICEF, WHO, the SDG framework — and when to adopt rather than invent
• Targets, baselines and the OECD DAC criteria a mid-term review will apply to your numbers

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Why an unweighted mean from a cluster survey is the wrong number, demonstrated on real weights
• Sample size for a prevalence estimate, and what precision the budget actually bought
• Design effect and intra-cluster correlation, and reading them off an existing survey report
• survey package in R and the equivalent survey-aware estimation in Python
• Non-response, replacement clusters, and what they do to the estimate you publish
• Reporting an estimate with its confidence interval, and refusing to disaggregate past the point the sample supports

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Data elements, category combinations and indicators, and how a DHIS2 pivot is assembled
• The org unit hierarchy, and aggregating up without double-counting a reassigned facility
• Reporting rate and reporting timeliness as denominators in their own right
• Pulling data through the Web API instead of exporting by hand every month
• Reconciling routine aggregates against a survey estimate of the same indicator

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Computing weight-for-height, height-for-age and weight-for-age z-scores against the WHO 2006 standards
• MUAC, oedema and the case definitions for SAM and MAM, and where the two admission criteria disagree
• SMART plausibility checks — flags, digit preference, age heaping, standard deviation of the z-score
• GAM and SAM prevalence with confidence intervals, and reading them against the IPC acute malnutrition phases
• CMAM performance — cured, defaulted, died, non-response — against the Sphere minimum standards
• Coverage estimation, and why programme admissions over expected caseload is not coverage

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Rates, ratios and proportions, and the person-time denominator that makes incidence meaningful
• The HIV treatment cascade and the TB care cascade as denominators chained together
• Immunisation coverage — administrative, survey and the dropout rate between doses
• Epidemic curves, attack rates and case fatality, computed on line-list data
• Age-standardisation, and why two districts are not comparable without it
• Confounding in routine data, and the limits of what an observational comparison can claim

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• The JMP service ladders for water, sanitation and hygiene, applied to household survey data
• Litres per person per day, and the queue and distance data the Sphere standard also requires
• Water quality — E. coli counts, free residual chlorine, and detection limits in a censored dataset
• Water point functionality over time, and separating seasonal failure from abandonment
• Open defecation, latrine coverage and menstrual hygiene management indicators

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Food Consumption Score — the weights, the recall period and the two competing thresholds
• Reduced Coping Strategies Index and the Livelihood Coping Strategies module
• Household Hunger Scale and Household Dietary Diversity Score, and when each is the right instrument
• Market price and terms-of-trade analysis feeding the same classification
• IPC evidence tables, convergence of evidence, and what a Phase 3 classification actually asserts

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Safe, ethical data handling — the GBV information management principles and what they forbid
• Disclosure risk, small cell suppression, and why a district table can identify a survivor
• Referral pathway analysis — time to service, completion rate, and where a pathway leaks
• Case management caseload, case closure reasons and average time to closure
• Incident data as a measure of reporting, not of prevalence, and stating that in the report

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Gross and net enrolment ratios, and the age-appropriate denominator each one needs
• Attendance from a daily register — average daily attendance versus the proportion regularly attending
• Dropout, repetition and survival to the final grade, computed across a cohort
• Learning assessment data — raw scores, proficiency bands and comparability across rounds
• Disaggregation by sex, disability and displacement status, and the sample it demands

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Distributions, and reading a histogram of a real indicator before computing anything on it
• Confidence intervals, and writing a result as an interval in the narrative of a report
• Comparing two groups — t-test, proportion test, chi-square — and checking their assumptions
• p-values, effect size, and why a significant difference can be operationally irrelevant
• Multiple comparisons across twelve districts, and the false positives it manufactures
• Correlation, and the sentence in your report that quietly turns it into causation

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Linear regression as a comparison of means with covariates, not as a prediction machine
• Logistic regression, odds ratios and risk ratios, and which one your audience will misread
• Adjusting for confounders, and the covariates you must not adjust for
• Clustered and multilevel data — children within households within villages
• Survey-weighted regression, so the model matches the sampling design
• Diagnostics, and reporting a model that does not fit rather than dropping it

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• The counterfactual, and why a before-after change is not an effect
• Randomised designs, cluster randomisation and what happens when randomisation is refused
• Difference-in-differences and the parallel trends assumption you have to argue for
• Propensity score matching, and honest reporting of the units it drops
• Regression discontinuity around an eligibility threshold — a MUAC cut-off, a poverty score
• Power calculation before the evaluation, and what an underpowered study is allowed to conclude

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Choosing the mark before the library — what comparison the chart is actually making
• Uncertainty on the page — error bars a programme audience will not misread
• Sector conventions — IPC phase colours, MUAC bands, and the palettes you must not repurpose
• Small multiples across districts, and when a map is worse than a sorted bar chart
• Accessibility and print — colour-blind safe palettes, greyscale survival, minimum type size
• Labelling a chart so it can be lifted into a report and still be correct

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Deciding what the dashboard is for, and refusing the twenty-tile request
• Layout for a decision — the three numbers first, the detail behind a click
• Refresh and provenance — every figure carries its as-of date and its source
• An indicator definition panel, so the number and its definition travel together
• Writing the narrative report — findings, limitations, recommendation, and the annex that shows the working
• Presenting to a steering committee when the finding is not the one they wanted

Not yet written. The syllabus below is a commitment: when the course ships, its lessons cover every line of it.

• Project structure — raw data read-only, derived data disposable, code the only thing edited
• Git for analysts, including what never belongs in a repository that holds beneficiary data
• Pinned environments with uv or renv, and reproducing a result on a colleague's laptop
• Quarto and parameterised reports — one template, twelve district reports, no copy-paste
• Automated checks on the pipeline, so a broken upstream export fails loudly
• Handover documentation, written for someone with your job and none of your context