Evaluate The Methodology of The 1997 Agriculture Census Nonresponse Survey

Evaluate Screening Procedure Used In The 1997 Census of Agriculture

Develop a Transparent "Census" File In Off-Census Years

The weighting procedure for the past several censuses integerized both the nonresponse and sample weights at the record level. This process was implemented to ease data review procedures and to ensure that publication totals added to exact figures. The goal of this project is to compare estimates calculated with the noninteger weights to the published estimates (which were calculated with the integer weights) and evaluate how different the estimates are and how this difference relates to the standard error. This project will examine census data from several states and focus on county level estimates for various characteristics (both sample and nonsample)

The goal of the nonresponse survey in the Agriculture Census program is to produce stratum-level estimates for each state (except Alaska and Rhode Island) of the proportion of census nonrespondents that quality as farms. These stratum-level proportions and the final number of census nonrespondents in the state are used to estimate the number of farms among census nonrespondents for each county. The county-level estimates are then used to compute nonresponse weights for the census respondents. 

The goal of this project is to determine an optimal stratification methodology for the 2002 Agriculture Census nonresponse survey. The 1997, the nonresponse survey strata were based primarily on total value of products (TVP) sold. This research will evaluate various stratifications of TVP and other available characteristics.

The 1997 Census of Agriculture utilized several different techniques, such as the screener operation and the advanced follow-up operation, in an attempt to reduce interviewing costs. 

The screener operation identified records which had a low probability of being a farm by using two criteria: automatic screeners (special list records without a good source) and screeners determined by a mathematical model to have a probability of being a farm of less than 30%. A short questionnaire was then administered to determine the scope of the screener records, and out-of-scope records were omitted from further census processing. The advanced follow-up operation identified records which had a high probability of being a census nonrespondent and a low probability of being a farm (i.e., 1992 census nonrespondents, 1997 screener nonrespondents). Records meeting this criteria for which a response to the mailed 1997 census questionnaire was not obtained were forwarded for computer assisted telephone interviewing. The goal of this project is to evaluate the effectiveness of these screening procedures in identifying low probability farms.

A potential opportunity exists for taking advantage of the census of agriculture conducted every fifth year to improve the estimation for off-census years, by creating a "sample census" in the intervening years. This could be done through sample "updating" of the most recent census file or previous "transparent" file to create a new file. Creating a transparent file involves modeling the multivariate relationships between NASS survey and census data and updating all census records with off-census year survey data based on these relationships. The U.S. Bureau of the Census currently uses this concept in creating public use files of demographic data. However, if the relationships among agricultural items are strong enough, the approach could also be used to help bridge our census year and off-census year estimates. This approach could also be used to support editing and imputation systems.

Top of the Page