S2 Thesis Design Under Access, Ethics, and Missing Confounders
A master's student in Public Health must choose a thesis design on childhood stunting. Only 3 of 12 subdistricts grant research access; the local ethics board requires parental consent and child assent; schools refuse data collection during exam month; the budget supports 200 households; four months remain. Option A is a cross-sectional survey of 200 households in the 3 accessible subdistricts, measuring height-for-age and household variables. Option B is a case-control study of 100 stunted cases and 100 non-stunted controls drawn from hospital records in the same district; records lack gestational age and birth weight for 40% of cases. The advisor wants a design that can support causal inference about maternal nutrition. The student must submit a proposal in two weeks.
1. Compare the two designs on measurement completeness, confounding, and generalisability.
Read the worked answer
Option A has complete exposure and outcome measurement and covers the accessible population, but it is cross-sectional, so temporality is unclear and it supports association rather than causal inference. Option B has better outcome ascertainment and is efficient for a relatively rare outcome, but 40% missing gestational age and birth weight means a key confounder is unavailable; hospital controls may also be selected on access to care. Recall bias is likely for maternal nutrition. Generalisability differs: A is limited to 3 of 12 subdistricts, while B is limited to hospital-attending children. Neither design as stated supports the advisor's causal goal strongly. A is stronger on measurement completeness; B is stronger on outcome validity but weaker on confounding and missing data.
2. Recommend a design and specify how the proposal should handle missing data, bias, and causal language.
Read the worked answer
Recommend Option A with a retrospective maternal-nutrition module and a pre-specified sensitivity analysis for unmeasured confounding. Add a small validation substudy of about 40 households comparing maternal recall with antenatal cards where available. Acknowledge limited generalisability to the 9 inaccessible subdistricts and avoid causal claims; use association language. If the ethics board permits, add a nested case-control later. Do not choose Option B unless the hospital can retrieve birth records for at least 80% of cases; otherwise the missing confounder makes causal claims indefensible. The proposal should state the design's limitations explicitly, justify the accessible-subdistrict sampling, and describe how consent and child assent will be obtained during the school-exam constraint.