Most Weekly Crude Production Is Forecast. The Monthly Figure Is an 85 Percent Sample.

Domestic crude oil production reaches the page on two different clocks, under the same four words. A reader who pulls the weekly figure and a reader who pulls the monthly figure are not looking at two vintages of one measurement. They are looking at two different estimators, built from different inputs, released on different lags, and carrying different revision rights.

The monthly estimator is a sample survey. The U.S. Energy Information Administration, EIA-914 Monthly Crude Oil and Lease Condensate, and Natural Gas Production Report Methodology sizes it precisely: “The total sample consists of approximately 375 operators out of roughly 15,000 oil and gas operators in the United States.” Cutoff rates are designed to yield “sample coverage of at least 85 percent of the total oil and gas production of each state.” The weekly estimator is not a survey at all for the region that produces most of the barrels. This piece follows that 85 percent to the end: what it covers, what stands in for the rest, and which of the two numbers a balance sheet is using.

One Label, Two Estimators

Start with the monthly route, because it is the one that gets called official. The U.S. Energy Information Administration, Petroleum Supply Monthly and Petroleum Supply Annual Explanatory Notes describe the instrument plainly: “We estimate U.S., PADD, and state crude oil and lease condensate production in the PSM based on Form EIA-914, Monthly Crude Oil, Lease Condensate, and Natural Gas Production Report. The EIA-914 is a sample survey that collects state-level crude oil and lease condensate production information from well operators each month.” The same notes fix the publication lag: “We report the estimates in the PSM about 60 days after the production month.”

Sixty days is the whole problem. A market that trades weekly cannot wait two months, so a weekly figure fills the gap, built from something else entirely. It is not an early draft of the survey; it is a forecast under the survey's name, and the appendix sets out its composition.

What the Weekly Line Is Built From

The U.S. Energy Information Administration, Weekly Petroleum Status Report Appendix B: Explanatory Notes and Definitions lists the components, and the first one matters most: “the most recent Short-Term Energy Outlook (STEO) model estimate (including interim estimates) for average daily production for the lower 48 States and the Federal Gulf of America (GOA).” The appendix counts four data elements in all, and the other three are physical. North Slope barrels enter as volumes delivered to the Trans-Alaska Pipeline System, a separate element carries natural gas plant liquids produced on the North Slope and delivered to that same pipeline, and the southern part of the state enters as “daily production for South Alaska estimated from monthly production reports (lagged by two months)” from the Alaska Oil and Gas Conservation Commission.

So the weekly line is a composite: measured pipeline throughput for a small share of national output, a liquids stream delivered into that same pipeline, a two-month-old state report for a smaller share, and a model output for everything else. The appendix places the error bar where that composition implies it should sit: “Most of the uncertainty in the weekly estimate is associated with the STEO forecast for lower 48 and GOA production.”

That sentence is load-bearing for anyone reading a week-over-week change, because the change is dominated by a forecast revision rather than by a count of barrels. Appendix B also sets expectations against the survey directly: “The weekly data are not expected to have the same level of accuracy as the preliminary monthly data when compared with final monthly data. However, the weekly data are expected to exhibit like trends and product flow characteristic of the preliminary and final monthly data.” Trend fidelity is the claim. Level fidelity is not.

What the Monthly Line Is Built From

The survey route has its own interior. The PSM notes state the coverage geography: “We collect data directly from operators in 15 states and the Gulf of America.” Those direct collections are grossed up rather than added up. “We estimate monthly production for the 15 individually reported states by determining the historical relationship of production by sampled operators and all operators using data from Enverus, a vendor of well-level data collected by state agencies. We model this relationship using a Weighted Least Squares (WLS) linear regression.”

Two features deserve attention. The scaling factor is fitted against a commercial well-level database rather than against a census of operators, and a fitted relationship has residuals that the published state total inherits. The methodology states the arithmetic: a state total is the sum of all operators' database values times the ratio of surveyed to sampled database production.

