Differential Privacy (also known as Disclosure Avoidance) is the Census Bureau's method that adds "statistical noise" to the population counts in a way that protects each respondent's identity. Differential privacy uses statistical noise to slightly alter data so that the link between the data and a specific person or business can't be certain. With differential privacy, the Census Bureau precisely controls the amount of statistical noise added using sophisticated mathematical formulas that allow them to assure that enough noise is added to protect privacy but not so much as to damage the statistical validity of their publications. The formulas allow the bureau to balance between two opposing extremes: total accuracy and total privacy. In practice, this means that areas that have a large number of people will have highly accurate statistics, but areas or subpopulations that have just a few people will have proportionately more noise and therefore less accuracy. In this way, the statistical noise prevents someone from learning anything meaningful about any particular individual. You can read more about differential privacy in this article.