Prior to 2000, respondents were asked to separately report the frequency with which they drank three different kinds of milk. These responses have been recoded and combined to create a harmonized variable (MILKYR; MILKTYPE provides data on the type of milk most consumed). Beginning in 2000 respondents were instead asked about the frequency for milk consumption in general (MILKMNO and MILKMTP), and then asked about type of milk they typically consumed (MILKTYPE). The types of milk available as response categories differ across years. Whole milk and 2% milk are available as response categories in all samples. However, prior to 2000, skim milk, 1% milk, ½% milk, and buttermilk are all included in a single response category. In 2000, response categories include 1% milk, ½% milk, non-fat or skim milk, and "other" in addition to 2% and whole milk. Beginning in 2010, low-fat milk (including 1% and ½%) are condensed into a single response category; other categories include skim or non-fat milk, 2% or reduced-fat milk, whole milk, and other.

Prior to 2000, respondents were asked about foods they usually ate in the past year. Beginning in 2000, this recall period was reduced to foods the respondent usually ate in the past month. The structure of the responses (recording number of times a food was consumed and the time unit associated with that frequency) makes it possible to compare response categories even with different recall periods. In 2005, this format of separate frequency and time units was not used; instead respondents' answers were organized into pre-defined categories. Researchers interested in using more years of available data may consider seeing the 2005 variable MILKMO, and modifying MILKMNO and MILKMTP (or MILKYR) to correspond to the pre-defined categories from 2005.