The 375 operators are also not a random draw from the 15,000. They enter by size, and the oil threshold is explicit: “all operators added because of their oil production have an additional selection criterion of producing at least 500 barrels per day in at least one state/area.” Selecting on size buys coverage cheaply and gives up representativeness at the small end, which is where drilling activity turns first when prices move.

The Share Outside the Sample

Follow the 85 percent. The design target is coverage of at least that share of a state's combined oil and gas production, so as much as 15 percent of it can reach the published total without any operator having filed a form for them. They arrive through the regression, and their size depends on how well the fitted relationship holds in the month being estimated.

Outside the 15 directly collected states the machinery changes again. “For all other states (the Other States group), we rely on data from state agencies, the U.S. Department of the Interior, Bureau of Safety and Environmental Enforcement, and Form EIA-182, Domestic Crude Oil First Purchase Report to estimate crude oil production. For states within the Other States group, we use the average lagged ratio (ALR) of the state-reported data to Form EIA-182 data, multiplied by the current EIA-182 data.” A first-purchase form is a commercial document about a transaction, and a lagged ratio is an assumption about stability. Neither is a well count.

Alaska sits outside the survey altogether, with production republished from the state commission rather than collected on the EIA-914. One national figure is therefore a stack of three estimation regimes, and a change in the national total can originate in any of them without the headline saying which.

The Export Row Belongs to Another Agency

A production figure only becomes useful inside a balance, and that balance pulls one of its largest rows from a different institution. The PSM notes are direct: “The U.S. Census Bureau compiles official U.S. export statistics.” The handoff is monthly and aggregated: “Each month, we receive aggregated export statistics from the Census Bureau (EM-522 and EM-594).” Then comes the conversion step a barrels reader should notice: “We aggregate Census data using Schedule B Codes. We map these onto our products and convert them into barrels.”

On the Census side the same movement is defined by customs geography rather than by geology. The U.S. Census Bureau, Foreign Trade: Definitions of Terms page defines the concept as “Exports measure the total physical movement of merchandise out of the United States to foreign countries whether such merchandise is exported from within the U.S. Customs territory or from a CBP bonded warehouse or a U.S. International Trade Zone.” Destination is filer knowledge rather than observation: it is the country where the goods are “to be consumed, further processed, or manufactured, as known to the USPPI at the time of exportation.”

The quantity caveat is the sharpest line on that page for anyone building a volume series: “Quantity is only used at the 10-digit HS level. Any higher aggregation may result in multiple units of quantity being included.” Barrels exist at the bottom of the classification and stop existing as you climb it. A reader who aggregates a petroleum category upward and compares the result to a production figure has changed the unit without touching the data. The same joint that produced two different July goods deficits is being crossed here, in volume rather than in value.

The Adjustment Row Is Where the Balance Closes

Because supply and disposition come from different collections, the identity does not close on its own. The PSM notes name the closing device: “We derive adjustment quantities for crude oil to balance crude oil supply and disposition.” They also give its usual sign: “Crude oil adjustment quantities are normally greater than zero, indicating crude oil disposition greater than crude oil supply.”

The notes record that the row once carried a franker name: “We formerly called the crude oil adjustment Unaccounted-for crude oil. The name change was effective beginning with data for January 2005.” A second row was later added to the supply side instead, covering unfinished oils and natural gas liquids blended into crude in pipelines and storage, which the notes say tends to bring the adjustment closer to zero.

Read together, those rows mean production is not the residual of the balance. The adjustment is. When a weekly balance moves, the mover can be the forecast component of production, the Census-derived export row, a blending reclassification, or the plug that absorbs whatever the other three leave over. That is the same structural pattern as the gas balance, where net change and implied flow part company at a reclassification line.

