United States Court of Appeals


                     FOR THE DISTRICT OF COLUMBIA CIRCUIT


              Argued April 20, 1998     Decided August 21, 1998 


                                 No. 95-1619


                             Secretary of Labor, 

                                  Petitioner


                                      v.


                    Keystone Coal Mining Corporation and 

              Federal Mine Safety and Health Review Commission, 

                                 Respondents


                     Southern Ohio Coal Company, et al., 

                                 Intervenors


                  On Petition for Review of an Order of the 

               Federal Mine Safety and Health Review Commission


     Edward D. Sieger, Senior Appellate Attorney, United 
States Department of Labor, argued the cause for petitioner, 
with whom J. Davitt McAteer, Acting Solicitor, and Nathani- el I. Spiller, Deputy Associate Solicitor, were on the briefs.




     R. Henry Moore argued the cause for respondent Keystone 
Coal Mining Corporation, with whom Heather A. Wyman was 
on the brief.  Norman M. Gleichman, General Counsel, Mine 
Safety and Health Review Commission, and Elizabeth Ebner, 
Attorney Advisor, entered appearances.
     Timothy M. Biddle, R. Timothy McCrum, H. Thomas 
Wells, J. Alan Truitt, Laura E. Beverage, L. Anthony 
George, R. Henry Moore and Heather A. Wyman were on the 
brief for intervenors Southern Ohio Coal Company, et al.  
Gail L. Simmons entered an appearance.

     Before:  Williams*, Sentelle and Randolph, Circuit 
Judges.
     Opinion for the court filed by Circuit Judge Sentelle.
     Sentelle, Circuit Judge:  The Secretary of Labor ("Secre- tary"), on behalf of the Mining Safety and Health Administra- tion ("MSHA"), asks us to reverse a November 1996 decision 
of the Mine Safety and Health Review Commission ("Com- mission"), affirming rulings by an Administrative Law Judge 
("ALJ") in a case involving citations for alleged tampering 
with coal dust samples.  The ALJ and Commission agreed 
that the Secretary failed to prove (1) in general, that an 
"abnormal white center" ("AWC") on a coal dust sample filter 
warrants an inference of intentional tampering;  and (2) in a 
specific test case, that defendant Keystone Coal Mining Corp. 
("Keystone") intentionally tampered with its samples.  The 
Secretary argues that the ALJ and Commission held it to an 
improperly high burden of proof in the first, common-issues 
proceeding, and that Keystone's exoneration in the second, 
case-specific proceeding was not supported by substantial 
evidence.  We affirm the Commission's ruling.

                                      I
     The case involves over 5000 citations, issued to over 500 
coal mines, alleging tampering with air filter samples.  These 
citations arose from a nationwide investigation by the Secre- tary which began in August 1989.  The citations issued 
between April 4 and June 7, 1991, and included 75 citations to 

_____________
     *Circuit Judge Williams did not participate in this 
decision.  He found it necessary to become recused after 
hearing oral argument.

Keystone's Urling No. 1 mine ("Urling").  Under the Federal 
Mine Safety and Health Act of 1977, 30 U.S.C. s 801 et seq. 
(the "Act"), coal mine operators must periodically sample the 
concentration of respirable coal dust in the mine atmosphere.  
The tests employ sampling devices and methods prescribed 
by the Secretary.  The devices are all manufactured by the 
Mine Safety Appliance Company ("MSA"), and involve basi- cally a pump and a filter cassette.  The pump pulls air at a 
defined rate through the filter, where respirable coal dust is 
deposited.  The filters are then sent to MSHA, within 24 
hours of collection.  In February 1989, MSHA noticed that 
some filters had unusual light areas in their centers which 
generally corresponded to the 6mm opening in the cassette.  
MSHA concluded that these abnormal white centers were 
likely caused by reverse air flow--specifically, by a person 
blowing through the cassette opening in order to dislodge 
dust from the filter and thereby decrease the sample weight.  
MSHA expanded the investigation to all mine operators in 
August 1989, thereafter examining all dust samples for 
AWCs.  Hundreds of mines had no AWCs, but 3900 AWC 
samples (about 6.5% of all samples received) were discovered 
by March 19, 1990.  On March 20, 1990, MSHA introduced 
the "AWC void code" which officially notified operators that 
AWC samples would no longer be accepted as sufficient to 
fulfill the operator's sampling obligations under the Act. 
Fewer than 1% of the samples submitted after that date 
exhibited AWCs.

     In August 1992, the ALJ consolidated the citations in order 
to try common issues (the "common issues" proceeding).  The 
relevant issue in this proceeding was whether deliberate 
conduct was the "only reasonable explanation" for the cited 
AWCs.  After a 47-day hearing, the ALJ decided against the 
Secretary, finding that case-by-case inquiry into dust sam- pling and handling procedures was required to determine 
whether intentional tampering caused AWCs on samples re- ceived from each mine.  The ALJ selected Keystone's Urling 
No. 1 mine for a case-specific trial regarding dust sampling 
and handling practices.  After an 18-day hearing, the ALJ 
vacated the Urling citations, holding that the Secretary had 



failed to prove that Keystone intentionally altered the weight 
of the 75 cited filters.

     The Secretary sought review of both the common issues 
and Keystone decisions before the Commission.  A divided 
Commission affirmed on November 29, 1995.  In re:  Contests 
of Respirable Dust Sample Alteration Citations, Keystone 
Coal Mining Corp. v. Secretary of Labor, 17 F.M.S.H.R.C. 
1819 (1995).  Dissenting Commissioner Marks argued that 
the ALJ had improperly interpreted MSHA regulations to 
require proof of intentional alteration (an interpretation not 
challenged here), and further contended that the ALJ had 
improperly "required the Government to prove that the only 
cause of the AWCs was intentional conduct, to the exclusion 
of all other causes! " (emphasis in original).  Commissioner 
Marks would have held that the Secretary had presented 
sufficient evidence to prevail in both the common-issues and 
the Keystone proceedings, and that the ALJ's conclusions to 
the contrary were not supported by substantial evidence.  We 
review the Commission's legal conclusions de novo, Donovan 
ex rel. Anderson v. Stafford Constr. Co., 732 F.2d 954, 958 
(D.C. Cir. 1984), and its findings of fact for substantial 
evidence, 30 U.S.C. ss 816(a)(1), (b).

