Source: U.S. Census Bureau
The SNAP data are used in the production of the Census Small Area Income and Poverty estimates (SAIPE). This release is located at production purposes (https://www.census.gov/programs-surveys/saipe.html). The SNAP data set is not raw SNAP data. Below are some details about the reference periods, averages computed, raking/scaling done, and imputations done, which make the data most suitable for SAIPE production purposes, but which also apply to the posted SNAP data, located here https://www.census.gov/data/datasets/time-series/demo/saipe/model-tables.html.
The county-level SNAP data are almost always in reference to July of the given year, except in a small number of instances another nearby month had to be used. The numbers are usually a snapshot as of the end of the month, start of the month, or middle of the month, and occasionally can be a measure of anyone enrolled any time during that month. Also, not all states compile their figures exactly the same way.
Annual county-level data are controlled data are controlled to sum to the twelve-month average of the state data in the following way: Add up the county-level July 20YY data in each state, and compare it to the 12-month average of the state-level July 20YY through Jun 20YY+1 data. Then scale the county-level data up or down in such a way that the state sum of county-level data exactly equals the 12-month average of the state-level data. For example, with the July 2012 SNAP data, the Census controls the county sums to equal the state 12-month average of July 2012 through July 2013 data. After this scaling (i.e., raking) is done, note that the resulting 12-month average county-level SNAP data, at the state level, are centered in time around Dec 20YY and Jan 20YY+1, which is about six months farther in the future than the original July 20YY county-level SNAP data. However, the county-level sub-detail still refers to the July 20YY timeframe.
Importantly, note that a small number of "anomalous" data points are "imputed" prior to posting. Sometimes, there may be large but temporary data drop-offs in a small number of counties that appear to have had reporting issues for the given time period. In such cases the Census may replace the anomalous county-level data as needed to run our SAIPE production processes. For example, the Census replaced anomalous data with an average of the prior and next year of data, or the Census may apply that county's last-year SNAP rate scaled up by the year/year percent change in the state SNAP rate, etc. Especially, when there are officially-declared natural disasters, states may loosen the normal SNAP eligibility criteria, and there may temporarily be many more SNAP enrollees. As a result, the more typical relation between SNAP and poverty status is broken. For SAIPE to still use such SNAP data in our models, the Census need to first remove such disaster-generated spikes and replace them with seasonally-adjusted predictions. For example, in the raw SNAP data, in Louisiana, there was a spike for a couple months in late 2005, however, in the posted SNAP data; the Census has removed those one-time spikes.
For further information, see the link provided below.
U.S. Census Bureau, SNAP Benefits Recipients in Twin Falls County, ID [CBR16083IDA647NCEN], retrieved from FRED, Federal Reserve Bank of St. Louis; https://fred.stlouisfed.org/series/CBR16083IDA647NCEN, October 15, 2019.