0

During the covid-19 crisis the number of unreported cases (dark figure) could be very high. To get a more accurate number of total infections the government could take a sample by testing n=1000 (?) households for infection and then calculate approximate total infections for the city. Right?

As an statistical enthusiast I'm wondering what would be a reasonable number of households to test, if the city has 1 million inhabitants.

And secondly, how do you decide which households you are going to test? Do you choose them randomly? If so, random by age and gender or random by geographic location?

user3200534
  • 101
  • As an issue in statistical design you might use formula for binom CI, guess at prevalence, and pick $n$ so margin of error is suitably small. Unless your $n$ might be more than 10% of total population, you can assume a sample without replacement is close to independent. – BruceET Apr 04 '20 at 00:22
  • There are unique difficulties using tests for COVID-19 to estimate prevalence. Little authoritative info publicized on the sensitivity $\eta$ and specificity $\theta$ of these tests. (One "expert " on TV yesterday said maybe $\eta \approx \theta \approx 0.7.)$ Unless $\eta = \theta \approx 1,$ the pct testing positive is nowhere near the same as the prevalence. // Maybe that's why current guidelines tend to restrict testing to people with symptoms or exposure. Perhaps [see](https://stats.stackexchange.com/questions/455129/trying-to-estimate-disease-prevalence-from-fragmentary-test-results). – BruceET Apr 04 '20 at 00:28

0 Answers0