Using Sample Weights

Sample weights are a critical methodological tool in survey research. The calculations that go into sample weights are typically done behind the scenes. As an end user of the dataset, you don't often need to know much about how exactly the weights are calculated. It's still helpful to understand what sample weights are and, in broad terms, how they are implemented in the National Crime Victimization Survey (NCVS).

Because surveys are randomized, even a tiny fraction of a large population can tell us about the whole. The foundation of a sample weight is a probability weight, which is the inverse of the chance that case was selected in the survey. To look at it another way, the probability weight counts how many individuals in the population are represented by a single case in the survey. Researchers multiply results from the sampled data by probability weights to see how those numbers look in the full population.

So if the adult U.S. population is 250,000,000 and 1,000 adults are surveyed via a simple random sample, each of those cases have an equal probability weight of 250,000 (250,000,000/1,000). When that survey finds a total of 320 cats owned in the sample, using the probability weight alone, the population estimate from the sample would be 80,000,000 cats nationwide (320 x 250,000).

In practice, however, data from surveys like the NCVS do not result from a simple random sample where every case had an equal chance of being selected. Cases in the NCVS are selected in a more complicated survey design, and each case does not have an equal chance of selection from the population. Very briefly, these are the most significant factors in the NCVS sample weights that you will use in this exercise:

• A weighting control factor adjusts for any changes in the field or in how the survey design was actually implemented.
• Weights for households and for people are adjusted to try to represent those households and people that did not respond to the interviews.
• Because we know demographic features of the U.S. population with some precision, adjustments are made to the weights to bring the sample in line with characteristics like age, sex, and race.

The goal of sample weight calculation is to maximize generalizability: the degree to which sample results can be projected to the population. Unweighted analyses have their place, but they do not provide the most representative results. Weighting the data appropriately, your results can provide the best estimates available of victimization in the U.S. population.

Want to read more detail on the weights used in the NCVS? Further information on weights and error in sampling can be found in our NCVS resource guide. Details on the broader sample design of the NCVS is available in a report from the Bureau of Justice Statistics.