                                      II


     The Secretary argues that in the common issues proceed- ing, the Commission and the ALJ erred as a matter of law by 
requiring a standard of proof higher than a preponderance of 
the evidence for the proposition that the presence of an AWC 
allowed an inference of intentional tampering.  With respect 
to the Keystone mine-specific proceeding, the Secretary as- serts that the Commission and the ALJ applied an improper- ly strict burden of proof and that the findings were not 
supported by substantial evidence.

                                      A


     In the common issues proceeding, the Secretary attempted 
to prove via statistical evidence that the presence of an AWC, 
without more, established intentional tampering with the 


sampling device.  Such a finding would have led to a pre- sumption that illegal tampering occurred whenever an AWC 
was found, perhaps subject to rebuttal by an individual 
operator who could show that other factors (for example, its 
handling of filters) caused the AWC in a specific case.

     The ALJ held that to prevail the Secretary must prove by a 
preponderance of the evidence that (1) the AWC definition 
had a coherent meaning and was consistently applied;  (2) the 
cited AWCs could only result from intentional acts;  and (3) 
the AWCs resulted in weight losses in the cited filters.  
Although concluding that any inconsistencies in applying the 
AWC definition were insignificant and that an AWC did 
result in weight loss, the ALJ found several potential causes 
of AWCs and received a wide range of expert opinion on the 
likelihood of each possibility.  For example, AWCs could be 
caused by tampering, by impact to the cassette, by impact to 
the air hose, or by snapping together the cassette.  The ALJ 
also found that the likelihood of generating an AWC by non- intentional causes depended upon filter manufacturing char- acteristics (filter-to-foil distance and filter floppiness), hose 
pliability, mine and dust characteristics (including type of 
coal, humidity, weight of dust on the filter, size and shape of 
particles, and quantity of rock dust or diesel dust on the 
filter), and cassette population (certain batches of cassettes 
manufactured by MSA had a greater likelihood of experienc- ing AWCs, as did all cassettes manufactured before Jan. 1, 
1990).  Thus, the non-random distribution of AWCs across 
the mining industry could have been related to tampering at 
certain mines, but also could have been related to characteris- tics of certain mine environments or operators' handling 
techniques.

     Therefore, even though the Secretary's statistical evidence 
demonstrated that AWCs did not occur randomly, the ALJ 
held that the Secretary had failed to prove that those AWCs 
were indeed caused by intentional tampering.  The Secre- tary's analysis failed to account for potential accidental 
causes, manufacturing variables, and mine environment varia- bles.  Further, even though the Secretary introduced evi- dence showing a sharp decline in the number of cited AWCs 



in late March, 1990, a date which correlated with the an- nouncement of the "AWC Void Code," the ALJ held that the 
Secretary had failed to prove that the decline was caused by 
mine operators responding to that announcement.  Thus, the 
ALJ concluded that the Secretary had "failed to carry [her] 
burden of proving by a preponderance of the evidence that an 
AWC on a cited filter establishes that the mine operator 
intentionally altered the weight of the filter."

     The Secretary first contends, as she did before the Com- mission, that the ALJ imposed an improper burden of proof 
in this ruling, despite the "preponderance of the evidence" 
language in both opinions.  The Secretary argues that the 
ALJ erred by requiring proof that "the cited AWCs can only 
have resulted from intentional acts," Brief for the Secretary 
of Labor ("Petitioner's Brief") at 41 (emphasis added), or that 
deliberate conduct "is the only reasonable explanation for the 
cited AWCs," id. (emphasis added).  Instead, she contends 
that she "should have prevailed by establishing on the weight 
of the evidence that intentional alteration was the more likely 
explanation for AWCs than other possible explanations."  Id. 
(emphasis added).  We reject this argument.

     In effect, the Secretary sought to establish in the common 
issues proceeding an evidentiary presumption:  that the exis- tence of an AWC, without more, compels (or, at least, allows) 
an inference that the mine submitting the filter with the AWC 
intentionally tampered with it in violation of the Mine Act. 
Such a presumption is only permissible if there is "a sound 
and rational connection between the proved and inferred 
facts," and when "proof of one fact renders the existence of 
another fact so probable that it is sensible and timesaving to 
assume the truth of [the inferred] fact ... until the adversary 
disproves it."  Chemical Mfrs. Ass'n v. Department of 
Transp., 105 F.3d 702, 705 (D.C. Cir. 1997) (quoting NLRB v. 
Curtin Matheson Scientific, Inc., 494 U.S. 775, 788-89 (1990)) 
(internal citation and quotation marks removed).  If there is 
an alternate explanation for the evidence that is also reason- ably likely, then the presumption is irrational.


     In making her argument that the evidence presented to the 
ALJ, and reviewed by the Commission, compelled the imposi- tion of the presumption that every AWC resulted from tam- pering, the Secretary ignores such cases as Curtin Matheson 
and Chemical Mfrs.  She instead relies on Concrete Pipe & 
Products of California, Inc. v. Construction Laborers Pen- sion Trust, 508 U.S. 602, 622 (1993), for the proposition that 
the preponderance of evidence standard governing the pro- ceedings "simply requires the trier of fact to believe that the 
existence of a fact is more probable than its nonexistence."  
It is most evident that the Concrete Pipe holding relied on by 
the Secretary is inapposite.  The question before the ALJ, 
the Commission, and now us, was not whether the Secretary 
had established by the preponderance of the evidence a 
simple evidentiary fact--e.g., whether a particular AWC re- sulted from tampering--but rather whether the Secretary 
had established that all AWCs result from tampering by 
some standard sufficiently compelling to require the Commis- sion to adopt it as a presumption.  By way of comparison, a 
plaintiff establishing that a defendant assaulted her is not the 
same as a litigant convincing a trier of fact that persons 
similarly situated to the defendant were so likely to have 
committed assault that liability could be presumed against 
them.

