1. Adjustment Methodologies for the Census of Agriculture Andrea C. Lamas, Denise A. Abreu, Shu Wang, Daniel Adrian, Linda J. Young National Agricultural Statistics Service (NASS) Census of Agriculture NASS conducts hundreds of surveys including the Census of Agriculture. NASS prepares reports covering every facet of United States agriculture. For example: Production and food supplies Prices paid and received by farmers Farm income and finances Number of farms and land in farms A farm is any place from which $1,000 or more of agricultural products were produced and sold or normally would have been sold during the year. Some special case examples are: Christmas trees "government payment" farms "pasture only" farms (at least 100 acres) nurseries and greenhouses exotic livestock Conducted every 5 years (years ending in 2 and 7) Count of all US Agricultural operations ($1000 or more in sales) Also collect information on agricultural operations’ commodities and operator demographics Only source of uniform, comprehensive agricultural data for every county or county equivalent in the U.S. Is a list-based survey. The list is referred to as the Census Mail List (CML). Primarily mail data collection Under-coverage Incompleteness of the list, which occurs when not all agricultural operations appear on the Census Mail List Nonresponse Not all agricultural operations on the Census Mail List respond Errors in Census reporting: Misclassification of Census non-farms Occurs when non-farms are classified as farms. This misclassification includes a subset of non-farms. Misclassification of Census farms Occurs when farms are classified as non-farms. This misclassification omits a subset of farms. Sources of Error U.S. Farms U.S. Agricultural Operations Operations on CML Census Respondents Census Farms June Area Survey (JAS) The JAS is an area-based survey. It is conducted annually and is a theoretically complete sampling frame with no overlaps or gaps. It uses a sampling rotation scheme every year with 20% of the sample replaced each year. A sample rotation remains for 5 years. It is a stratified sample based on land-use and percent of cultivation Segments of land are sampled. Sampled segments are divided into tracts that represent unique land operating arrangements. The survey is conducted with in-person interviews. Crop and livestock information is collected only on the agricultural tracts. Capture Recapture Methodology Dual system estimation (DSE) requires two independent surveys. The goal is to get estimates for the Census of Agriculture. Sample 1: Census of Agriculture records overlapping JAS tracts (not all census records) Sample 2: June Area Survey tracts DSE has a primary assumption that there is independence of the Census and the JAS. DSE also assumes that the proportion of JAS farms capture by the Census is equal to the proportion of U.S. farms captured by the Census. DSE adjusts for farms that are not captured by either the Census of the JAS. Data required for the DSE are the matched Census records and JAS tracts. In order to obtain this, the 2012 CML and JAS tracts are overlapped. The records kept are CML respondents/non-respondents matching JAS tracts. Also farms on the JAS that are not on the CML are kept. U.S. Agricultural Operations CML Records JAS Tracts Census Sample Unresolved Records In the matched dataset, farm status based on the Census and the JAS agree in most cases. Resolved farm status. Some records are identified as farms (non-farms) on the JAS and non-farms (farms) on the Census. Unresolved farm status To account for this, logistic model of the probability of an operation being a farm based on records with resolved farm status is developed. The final model is used to estimate the probability that each of the agricultural operations with unresolved farm status is a farm. Normalized JAS weights are used in the model. The probability that an unresolved record is a farm will be reflected in a reduction of the associated JAS weights. where = initial JAS weight = predicted probability that a record is a farm = adjusted weight that will be used in the regression models Logistic Regression The matched dataset is used to model the probabilities of a farm being on the CML, of responding, and of responding as a farm.  C =  (CML|Farm)  R =  (Responded|CML,Farm)  CCF =  (CML Farm|CML,Responded,Farm) Some farms on CML may be misclassified and may be non- farms. Use matched dataset to also model the probability of being classified correctly  CCFC =  (Farm|CML Farm) Logistic regression is used to model the probabilities. JAS survey weights are used in the models. The weights are applied to the in-scope records (CML respondents that are farms). weight = Example Conclusions Dual System Estimation adjusts for farms not captured by either the Census or JAS. It identifies farms that have been undercounted or previously missed by the Census (small and minority farms). JAS Tracts Census List P(R,CML, Census Farm | Farm) P(Farm | Census Farm) Total Farms = Sum Weights CML Respondents Matches Resolve Weights for Unresolved Records