5.2 Survey Data

Survey data, often collected for research purposes, are collected differently than administrative data and these differences should be considered in decisions about security, confidentiality and data release.

Administrative data sources (non-survey data) such as: vital statistics (e.g. births and deaths), healthcare administrative data (e.g. Medi-Cal utilization; hospital discharges), reportable disease surveillance data (e.g. measles cases) contain data for all persons in the population with the specific characteristic or other data elements of interest. Most of the discussions in this document pertain to these types of data.

On the other hand, surveys (e.g. the California Health Interview Study) are designed to take a sample of the population, and collect data on characteristics of persons in the sample, with the intent of generalizing to gain knowledge suggestive of the whole population.

The sampling methodology developed for any given survey is generally developed to maximize the sample size with the available resources while making the sample as un- biased (representative) as possible. These sampling procedures that are a fundamental part of surveys generally change the key considerations for protection of security and confidentiality. In particular, the main “population denominator” for strict confidentially considerations remains the whole target population, not the sampled population. But, if persons have special or external knowledge of the sampled populations (e.g. that a family member participated in the survey), further considerations may be required. Also, it is in the context of surveys that issues of statistical reliability often arise—which are distinct from confidentially issues, but often arise in related discussions.

Of particular note, small numbers (e.g. less than 11) of individuals reported in surveys do not generally lead to the same security/confidentiality concern as in population-wide data, and as such should be treated differently than is described within the Publication Scoring Criteria and elsewhere. In this case a level of de-identification occurs based on the sampling methodology itself.