Mary S. Fowler

Worcester State College

Joseph B. Kadane

Carnegie Mellon University

Journal of Statistics Education Volume 14, Number 3 (2006), jse.amstat.org/v14n3/kadane.html

Copyright © 2006 by Mary S. Fowler and Joseph B. Kadane all rights reserved. This text may be freely shared among individuals, but it may not be republished in any medium without express written consent from the authors and advance notification of the editor.

**Key Words:** Expectation; Law; Location-scale family

We develop a probability model that estimates the money due to the tribes for a given distribution of oil and gas prices. The case was settled before data could be used to estimate this distribution; hence, this example demonstrates the power of applying probability in a legal application. This problem offers two things to a calculus-based probability class; first, it is nice to have a non-textbook example of the relevance of probability both through direct calculus application and computer simulation and second, it introduces the class to the application of probability and statistics to the legal field.

In Section 2, we present the historical background of Indian reservations and US involvement in the sale of oil and gas from Indian land. In Section 3, we describe Major Portion Analysis, the federally regulated method to establish royalty payments for gas and oil. In Section 4, we give an example of Major Portion Analysis and present a probability model for the loss of revenues due to a failure to perform Major Portion Analysis. Section 5 uses the probability model to develop an expression for the expected value of the revenue lost by the Indians, in general and in the special case of the normal and t-distributions. In Section 6, the losses are estimated through a computer simulation for a distribution that has a form for loss of revenue that is not mathematically tractable. Section 7 gives some additional problems appropriate for a student in a calculus based probability class.

http://digital.library.okstate.edu/KAPPLER/index.htm,

http://www.hanksville.org/NAresources/indices/NAhistory.html,

http://www.csulb.edu/projects/ais.

By the mid-1800s, there was a migration of settlers from the eastern US to and across tribal lands. The native tribes resisted and it soon became clear that there was a need to settle and clarify land rights and other issues between the US government and the tribal people. In 1863, the first treaty between the US government and the leaders of the Shoshone tribe was signed. [See Kappler (1904).]

In 1878, the federal government authorized a band of starving Northern Arapahos to winter on the Wind River Reservation against the wishes of the Shoshone tribe. The Northern Arapaho tribe remains there to this day. In 1937, the Shoshone tribe successfully sued the U.S. government in the Court of Claims of the United States for half the value of the reservation because the US had essentially given half the Reservation to the Arapaho in violation of the Shoshone treaties. The Shoshone received cash and the Shoshone and Arapaho tribes were confirmed as joint owners of the Wind River Reservation.

In 1979, the Shoshone and Arapaho tribes of the Wind River Reservation sued the U. S. government in connection with the selling of oil and gas, claiming that the federal government had not followed the method prescribed by regulation for valuing oil and gas for royalty purposes. One of the questions in this case was whether and to what extent the tribes would have benefited had it done so. For material on the specific case, see www.usdoj.gov/osg/briefs/2004/2pet/7pet/2004-0929.pet.app.pdf.

For at least 50 years, federal regulations, and Indian leases themselves, have provided that value for royalty purposes of oil and gas be determined using a “major portion analysis.” In a major portion analysis the value of the oil or gas is to be no less than the highest price “paid or offered” at the time of production for the “major portion” (50% plus one barrel of oil or one thousand cubic feet of natural gas) of production. The highest price “paid or offered” means the price at which the transaction was conducted, unless a higher price was offered and refused, in which case the highest price “paid or offered” is the highest price offered. The relevant transactions used to determine the price of the major portion include only transactions within a given month between non-affiliated entities, also known as arm-length transactions, of like quality oil or gas, from the same field or area. Product sold at less than the major portion price is to be valued at the major portion price; product sold at more than the major portion price is to be valued at the sale price, when calculating royalty payments to landowners. These rules are set forth in the Code of Federal Regulation, specifically at 30 C.F.R. section 206.52 (oil) and 30 C.F.R. sections 206.172 and 173 (gas) www.gpoaccess.gov/cfr/index.html. Because the major portion value depends on other transactions going on throughout a given month involving multiple operators who typically do not share pricing information with each other, it must be calculated by the federal government well after the transaction occurs.

For much of the period in dispute, the federal government did not perform a major portion analysis on the oil and gas sold on behalf of the Shoshone and Arapaho tribes of the Wind River Reservation. Instead, royalty payments were based on actual sale price, including non-arm's length sales between affiliates. As a result, the question arose whether and to what extent the tribes would have benefited from increased royalties had it done so.

We can address this question in a two-stage process:

- Explore the variables that affect the loss of royalties and how the total loss of royalties is related to these variables. We need to find a functional form that represents lost royalties as a function of the distribution of sale prices for gas and oil.
- To investigate sales data to determine the distribution of sale prices and use this in the functions described in stage 1 to determine the actual loss of royalties.

In this article, we address only the question in stage 1. A settlement was achieved before the second stage of the analysis was reached. Carrying out a major portion analysis would have been controversial, as quarrels were likely to surface about what oil is of “like quality” and what the “same field or area” means operationally. Despite these practical difficulties and ambiguities, some useful information can be gleaned from the mathematical development of royalty loss as a function of the distribution of sale prices.

