The purpose of this paper is to stimulate and support broad discussion within the Bureau of alternative disclosure-avoidance techniques for summary data from the 1980 census, specifically: suppression, random rounding, and other forms of introducing random variation.
The degree of protection required against statistical disclosure is necessarily a matter of judgment. The recently published Statistical Policy Working Paper 2 "Report on Disclosure and Disclosure-Avoidance Techniques" makes the important point that it is unreasonable to attempt to absolutely prevent any disclosure. For example, a statistic indicating that a block is 100% Black discloses a characteristic of every resident of that block. Disclosure can be probabilistic as well, e.g., a figure of 90% discloses an individual's characteristic with high probability. Some level of disclosure must be tolerated in order to achieve important societal benefits. The task that this paper addresses is which technique most effectively maximizes utility of the data while reducing the risk of disclosure about individuals to an acceptable level.
One may note at this point that the use of the term suppression in this paper should not be confused with the practice of withholding certain derived statistics (e. g. , medians or percents) if there were fewer than 100 persons, families or households in the distribution; or the practice of showing race/ethnic detail in the Census Tracts, reports only if there were 400 or more persons in the race/ethnic category in a particular tract. In these cases the motives were statistical or practical and not related to the avoidance of disclosures about individuals.