Journal of Business and Accounting

Vol. 9, No. 1; Fall 2016

USING DIFFERENT PROBABILITY DISTRIBUTIONS

FOR MANAGERIAL ACCOUNTING TECHNIQUE: THE
COST-VOLUME-PROFIT ANALYSIS
Hasssan A. Said
Austin Peay State University

ABSTRACT: The stochastic cost-volume-profit (CVP) analysis has received

ample attention in the accounting, finance, and economics literature, not only
because it is a pivotal technique, but also because methods developed in these areas
are often transferable to stochastic applications of decision science problems.
Because CVP analysis is based on statistical models, decisions can be broken down
into probabilities that help with short-term decision-making objectives. This study
explores, investigates, and applies the CVP model to four different statistical
distributions. Rather than rendering an exact mathematical model, the analysis is
based on specific input information and requires tremendous attention to details,
the best that CVP can do is provide approximate answers to practical problems.
The CVPs assumptions embodies sacrifices of the models pragmatism and
accuracy, however, advancements in software technologies have made cost, effort,
and time inexpensive to estimate the variables making solutions stochastically
more feasible. Undoubtedly, an exact solution to an unrealistic model has very
little practical value. Ultimately, managements acumen has to be made after
careful consideration of inputs and not just rely solely on the model statistical
outcomes.

Keywords: CVP analysis; probability distribution; Beta-PERT; Skewness;

Kurtosis; EasyFit

INTRUDUCTION

The use of cost-volume-profit (CVP) analysis has application not only in

the manufacturing sector but also for financial services entities (Basu et al. 1994).
Despite a considerable research literature progress on CVP analysis, that has
accumulated since the seminal contribution of Jaedicke and Robichek (1964), this
advancement has been almost entirely unheeded by textbooks authors of
accounting and finance. Like all financial models, CVP, is based on a set of
simplifying assumptions that reduce the complexity of input and output variables
to make decision making more tractable. To understand a financial model and its
usefulness, its assumptions and their role in a decision must be understood.
According to Horngren and Foster (2010), the basic CVP model is subject to ten
essential assumptions and limiting conditions: behavior of costs and revenues is
linear, selling prices are constant, prices of production inputs are constant, all costs
can be categorized into their fixed and variable elements, total fixed costs remain

3
Said

constant, total variable costs are proportional to volume, efficiency and

productivity are constant, the model involves a constant sales mix or a single
product, revenues and costs are being compared over a unit-volume base, and
volume is the only driver of costs. Learning the basic deterministic CVP model is
fortunate for students, but an understanding of the generalization of the model to
uncertainty situations and relaxing some of its limiting conditions is an added
improvement. A CVP model that incorporated uncertainty would hence provide a
good entry point into the essential but challenging topic of decision-making under
uncertainty. Virtually all real-world business decisions take place under conditions
of uncertainty, and that at least some modest degree of familiarity with analytical
approaches to decision-making under uncertainty could well benefit the future
business leaders. The seminal application of uncertainty to the CVP model was
first introduced by Jaedicke and Robichek using the basic CVP equation:

Z = Q (P-V) - F (1)
Where Z = Profit, Q = Unit Sales, P = Price/Unit, V =Variable Cost/unit, F=
Total Fixed Cost
Various statistical distributions has been investigated perilously such as
normal (Jaedicke and Robichek, here after JR, (1964)), log normal (Hilliard and
Leitch (1975)) and several distribution-free methods such as the Tchebycheff
Inequality (Buzby (1974)), model sampling (Liao (1975), and Kottas and Lau
(1978)), and additional improvements are examined to the CVP model such as
multiproduct (Johnson and Simik (1971), and cost of capital and degree of
operating leverage (Guidry, Horrigan and Craycraft (1998)), all have been
employed by these and other authors (Shih (1979), and Yunker and Yunker (2003)
and Banker, Dyzalov, and Plehn-Dujowich (2014)) to analyze the demand
uncertainty, cost behavior and the random behavior of profits. The application of
these works was largely confined to the assessment of probability distribution of
profit and the calculations of their central tendency (mean) and spread (variance)
to identify the "best"' choice among alternative measures of profit.
Thus far, this extensive literature has been virtually ignored by managerial
and cost accounting authors, e.g., Garrison, Noreen, and Brewer (2011),
Zimmerman (2013), Warren, Reeve, and Duchac (2014), and their reluctance to
undertake CVP models under uncertainty may be attributed to the diversity and
complexity of the research literature, i.e., multi-product, multiple uncertainty
sources, the assumption that demand exceeds, equals, or less than production sales,
use of the basic accounting CVP model versus economic demand relating
quantity sold to price and/or unit cost functions. The CVP analysis is expected to
be complicated, connecting as it does to various concepts from economics and
mathematical statistics. However, Bhimani, et al. (2008) cautioned that, in
situations where revenue and cost are not adequately represented by the
simplifying assumption of CVP analysis, managers should consider more
sophisticated approaches to their analysis. Notwithstanding, it is the belief here
that the CVP model provides an excellent context for introducing these analytical
approaches. The extreme simplicity of the basic deterministic CVP model enables