- The larger of the best price offered and the price paid for each
transaction is calculated (for “like quality” product in the “same field or
area,” and including only “arms-length” transactions, called here the
relevant transactions). These relevant transactions are weighted by the
volume of product sold, and ordered by price. The median
*m*of these best prices offered or paid is calculated. (The effect of the “plus one barrel of oil” and “plus 1000 cubic feet of gas” is trivial in the calculations that follow, and is omitted for simplicity). - For each barrel of oil, the larger of
*m*and the best price offered or paid for that barrel, is calculated. This is the royalty value of the barrel of oil. Royalty is due on this amount.

A typical royalty rate at the time was 1/6 the value of the oil. In this hypothetical example, the 500 barrels sold at $14 had a royalty value of $16. Hence the Tribes should have been paid ($16 - $14)(500)/6 = $166.67 more than they were for this month. Over many years, and with interest due from the time the money should have been paid to the present, this can add up to a substantial sum of money owed to the Tribes.

Necessarily, the application of (b) cannot lower the royalty
value to which the Tribes are entitled. Suppose that the best prices
paid or offered for relevant transactions for a given field in a given
month are *X*_{1}, ... , *X*_{n}. Necessarily
. When and only when this inequality is strict,
the Tribes gain by the application of major portion analysis. How much
can this be expected to amount to? This depends on the distribution of
the *X*_{i}'s, as the analysis below shows.

Let *X* be a random variable whose
cumulative distribution function (cdf) is *F*. Also let

(1) |

Then *Y* has the
distribution of the royalty value of the oil or gas, and *Y - X* is the
gain in royalty value due to the performance of major portion
analysis. The next section studies the expectation of *Y - X*.

(2) |

Let *t* = (*x - m**)/*s*. Then

(3) |

In a location-scale family, if *Y* has a density *f*(*y*) and *X = m** +
*sY*, where *m** and *s* > 0 are chosen numbers, then *X* has density
*sf*(*sy + m**). In such a family, the density satisfies

(4) |

independent of *m** and *s*. Thus *g*(*t*) is the density of that
member of the location-scale family with *m** = 0 and *s* = 1.

If *X* is in such a family,

(5) |

This representation shows that the expected gain to the tribes from
major portion analysis depends critically on the scale parameter
*s*. The advantage to the tribes of major portion analysis
is proportional to *s*,
with the constant of proportionality dependent on the shape of the
distribution. Next, we evaluate the constant of proportionality for
some examples.

Suppose *X* ~ *N*(*m*, *s*^{2}). Then *X* is a member of the location-scale
family of normal distributions. Hence equation (5) applies. The
integral in (5) is evaluated as follows:

Let *y* = *t*^{2}/2, so *dy* = *t dt*.

Then

(6) | |

For example, an average of $15 per barrel and a standard deviation of $2 per barrel, would lead, by formula (6), to a loss of royalty value of = .80 per barrel. If the oil produced on Tribal Lands were 100,000 barrels in a year, the loss in royalty value would have been $80,000. At a royalty rate of 1/6, this would have come to 80,000/6 = $13,333 for that year.

**Example 2** *Now suppose X has a t-distribution with degrees of
freedom*.

The *t*-distribution most readers may be familiar with is a standard
*t*-distribution which has median 0 and variance where
is the degrees of freedom (see
DeGroot and Schervish (2002),
p. 407). (For degrees of freedom less than 2, the variance of the
*t*-distribution does not exist). However, the *t*-distribution can be
extended to be a location/scale family by allowing a linear function
of a standard *t*, as follows:

Let

where *W* has a standard *t*-distribution with degrees of
freedom. Then *X* has median *m**, and variance . (See
DeGroot (1970, 2004), p. 42) for more on the location/scale
extension of *t*-distributions).

While the *t*-distribution is defined for all positive
, it has a mean only if > 1.
Hence we restrict to
that domain for this calculation. Again, *X* is in a location-scale family, so again
(5) applies, and the integral is

Where *B*(*a*,*b*) is the beta function defined by
for *a* > 0, *b* > 0.

Let , and , so .

Then

Then

(7) | |

Figure 1 displays the expected gain from major portion analysis in
royalty valuation, *E*(*Y - X*), for *t*-distribution (as a function of
the degrees of freedom, ), and for the normal distribution. Both
calculations take the scale parameter *s* equal to one. The star for
the normal distribution is displayed on the extreme right of Figure 1,
because a normal distribution is the same as a *t*-distribution with
infinite degrees of freedom.

Figure 1: *E*(*Y - X*) as a function of degrees of freedom.

That the location parameter *m*^{*} is irrelevant to *E*(*Y - X*) should not
be a surprise, since adding a constant to *X* adds a constant to *Y*,
and hence leaves *Y - X* unchanged. Similarly it should not be a
surprise that as the scale *s* increases, the expected gain to the
tribes increases and conversely as *s* decreases so does the expected
gain. As a limiting case, we note that as *s* approaches zero, all
prices approach *m*^{*} and hence the gains to the tribes would go to
zero. In every location-scale family with a mean, the expected gain in
royalty value due to the tribes by conducting a major portion analysis
is a constant times the scale. Like exact derivations using
calculus, simulations are important tools for modern statisticians.

