Sunday, June 19, 2011

Praise for ASTM Committee E11

Praise for ASTM is also due for correcting the first name of Dr Jan Visman. I had reported on May 22, 2011 to ASTM’s President that it was misspelled. Catharine Allan, Administrative Assistant, Technical Committee Operations, made the correction and kept me posted. Now that’s the iconic society I got to know so well. ASTM Committee D05 on Coal and Coke has been working with applied statistics ever since Greg Gould became its driving force. One reference in Visman’s 1947 PhD thesis reads, Gould, G B, How to use laboratory tests in judging coal values, Combustion 11, 31-37, 1939-1940. How about that? Visman was already aware of Gould’s work! Gy’s 1967 and 1971 works do refer to Visman but his 1979 Sampling of Particulate Matter no longer does. On the contrary, Gy praised in his Introduction not only Matheron’s theories but also David’s 1977 textbook. Gy’s praise may well be the reason why David blew a fuse when Merks & Merks showed how to test for spatial dependence between gold assays of ordered round in a drift.

Greg Gould’s friends called him GG. When I met Greg for the first time he told me that his number one textbook was Volk’s Applied Statistics for Engineers. I still have my first copy. Volk explores all properties of variances in a chapter called Analysis of Variance. Visman applied the additive property of variances to partition his sampling variance into its composition and distribution components. I divided the set of primary increments into interleaved pairs so as to get a single degree of freedom. Gy would put his set of primary increments in a single basket. Clearly, the French sampling school didn’t respect the concept of degrees of freedom nearly as much as do statisticians. That’s why I put together a paper called The Properties of Variances. I wanted to prove that kriging variances and classical variances are as different as night and day.

GG’s grasp of sampling and statistics made him a valuable member of ASTM Committee E11 on Quality and Statistics. He didn’t attend the meeting of ISO/TC27 at Vancouver, BC in 2009. Here’s where and when SGS’s Charles D Rose talked about multivariate analysis when testing for bias between paired data. He didn’t talk about trace elements in rocks or soils but about bias test data for total moisture and dry ash in coal. The power of Student’s t-test to detect a bias is best defined in terms of Bias Detection Limits and Probable Bias Ranges. C D Rose and R M Srivastava seemed to have solved in 1993 some sort of sampling problem. They did it with a fractal correlation function but without degrees of freedom. Scores of degrees of freedom deprived papers were presented in 1993 at David’s bash. Thus it came to pass that David’s 1977 Geostatistical Ore Reserve Estimation was praised for no reason whatsoever.

British Columbia got stuck with environmental guidelines cooked up by R Mohan Srivastava and his FSI pals. Mo ticked me off by dismissing degrees of freedom. He put up a squabble when Sandra Rubin’s “Whistleblower Raises Doubts over Ore Bodies” was published in the National Post of September 30, 2002. Mo beats around the bush with the best when talking about degrees of freedom. All the same, a gifted geostatistocrat thought Mo was witty.

GeovariancesDr M Armstrong
Past Editor: De Geostatisticis
Coauthor: A Study on Kriging Small Blocks

Both Armstrong and her coauthor failed to grasp why kriging variances rise and fall. They thought mine planners were over-smoothing small blocks. So they cautioned against over-smoothing. As luck would have it, testing for spatial dependence didn’t play much of a role in Matheronian geostatistics.

Stanford’s Journel made geostatistics a piece of cake by assuming spatial dependence between measured values in ordered sets. So, he didn’t even teach his students how to apply Fisher’s F-test.

Stanford’s Professor Dr A G Journel
Unencumbered with Fisher’s F-test

Mineral analysts, too, are blamed when metal grades of mined ores are lower than geostatistically “engineered” grades. Mineral analysts do know that interpolation without justification makes no sense in any science. I have talked about sampling and statistics at several of their annual meetings. I even put together a paper on Self-defense for Mineral Analysts.

ASTM awarded me in 1995 for 25 years of continuous membership. Peter S Unger, Vice Chairman, Committee E-11, had written on March 4, 1995, that Ricardo Stone, 1st Vice Chair, would review my notes and contact me. Hennie and I enjoyed luncheon with Carol and Ricardo. I explained that measured gold grades of mined ore were lower than geostatistically engineered gold grades at Hecla’s Grouse Creek mine. Hecla's chief geologist and his dad were both into geostatistics. That’s why the assay laboratory was blamed for low gold grades. I had met its chief assayer at some other mine in the USA. He asked me to visit the Grouse Creek mine and figure out what was wrong. All I did was apply Fisher’s forbidden F-test to gold grades of ordered blast holes. Spatial dependence dissipated into randomness between 20 and 30 m. Yet, the geostatistical model was based on assumed spatial dependence at 100 m. Hecla’s Grouse Creek never made the predicted grade. As fate would have it, geostatistical software was already converting Bre-X’s bogus grade and Busang’s barren rock into a massive phantom gold resource. ASTM Committee D-18 on Soil and Rock is still kriging and smoothing as much as does the world’s mining industry.