     Unsurprisingly, none of the authorities offered by the 
Secretary, and none that we have located, hold that a litigant 
can, even by powerful evidence, compel an adjudicating com- mission to adopt a presumption favoring the litigant in an 
entire universe of cases.  Generally, the authorities offered 
by the Secretary and discussed by us concern either the 
validity or the application of presumptions created either by 
an administrative body or by statute.

     For example, in Chemical Mfrs., we upheld a presumption 
established by regulation of the Department of Transporta- tion which allowed an inference of inadequate pre-trip inspec- tion from the presence of loose closures on railroad tank cars.  
105 F.3d at 703-04.  We held that the agency had articulated 
its reasons for establishing the presumption, and noted that 
the presumption only shifted the burden of producing evi-



dence.  We concluded that the Department had articulated a 
reasonable evidentiary basis even though it did not consider 
"every possible intervening event" that could cause a loose 
closure.  Id. at 706.  Further, we held that such "administra- tive presumptions" could be sustained without an evidentiary 
showing to support the rule, so long as the agency articulates 
a rational basis.  Id.  The presumption did no more than 
"eliminate[ ] the need to call an expert witness in each 
enforcement proceeding to establish that properly tightened 
closures generally do not loosen of their own accord in normal 
transportation, and that loose closures often reflect inade- quate pre-trip inspections."  Id.  Those facts had been ade- quately established in the record.  We also recognized that 
because closures were designed "so that, once properly tight- ened, they will not loosen as a result of vibrations or other 
conditions normally incident to rail transportation," it was 
reasonable to presume failure to inspect properly, absent 
evidence of some intervening event.  Id.

     The present record does not remotely parallel Chemical 
Mfrs. If an appropriate government agency charged with 
mine safety regulation had held a rulemaking, established a 
proper foundation for the presumption advanced by the Sec- retary, and adopted it, we might well uphold the presumption.  
At the very least, Chemical Mfrs. would be appropriate 
support for the Secretary's argument.  But that is not what 
happened.  A trier of fact took evidence and weighed it.  This 
case turns not on the construction of regulations or on 
statutory interpretation, but on the weighing of evidence and 
reasonable inferences made therefrom.  Thus, our deference 
runs not to the policymaking body, MSHA and the Secretary, 
but to the ALJ, the factfinder who oversees the adjudicatory 
proceedings.

     Curtin Matheson and Concrete Pipe are even less appro- priate precedents for this controversy than Chemical Mfrs.  
In Curtin Matheson, the Supreme Court reversed the at- tempt of a circuit court to impose upon an administrative 
agency the duty of adopting a presumption.  494 U.S. at 781.  
In no sense did it attempt to set forth terms under which the 
courts could impose upon an adjudicating commission the 



duty to adopt a presumption based upon a certain level of 
proof offered by a litigant, as the Secretary asks us to do 
here.  Concrete Pipe involved the application of a particular 
set of presumptions created by statute to a particular sort of 
factual dispute, 508 U.S. at 630-31, and again offers no 
support for the Secretary's attempt to impose upon the 
adjudicators before whom it appeared the duty of presuming.

     In another important respect, the presumption sought by 
the Secretary in this case is far more troubling than the one 
at issue in Chemical Mfrs. In that case, the Department of 
Transportation established a rational connection between two 
concrete facts:  the fact of a loose connection allowed inferring 
the fact that the connection had not been inspected.  Absent 
evidence of an intervening event, such a presumption seems 
ironclad, especially since a shipper was strictly liable for 
failure to inspect, without need to prove negligence or intent.  
But in this case, the Secretary seeks to establish a connection 
between a fact and an intentional act, namely, to infer from 
the presence of a light area in a filter's center that the mine 
operator intentionally and illegally tampered with the sam- pling device.  Distinctions between accidental, negligent, 
reckless, and intentional conduct, not relevant in Chemical 
Mfrs., make all the difference between an innocent act and a 
citable offense in cases involving the Secretary's proposed 
presumption.

     In considering the evidence presented in the common is- sues proceeding, we cannot say that the ALJ reached an 
unreasonable conclusion in holding that the Secretary had 
failed to prove by a preponderance of the evidence that the 
existence of an AWC established the deliberate conduct re- quired to sustain a citation under the Mine Act and associated 
regulations.  The ALJ certainly did not require that the 
Secretary prove impossible all other potential causes of 
AWCs at the hearing.  But because AWCs could result from 
a variety of non-intentional causes, the ALJ found more than 
a mere "element of doubt" that the Secretary had carried her 
burden of proof.



     To sum up, the Secretary is mistaken in her assertion that 
under a "preponderance of the evidence" burden, the Com- mission is required to adopt her presumption when she 
proves that intentional alteration is merely the "more likely 
explanation for AWCs than other possible explanations."  We 
therefore affirm the judgment in the common issues proceed- ing.

                                      B


     In the Urling mine-specific proceeding, the Secretary 
sought to establish by a preponderance of the evidence that 
Keystone had unlawfully tampered with sampling devices.  
Both parties introduced a volume of statistical evidence along 
with the testimony of several experts and witnesses regarding 
mine conditions and the handling of the filters.