4
Journal of Business and Accounting
a clearer perception of the elements added by generalizing the model to a stochastic
one. While the full mathematical derivations and statistics shown herein are
probably too complicated for most undergraduates, the results themselves are fairly
straightforward, and they facilitate a precise focus on such fundamental concepts
in decision-making under uncertainty as the tradeoff between expected profits and
breakeven probability.
There is tradeoff between the comprehensiveness and accuracy of a model
that tends to generate mathematical complexity and its applicability and ease-of-
use to which it can readily provide convincing answers to particular questions. The
purpose of this research is an attempt to strike an appropriate balance between
these two competing criteria. Statistics is the branch of applied mathematics
concerned with the collection and interpretation of quantitative data and the use of
probability theory to estimate parameters that find use in science, engineering,
business, computer science, and industry. Its importance is given to definitions of
concepts, derivation of formulas and proofs of lemmas and theorems. In business,
emphasis is placed on the concepts, use of formulas without their derivations and
practical applications in all areas of business. Technology and its applications in
accounting, finance, and statistics is trying to orient decisions about business and
economic applications, functionality or, in the case of academic software,
pedagogy. According to Nolan and Lang (2009) approaches to the teaching of
statistics for business have changed dramatically. The advancement in use of
computer technologies in the class room made it easy to use the formulas and
computer software that give various kinds of probabilities, random samples
estimations, confidence intervals, descriptive information that are be able to test
hypotheses make fitting distributions instantly (Madgett,1998).

The American Institute of Certified Public Accountants (2005) states that

technology is pervasive in the accounting profession, stressing that leveraging
technology to develop and enhance functional competencies though appropriate
use of electronic spreadsheets and other software to build models and simulations.
Therefore, what the business students and future professionals should learn, with
the help of computer technologies, is to understand statistical concepts and use
them in analyzing practical data and make appropriate conclusions. This paper sets
forth to analyze and applies CVP models intended specifically for pedagogical use
in managerial and financial accounting progressions as a gateway to decision-
making under uncertainty applying four different distributions: Normal,
Lognormal, PERT, and Kumaraswamy. Section 2 will portray the basic concepts
of CVP model under uncertainty and distribution fitting. In section 3 will detail the
uncertainty in CVP and apply the four distributions above using the same
numerical example, and finally, section 4 briefly summarizes and evaluates the
contribution to business professionals and pedagogy.

5
Said

THE BASIC CONCEPTS OF STOCHASTIC CVP MODEL AND

DISTRIBUTION FITTING
CVP analysis in the certainty case is its popularity as a decision tool used
to determine the breakeven volume or sales, its usefulness is, however, limited by
the deterministic nature of the relationship assumed (see (1) above). The
breakdown point at which sales equal total costs and profit equals zero may be
found as follows:
P x Q* - (F + V x Q*) = Z= 0, where Q* is the breakeven volume (sales in
units).
Consequently, Q* can be written as: Q*= F / (P-V), where: P - V = Contribution
margin per unit = C, and Total Contribution margin is TC =Q x C, thus at
breakeven level TC = F. To convert Q* to dollar sales, multiply both sides of the
Q* formula by P, yields breakeven in sales = S*. S* = F / (1 V/P), where the
denominator is called the contribution margin ratio.
Consider a previous example used by JR (1964), which will be using
throughout the paper, where F = $5.8 x 106, V = $1,750, Q = 5000, P = $3000, then
Q*= 4,640 units and S* = $13.92 x 106, thus, the manager would make sure that
the sales level needs to exceed these thresholds to generate any profit. JR surmised
that the assumptions or simplifications implied in the deterministic model are
justified if they are assumed to lead to the same or better decisions than might be
provided by more intricate yet workable uncertainty models. A realistic approach
model would be to examine the usefulness and the implementation of the model
under uncertainty conditions. Thus, most of the input variables included in the
breakdown formula are subject to a wide range of possible outcomes due to chance
variations. These input variables are: P, Q, V, C, and F that yield the output
variables Z. In a probabilistic CVP analysis, one or all of these input variables may
be treated as a random variable. It is assumed that all input and output variables
are having unimodal (one mode) distributions.
For each random variable it is possible to estimate (fit) the probability
distribution indicating the likelihood that it will take on various possible values.
Raw data is almost never as well behaved as we would like it to be. Consequently,
fitting a statistical distribution to data is part art and part science, requiring
compromises along the way. In a typical managerial accounting and finance
textbook one finds two or three summary measures of the distribution that
generally provide value to a decision maker: the mean (), the standard deviation
(), and the coefficient variation (CV= /). However, additional statistics that are
shown to have importance in explaining the distributions properties are: skewness
(SKW- third central moment about the mean) is a measure of asymmetry about the
mean) and kurtosis (KUR- the fourth central moment about the mean) is primarily
peakedness (width of peak), tail thickness, and lack of shoulders, a higher peak
(higher kurtosis) than the curvature found in a normal distribution. To provide a
comparison of the shape of a given distribution to that of the normal one, the excess
kurtosis measure is usually used instead; distributions with negative or positive
excess kurtosis are called platykurtic or leptokurtic distributions respectively. A

