Salt Marsh Mosquito Survey
A biology team surveys mosquito larvae in brackish pools near a coastal town during the rainy season. They sample 20 pools of equal surface area. In 12 pools with salinity above 15 parts per thousand they find a mean of 40 larvae per dip; in 8 pools with salinity below 5 parts per thousand they find a mean of 5 larvae per dip. The team also notes that the high-salinity pools are all closer to the mangrove edge and further from a recently cleared drainage channel. A local health officer wants to know whether to recommend draining the low-salinity pools or the mangrove-edge pools to reduce the adult mosquito population.
1. The team concludes that high salinity causes higher larval density. Identify the weakness in this reasoning.
Read the worked answer
The comparison is observational, not controlled. Salinity is confounded with location: all high-salinity pools sit near the mangrove edge and far from the drainage channel, so the difference in larval density could be driven by shade, organic matter, predator absence, or proximity to breeding adults rather than salinity itself. There is also no measure of variance or sample-size control, so the means (40 vs 5) are presented without any indication of spread. The claim of causation is not supported; at most the data show an association.
2. Recommend a sampling or intervention design that would let the officer act on evidence rather than association.
Read the worked answer
A defensible design would:
- Stratify and pair pools so that salinity classes are represented at both near-mangrove and far-from-channel locations, breaking the confound.
- Randomise which pools are drained, keeping a matched set of undrained controls, and measure larval density before and after.
- Record covariates (shade, predators, organic content, distance to channel) so their effect can be separated statistically.
- Measure the outcome that matters - adult mosquito catch rates, not just larvae - since larval density does not translate directly into adult nuisance.
The officer should not drain based on the current survey alone; the paired, randomised design gives a causal answer within one or two seasons.