Peer Review
Read fellow learners' assignments and case analyses, give structured feedback, and see what peers thought of your own work. Reviewing others sharpens your own analysis.
See peer review in action
Illustrative examplesHere’s what a strong submission and the feedback it draws look like. When you sign in, you review real submissions from other learners the same way — three quick ratings and a few honest lines. Reviewing others is where most of the learning happens.
Weekly challenge · Design one outcome indicator
Program: a maternal-nutrition programme distributing IFA supplements to pregnant women across 40 villages. Outcome indicator: the proportion of enrolled pregnant women who are moderately or severely anaemic (Hb < 11 g/dL) at their third-trimester ANC visit. Numerator: enrolled women with Hb < 11 at the 3rd-trimester visit; denominator: all enrolled women who reached a 3rd-trimester ANC visit in the period. Source: ANC register + point-of-care Hb, pulled quarterly. Why outcome, not output: supplements distributed (an output) says nothing about whether anaemia actually fell; Hb at the 3rd trimester captures the change we care about. Attribution caveat: anaemia is seasonal and multi-causal, so I'd read this against a comparison group of non-enrolled ANC attendees rather than claim the whole fall is ours.
Peer reviews
Clarity ★★★★★Rigor ★★★★☆Usefulness ★★★★★
Genuinely strong — numerator and denominator are unambiguous and the output-vs-outcome reasoning is exactly right. One push: ‘reached a 3rd-trimester ANC visit’ bakes in a selection effect — women who drop out earlier vanish from the denominator, and they may be the most anaemic. Add a coverage indicator alongside so the outcome isn’t flattered by attrition.
— Peer reviewer
Clarity ★★★★☆Rigor ★★★★★Usefulness ★★★★☆
The comparison-group instinct is the best part; most submissions stop at the indicator and just assert attribution. Tighten ‘moderately or severely’ to a single pre-registered Hb cutoff so the threshold can’t drift between rounds — and name who pulls the register and when, or the quarterly cadence will slip.
— Peer reviewer
Weekly challenge · Test an assumption in your Theory of Change
Assumption: community health workers will conduct regular home visits. Evidence that would support it: visit registers cross-checked by a 10% supervisor back-check; a monthly SMS/IVR beneficiary-confirmation poll; GPS timestamps where phones allow. Evidence that would refute it: registers with visits clustered at month-end (a sign of back-filling), or beneficiary recall contradicting the register. Decision rule: if the verified visit rate is below 70% for two consecutive months, the assumption fails and we escalate to a CHW workload/incentive review — rather than assume the downstream outcomes are still on track.
Peer reviews
Clarity ★★★★★Rigor ★★★★★Usefulness ★★★★★
This is what an evidence plan should look like — every assumption paired with what would confirm and what would refute it, plus a decision rule with an actual number. The ‘clustered at month-end’ signal is a sharp, cheap red flag. Only note: the SMS/IVR poll under-represents no-phone households, so lean on the supervisor back-check for those.
— Peer reviewer
Clarity ★★★★☆Rigor ★★★★☆Usefulness ★★★★★
Very usable — the 70%/two-months rule is exactly the kind of thing teams forget to pre-commit to. Define ‘regular’ up front (visits per beneficiary per month) so the register audit has a target to measure against.
— Peer reviewer
Weekly challenge · Draft a sampling approach for a baseline
Baseline for a livelihoods programme across 6 districts. Design: two-stage cluster sample — villages (PSUs) selected PPS by number of eligible households, then a fixed 12 households per village by systematic random walk from a random start. Size: powered to detect a 0.2 SD difference at 80% power, α=0.05, inflated by a design effect of ~2 (ICC ≈ 0.08, cluster size 12) → roughly 60 clusters / ~720 households per arm. Practical notes: over-sample 10% for attrition and refusal; keep one participant list per household across all interventions so the same family isn’t double-counted.
Peer reviews
Clarity ★★★★☆Rigor ★★★★★Usefulness ★★★★☆
The design effect is handled properly — most baselines quietly ignore ICC and end up underpowered. PPS at stage one is right for unequal village sizes. One correction: a fixed 12 per village under PPS gives a self-weighting sample — state that explicitly, it simplifies the analysis and is a genuine strength worth claiming.
— Peer reviewer
Clarity ★★★★★Rigor ★★★★☆Usefulness ★★★★★
The single-participant-list point is the most valuable line here — it’s an operational failure that sinks more baselines than any sampling error. I’d sanity-check the ICC 0.08 against a comparable survey before locking 60 clusters; if the true ICC is 0.12 the sample is short.
— Peer reviewer