6
Journal of Business and Accounting
leptokurtic distribution (e.g., students) is having a fatter tail and higher
peakedness than normal (see Figure 1).
Figure 1

Distribution fitting is the procedure of selecting a statistical distribution

that best fits to a data set generated by some random process. Random factors affect
all areas of business striving to succeed in today's highly competitive environment
need a tool to deal with risk and uncertainty involved. Using probability
distributions is a scientific way of dealing with uncertainty and making informed
business decisions. In many industries, the use of incorrect models can have
serious consequences such as inability to complete tasks or assess projects in time
leading to substantial time and money loss. Distribution fitting allows the
development of valid models of random processes.
When one is confronted with data that needs to be characterized by a
distribution, it is best to start with the raw data and answer four basic questions
about the data that can help in the characterization. The first relates to whether the
data can take on only discrete or continuous values. Most CVP models have used
continuous distributions and this paper follows that convention. The second looks
at the symmetry of the data and if there is asymmetry, in other words, are positive
and negative outliers equally likely or is one more likely than the other. The third
question is whether there are upper or lower limits on the data; there are some
variables like Q, S, V and F that cannot be lower than zero (non-negative
distributions, i.e., one side bounded) whereas there are others like Z that can be
any amount (unbounded, and if it is bounded its value is unknown). The final and
related question relates to the likelihood of observing extreme values in the
distribution; in some data, the extreme values occur very infrequently whereas in
others, they occur more often. The Normal distribution is defined on the entire
real axis (- , + ), and if the nature of the data is such that it is can only take on
positive values, then this distribution is almost certainly not a good fit.

The shape of the Normal distribution does not depend on the distribution
parameters (; location and ; scale). Even if the data is symmetric by nature, it is
possible that it is best described by one of the heavy-tailed models such as the
Cauchy distribution (See Figure 2). Similarly, one cannot "just guess" and use any
other particular distribution without testing several alternative models.

7
Said

The use of probability distributions involves complex calculations which

are practically impossible or very hard and time consuming to do by hand.
Distribution fitting software helps automate the data analysis and decision making
process, and enables managers to focus on the core business goals rather than

technical issues. In particular, one may decide to settle for a distribution

that less completely fits the data over one that more completely fits it, simply
because estimating the parameters may be easier to do with the former. This may
explain the overwhelming dependence on the normal distribution in practice,

Figure 2

notwithstanding the fact that most data do not meet the criteria needed for the
distribution to fit. Nowadays, there are many low-priced software packages
available in the market to estimate distributions and generate random numbers
fitting these distributions (Excel, Stat::Fit, CumFreq, EasyFit, NetSuite, Vose
Software, Risk Solver, @Risk MATLAB and R). All the results in this research
are obtained using either Excel or EasyFit, employing 100,000 randomly generated
variables to fit the four distributions used in the stochastic CVP model.

UNCERTAINTY AND THE CONVENTIONAL CVP MODEL

A. Normalcy of Profit Model

A probability density function of a Normal distribution is characterized by
location and scale parameters. Location and scale parameters are typically used in
modeling applications. For the normal distribution, the location and scale
parameters correspond to the and , respectively, and its SKW and KUR are
zeros. However, this is not necessarily true for other distributions. JR introduced
uncertainty into the conventional CVP model by assuming first that only one
independent variable, volume (Q) that is independent and normally distributed
while all other inputs are given, known values with certainty (deterministic), thus,
the profit equations may be written as E (Z) = E (Q) (P - V) - F, where E is the
expectation operator. However, if all components are normally distributed and
independent of each other and that the resulting profit is also normally distributed.
They defined the expected value of profit as:

8
Journal of Business and Accounting
E (Z) = E (Q) [E (P) E (V)] E (F) (2)

It is known that the subtraction of one normal variable from another yield
a normal variable, therefore, the distribution of the resulting profits (Z) is close to
normal because a normal variable (F) is subtracted from an approximately normal
variable (Q x (P V = C)). Nevertheless, JR have assumed the multiplication of
two normal variables (Q and C). The issue of (Z) being normally distributed
random variable was then questioned by Ferrara, Hayya and Nachman (FHN
thereafter, (1972)) arguing that the total contribution (Q x C) cannot be distributed
normally unless the sum of the coefficients of variation of the two variables (CVQ
and CVC) is less than or equal to 12 percent at the 0.05 significance level. Since all
input variables are mutually independent, there are no correlations between them,
and the variance of profit (Z) is 2 (Z) = 2 (Z), 2 (Z) is the second moment about
the mean (Z) = E (Z). Throughout the paper the use of first (central) moment is
to represent the mean and the second moment about the mean is to represent the
variance of the random variable, and the third and fourth moments are representing
its skewness and kurtosis respectively. Kottas and Lau (1978) have developed
formulas for computing the second to fourth moments of two random variables, in
the following example these formulas will be used to illustrate relationships and
distribution properties that are govern by at least their four moments.
Consider the previous example that is used by JR (1964) with added information:
Q ~ N (5000, 4002), P ~ N (3000, 502), V ~ N (1750, 752), F~ N (5800000, 1000002)
The coefficients of variations (CV = / ) are CVQ = 8% and CVP = 1.67%, CVV
= 4.29%, CVF = 1.72%. Given the uncertainty situation the expected profit is E (Z)
= 5000 [3000 1750] 5800000 = $450,000 = the first central moment = (Z) =
Median = Mode of Z.
Using central moments notation, remember that all input variables are pairwise
independent, thus contribution margin per unit is
(C) = (P) (V) = 3000 1750 = $1250, and (C) x (Q) = (TC) =
Total Contribution margin, (TC) = 5000 x 1250 = $6,250,000, then expected
profit is
(Z) = (TC) (F) = (1250) (5000) (5800000) = $450,000.
The second moments about the mean for output variables are 2 (C) = 2 (P) + 2
(V) = 502 + 752 = 8,125, and
2 (TC) = ( (Q))2 x (2 (C)) + ( (C))2 x (2 (Q) + 2 (Q) x 2 (C)
2 (TC) = (5000)2 (8125) + (1250)2 (400)2 + (400)2 (8125) = 4.54425 x 1011, and
2 (Z) = 2 (TC) + 2 (F) = 4.54425 x 1011 + 1000002 = 4.64425x1011
Since the input variables are statistically independent, therefore the profit standard
deviation can be written as:

(Z) = {2 (Q) [2 (P) + 2 (V)] + (E (Q))2 [2 (P) + 2 (V)] + [E (P) E (V)]2

2 (Q) + 2 (F)}1/2
= 2 (Z) (3)

(Z) = {4002 (502 + 752) + 50002 (502 + 752) + [3000 1750]2 (4002) + 1000002}1/2

9
Said

= [4.64425x1011]1/2
(Z) = 681,487 = [2 (Z)] 1/2
The CVZ = 151.44%, it is very high relative to the CV of input variables. The
measure is useful because the standard deviation (spread) of data must always be
understood in the context of its mean. That is, the actual value of the CV is
independent of the unit in which the measurement has been taken, so it is a
dimensionless number that allows comparison of risk versus expected profit. At
the 95% confidence level the probability to generate a loss P(Z<0) is 25.45%, and
the chance of breaking even is 74.55%, and the probability of generating profit
more than $450,000 and less than a million dollars is 29.02%. If the input means
stayed the same but their spreads () increased then the risk (probability of upside-
profit and downside-loss) of Z is increased, but the breakeven of Z stays the same.
The 95% of the area under a normal curve lies within z (=1.96) standard deviations
of the mean, i.e., P (Z1 < Z < Z2) = 95%, thus (Z) z x (Z) = $450,000 (1.96)
$681,487, representing the range of Z from Z1 = $-885714.52 to Z2 = $1,785,714.
(See figure 3)
Figure 3

The skewness of Z (SKW (Z)) resulting from the multiplication of C and

Q is dependent on the coefficients of variation of C and Q, accordingly, the larger
of the confidents the more skewed is the distribution of TC and consequently Z.
Ware and Lad (2003) show that normalcy of the product of the two products
function depend on the CVs of the two variables and their correlation, but since
correlation= 0, since Q and C are statistically independent, it will only be necessary
to determine how large are the two coefficients must be in order for the frequency
distribution of Z to approach normal distribution. In the above example, the
coefficients of variations are CVQ = 400 / 5000 = 8% and CVC = [(8125)1/2] / 1250
= 7.21% and their sum is 15.21%, which is more that 12% stated by FHN above.
It is still normal but slightly skewed to the right. The calculation of SKW and Kur

10
Journal of Business and Accounting
are done through computer program for ease of time and efforts. If Z is strictly
normal we would have zeros for both measures, but we know this Z distribution is
slightly skewed to the right.
SKW (Z) = E [(Z - (Z) / 3 (Z))], this measure also written as (3 (Z) / (Z)) 3 =
4.875 x 1016 / (681487)3
SKW (Z) = .154029 that is positively skewed i.e., the right tail is longer and the
mass of the distribution is more concentrated on the left. To see the fatness of the
tail and whether the distribution has higher peak than normal distribution we look
at the KUR measure.
KUR (Z) = {E [(Z - (Z)] 4} / {[E (Z - (Z)] 2 }2
= [4 (Z) / 4 (Z)]
Excess Kurtosis (EKUR) however is more commonly defined as the fourth
moment around the mean divided by the square of the variance of the distribution
minus 3, thus, EKUR = KUR (Z) 3= {[6.5415 x 1023] / [(681487)2]2} - 3 =
3.03282 -3 = 0.03282. Since EKUR for normal distribution is zero, the Z
distribution is slightly having more tail or higher peak. Note that SKW and KUR
are uneven measures of normalcy of unimodal distributions. A distribution could
be perfectly symmetrical (zero skewness) yet may be very peaked (e.g., Cauchy
Distribution). In such case the distribution being tested would not be normal, but
use of skewness alone would shows opposite suggestion, thus, both tests yield
better conclusion of departure from normalcy and ignoring both is misjudgment.