     Rochester and Pittsburgh Coal Co. ("R&P") operates 13 
mines, including Urling, through several subsidiaries, includ- ing Keystone.  For all these mines, the independent R&P 
Environmental Safety Department ("ESD") conducted a coal 
dust sampling program.  From 1970 until 1991, Donald Eget 
supervised ESD, and Shawn Houck and Douglas Snyder 
worked with him as laboratory technicians.  Normal operat- ing procedures at ESD between 1989 and 1991 had the dust 
technicians picking up pumps and sampling assemblies in the 
morning and delivering them to R&P's mines for use that 
day.  Each morning, Eget drove to all 13 R&P mines to 
retrieve pumps and samples from the previous afternoon and 
midnight shifts;  and each afternoon, the dust technicians 
returned to ESD with pumps used during the day shift that 
day.  While Eget collected pumps, Houck processed those 
from the previous day by removing the sampling head and 
hose, filling out data cards, cleaning the sampled units, recal- ibrating and reassembling the units, and inserting a new filter 
cassette.  When Eget returned, he inspected the used cas- settes, checked the data cards, looked into the inlets and 
recorded the filter appearances in a logbook for each mine.  
The cassettes were then packaged and mailed to MSHA.


     Robert Thaxton, the MSHA supervisory industrial hygien- ist responsible for analyzing, monitoring and classifying 
AWCs, testified that, in his opinion, AWC patterns on Key- stone's 75 cited and 3 "no-call" filters resulted from deliberate 
acts.  The Secretary's scientific expert Marple examined and 
classified the 78 filters, opining that none could result from 
impact to the cassettes, but that 71 or 72 resulted from 
reverse air flow, 2 or 3 from a vacuum source introduced to 
the cassette inlet, and 1 from water introduced into the filter.  
The Secretary's statistical expert Miller testified that Urling 
had an AWC citation rate of 43% before the void code notice 
issued on March 26, 1990 (compared to 6% for other mines), 
and that the rate dropped to 0.18% after March 26.

     Keystone's scientific expert Lee concluded that most of the 
cited filters indicated lesser forces than would have occurred 
with deliberate reverse air flow, that the AWC patterns were 
consistent with a mixed mechanical pulse/reverse air pulse, 
that humidity reduced the susceptibility to dislodgement, and 
that water sprays and scrubbers introduced at Urling in 1989 
and 1990 contributed to the decline in AWCs.  Keystone's 
statistical expert Roth examined the citation rates of Urling 
and of all R&P mines combined on a bimonthly basis and 
concluded (1) that the data showed a gradual decline in AWCs 
from August 1989 through March 1992, with no significant 
change in March 1990;  (2) that manufacturing variables may 
have been a factor in AWC formation;  and (3) that high 
incidence rates may be attributable to cassettes manufactured 
by MSA on four consecutive dates in mid-1989 (for all R&P 
mines, cassettes manufactured on those four dates were cited 
at a rate of 50% as opposed to 6% for all other dates of 
manufacture).  Thirty-three R&P employees testified, includ- ing ESD personnel Eget, Houck, and Snyder, who described 
their role in the dust sampling program and uniformly denied 
tampering or observing anyone else tampering with cassettes.

     The Secretary's first argument, much like that advanced 
and rejected with respect to the common issues proceeding, is 
that the ALJ improperly held any doubt as to the cause of an 
AWC sufficient to vacate the citation.  Applying such a 



burden of persuasion, higher than a "preponderance of the 
evidence," would constitute reversible error.  The Secretary 
argues that she did prove that tampering was the most likely 
cause of Keystone's AWCs, even though competing causal 
theories had not been completely ruled out.  In her view, the 
ALJ should have explicitly determined the probability that 
rough handling or other non-intentional conduct caused Key- stone's AWCs.  Without such a determination, according to 
the Secretary, the ALJ could not have adequately addressed 
the question of whether the cited filters were more likely than 
not caused by tampering.  We disagree.

     The ALJ recognized and the Commission affirmed that the 
Secretary bore the burden of proving by a preponderance of 
the evidence that tampering actually occurred, and both 
agreed that the Secretary had not met that burden.  In the 
process of weighing the vast amount of sometimes conflicting 
evidence, including the often divergent interpretations by 
experts, it is simply unreasonable to require that a factfinder 
determine the mathematical probability of the various differ- ent explanations of that evidence.  We know of no case in 
which a reviewing court has required that sort of mathemati- cally nice analysis, nor has the Secretary cited any.  Rather, 
the factfinder must assess whether, on the whole, he is 
convinced that greater weight of the evidence supports the 
plaintiff's account.  See, e.g., Steadman v. SEC, 450 U.S. 91, 
101 (1981).  So long as that determination is properly made, 
no further precision or subdivision in specification of probabil- ities is required.  The record indicates such a finding.

     The Secretary's second argument reveals the heart of her 
position:  that her evidence showed that tampering was in 
fact the most likely cause of Keystone's AWCs, despite the 
ruling of the ALJ and Commission to the contrary.  In 
essence, the Secretary seeks to have this Court review the 
entire trial record, reweigh the evidence, and decide the case 
differently.  But this Court's duty is to determine whether 
the findings below were supported by substantial evidence.  
This sensibly deferential standard of review does not allow us 
to reverse reasonable findings and conclusions, even if we 
would have weighed the evidence differently.  We must 


therefore examine the Secretary's allegations regarding spe- cific inconsistencies between the evidence presented and the 
conclusions of the factfinder, and determine whether a theo- retical "reasonable factfinder" could have reached the conclu- sions actually reached by the Commission and the ALJ. 
United Steelworkers of America v. NLRB, 983 F.2d 240, 244 
(D.C. Cir. 1993). AWCs Not Random Events

     The Secretary presented statistical evidence showing that 
AWCs were not randomly distributed across all coal mines.  
Out of samples from 2677 coal mines, about 1300 mines had 
no AWCs between August 1989 and March 1991.  Other 
mines, like Keystone, had AWCs on more than 40% of their 
samples submitted during this period.  The Secretary insists 
that this evidence forces the "inescapable conclusion" that 
"random events do not cause AWCs and AWCs are not 
inherent in coal mine respirable dust sampling."  From this, 
she concludes that random events (like accidentally dropping 
a toolbox on an airhose) cannot explain the occurrence of any 
AWC at any mine, and that the ALJ could not reasonably 
have relied on random events to explain Urling's high fre- quency of AWCs.