Monday, June 13, 2011

ASTM ought to shred geostats standards

ASTM Committee D18 on Soil and Rock had arranged an International Symposium on Geostatistics for Environmental and Geotechnical Applications. The stage was set at the Hyatt Regency, Phoenix, Arizona on January 24-25, 1995. One of its co-chairs was R Mohan Srivastava. His point of view on geostatistics was already an integral part of BC Environment Guidelines. Environment Canada has a handbook called The Inspector’s Field Sampling Manual. I read it when EC took a client of mine to court. Here’s what EC’s inspectors are taught, “Systematic samples taken at regular intervals can be used for geostatistical data analysis, to produce site maps showing analyte locations and concentrations. Geostatistical data analysis is a repetitive process, showing how patterns of analytes change or remain stable over distance or time spans”. It refers to shellfish samples taken at 1-km intervals along a shore, and to water samples taken from varying depths in a water column. It’s just as short of primary data sets and derived statistics as is Matheron’s whole magnum opus. The question is then why geostatistical data analysis underpins the joint study of the Great Lakes by Canadian and US governments. It boils down to blind ambition and blatant contempt for the properties of variances.

Founder of Geostatistics
Founder of Spatial Statistics

ASTM Committee D18 was set up after Geostatistics for the Next Century came about somewhat early in 1993. That’s when geostatisticians from far and wide had flocked together at Montreal, Canada. They had come to praise David’s 1977 Geostatistical Ore Reserve Estimation. Nobody asked David why he had not derived the variance of the distance-weighted average. The more so since Dr Isobel Clark did derive this variance in her 1979 Practical Geostatistics. Sadly, she wasn’t present at McGill. All I had wanted to do was point out that functions do have variances. Alas, my view was as unpopular at Geostatistics for the Next Century as it was when Professor Dr Michel David was the chief enforcer of geostatistics with CIM Bulletin.

David’s peers had come not only to praise his 1977 textbook but also to peddle their own geostat stuff. For example, Journel peddled Modeling Uncertainty: Some Conceptual Thoughts. What David had done in this textbook was derive sixteen (16) distance-weighted averages from nine (9) boreholes. He didn’t derive the variance of each distance-weighted average. He didn’t test for spatial dependence between ordered boreholes. Neither did he count degrees of freedom. David did come up an infinite set of what he then called “simulated values”. Journel derived the zero kriging variance of David’s infinite set of simulated values AKA kriged estimates. Here’s in a nutshell what Professor Dr Georges Matheron has taught all of his disciples. Assume spatial dependence between measured values in ordered sets, interpolate between measured values, smooth the least biased subset of some infinite set of kriged estimates, and rig the rules of applied statistics with impunity. He did all of his thinking about Brownian motion along a straight line in this gloomy edifice.

Centre de Géostatistique
Centre de Morphology Mathématique

Matheron invoked it on this continent in June 1970. Brownian motion set the stage to assume spatial dependence between measured values in ordered sets rather than test for it by applying Fisher’s F-test. One of Matheron’s most gifted disciples was Stanford’s Journel. Not surprisingly, Journel never did what Matheron had not taught him to do. But surely, Journel knew a bit of spatial stuff!

Merks and Merks Precision Estimates for Ore Reserves was praised by and published in Erzmetall. All we did was test for spatial dependence between gold grades of an ordered set of twelve (12) rounds in a drift. Here’s what Stanford’s Journel wrote on October 15, 1992 to Professor Dr Robert Ehrlich, JMG’s Editor, “…Mr. Merks’ anger arises fro [sic!] a misreading of geostatistical theory, or a reading too encumbered by classical “Fischerian” [sic!] statistics.” Journel beat a bit more around the bush when he wrote, “The very reason for geostatistics or spatial statistics in general is the acceptance (a decision rather) that spatially distributed data should be considered a priori as dependent one to another, unless proven otherwise.” He ponders on page 6 of his clarification, “I’ll leave it to you to decide whether this letter should be sent to J. W. Merks; however, I strongly feel that Math Geology has had more than its share of detracting invectives.” Professor Dr Robert Ehrlich, JMG’s Editor, wrote on October 26, 1992, “Your feeling that geostatistics is invalid might be correct.” JMG's Editor did attach a copy of Journel’s 6-page letter. Distracting invectives? I’ll say!

How many Stanford students did he teach about Brownian motion along a straight line? What a silly notion when measuring mineral inventories. What’s more, Fisher’s F-test is forbidden where spatial dependence is deemed to exist a priori. That’s why ASTM ought to shred all standard methods cooked up by ASTM Committee D18 on Soil and Rock. These letters and many others are or will soon be posted on my website under Correspondence. Of course, ASTM Committee D18 ought to be dismissed. Soon I’ll post where we were before we met Dr Ricardo R Stone and his partner. He passed away shortly after we met. Ricardo was an IBM Fellow and a Member of ASTM E11 on Quality and Statistics.