B. The Lognormal Distribution Application

The question of normally distributed and statistically independent input
variables under conditions of certainty of the CVP model of JR has been criticized
by FHN (1972), Hilliard and Leitch (1975), Lau and Lau (1976), and Kottas and
Lau (1978) on the ground that, at 5% confidence level, Z is not likely to be
normally distributed unless certain restrictive conditions are met, and there may be
natural dependency between the input variables. The independency assumption of
inputs severely restricts realism in application of the model since Q, P, and V are
often correlated, and the normalcy assumption raises the possibility of negative S,
P, V, and F since the lower bound of a normal distribution could have negative
values. Hilliard and Leitch proposed that the input variables in the CVP model be
assumed log normally distributed. There are two major advantages for using
lognormal distribution: (1) Using it solves the problem associated with the
multiplicative operation Q x C. (2) The distribution is more appropriate for
describing the behavior of the input variables. Hilliard and Leitch assumed that the
inputs Q and C are bivariate log normally distributed and F is deterministic.
Similar to the normal the lognormal distribution is completely defined by its two
parameters values: the first moment about zero and the second moment about the
mean (location = , and scale = , parameters respectively), the log normal
distributions coefficient of variation (CV = {[exp (2)] -1}1/2 and it is independent
of its arithmetic mean.

11
Said

The normal and lognormal distributions are closely related. Assuming that the
input variables Q and C are log normally distributed that means the log of Q and
the log of C are normally distributed. For example, if Q L is distributed log normally
with parameters L and L, then log (Q L) is distributed normally with mean N and
standard deviation N. Similarly, if QN has a normal distribution, then QL = exp
(QN) has lognormal distribution. Note that in all calculations the natural log (ln) is
used and exp term represents the natural exponential function. Olsson (2005)
shows that if Q is log normally distributed, then the median of log (Q) is equal to
the median of Q, but the mean of log (Q) does not equal the antilog (exp) of the
log (Q). Since lognormal and normal are related, MATLAB of MathWorks, Inc.
has developed formulas (Lognstat) that can be used to convert the L and L of log
normal distribution to N and N of normal distribution and vice versa. These are
as follows:
L = exp [N + (N 2) / 2] (4)
L = { exp [2 N + N 2] . exp [N 2] 1}1/2 (5)
and
N = LN {[(L) 2 / [L 2 + L 2]1/2} (6)
N = {LN [L 2 / [L 2 ] + 1)] }1/2 (7)

Using the same information of Q in RJ example before, but now Q is log

normally distributed, that is Q ~ LOGN (5000, 4002), we can generate the normally
distributed N and N using (6) and (7) above.
N (Q) = LN {50002 / [4002 + 50002] 1/2} = 8.5140034, and N (Q) = {LN [4002
/50002] + 1}1/2 = 0.07987442
Similarly, the L and L can be converted (using (4) and (5) above) and the results
above to generate the log normal Q parameters:
L (Q) = exp [8.5140034 + (0.07987442) 2 /2 = 5000, and
L (Q) = {exp [2(8.5140034) + (0.07987442) 2] * exp [(0.07987442) 2] 1} 1/2 =
400
Similarly, attempt is made matching pervious result of L (Z) = $681,487 using (5)
for calculated values of
N (Z) = 12.421034 and N (Z) =1.091759
L (Z) = {[exp 2(12.421034) + (1.091759) 2] [exp [(1.091759) 2] 1]} 1/2 =
$681,487

If inputs Q and C are log normally distributed then Z is log normally

distributed since the product of bivariate log normal is also log normal, hence, the
expected values are
E [LN (Q)] = L (Q) and E [LN (C)] = L (C), where L (C) = L (P) - L (V) =
$1250
L (C) = [L2 (P) + L2 (V)- 2*COV (P, V)]1/2, where COV is the covariance
between P and V. If the COV(P,V) is zero then L (C) = [L2 (P) + L2 (V)] 1/2 , and
then 2L (C) = [502 + 752] = 8125
Previously stated in the deterministic CVP model that TC = Q x C, and Z = TC
F, then TC= Z + F, therefore

12
Journal of Business and Accounting
E [LN (TC)] = E [LN (Q) + E [LN (C)], and hence
L [LN (TC)] = {L2 [LN (Q)] + L2 [LN (C)] + 2 COV [LN (Q), LN (C)]} 1/2, where
COV is the covariance between the log of Q and the log of C, the last term is
positive because the correlation between Q and C is expected to be positive, if it is
assumed zero that means the input variables are independent (i.e., the correlations
between Q and P and between Q and V are zeros), thus the expected profit is

E (Z) = L (Z) = L (Q) [L (P) - L (V)] - L (F)

(8)

This is similar to JRs expected profit of $450,000.