     But the Secretary overstates the record evidence and mis- understands the implications to be drawn from the fact of 
non-random distribution across mines.  Before the ALJ, the 
Secretary's experts Marple and Thaxton conceded that the 
Urling AWCs could have been accidentally caused, and that 
the evidence could not establish whether the pattern on any 
particular filter resulted from tampering.  Miller, the Secre- tary's statistical expert, did not conclude that intentional 
misconduct caused the Urling AWCs, but testified only that 
his conclusions were not inconsistent with tampering.

     At best, this evidence demonstrates nothing more than that 
the likelihood of finding an AWC on a randomly selected filter 
sample is affected by the mine from which the filter is drawn.  
In the universe of possible AWC causes, intentional tamper- ing by certain operators is only one of many possibilities that 
could explain why AWCs occur more frequently at certain 
mines.  Even if all AWCs resulted from purely accidental 



causes which were randomly distributed across all mines, the 
fact that AWC likelihood is affected by environmental condi- tions like humidity would lead one to expect a non-random 
distribution of AWCs across mines. The AWC Rate Decline in Late March, 1990

     The Secretary argues that the drop in AWC rates in late 
March 1990 was statistically significant and interprets it as an 
indicator of intentional tampering.  Because of the correlation 
between the drop and the date of issue of the AWC void code, 
the Secretary speculates that Keystone had been tampering 
but stopped once it learned of the void code.  It is undisput- ed, the Secretary asserts, that Keystone learned of the new 
void code on March 26, 1990.  Miller, the Secretary's statisti- cian, testified that between August 1989 and March 26, 1990, 
Keystone's weekly AWC rate fluctuated between 40% and 
45%, but after March 26, the rate dropped to near zero and 
stayed there.  According to the Secretary, the "obvious infer- ence" from this is that Eget and Houck, who had sometime 
earlier learned that MSHA was investigating Keystone's sam- pling, discovered on that date that MSHA would no longer 
accept AWCs on dust samples.

     The Secretary claims that the ALJ reached his conclusions 
based solely on Keystone's proffered methodology which ana- lyzed AWC rates on the basis of a bimonthly average.  The 
Secretary argues that there was no good reason for analyzing 
AWC rates over such a long period, where samples were 
collected continuously.  Of course, such a bimonthly sample 
interval could make the reduction in AWC rates appear much 
more gradual, washing out evidence of a sudden change.  But 
the ALJ did not simply adopt Keystone's statistics, as the 
Secretary argues.  Rather, the ALJ weighed all of the statis- tical evidence, and found that on balance no conclusion could 
be drawn that there was a dramatic change in AWC rates on 
or around March 26 that was caused by the issuance of the 
AWC void code.

     On this point, the Secretary advances one reasonable inter- pretation of the March 26 data.  Were we reviewing the 
evidence de novo, we might (or might not) favor her interpre-



tation.  But she falls far short of establishing that the ALJ 
lacked substantial evidence to reject her interpretation.  
There is strong evidence that well before March 26, ESD 
personnel were aware of the MSHA investigation.  The ALJ 
could have reasonably agreed with Keystone that, if truly 
motivated to stop tampering because of fear of discovery, it 
would have more naturally done so well before March 26--in 
fact, 89 citations had already issued by that date.  The 
Secretary responds that Keystone must not have stopped 
tampering until the date the void code issued, because "[t]he 
statistical evidence points unequivocally to March 26."  This 
sort of circular argument, assuming the conclusion, is typical 
of the analysis the Secretary has advanced in this case and 
does not present an adequate basis to reverse the judgment 
below.

     Even if the March 26 date is ascribed the statistical signifi- cance urged by the Secretary, it is a stretch, given the other 
record evidence, to conclude on that basis that the change in 
AWC rate is explained by cessation of intentional tampering.  
The ALJ found that there were other changes around that 
time, not adequately ruled out by the Secretary's analysis, 
which also could have lowered the AWC rate.  For example, 
in the relevant period, the ALJ found that there were 
changes in filter-to-foil distances and other manufacturing 
variables, increasingly stringent AWC selection criteria, 
changes in sample handling at Urling, changes in sample 
handling by ESD personnel, changes in continuous mining 
machines at Urling, changes in mining conditions, and 
changes in sampler hose softness.  The ALJ evaluated and 
balanced all these factors to conclude that the Secretary had 
not demonstrated that any abrupt change occurred on March 
26 or that changes in AWC rate justified an inference of prior 
tampering.  The evidence demonstrated that Keystone in- creased its use of scrubber miners;  that the U.S. Attorney's 
investigation and obvious scrutiny itself might have caused 
more care in the handling of samples;  that R&P heightened 
its own internal scrutiny and reported actual instances of 
tampering during this period;  that Eget, the roughest han- dler of pumps at Urling 1, did not transport samples between 



April 9 and May 10 because of a bad back;  and that after 
Eget's return, R&P had used up its stock of cassettes with 
shorter filter-to-foil distances and began using new transport 
boxes.  Thus, substantial evidence supported the ALJ's rejec- tion of the Secretary's interpretation of the declining AWC 
rate in late March of 1990. Cassette Manufacture Date

     The Secretary rejects as "speculation" the ALJ's conclusion 
that cassettes manufactured on four consecutive "key dates" 
in 1989 were responsible for significantly more AWCs.  The 
Secretary contends that when used at mines other than 
Keystone, cassettes manufactured on those dates actually had 
a lower than average (2.5%) AWC rate.  The Secretary 
argues that this data suggests nothing more than mere 
correlation:  cassettes manufactured on those dates were used 
in large numbers when AWCs were occurring at high rates 
for other reasons.  For cassettes manufactured on September 
26, 1989 (one of the four dates), 29 of 81 had AWCs before 
March 26, but 0 of 175 had AWCs after March 26.

