As you extend testing you should move from riskiest populations to less risky (due to more tests). But if you chase hotspots & wait until things are bad to test, you get weird data.

Tests vs Cases in India & S. Korea (from http://ourworldindata.org ).

Implications below.

The South Korea curve makes sense...as testing grows, the probability of finding new cases drop. In India, it's linear. Bc there is a linear relationship btw cases & tests in India, case increase data makes little sense. Had tests expanded exponentially, cases might have as well.

This also explains why we cannot easily use the existing case and mortality data to understand COVID prevalence in these matters.

So should you just do random sampling? Not exactly. You need a more careful strategy. Let's say 1million Indians are infected (which would be super high), that still only 0.07% of the population, so you would have to do a lot of tests to get prevalence rates w serious precision.

My solution (with Jishnu Das and Partha Mukhopadhay) for these problem. Tomorrow in @htTweets: https://www.hindustantimes.com/analysis/a-blueprint-for-a-testing-strategy-analysis/story-OEx409xxBbAXbzLDV1diGI.html