If, however, the input variables are dependent then the expected value of Z is

E (Z) = L (Z) = L(Q) [L (P) - L (V)] - L (F) + L (Q) [QP . L (P) - QV . L

(V)] (9)

Where QP is the correlation between Q and P and QV is the correlation between

Q and V.
If Q= 0 (or C 0; shutdown point) then E (Z) = - L (F) =$5,800,000, which is the
lower bound of Z, highly unlikely. If the correlations in (9) are zeros then equation
(8) equals (9). However, if L (Q) differs considerably from zero then its effect is
very pronounced. Economies of scale and laws of supply and demand rubrics
would suggest that the above correlations are likely to be less than zero. Thus,
expected profit is likely to be a decreasing function of L (V) and an increasing
function of L (P). In the log normal CVP model with correlations of zeros (no
dependency) the probability of generation less than zero profit (i.e., loss) is 26.1%,
at 95% confidence level, slightly higher than the normal model result of 25.45%.
Using (9), the model shows a lower expected profit of $435,000 when QP = -0.9
and QV = -0.1 and PV =0.
E (Z) = 5000 [3000 -1750] 5800000 + 400 [(- 0.9) (50) (- 0.1) (75)] = $435,000
with probability of 20.1%. Under normal model the expected profit at 95% level
the range is between $-885,690 and $1,785,690 and for the log normal with zero
correlation the range at the same confidence level the range is between -$767,000
and 1,872,000 (see figure 4)

C. Beta-PERT Distribution Approach

Under uncertainty conditions, managers appreciate and prefer the information
that has predictability within a range rather than a point estimate. Because it is
highly effective, simple, cost efficient, and computationally convenient, the Beta-
PERT technique utilizes predictive information that helps managers to incorporate
the uncertainty element into the conventional CVP analysis. It integrates the
managers business acumen, and knowledge of statistics and accounting into a
framework useful for practical applications. Many managers are familiar with
PERT because it is used in Critical Path Method, first discovered by Malcolm et

13
Said

al. (1959) and used in the Program Evaluation Research Task (PERT). According
to Greer (1970), it originally postulates the time estimates for completing a project
that follows Beta distribution. In making decisions managers are not infallible, for
them the technique provides a mechanism for developing three cost estimates with
a range based on their (background) experience, insights, intuition, and forecast of
economic conditions (Berger, 2006).

Figure 4

To simplify the process, the manager collects (estimates) information for three
scenarios having three cost estimates: Optimistic (O), Most Likely (M), and
Pessimistic (P). These estimates are derived subjectively and describe three
measures required for the PERT application; the maximum (highest costs, P), the
minimum (lowest costs, O), and the in between (most likely, M) that could prevail
for either new or existing products or services (see Figure 5). Assuming that the
variable in question is an economic 'good' like profit, so that it makes sense to set
O > P, however, the technique can be applied equally well to variables like costs
by reversing the roles of P and O. Using the PERT method introduces uncertainty
into a CVP by treating each input or output (Q, P, V, C, F, or Z respectively) as
random variables. The probability distribution of PERT random variable follows
Generalized Beta Distribution. PERT Distribution is a particular case of Beta
Distribution, encompassing a wide range of distributions with values within the
defined range.

In the standard textbook PERT method (Hillier and Lieberman, 2009), the
three estimates are called the PERT parameters and are fed into the mean and
variance formulas for PERT. Letting [a, m, b] be the three assumed PERT
parameters as they are elicited from experts representing minimum, maximum, and
mode of the input variable (e.g., variable cost), then the standard PERT expected
value is = (a + 4 * m + b) / 6.

14
Journal of Business and Accounting

Figure 5

The mean formula is giving weights to the mode (m) twice as much as the ends
(a, b), and that the value of the mean is different from m in all unimodal-
asymmetric PERT distributions. If the mode is closer to a, the tail is longer to the
b direction, bringing mean for the b side and vice versa. The statistician David
Vose, (of Vose Software) has proposed the Modified PERT (adopted by
Mathematica from Wolfram|Alpha). This distribution is more versatile for
applications, because the mean is calculated in a more flexible way.
Mean = = (a + * m + b) / ( +2)
(10)
In this model, the higher lambda () is the steeper the function in the mode
neighbor (higher kurtosis), and the smaller is the distance between mean and mode.
The model also makes the density near the ends (a, b) less important (having less
mass). Obviously Modified PERT becomes standard PERT when lambda () is
equal 4, and that will be the worth with the calculations that follows. The second
formula from the standard PERT is the variance:
2 = (b a)2 / 36
(11)
Farnum and Stanton (1987) show that this combination of =4 and the
denominator of 36 in 2 is limited (i.e., having a constant that is 1/6 of the range
(b - a)), but indeed optimal for a wide range of m, Herreras-Velasco, et al. (2011).
When the mode is close to the middle between the two ends, the density is
symmetrical, and moving to the sides there is an increasingly asymmetrical feature.
This versatility, different than symmetrical Normal distribution, is what makes
Beta-PERT Distribution so convenient distribution to model many metrics from
business world. It is a very common situation in which one needs to assign a
variable within a specified range, where the mode is approaching the two ends.
The Beta Distribution is defined by four parameters, two of them are (
and , called the shape parameters) that defines the make of the Beta function, and
the other two are (a=Min and b=Max), within which there is possibility of having