     The Secretary's expert Miller conceded that the fact that 
R&P mines had different citation rates with the cassettes 
from these dates shows only that something is different in 
the way ESD samples.  The Secretary, of course, attributes 
this difference to intentional tampering by ESD.  That is 
perhaps one reasonable interpretation of the evidence.  On 
the other hand, it is not the only one, and we are obligated 
not to compel adoption of the Secretary's proffered explana- tion if the ALJ reached a different conclusion based on 
substantial evidence.

     The ALJ found that Keystone was different from other 
operators in the way samples were handled and processed.  
Further, evidence supports the finding that cassettes manu- factured on the four key dates in 1989 were responsible for a 
disproportionate number, over half, of R&P and Urling 
AWCs.  The Secretary's data showed that cassettes from 
those dates had shorter filter-to-foil distances than later 
filters, a factor that the ALJ found contributed to the likeli- hood of a non-intentional AWC.  Overall, there is substantial 
evidence in the record to support the ALJ's conclusion that 


the Secretary did not prove by a preponderance of the 
evidence that intentional tampering, rather than some combi- nation of the filter manufacturing, handling by ESD, and 
Urling mine characteristics, caused the Keystone AWCs. 

Quartz Sampling Data

     The Secretary finds error in the ALJ's decision to disre- gard MSHA data on sampling of quartz between August 1989 
and March 1991.  Quartz samples were collected in the same 
fashion and with the same equipment as the coal dust sam- ples, and were transported and processed by ESD in the 
same fashion.  The Secretary claimed that while 44% of the 
dust samples had AWCs, none of the quartz samples had that 
appearance.  The Secretary's explanation was simple:  with 
quartz samples, it is not to the operator's advantage to reduce 
the weight of the dust collected by the device.  The ALJ and 
Commission refused to give any weight to this evidence 
because the filters at issue were not in evidence, having been 
destroyed in the normal process of MSHA's quartz analysis;  
because the filter's appearance had not been preserved 
through photographs or other records;  and because the 
Secretary had failed to call as witnesses any of the personnel 
who actually analyzed the quartz filters.

     In another attempt to shift the burden of proof, the Secre- tary notes that one of the actual testers was on Keystone's 
witness list, but was never called.  She forgets that it is the 
government's burden to prove the existence of a violation.  
Here, the Secretary introduced no direct evidence--not even 
photographs or descriptions of the examined filters--to back 
up these claims.  The ALJ and the Commission did not err in 
refusing to draw any conclusion from this evidence. ESD Employee Conduct and Witness Testimony

     The Secretary introduced direct testimony regarding ESD 
employees looking into dust filters and talking about what 
might happen if they blew into them, arguing that this 
evidence justified the conclusion that they were in fact blow- ing into them.  Keystone offered the testimony of the em- ployees who handled the cassettes to the effect that they did 
not tamper with them.  The Secretary asserts that the ALJ 



should not have credited ESD employees' denials of tamper- ing.  The Secretary describes as "insupportable" the ALJ's 
stated reasons for crediting Eget and Houck:  the absence of 
motive for tampering and the strong disincentive from their 
knowledge of possible sanctions.  The Secretary also argues 
that these witnesses contradicted themselves and each other.  
The Secretary asserts that it was error for the ALJ to have 
believed those witnesses, arguing that "[b]ecause denials of 
tampering by ESD witnesses are inconsistent with the other 
evidence, the ALJ's credibility findings would not stand even 
if they had been based on demeanor."  For this remarkable 
proposition, the Secretary cites two cases, Bishopp v. District 
of Columbia, 788 F.2d 781, 785-86 (D.C. Cir. 1986);  and 
Millar v. FCC, 707 F.2d 1530, 1539 (D.C. Cir. 1983).  Unsur- prisingly, neither of these cases goes anywhere nearly so far 
as to say that a trier of fact commits error by believing a 
witness whose evidence is inconsistent with other evidence.  
Logically, of course, the Secretary's proposition could not 
stand.  If evidence could not be credited when it was contra- dictory to other evidence, then presumably neither could the 
other evidence be credited since it is contradictory to that 
rejected in the first instance.  As one might expect, neither 
the Bishopp nor the Millar case stands for the proposition 
which the Secretary asserts.

     What we actually held in Bishopp was that "we must be 
particularly careful to defer to the district court's credibility 
findings...."  788 F.2d at 786.  Obviously, that is the very 
opposite of what the Secretary asserts.  With due charity to 
the Secretary, we note that we went on to say that in "the 
rare case" we would reverse even "under this very restricted 
scope of review," when "the judge below credited a witness 
whose testimony was so internally inconsistent or implausible 
on its face that a reasonable factfinder could not credit it."  
Id.  Millar is to the same effect, allowing for a reversal 
where a witness's testimony is "so incredible," or is faced by 
"contrary evidence ... so overwhelming," that a reasonable 
factfinder could not believe the testimony regardless of the 
witness's demeanor.  707 F.2d at 1539.  Thus, both of the 
cases upon which the Secretary relies are little more than 



restatements of the "reasonable factfinder" standard of re- view as applied to credibility determinations.  We have allud- ed to the Secretary's misunderstanding of that standard 
above, and will discuss the same further infra.  As to this 
argument, it is sufficient to say that the Secretary has fallen 
far short of that demanding standard.

