Lesson 7: Using intervals to judge claims
Modelling a koala disease survey
SkillsModelling with interval estimatesJudging claims with intervalsChoosing a sample size
Worked example
Researchers test 180 randomly selected koalas in a South East Queensland region and find that 63 have chlamydia.
(a) Calculate an approximate 95% confidence interval for the proportion of koalas in the region with chlamydia.
(b) A news story claims that "at least half of the region's koalas are infected". Evaluate this claim.
(c) The researchers want a margin of error of at most 0.05 in next year's survey. Using this year's result, how many koalas should they test?
(d) A volunteer suggests saving money by testing koalas brought in to a wildlife hospital. Explain why this would be a problem.
Step 1: Interval
. , so the interval is .
Step 2: Check the conditions
and are both large, and the koalas were randomly selected, so the approximation is reasonable.
Step 3: Evaluate the claim
The whole interval lies below 0.5, so the data give evidence against the claim: plausibly 28% to 42% of the koalas are infected, which is serious but less than half.
Step 4: Sample size
, so test at least 350 koalas.
Using this year's as the prior estimate gives a smaller than 0.5 would (385). If the true proportion is nearer 0.5 the margin will be slightly bigger than 0.05.
Step 5: Bias
Koalas brought to a wildlife hospital are more likely to be sick, so they are not a random sample. Their proportion with chlamydia would overestimate the region's, and the margin of error can't correct for this bias.
Answer
(a) (b) The interval is entirely below 0.5, so the data are evidence against the claim. (c) 350 koalas. (d) Hospital koalas are a biased sample, likely to overestimate the infection rate.
Try a similar one
A fresh question of the same kind, checked as you answer, with hints and a full solution.
Method
Calculate and check the interval.
Compare a threshold claim with the whole interval.
Use a prior for the sample size, rounded up.
Comment on randomness: the interval assumes a random sample.
Watch out for
Using the margin of error to "fix" a biased sample: it measures random variation only.