15
Said

a value. When Min=0 and Max=1, it is called Standard Beta Distribution and is
suitable for modeling percentages (e.g., the proportion of defective items in a
production cycle). The Betas first to fourth moments are dependent only on it
shape parameters ( and ). The Generalized Beta Distribution (hereafter, Beta),
is not limited by the limited (a=0 and b=1) range, but it could take both positive
and negative values of a and b providing a < b.
In Standard Beta, = ( / ( + ), if >1 and > 1, and the mode = m = (( 1) /
( + 2), for any values of and . The parameter is a homogeneity indicator,
that is as increases the distribution masses around the mode. Beta has positive
skewness (right-tailed) for < , and negative skewness (left-tailed) for > . If
= , then m= == median and these location parameters have the highest point
(peak) on the probability density function, and Beta is symmetrical. If m moves to
the left, then (i.e., ( / ( + )) < ), and the distribution is positively skewed.
Figure (6) shows various Beta-PERT density functions by letting m vary
in the range of a=0 to b=10 and m undertakes only integer values from 1 to 9.
Rescaling or shifting of the range does not have an effect on and or their sum.
The parameter is a homogeneity indicator, that is as increases the distribution
masses around the mode. If = 4, then in equation (10) results in the symmetric
density case ( = ), that is when m= 5 = = [(0 + 4 (5) + 10) / 6] = 5= median.
One can see when using (10) (a deterministic model) that is from a low of 2 1/3
to a high of 7 2/3 for m = 1 and m = 9 respectively, which is different from using
the Beta function with ranges from a low of 1 2/3 to a high of 8 1/3 for the same
m values (see Regnier 2005 and Davis 2008). That is why one has to figure out the
mean and variance and proper and that fit the Beta-PERT distribution.
Many of the software packages that might be used for simulation do not have
the Beta-PERT built in. In these cases, a transformation is required to calculate the
four Beta parameters that will produce the Beta-PERT distribution or other desired
beta distribution. The mathematics of the relationship between the general beta and
the Beta-PERT are hammered out by Golenko- Ginzburg (1988) and Davis (2008).
Figure 6

16
Journal of Business and Accounting
Beta-PERT is widely used to fitting probability distributions of variables in
many areas of research investigations. For example, besides its applications in
engineering systems, and decision science, it is also used in risk analysis for
strategic planning, accounting, finance, and marketing research, and even
subjective (Bayesian) probabilities (Fienberg, 2006). The reason why the beta
distribution is so broadly utilized is that it is extremely versatile, in that a variety
of uncertainties can be usefully modelled by it. For example, it can accommodate
a variety of skewnesses, both positive and negative, and thus, when skewness is an
important factor, in the case of CVP analysis, the beta distribution is often put to
use.
The mathematics of the relationship between the general Beta and the Beta-
PERT are hammered out in, among others, Golenko-Ginzburg (1988). The method
of estimating the Beta-PERT distribution from elicited values (Max = b=
Optimistic, Min = a = Pessimistic, and m = most likely) is first to obtain the mean
and variance, using (10) and (11). And the second step is to use the and 2 to
obtain the shape parameters ( and ), using the following two equations develop
by Regnier (2005):

= [( - a) / (b - a)] [ + ], OR = [( - a) / (b - a)] {[( - a) (b - )] / 2]

1} (12)
= [(b - ) / (b - a)] [ + ] = [(b - ) / ( - a)] = ( + ) , OR
= [(b - ) / (b - a)] {[( - a) (b - )] / 2 -1}
(13)

Defining the values of a and b for the CVP model, using the profit variable (Z)
only (for other variables one could follow the same procedure), the Normal
distribution results are utilized. In the Normalcy model it was shown that: (Z)
z x (Z) = $450,000 (1.96) $681,487, representing the range of Z from Z1 = $-
885714 to Z2 = $1,785,714. Consider a = Z1 = $-885714 and b = Z2 = $1,785,714
as the upper and lower bounds. First the median (= mode = mean) of the Normal
distribution for variable Z will used as the mode for Beta-PERT calculations, if
this is done then one would expect that and be equal. Equations (10) and (11)
are utilized for the calculations of (Z) and then (Z) then using them in (12) and
(13) to obtain and that fit the Beta-PERT distribution.
(Z) = (-885714 + 4 * 450000 + 1785714) / (4 +2) = $450,000
2 = [1785714 (-885714)]2 / 36 = 1.982368767 * 1011
= [(450000 - (-885714))] / [1785714 - (-885714))] {[(450000 - (-885714))
(1785714 - 450000)] / 1.982368767 * 1011] 1} = 4
= [(1785714 - 450000) / (1785714 - (-885714))]
{[(450000 - (-885714)) (1785714 - 450000)] / 1.982368767 * 1011 -1} =
4

17
Said

Figure 7

As expected and are equal that signifies symmetrical distribution. The

bound of Z will be retained since they represent fair estimation of the extreme
values, but consideration of the mode as an elicited value is in question.
Experienced experts may have different opinions as to the mean value but largely
agree on the most likely value of the mode, that is the motivation behind the use
of the mode. And since positive skewness is to be expected the mode is elicited at
$400,000 this time and a repeat of the calculations above are in order.

