Key omissions

I can only touch on a few of the problems with Dr Wendel’s analysis (Speaker: Study of water wells near gas drilling is encouraging, 12/19/13).

First, his claim that the expected probability of groundwater contamination by gas extraction is at most 1:1,000,000. He based this on a 1:100 probability that a casing will fail and the fact that gas wells have 3 concentric casings where they pass through strata with fresh water aquifers. How did he square this with a review of DEP records for 2010 – 2012 which shows a minimum annual immediate loss of well integrity of over 6:100?

He didn’t. He was either unaware of those statistics or chose not to mention them. However, the answer is very straightforward. The most likely mechanism for loss of well integrity is not the simultaneous failure of 3 casings, but rather flaws in the cement that allow gas and other chemicals to travel along the outside of the well. It is worth noting that most of the vertical leg of these wells is not even cemented.

While one doesn’t need an understanding of statistical analysis to follow the above, I suspect those who don’t will have trouble understanding the next paragraph. It was wrong for Wendel’s work to have been presented to the general public before it underwent peer review to vouch for the science being sound.

Taken together, the following facts completely discredit Wendel’s analysis.

He showed several graphs that reveal weak negative correlations. (Evidence that gas wells are contaminating ground water.)

He gave a confused explanation for dismissing these “apparent” correlations; the randomization process supposedly selected over 5 percent bad data points.

The actual rationale for dismissing these correlations is that the data fails to meet acceptable alpha levels; which indicates a non-representative sample.

But alpha levels only apply to normally distributed data.

And Wendel’s presentation included several histograms showing that his data is not normally distributed.

At least get Wendel to justify applying a test that requires normally distributed data to data which he himself admits isn’t?

John Kesich


Submitted by Virtual Newsroom