     The record demonstrates that the ALJ specifically and 
carefully assessed the credibility of the employee witnesses, 
and found that their denials of tampering were not only 
believable, but consistent with other evidence.  The Secretary 
simply has not explained to this Court why we must depart 
from the rule that a factfinder's determinations of credibility 
are entitled to great deference.  Nothing justifies the ex- traordinary step of overturning these findings.  See Chen v. 
GAO, 821 F.2d 732, 738 (D.C. Cir. 1987). Accidents, Rough Handling, Filter Manufacturing, Mine 
Environment

     The Secretary would have us reject the ALJ's findings that 
accidents and rough handling of samples could have contrib- uted to or explained Keystone's high AWC rate.  These 
conclusions were based on the testimony and theories of 
Keystone's scientific expert Lee, who concluded that handling 
could have accounted for many AWCs;  that short filter-to-foil 
distance increased AWC likelihood;  that increased humidity 
decreased AWC likelihood;  that increased use of scrubbers 
made it more difficult to dislodge dust from filters, decreasing 
the AWC rate;  and that the AWCs on Urling filters resem- bled dislodgements caused by impact, not reverse air flow.  
The Secretary argues every detail of the evidence at length.  
In essence, she contends that her scientific evidence was so 
overwhelmingly correct and so clearly compelled her conclu- sion that the ALJ could not lawfully have found against her.  
But the record does not support this proposition.

     The ALJ found that in many instances the Secretary's 
scientific evidence was inconclusive or otherwise could not be 
adequately evaluated.  All of these issues involve conflicting 
expert testimony, and this Court must defer to the reasonable 
determination of the trier of fact regarding not only the 
relevance but the reliability of expert testimony presented at 


trial.  See General Electric Co. v. Joiner, 118 S. Ct. 512 
(1997);  Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 
U.S. 579, 589 (1993);  see also Millar, 707 F.2d at 1539.  We 
cannot deem unreasonable the conclusion that the Secretary 
failed to meet her burden of proof.

     The record clearly supports the proposition that accidental 
events caused at least some citable AWCs at Urling.  Hose 
impacts, for example, occurred routinely in the transport of 
sampling apparatus.  Further, because the cassettes were not 
removed and transported separately from the testing appara- tus, R&P's filters may have had a greater potential for such 
impacts than other mines.  And, the Secretary conceded that 
filter samples collected by MSHA field personnel sometimes 
contained AWCs, apparently caused by opening and reclosing 
of the filter cassettes at MSHA.  Although there was appar- ently no evidence on this point, it is at least possible that the 
Keystone filters might have been opened and reclosed after 
delivery to MSHA. The likelihood of these various possible 
causes cannot be established with mathematical precision.  
The Secretary's burden is to demonstrate, by a preponder- ance of the evidence, that intentional tampering actually 
caused the dust dislodgment on the particular filters at issue 
in each citation.  The ALJ and Commission reasonably con- cluded that she had not carried that burden.

                                     III 


     Ultimately, the Secretary's position is fraught with misun- derstanding of the nature of her burden of proof and of the 
danger of relying on a probabilistic estimate of the correlation 
between some observation and a proffered explanation of its 
cause.  In the first instance, the Secretary never seems to 
accept the fact that we review this case under the standard of 
the reasonable factfinder.  That standard, as we have noted, 
renders the Commission's "findings of fact ... 'conclusive' 
when supported by substantial evidence on the record consid- ered as a whole."  United Steelworkers, 983 F.2d at 244.  
Occasionally, though rarely, we do hold that record evidence 
is not sufficient to support a decision in favor of a party with 



the burden of proof, even in the face of that deferential 
standard.  Even less frequently have we held that evidentiary 
support for the party with the burden of proof was so 
overwhelming that a trier of fact erred by ruling that the 
burdened party had not carried its load.  The Secretary has 
pointed to no such case and our research has uncovered only 
one.  See Gibson Greetings v. NLRB, 53 F.3d 385, 389 (D.C. 
Cir. 1995).

     The closest the Secretary comes is Bishopp, supra.  There, 
the plaintiff had lost in the trial court in an employment 
discrimination case.  Obviously, the plaintiff ultimately bore 
the burden of persuasion.  St. Mary's Honor Ctr. v. Hicks, 
509 U.S. 502, 507-11 (1993).  But in Bishopp, the plaintiffs 
had presented a prima facie case under McDonnell Douglas 
Corp. v. Green, 411 U.S. 792 (1973).  The issue upon which we 
reversed the district court was whether the defendant had 
come forward with legitimate nondiscriminatory reasons for 
the commission of the allegedly discriminatory acts which 
made out the prima facie case.  On that issue, the appellees 
had borne at least the burden of production, and it was on 
that issue that we reversed.  Bishopp, 788 F.2d at 789.  This 
is not to say that we would never find a record so overwhelm- ing as to require us to "direct a verdict" in favor of the party 
with the burden of proof, but it is to say that given the 
deferential standard of review such a case would be rare 
indeed.  This is not such a case.

     Although she picks at various items of evidence, the Secre- tary principally relies on her evidence of probability--that it 
was more likely than not that the cause of any given AWC 
was intentional tampering.  This falls far short of the compel- ling case in which a reasonable finder of fact must find for the 
party with the burden of proof in the face of direct evidence 
supporting the other litigant.  There is a false sense of 
security that comes from the use of numbers, which in this 
context can appear much like scientific data.  But any useful 
scientific measurement must be accompanied by an estimate 
of its uncertainty, and when the entire body of evidence has 
been considered, the Secretary fails to persuade that she has 
established with any certainty that AWCs in general, or 



Keystone's AWCs in particular, were in fact caused by inten- tional tampering.

