Missing data can be a big problem in survey data; missingness is particularly important in studies of cognitive aging, where there may be higher levels of missing data due to participants being unable or unwilling to complete testing. Although the magnitude of missing data is often lower than other indicators, such as economic variables on income and wealth, this non-random pattern of missing data can lead to bias. Imputation methods create datasets that are easy to analyze for researchers and that result in estimators that properly reflect the relations in the data. Imputation does this by modeling a variable that has missingness as a function of other variables in the data, then replacing missing data with draws from the model. See, for example, Donders, ART, van der Heijden, GJMG, Stijnen, T, & Moons, KGM (2006). *Review: A gentle introduction to imputation of missing values.* *Journal of Clinical Epidemiology*, 59, 1087–1091.

In imputations for harmonized HCAP data we use information on a variety of topics, which may come from either the core HRS family study or the HCAP study:

• Demographics
• Socio-economic indicators
• General health status
  • Self-reported health
  • Activities of daily living and instrumental activities of daily living
  • Chronic conditions
  • Sensory health
• Prior cognition (from core HRS family study or earlier wave)
• Other items from cognitive assessments and informant report

General methods: On cognitive test items, responses of “don’t know” are typically recoded as incorrect (0), whereas refusals are imputed. For informant reports, both are imputed. Because information on the abovementioned covariates was sometimes missing, imputations were conducted in a sequence of blocks of variables. Generally, these blocks were constructed as follows:

1. All core HRS family data
2. Demographic and socio-economic variables from the HCAP study
3. Non-cognitive health variables from the HCAP study
4. Cognitive health variables from the HCAP study (objective cognitive testing and informant reports)

Within each block, we used chained imputation (also known as fully conditional specification) to impute missing data. This involves cycling over the set of variables, using the latest updated imputations of the other variables in the block as covariates. Imputations were pseudo-random draws from regression models predicting each variable with missing data based on all observed information in previously imputed blocks and the current block. More specific methodological details on this process are available in the codebooks for the harmonized HCAP data available here.