Okay, so this part of Statistics gets complicated real quick, but I’m going to try and explain it.
Now, I’m not really sure which study we’re referencing here, since I think @Vektor was speaking in generalizations about the findings of multiple studies. Here is how one of the those studies could represent the population though.
So let’s I’m doing a study on wether or not people in a certain town like the color orange. We’ll say my fake town has a population size of 10 million people. I’m going take a 1,000 person sample for my study. To get the 1,000 people for this sample, I use a method known as “random sampling.” Essentially people are chosen a pure random opportunity to participate in this study. Obviously people in town are very diverse. They come from different social classes, some might have immigrated to my town rather than being born it in, and there are different sexualities, races, religious beliefs, political beliefs, genders, etc. represented in the population. Hopefully, when I take my random sample, the percentage of those involved in the study who identify as let’s say, transgender, matches the percentage of those who identify as transgender in my town. If the percentages match, then the sample is probably an accurate representation of my town.
Obviously, this is the tricky part of these studies. A random sample can be an accurate representation of a population, but there’s some caveats to that. So, if we assume the studies that have been published about the Japanese public’s opinion on Tokyo 2020 used random samples, and those sample were representative of the target population, then those studies are most likely accurate.
Statistics can go from very simple to understand, to incredibly complicated in the blink of an eye. This is a simple principle at heart, but there are a lot of factors at play. I’m going to leave a link to an article from the The BMJ (a peer-reviewed journal) that explains sampling, in case you don’t get (or don’t like) my explanation. Link: https://www.bmj.com/about-bmj/resources-readers/publications/statistics-square-one/3-populations-and-samples