Without knowing the specifics (because it's not mentioned in the Yale press release), I can only guess. But it's something that we did in grad school to reduce

Without knowing the specifics (because it's not mentioned in the Yale press release), I can only guess. But it's something that we did in grad school to reduce sample processing costs looking for rare genetic variants. The main idea is that if you are looking for a rare event, but testing an individual is expensive, time consuming, or difficult, then you can save some of that time and money by performing multiple tests at one time. So, instead of using a single test to see if a single person is infected, you'd use a single test to see if a pool/batch of people are infected. Imagine a university with 20,000 students that has decided to test everyone once a week. Ordinarily, they'd need to run 20,000 tests a week. Now, assume they are going to use pooled testing. If your student population is 20,000 and the pool size is 100, instead of 20,000 tests a week, you'd only need to run 200 a week. 200 tests a week is manageable... 20,000 is not (from time, cost, and interpretation standpoints). The machines that run the qPCR test run in batches of 96 or 384 (not including controls), so you can see how reducing your total numbers of tests would be useful. Pooling then answers the question of if anyone in a particular pool has Covid-19. If you have a positive signal from a pool, you should then go back and re-test all of the members of the pool. Even if you have multiple pools come back as positive and need to retest entire pools, you'd still be saving tests (and time/money). 20,000 vs 200 + 100 from each positive pool. There are a few variations of pooling that can help make this even more efficient. If instead of 200 pools for 20,000 students, what if you had each student be part of 2 pools? Then you'd need twice as many tests per week (400), but if you have positive signals, then you'd only need to retest the students that were in both pools, which should be a very small number. If you went from having to retest 200 people to 5 people, then you're looking at even more savings. The requirement for 2 pools to have a hit is also a quality control measure -- what if you had a false positive in one pool, but no second hit from another pool... then you might assume that it was a false-positive and skip re-testing entirely. This is only something that really works with rare events. But if you have a test with known false-positive and false-negative rates, then you can really save a lot with pooled testing... even with lower rates of pooling than 100. If you had pools of 5-10, you'd still be saving a lot of test reagents and costs. 20000 vs 2000-4000 is still something that would be a big advantage.