Improving Crop Estimates by Integrating Multiple Data Sources (2017) / Chapter Skim
Currently Skimming:
2 A Vision of NASS in 2025
2 A Vision of NASS in 2025
Pages 17-54
The Chapter Skim interface presents what we've algorithmically identified as the most significant single chunk of text within every page in the chapter.
Select key terms on the right to highlight them within pages of the chapter.
From page 17... ...
First, NASS prepares its county estimates using a transparent and well-documented process, publishing measures of uncertainty along with point estimates. Second, the NASS list frame is a georeferenced farm-level database, serving as a sampling frame for surveys and facilitating the use of farm data in statistical analysis.
|
From page 18... ...
Instead, as summarized in Chapter 1 and described in more detail later in this chapter, NASS uses the Agricultural Statistics Board (ASB) process or expert judgment to determine the official estimate given a variety of auxiliary indications.
|
From page 19... ...
2. se auxiliary data to improve precision and/or reduce the error associ U ated with direct survey estimates.
|
From page 20... ...
Especially for crop estimates, the auxiliary information available to NASS provides strong predictors. Vision of a Georeferenced Farm-Level Database Development of a georeferenced farm-level list frame, the second component of the vision, would be done most expeditiously by adopting the geo spatial convention already in use by the Farm Services Agency (FSA)
|
From page 21... ...
As discussed, multiple data sources, along with the accompanying measures of uncertainty, will be integrated through formal statistical models that will provide the basis for ASB's decisions. ASB will exercise its judgment in deciding when model estimates can be improved by consideration of unforeseen events (e.g., a drought or a hurricane)
|
From page 22... ...
Published FIGURE 2-1 Summary of information flow according to the panel's vision of the National Agricultural Statistics Service in 2025. NOTES: ASB = Agricultural Statistics Board; FSA = Farm Services Agency; RMA = Risk Management Agency.
|
From page 23... ...
After giving a basic definition of a model in Box 2-2, this section describes the current ASB process for estimation, as well as four formal statistical models developed by NASS, three of which are already used to provide input to the ASB process. The Agricultural Statistics Board Process For national- and state-level estimates, ASB meets in a carefully controlled "lock-up" to review all available indications (survey, administrative, and/or model-based)
|
From page 24... ...
To incorporate survey and auxiliary information into its crop estimates, NASS computes composites of available inputs for its acreage and production totals. The ASD-level composites are ratio adjusted to the published state total, and then rounding rules are enforced in producing the official ASD statistics.
|
From page 25... ...
For cash rents, the direct survey estimate is considered most important, and the only alternative indication available is a model-based result. Estimates established by RFO staff are submitted for review to ASB, with headquarters and RFO staff working from their respective duty stations.
|
From page 26... ...
The final review of well-defined estimates is commonly performed by analysts in other statistical agencies.5 The process for reviewing estimates needs to follow clear guidelines and result in a limited number of welldocumented adjustments to otherwise automatically produced estimates, thus ensuring the transparency of the process. RECOMMENDATION 2-1: The National Agriculture Statistics Ser vice should evolve the Agricultural Statistics Board role from one of integrating multiple data sources to one of reviewing model-based predictions; macro-editing; and ensuring that models are continually reviewed, assessed, and validated.
|
From page 27... ...
is the product of the direct survey estimate for the ratio of harvested acres to planted acres and the composite estimator for planted acres from a row above. The rows reflect the indications that will be determined during the ASB process: planted acres, harvested acres, and production.
|
From page 28... ...
NOTES: FSA = Farm Services Agency; RMA = Risk Management Agency. SOURCE: Data from presentation by Lance Honig, National Agricultural Statistics Service, to the Panel on Methods for Integrating Multiple Data Sources to Improve Crop Estimates, November 13, 2015, Washington, DC.
|
From page 29... ...
model, along with classified pixels from the CDL and segment mapping data from the June Area Survey, to produce remote sensing indications of planted acres by crop. According to Johnson (2014)
|
From page 30... ...
NASS implemented this model and used the results, along with direct survey estimates (the primary indication) , as an indication for cash rents in the ASB process during 2013, 2014, 2016, and 2017.
|
From page 31... ...
methods and compared the model results with survey estimates, estimates prepared by ASB, and either FSA planted acres or remote sensing estimates where applicable. Ratio benchmarking was applied to posterior iterations.
|
From page 32... ...
Note that achieving this goal requires access to statistical/ modeling talent through in-house staff, collaboration with other USDA and government agencies, and collaboration with academic researchers. RECOMMENDATION 2-2: The National Agricultural Statistics Ser vice should achieve transparency and reproducibility by developing, evaluating, validating, documenting, and using model-based estimates that combine survey data with complementary data in accordance with Office of Management and Budget standards.
|
From page 33... ...
These model-based single-year estimates are more reflective of current conditions relative to multiyear survey estimates.8 The Census Bureau's county estimates of poor school-age children are produced using a county and a state regression model (National Research Council, 2000b, pp.
|
From page 34... ...
, cross-validation and out-of-sample comparisons) are described in the literature and should be part of the standard toolkit of researchers who work on developing models for NASS county estimates.
|
From page 35... ...
for the year of the census. The Census of Agriculture provides no information with which to assess cash rents or planted acres.
|
From page 36... ...
