I'm interested in the results of Objective Yield Survey by USDA NASS in a raw form of plot-level observations. As far, as I understand, quickstats (https://www.nass.usda.gov/Surveys/Guide_to_NASS_Surveys/Objective_Yield/index.php) provides data, aggregated for counties and states, but not directly field observations. Is this data publicly available?
Aleksandra -
I don't know, but will share your question with someone who might know. This question interests me because I'd like to know what methodology was used to arrive at aggregate levels. Do you know that? Are you interested in sample level data so that you can try out another method? It sounds like a similar problem to arriving at the official energy statistics which I developed, and thus my interest. Thanks.
Cheers - Jim
James, Thanks a lot! From this paper Article USDA Corn and Soybean Acreage Estimates and Yield Forecasts:...
and this document https://www.nass.usda.gov/Education_and_Outreach/Understanding_Statistics/Yield_Forecasting_Program.pdfI can conclude that aggregation is a simple average of plot-level observations. Before averaging, all data is checked for consistency with previous years. This straightforward method seems to be reliable because of the proportional sampling of farms and fields by crop and size within the studying area. I'm looking for actual field data for studying a field to field variability.
Aleksandra -
That second document is available, thanks, but at a glance at some formulas, it was hard for me to see when they were talking about estimating the current total yield, and when they meant forecasting. I saw some ratios of past month to current month, which is what I would use to 'estimate' a current total. (Technically the term would be 'predict' for me because when a ratio estimator is used in a simple regression, and we are estimating something current using regression, and thus using a random variable, y, then this is actually called "prediction," not estimation. So we can use regression to 'predict' a current total, or estimate that total using probability sampling, but we forecast future yields using methods including regression. - Sorry for that diversion. I just wanted to make sure we are talking about the same things.)
I worked a great deal with estimating, really "predicting" totals for official energy statistics, but not with forecasting.
I looked to see what they said about heteroscedasticity, but did not see anything. That was a big concern for my work. I searched for autocorrelation, with which I have not worked, to see if that was mentioned, but did not see anything. On the top of page 24 they mention "forecasting classes," and show a simple linear regression. In glancing at this, I did not see the point in an intercept term there. Without it, we have ratios again. But I have found ratios to be very helpful. Also, I think I saw there what you meant by "averaging." But if they used unequal probability sampling, I'd expect weighted averages. ???
It looks like you could spend a long time going through all of this, and trying different combinations of methods.
Thank you for your question and the document links.
I have not heard from the person I thought might know about this. I think you could go right to NASS though. I have sent people to my old agency for data, so I looked to see where you might write to NASS to ask for the data you want. I found this: [email protected]
It was on https://ask.usda.gov/s/.
I suggest that you write to them to request the data that you want from NASS.
Best wishes - Jim
- Badges
- Science topic
- Similar topics
- Statistics
- Survey