The American Community Survey provides estimates based on a sample of US households. The Census surveys about 3.5 million households *per year*, so over 5 years, about 17.5 million households have been surveyed. The boffins at Census (who are really quite good statisicans) then roll up this data to predict things like Internet Subscription Characteristics across the country. The estimates are really quite good, but they are just estimates.

Each estimate comes with a margin of error — basically the range over which there is a 90% probability that the actual value lies. So if the estimate is 90% ± 2%, you have a 90% probability that the actual value is between 88% and 92%. In order for you to properly understand the estimate, *you have to look at the margin of error*. That’s why they’re provided, and hard to ignore.