Three Revision Clocks, One Series

The last place the two readings separate is time. The survey side runs two clocks. The near one is short: “We allow EIA-914 survey respondents to revise submitted production data for two months prior to the current collection month.” The far one is long: “Once a year, with the release of the PSA, we will revise up to 10 years of historical crude oil production estimates.” The methodology document describes the rolling version of the near clock, in which each month's run regenerates the current month and the two months before it against the latest vendor database.

On the weekly side the reconciliation is scheduled rather than continuous: “The weekly estimates of domestic crude oil production are reviewed monthly when the Short-Term Energy Outlook (STEO) is released to identify differences with recent trends in survey-based domestic production reported in the Petroleum Supply Monthly (PSM).” The appendix also characterises the weekly series as sample estimates stemming largely from preliminary company data that serve as leading indicators of the monthly data.

For contrast, take a physical-volume statistic maintained by another agency on a published window. The Board of Governors of the Federal Reserve System, G.17 Industrial Production and Capacity Utilization: Explanatory Notes state it for a monthly estimate: “The estimate is preliminary (denoted by the superscript ‘p’ in tables) and subject to revision in each of the subsequent five months as new source data become available.” Five months of stated openness against two months of respondent openness plus a ten-year annual window is not a difference in care. It is a difference in what the series promises, which is also the argument behind reading capacity utilization as a ratio rather than a level.

Concrete Framework

Four questions settle which number is on the screen.

First, which release the figure came from. From the weekly balance sheet, the lower 48 and the federal offshore Gulf inside it are a model estimate, and the appendix assigns most of the uncertainty there. From the monthly report, it is a survey estimate roughly two months behind.

Second, whether the comparison is a level or a trend. Appendix B declines the level claim and keeps the trend claim, so a week-over-week direction is inside the series' stated purpose while a level differenced against a monthly survey figure is not.

Third, which row absorbs the residual. Production is not the balancing item; the adjustment is, and it normally runs positive. A weekly balance that moves while production stays flat and the adjustment grows is telling a story about collection rather than about wells.

Fourth, whose unit sits in the denominator. Exports entered as Schedule B quantities converted to barrels, and quantity is defined only at the ten-digit level, so aggregating above it and then differencing against production mixes units silently.

Applied to one reading: the weekly figure moved because a model estimate was refreshed, the monthly figure will move when sampled operators exercise a two-month revision right, and the two are reconciled on the schedule Appendix B describes, not when a reader compares them.

What to Watch Next Week

Three observable checks, all doable from the releases themselves.

Record the weekly balance sheet's domestic production line together with its adjustment line, not the production line alone. The pair shows whether a change came from the estimate or from the plug that closes the identity.

Watch the monthly outlook release, because that is when the model behind the weekly production component is refreshed and when the appendix says the weekly series is reviewed against the survey. A step in the weekly level that coincides with that release is a model update, and treating it as drilling news is the error this structure invites.

Watch the monthly supply report for the production month sitting about two months back, and when it lands, check the two months before it as well. Those earlier months are still inside the respondent revision right, so the interesting movement is often in rows that go unquoted that day.

What Would Invalidate This

The argument rests on the agencies' own methodology text, so it fails where that text is superseded or where the reading did not run far enough.

Three gaps are open. A rounding or precision convention for the weekly production line was not located in the captured text of Appendix B, so nothing here rests on one. The coverage figure is a target, and realised coverage for a specific state and month was not verified, so the modeled remainder could be smaller or larger than the target implies. And the Federal Reserve comparison is limited to the revision window its explanatory notes state; no passage naming the source data for the oil and gas extraction index was located, so no claim is made about where those barrels come from.

One value was deliberately left out. The monthly production series page was opened to cite a specific figure, and on a single read it showed a release date at the end of September with values running only through June, which is inconsistent with the stated publication lag. Because that cell could not be pinned down on one reading, no monthly value is quoted here, and the structural argument is made without one.

The piece would be wrong in substance if the weekly production component stopped being a model output: were the weekly release to carry survey-based lower 48 production, the two readings would converge and the trap would close. Nothing in the appendix points that way.