     Over and over, the Secretary insists that she established 
that the mathematical probability of tampering was some- thing greater than 50%.  Arguing from precedents involving 
employment discrimination, she contends that similar statisti- cal evidence may be deemed sufficient to establish a prima 
facie case of intentional discrimination or to rebut a defen- dant's explanation as pretextual.  See Palmer v. Shultz, 815 
F.2d 84, 90 (D.C. Cir. 1987);  McDonnell Douglas, 411 U.S. at 
792.  Statistics alone may suffice to show illegal discrimina- tion "if they are condemning enough," Berger v. Iron Workers 
Reinforced Rodmen Local 201, 843 F.2d 1395, 1413 (D.C. Cir. 
1988) (citation omitted), and cannot be dismissed "on mere 
conjecture," Palmer, 815 F.2d at 106.  The Secretary notes 
that in those cases, a result more than two standard devia- tions from the mean (indicating a 95% probability that the 
relationship is not random) suffices in most instances to give 
rise to an inference of intentional action.  Berger, 843 F.2d at 
1412.

     These precedents lend little aid to the Secretary's cause.  
Statistics may show a correlation between some characteristic 
(for example, age) and some unequal treatment (for example, 
refusal to hire), yet a finding of discrimination is allowed only 
(1) if the employer fails to present a legitimate justification or 
(2) if the factfinder concludes that the greater weight of the 
evidence, including the statistical data, supports a conclusion 
that the particular employee suffered illegal discrimination.  
In situations where direct evidence is difficult or impossible to 
obtain, a party may meet his burden of proof with statistical 
evidence alone.  (This may account for its acceptance as such 
in some employment discrimination cases.  See, e.g., Berger v. 
Iron Workers Reinforced Rodmen Local 201, 843 F.2d at 
1413;  Palmer v. Shultz, 815 F.2d at 90.)  Even then, statis- tics must reasonably control for a variety of factors to proper- ly define similarly situated employees, and in any event may 
be counterbalanced by evidence providing an alternate expla- nation of the pattern or of the particular action in question.  
The weight given to statistical evidence in such cases is not 



absolute, but depends on the degree to which it rules out 
legitimate explanations and how the statistics factor into the 
balance with the other available evidence.  See, e.g., Coward 
v. ADT Security Systems, 140 F.3d 271, 276-77 (D.C. Cir. 
1998) (Sentelle, J., concurring).  Here, it is true that AWCs 
are not randomly distributed across all mines, and that 
something probably explains the higher frequency of AWCs 
at Urling.  But without direct evidence of tampering, and 
given the substantial basis in the record for alternate theo- ries, there are no statistics "condemning enough" to require 
reversal of the judgment below.

     The Secretary throughout this case assumes that proving 
probability is the same thing as convincing a trier of fact by 
the greater weight of the evidence.  While the two proposi- tions may sound superficially similar, they are not the same.  
This case well illustrates why.  When the Secretary has cited 
a responding mine for tampering with a particular filter, 
certainly evidence of the probability of the cause of the AWC 
on that filter is relevant.  This relevant evidence does not 
mean that the trier of fact must be convinced to any degree 
that the mine operator's employees tampered with that par- ticular filter.  An hypothetical that reverses the facts of this 
case demonstrates why.  If it were the burden of the mine 
operators to prove their innocence, and they came forward 
with evidence that 99% of all filters had never been tampered 
with, this would not mean that they would be entitled to an 
acquittal as to particular filters on which the Secretary could 
offer direct evidence of tampering.  For example, if the same 
witnesses who came forward here to testify that they had 
committed no such acts instead came forward and swore that 
"we tampered with these filters," we could hardly say that a 
reasonable trier of fact would have to disbelieve them because 
statistical data proved that such tampering was extremely 
unlikely.  The same is true here.

     Perhaps the Secretary is right that a majority of the AWCs 
were caused by tampering.  Perhaps she is not.  Either way, 
it is not unreasonable for the finder of fact to conclude that 
the Secretary did not establish that a particular filter in 
evidence fell into the majority rather than the minority group.



     To offer one further hypothetical illustrative of the Secre- tary's misconception, we recall the example created by Pro- fessor L. Jonathan Cohen.  He posits a situation in which 
uncontroverted evidence establishes that something over half 
of 1000 attendees at a rodeo entered without paying the 
admission fee.  He rightly concludes that even though that 
evidence suggests that it would be "more likely than not" the 
case that a randomly selected attendee had not paid, that 
evidence would be legally insufficient to allow judgment 
against a specific selected attendee for the price of admission.  
Most likely such evidence without more would not even be 
submitted to a jury.  See L. Jonathan Cohen, The Probable 
and the Provable 75 (1977).  In our case, the problem is not 
merely that it is difficult to state with precision the probabili- ty that a randomly selected AWC was caused by intentional 
tampering.  The problem here, as in the gatecrasher hypo- thetical, is that the uncertainty arising from all of the infor- mation not presented to the factfinder (e.g., evidence regard- ing potential alternative causes for each AWC, its course of 
handling, mine conditions, and so forth) is of such degree that 
the factfinder cannot confidently say that the weight of the 
evidence supports the proposition.  In other words, the 
weight ascribed to the evidence is affected, in part, by the 
factfinder's judgment about the volume and significance of 
relevant information that is not available for examination.  
See Neil B. Cohen, Conceptualizing Proof and Calculating 
Probabilities:  A Response to Professor Kaye, 73 Cornell L. 
Rev. 78, 86 (1987) ("Convincing the factfinder of such a 
probabilistic judgment requires more ... than simply noting 
that the best guess of the probability exceeds 0.5;  rather, ... 
the factfinder also takes into account its judgment as to how 
likely the best guess is to 'hold up.' ").

                                  Conclusion


     In each of these proceedings, whether we would have 
reached the same conclusion as the ALJ is irrelevant.  We 
might have upheld a ruling in favor of the Secretary on the 
basis of this record.  But the Secretary has not come close to 
proving that the decisions below were unreasonable or not 



supported by substantial evidence.  Indeed, we find it highly 
unlikely that the government would desire a standard of 
review that would allow us to reverse such a decision based 
on nothing more than our distant and inexpert view of the 
record evidence.  We therefore affirm the decision of the 
Commission and deny the petition for review.