Besides some surveys I worked on extensively with help at the US Energy Information Administration (EIA), the Canadian Monthly and Annual Miller's Surveys noted in the poster linked below for an upcoming conference (the paper is not yet finished) would be an example that appears to be a good response to this question.

https://www.researchgate.net/publication/361024156_Poster_for_Application_of_Efficient_Sampling_with_Prediction_for_Skewed_Data

..............................

Monthly Miller's Survey:

https://www.statcan.gc.ca/en/survey/agriculture/3403

Annual Miller's Survey:

https://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&SDDS=3443

PS -

Here is the location for the paper to go with the poster above:

https://www.researchgate.net/publication/362370770_Application_of_Efficient_Sampling_with_Prediction_for_Skewed_Data 

Hi James,

if I understand your question, the nearest I can think of is the monthly retail sales index (which collects turnover). The annual equivalent is collected by the structural business statistics.

Steve

Steve MacFeely -

Thank you. I found the attached using your response. It has just two questions: value of business off and on the internet, and appears to be for UK businesses for a month (actually, five weeks in this case). If it makes sense that one would want to know the total across all UK businesses, and that one has a past year census and a current month sample, then one could use the methodology found in my paper. But I did not see an annual form to see if it had the same data items. (I suppose this may be found somewhere in https://ec.europa.eu/eurostat/web/structural-business-statistics.)

If the monthly is also a census, it could then at least be used as a hypothetical case, as with the Toxic Release Inventory example I used. One could collect some key results for large businesses and have an early "prediction" (estimate for the random variable) before the ma and pa businesses all finish replying to the current survey.

Thank you - Jim.

I just stumbled across the question. The methods I believe are also related to nowcasting approaches utilising microdata that is accumulated from surveys.

We use the said retail,industry,services sts-survey data to forecast missing data and aggregate indeces (we notice this also can correct for bias on our non-probability sample, data comes from large firms primarily.

See :https://link.springer.com/article/10.1007/s00181-019-01809-y and

https://www.etla.fi/en/publications/nowcasting-finnish-turnover-indexes-using-firm-level-data/

There could be generalisations that could be developed and explored

Thank you. For repeated production Official Statistics surveys, as in my experience, as Cochran pointed out long ago, a very good size measure for a data item in a sample survey is the same data item in a recent census. When the model is simple and accurate like that, this can be very good. I think that the more complex it has to be, the more additional problems you face. It may be hard to get a good estimate of predicted-y when things are necessarily complex.

Do you mean things like labor force surveys:

https://www.ilo.org/dyn/lfsurvey/lfsurvey.list

Good point Daniel. Those would be repeated establishment surveys. I was thinking in terms of establishment surveys like those for energy producers and the Candian grain mill surveys I mentioned, but that might be a possibility.

I had originally written this question before I finished writing

https://www.researchgate.net/publication/362370770_Application_of_Efficient_Sampling_with_Prediction_for_Skewed_Data

but I've been looking at other uses for that paper, such as sawmill surveys, mineral surveys, forest products, etc. I'm not familiar with labor force surveys, but that might be a possibility too. Thank you.

I guess labor force surveys can be household or establishment surveys???

This is for the crushed stone industry in the US:

https://www.usgs.gov/centers/national-minerals-information-center/crushed-stone-statistics-and-information

I'm told these quarterly surveys are for "a select set of companies" which reminds me of how the quasi-cutoff sample of electric sales in the US got started. The electric sales survey of a select group of entities was later modified and used as a sample, first a stratified random sample with a large company censused stratum, and then only the censused stratum as a quasi-cutoff sample, all after starting an annual census of all electric sales by economic sector (residential, etc.), from the production/supply side. If one wanted to monitor the crushed stone industry the same way, I would suggest this approach using a quasi-cutoff sample with a ratio model for prediction, as is done at the US Energy Information Administration (EIA).

Does anyone know of other surveys, each of a select group of larger establishments being followed, where there is a chance to instead have an occasional census to be used for regressor data for the same data items in a more frequent sample?

In Guadarrama, Molina, and Tillé, https://www.researchgate.net/publication/342657185_Small_area_estimation_methods_under_cut-off_sampling, they found that regression modeling appears to outperform calibration when considering small domains, for cutoff sampling. This is an important finding.

Additionally, in my JSM 2022 paper,

https://www.researchgate.net/publication/362370770_Application_of_Efficient_Sampling_with_Prediction_for_Skewed_Data, I included a section on the Guadarrama, Molina, and Tillé Survey Methodology article, in which I explained how I differ regarding the model structure they used in an example.  Their example was interesting, although it was for only one item, not for a multiple attribute survey. 

The example in the Guadarrama, Molina, and Tillé article was for a one-time cutoff sample survey, but there was a census on the same data item, for a longer time period, which preceeded it. The sample employed a special device for use by large retail stores to collect the data. If this were repeated with frequent cutoff sample surveys, and the occasional implementation of a census for predictor data, then more testing could be done to ensure high quality results for totals and relative standard errors, and more would be known about what changes in the market might need to be considered, perhaps even using a multiple regression. If there were multiple items, there would be tradeoffs in determining which stores would be included in the 'quasi-cutoff sample,' with some stores having larger sales of one item over another.

Repeated Government Establishment Census and Sample Surveys (Official Statistics) for Market Monitoring  ("How to") - 

The crushed stone potential example of January 13 is another one with multiple products.  That is why quasi-cutoff sampling is important.  High coverage for cutoff sampling does help lower estimated relative standard errors for predicted totals, although that is not all that is important.  A low estimated variance of prediction error for the remainder, is important, as is good stratification to avoid bias.  With multiple attributes/items, coverage may be somewhat lower for a given item when one of the respondents which has a large measure for that item may not be included because it isn't large for any other item.  Compromises may be necessary.  But that may not be difficult to decide upon when establishment populations have relatively few very large members (although size may change over time in future uses).  The important part is to have reasonably small estimated relative standard errors for the totals for each of the items.  Here are some references for doing this (and note that even when multiple items are not mentioned, that is still important to keep in mind), such that the following is something of a "How to do" guide: 

..........

"Application of Efficient Sampling with Prediction for Skewed Data," JSM 2022: 

https://www.researchgate.net/publication/362370770_Application_of_Efficient_Sampling_with_Prediction_for_Skewed_Data 

I have noted to some people that "... Guadarrama, Molina, and Tillé found that regression modeling appeared to perform better than calibration in small domains, for cutoff sampling, in their article found at