Appendix B provides steps NASS might consider if it decided to pursue REEP. NASS Publication Standard The current NASS publication standard is related specifically to the publication of estimates derived from probability sample surveys.
|
From page 37... ...
• other county-level estimates will be published, along with their All measure of uncertainty. BOX 2-3 The Quality of NASS Estimates The quality of a National Agricultural Statistics Service (NASS)
|
From page 38... ...
A popular Bayesian approach is the construction of posterior predictive p-values. RECOMMENDATION 2-4: The National Agricultural Statistics Ser vice should develop and publish uncertainty measures for county-level estimates.
|
From page 39... ...
They differ in the extent to which they view their data as being confidential, but both agencies share individually identifiable data with other USDA agencies. RECOMMENDATION 2-5: The National Agricultural Statistics Ser vice (NASS)
|
From page 40... ...
For example, even though FSA data are used in classifying pixels, and June Area Survey data are used in the Battese-Fuller regression estimate for planted acres, remote sensing estimates of planted acres by crop for a county would difficult to link either to a respondent to the June Area Survey or the FSA administrative data. The current remote sensing yield estimates rely on having a time series (6 years)
|
From page 41... ...
RECOMMENDATION 2-6: The National Agricultural Statistics Ser vice (NASS) should prepare documentation for any indication submit ted to the Agricultural Statistics Board for review.
|
From page 42... ...
Encouraging collaboration between ASB members and NASS modelers will be a key way to facilitate communication between the two groups. Modelers need to understand how ASB reviews and assesses indications from multiple data sources.
|
From page 43... ...
This issue could be addressed if USDA adopted a practice of snapping CLU boundaries to county boundaries, and functions exist in GIS to do this automatically. Finally, for CLUs to function as basic spatial units, it would be desirable to regard them as permanent to the extent possible, freezing them and minimizing any changes over time.13 While challenges are associated with using the CLU as a basic spatial unit, the approach offers advantages as well: CLUs were intended to serve as a standardized GIS data layer 14 that would ultimately include all farm fields, rangeland, and pastureland in the United States; they have unique identification numbers;15 they are updated annually by FSA;16,17 and they undergo a certification process.18 They also are widely used within USDA and are readily available to NASS.
|
From page 44... ...
Although the panel is recommending that CLUs be used as the NASS basic spatial unit, NASS should plan to maintain RLUs and, when they become available, precision agriculture reports of field location, in its databases. (See Chapter 3 for additional discussion of precision agriculture.)
|
From page 45... ...
There are many advantages to incorporating CLUs into NASS list frames. First, CLUs/RLUs will provide unique linkages among FSA, RMA, and NASS reported data, providing valuable information on current-year planted acres by crop that would improve NASS surveys (the panel's ideas are discussed at the end of this chapter)
|
From page 46... ...
for both crops and cash rents are based on probability sample survey data collected using samples from the list frame. Benchmarking of county-level estimates to agree with previously published state totals accounts for list frame undercoverage, and is done by using an identical multiplicative factor applied to each county.22 The success of NASS surveys and the Census of Agriculture depends on the quality of the sampling frames (area and list)
|
From page 47... ...
, and identification number. The POIDS includes identification number, current operating status, and whether the unit is a target for operator-dominant surveys such as the County Agricultural Production Survey (CAPS)
|
From page 48... ...
(7 CFR 718.2) • hese definitions of "field" are maintained as geospatial units that are T called Common Land Units (CLUs)
|
From page 49... ...
As NASS well knows, achieving this vision will not be easy; however, it will be extremely valuable. Once it has been achieved, NASS will have gained the ability to keep a high proportion of the frame (better than 95% of planted acres in the United States)
|
From page 50... ...
The Nebraska matching experiment indicated that there was an "easy" one-to-one exact match for almost 50 percent of farms. NASS needs to develop revised survey forms for matched farms that take advantage of administrative data to simplify reporting and reduce respondent burden.
|
From page 51... ...
Internet Survey Form to Facilitate Reporting by FSA/RMA Linked Farms As more of the FSA and RMA data are linked to the list frame, it may be useful for NASS to develop an easy-to-use Internet survey form to facilitate reporting by farms that also participate in FSA and RMA programs. The administrative data available prior to collection of CAPS data identify planted acres, crops, and some failed acres, but do not provide information on production or yield.
|
From page 52... ...
RECOMMENDATION 2-12: As farm-level administrative data and precision agriculture data become available, the National Agricultural Statistics Service should consider imputing for nonresponse based on these auxiliary sources of information. Improving Estimation Both FSA and RMA data define subpopulations of all farms, and NASS could potentially obtain access to relevant data from precision agriculture databases on an ongoing basis for a substantial fraction of the total farm population.
|
From page 53... ...
This relative precision can be substantial if the coverage of the precision agriculture database is large. Now suppose that the variable available from the auxiliary database is not the same as that needed for estimation, but is correlated with it.
|
From page 54... ...
RECOMMENDATION 2-13: Once alternative data have been linked to the National Agricultural Statistics Service's list frame for a suffi ciently large percentage of farms, alternative estimation methods that make use of the linked data should be evaluated.
|
Key Terms
This material may be derived from roughly machine-read images, and so is provided only to facilitate research.
More
information on Chapter Skim is available.