We can approximate *E*(*Y - X*) by simulating the process of major portion
analysis. The following **R**-code achieves such a simulation:

n = 100 x = rgamma (n, shape = 15, scale = 1) #this draws n=100 independent observations from a gamma #distribution with alpha = shape = 15 and scale = beta = 1. #The mean is alpha/beta = 15 and the variance is alpha/beta^2 = 15 m = median(x). y=pmax (x,m) # this yields a vector of length n whose ith # elements is the larger of x[i] and m. mean(y-x) # this is the estimated average amount per observation # of underestimation of the royalty value of the oil by #virtue of not conducting major portion analysis. sd(y-x) # this computes the standard deviation of (y-x).

We did this and obtained: mean(*y - x*) = 0.942 and *sd*(*y - x*) = 1.29. If the
simulation is repeated, different values for the mean and standard
deviation will be obtained, because the random numbers drawn will
differ.

The observations simulated from the gamma distribution are independent
and identically distributed, and so we can apply the central limit
theorem, which tells us that the standard deviation of the mean of
*Y - X* is estimated by . Hence the larger our *n* is
in our simulation, the more stable our estimate of the mean of (*Y - X*)
will be. To see this more clearly we repeated the above simulation for
*n* equal 10, 100, 1000, 10000, 100000 and obtained the following
results

n | mean(Y - X) | sd(Y - X) | |
---|---|---|---|

10 | 1.45 | 2.35 | 0.743 |

100 | 1.19 | 1.82 | 0.182 |

1000 | 1.35 | 1.93 | 0.061 |

10000 | 1.35 | 1.89 | 0.0189 |

100000 | 1.37 | 1.92 | 0.00607 |

The results reported in Table 1 are plotted in
Figure 2. Each line
represents a 2-standard deviation interval around the mean. As *n*
increases from 10 to 100,000, *i.e.*, as log(*n*) increases from
1 to 5, uncertainty is reduced, as is predicted by the central limit
theorem.

Figure 2: 95% confidence intervals for *Y - X* as a function of sample size for Gamma (15,1).

The above **R**-code can be used to simulate a major portion analysis from
other distributions by substituting another random number generator for

x = rgamma (n, shape = 15, scale = 1).

For example, to simulate Example 1 numerically, where *X* has a normal
distribution, we replace the above line with

x = rnorm (n, 15, 1).

Here the mean is equal to 15 and the variance is equal to 1.
According to equation 6, the exact value of
.
We ran the simulation in **R** and obtained mean(*Y - X*) = 0.3738. If you try the
same simulation, you will draw a different set of random numbers and
hence your value for mean(*Y - X*) will differ.

Again, if you wish to simulate from a beta distribution with mean
*a/(a+b)* = 1/3 and variance = *ab/(a+b)(a+b+1)* = 1/18, this implies
*a* = 1 and *b* = 2. Then the **R**-code would be

x = rbeta (n, 1,2).

- Suppose that
*X*is symmetric around its mean*m*, and has variance . Then- .
- .
- As a consequence, show that the correlation between
*X*and*Y*is given by , where is the standard deviation of*Y*.

- Since the normal distribution is the limiting case of the
*t*-distribution as , it is natural to conjecture that the constant multiplying the scale factor in equation (6), , is the limit as of the constant multiplying the scale factor in equation (7), . Prove that this is true. -
- Show that the moment generating function of
*Y - X*when*X ~ N(m, s*is^{2}). - Use derivatives of this result to check the mean and
second moment of
*Y - X*.

- Show that the moment generating function of

The case was settled after the presentation of the above model. Documents from the case are available on line through the PACER system. You can register for PACER at pacer.psc.uscourts.gov/register.html. Once registered, go to pacer.psc.uscourts.gov/psco/cgi-bin/links.pl and choose the United States Federal Claims Court website: ecf.cofc.uscourts.gov/ and enter case number 79-458. We hope to demonstrate to students the power of probability modeling, both through calculus based theory and computer simulation and that student will begin to consider the field of legal statistics and a source of interesting applications with importance to our society.

DeGroot, M. H. (1970, 2004), *Optimal Statistical Decisions*, Hoboken, NJ: J. Wiley and Sons.

Kappler, C. J. (editor) (1904), *Indian Affairs: Laws and Treaties, Vol. II, Treaties*. Compiled and edited by
Charles J. Kappler. Washington: Government Printing Office, 1904.

Mary S. Fowler

Department of Mathematics

Worcester State College

Worcester, MA 01602

U.S.A.
*mfowler@worcester.edu*

Joseph B. Kadane

Department of Statistics

Carnegie Mellon University

Pittsburgh, PA 15213

U.S.A.
*kadane@stat.cmu.edu*

Volume 14 (2006) | Archive | Index | Data Archive | Information Service | Editorial Board | Guidelines for Authors | Guidelines for Data Contributors | Home Page | Contact JSE | ASA Publications