Haskell County, Kansas

Average daily production per well, for wells grouped by when they first produced. Includes every well of the cohort that is old enough at each month. Wells that are shut in or have stopped producing count as zero, so the average reflects the full cohort, not just survivors. Rates are calendar-day (volume ÷ days in month). The cumulative axis accumulates those same monthly averages at the average month length. Months not yet fully reported are excluded. The 2010-2014 curve draws those years as one group. The chart opens on average daily rate against months on production, the rate on a log scale. The view control switches to rate against cumulative production per well, on the same log rate scale, or to cumulative production per well against months on production, on a linear scale. Every view draws the same cohorts from the same production history. Each cohort with a dashed curve is solid through the last month all the wells in its curve sit inside fully reported data. Past it the dashed curve carries the same wells forward, on our forecast where a well’s own record ends, for up to five years. A dashed curve runs as far as at least nine in ten of its wells are accounted for. A cohort with no dashed curve is drawn only through the months where at least 50% of its wells are old enough for the month to be fully reported. A well that is old enough but shut in still counts, as zero. The measured months are what these wells reported. In the forward months each well counts its own reported figure for each month it has one, our forecast for each month it has only that, and zero for the rest. The forward average counts every well in the curve. The drawn forecast runs up to five years past the measured months. Line thickness tracks how many wells each curve is drawn from. 805 wells are counted here but not shown on the age axis: they started producing in 1989 or earlier, too close to the start of our records for their region to verify their true age.

The two are produced differently: the EIA estimates from a survey of the largest operators and a model built to anticipate late filings, while the series compared here are rebuilt from the source’s own records as they arrive. Recent months can differ without either being wrong; the methodology page explains both methods.