EIA’s weekly natural gas storage data now include measures of sampling variability
Starting with the release of data for the week ending January 20, 2017, EIA’s Weekly Natural Gas Storage Report (WNGSR) now provides information about the statistical properties of published estimates of weekly working natural gas levels and their net changes. The methodology for the storage data itself, however, is unchanged.
The WNGSR provides estimates of working gas in storage—the amount of natural gas that can be withdrawn to satisfy market demand—based on a sample of storage operators in the Lower 48 states. Like other sample surveys, the WNGSR is subject to sampling error. To help users better understand sampling variability in WNGSR values, a new table provides
The standard error for estimates of weekly net changes in working gas levels. Standard error is a measure of the sampling variability of an estimate based on all possible samples that could have been selected using the chosen sample design.
The coefficient of variation for estimates of weekly working gas storage levels. The coefficient of variation, or relative standard error, presents the standard error as a percentage of the published working gas storage estimate.
EIA has also estimated these measures of sampling variability for all weeks since April 2015. Since that time, the standard errors for estimates of weekly net change have averaged about 2 billion cubic feet (Bcf) for the Lower 48 states and about 1 Bcf for each region. The coefficients of variation for stocks are typically about 1% for the Lower 48 states, and they range between 1% and 5% for the regions. For example, the larger coefficients of variation for the working gas estimates at the salt facilities in the South Central region likely resulted from the variable activity at these high-deliverability natural gas storage facilities. In contrast, the comparatively regular activity at aquifer storage sites in the Midwest contributes to the lower coefficients of variation for the Midwest region.
The standard error can be used to construct a confidence interval centered about the estimate. For example, the published weekly estimate of the net withdrawal for January 20, 2017, is 119 Bcf, and the standard error of the net withdrawal is 2.9 Bcf. The 95% confidence interval ranges from 113 Bcf to 125 Bcf.
For the Lower 48 states, the standard errors tend to be slightly larger during weeks when net changes in natural gas inventories are relatively large in absolute value, such as those weeks when changes exceeded 100 Bcf. For the 53 report weeks ending in 2016, none had net injections that exceeded 100 Bcf, but in 10 of those weeks, net withdrawals exceeded 100 Bcf. For those 10 weeks, the standard error ranged from 2.2 Bcf to 5.5 Bcf, while the standard error ranged from 1.0 Bcf to 3.2 Bcf for the other 43 weeks.
Estimates of working gas storage reported in the WNGSR for the last week of each month precede the estimates of working gas that are reported in the Natural Gas Monthly by roughly two months. Because the estimates reported in the Natural Gas Monthly come from a census of all known storage operators, the monthly values have no sampling errors. Since April 2015, all working gas levels reported in the Natural Gas Monthly did not deviate noticeably from the 95% confidence band constructed from the WNGSR weekly estimates and their standard errors.
Photos and recommended reading source: EIA