Search

Additional file 11:Bland Altman plots for each biological replicate. (PDF 4 MB)

Additional file of RNA-seq: technical variability and sampling

Excerpts from the report Introduction: As response rates decline, the costs associated with how the survey will be designed, conducted and analyzed to ensure the creditability of results increase. Data collection becomes more difficult and costly, and questions are raised about the best way to address the increased nonresponse bias due to low participation rates. The following report provides an in-depth overview of NASS’s subsampling design to address nonresponse in the 2017 Census of Agriculture. The theory and methodology of subsampling in the survey methodological literature is presented, as well as examples of large, nationally representative sample surveys and censuses that use subsampling to improve response and quality metrics. The subsampling method developed by NASS and the conclusions drawn from this inaugural effort are detailed near the end of the report.

Each year, the National Agricultural Statistics Service (NASS) publishes an estimate of the number of farms in the United States based on the June Area Survey (JAS). Independent studies showed that the JAS number of farm indications have significant undercount due to misclassification. To adjust for this undercount, a follow-on survey to the JAS called the Annual Land Utilization Survey (ALUS) has been proposed. ALUS is designed and developed based on the Farm Numbers Research Project (FNRP). NASS conducted the FNRP in the fall of 2009 (Abreu, McCarthy and Colburn, 2010). ALUS samples from all JAS segments containing any estimated or non-agricultural JAS tracts. For a selected segment, all estimated and non-agricultural JAS tracts will be re-evaluated. The collection of eligible segments in a particular year will be called the ALUS population. The sample allocation of ALUS segments to each state-stratum combination considers two factors: the proportion of the ALUS population in the stratum and the proportion of the FNRP adjustment from non-agricultural tracts in the stratum. ALUS can be treated as a second phase to the JAS. The two-phase stratified design, JAS-ALUS, can be applied to any estimate produced by the JAS. However, ALUS has non-response. In this paper, methodology for a three-phase sampling design is developed by extending the two-phase sampling design methodology proposed by Sarndal and Swensson (1987). A general sampling design is allowed in each phase; that is, the inclusion probabilities in each phase are arbitrary. The estimator is unbiased, and an unbiased estimator for the variance is provided. Here, this method is applied to the two-phase JAS-ALUS with the third phase being response/non-response.

Each year, the National Agricultural Statistics Service (NASS) conducts the June Area Survey (JAS), which is based on an area frame. The JAS provides information on U.S. agriculture, including an estimate of the number of farms in the U.S. NASS also conducts the Census of Agriculture every five years in years ending in 2 and 7. The census, which uses both a list and the JAS area frame, also produces an estimate of the number of U.S. farms. In 2007, the two estimates were further apart than could be attributed to sampling error alone. Previous studies of the JAS identified misclassification of JAS sampled units as a source leading to an undercount in the number of farms in the U.S. Using data from the 2007 JAS and the 2007 Census, misclassification of tracts as agricultural or non-agricultural were identified. Research has also identified the estimation of agricultural activities for sampled tracts as another factor that contributes to the discrepancy in the JAS number of farms estimate. This research report presents methodology that adjusts for two known sources of error on the JAS: misclassification and estimation (which later will be addressed as non-response).