(Z) = (-885714 + 4 * 400000 + 1785714) / (4 +2) = $443,333

2 = [1785714 (-885714)]2 / 36 = 1.982368767*1011
= [(400000 - (-885714))] / [1785714 - (-885714))] {[(400000 - (-885714))
(1785714 - 400000)] / 1.982368767 * 1011] 1} = 3.8442
= [(1785714 - 400000) / (1785714 - (-885714))]
{[(400000 - (-885714)) (1785714 - 400000)] / 1.982368767 * 1011 -1} =
4.143191 (see figure 7)
Again and as expected the result is < . The four parameters (a, b, , ) are fitted
into the Beta-PERT distribution function to get the median, shape of the density,
higher moments and other relevant information. The median of Z is $395,627 and
as a measure of relative variability CVZ = 111.3%, relatively less than that
obtained by the Normal result. At the 95% confidence level the probability to
generate a loss P(Z<0) is 20.10% and the chance of breaking even is about 80%,
and the probability of generating profit more than (Z) = $443,333 and a
probability for less than a million dollars is 0.36.32%. Skewness is small but
positive at 0.045 level and Excess Kurtosis is -0.54332.

D. Kumarasway Distribution Approach

As an alternative to the beta distribution, Kumaraswamy (hereafter K, (1980)
proposed a two-parameter Kumaraswamy distribution on (0, 1), and denoted by K

18
Journal of Business and Accounting
(q, p). The K distribution was initially proposed for applications in hydrology, but
in the few years it has often been used in several other areas e.g., engineering and
simulation studies and hydrology. This distribution is applicable to many natural
phenomena whose outcomes have bounded limits, such as the height of
individuals, scores obtained on a test, atmospheric temperatures daily stream flow,
, daily rain fall, reliability, cash flow, etc. One factor for this increased interest in
this distribution is due to its simple mathematical form and closed functions. And
this research is attempting to proclaim it in the business area.
Nadarajah (2008) has discussed that the K distribution is a special case of the
three parameter Beta distribution. The basic properties of the distribution have
been given by Jones (2009). Garg (2009) considered the generalized order statistics
for K distribution. While the interest is mainly in the unimodal distributions, Jones
has shown that the K distribution has two boundary parameters (c and b) and two
shape parameters (p and q), and the following density function in its generalized
form:
1 (zc) p1 () 1
() = [] [(bc)] [1 [ ] ,c<z<b

If p > 1, q > 1, then K distribution is unimodal, if p =1= q = 1 then K distribution

is uniantimodal (uniform), If p < 1, q < 1, then K distribution is uniantimodal, if p
1, q 1, then K distribution is increasing, and If p 1, q 1, then K distribution
is decreasing. The interest here is in the K unimodal (both symmetrical and
skewed) distribution, thus one is looking for shape parameters with p > 1, q > 1.
Figure 8

Conceivably the least attractive feature of the K distribution is that, unlike the beta
distribution, it has no symmetric special cases other than the uniform distribution.
The issue here is finding the proper shape parameters for the variable profit
variable, given the results in the normal distribution. In Beta distribution if =
one has a symmetrical shape, but with K distribution if we set p = q > 1, we have
negative skewness that is increasing with the rise in parameters, shifting the mode
to the right. According to Jones (2009) it is possible that skewness to the right is
increased with decreasing p for fixed q, however, there is no simple property for

19
Said

changing q and fixed p. Jones also listed few shape parameters for the symmetric
case. The attempt here is to find shape parameters that fit close enough
symmetrical K distribution that has the same median value that equals pervious
result obtained with the Normal distribution for the profit variable (Z). After few
attempts experimenting with Joness results, recognition of these values was more
appropriate for CVP calculation that follows: p = 2.468 and q = 5.
Armed with these figures and using EasyFit software (because K is not available
in Excel) the following CVP results are obtained:
(Z) = $ 284,125, Median = $281,038, Mode = $281052, 2 (z) = 2.119*1011,
CVz = 1.62%, Skewness = 0.0517, Kurtosis = - 0.578.
Figure 8 shows that all location indicators are close to the median, and that the
portability difference between the mean of Z in the Normal versus K distribution
is P(284,125< Z< 450000) 12.85%. At the 95% confidence level the probability to
generate a loss P(Z<0) is 28.75% and the chance of breaking even is about 71.24%,
and the probability of generating profit more than (Z) = $284,125 and less than
a million dollars is 43.37%. Nowadays, the Beta and Kumaraswamy distributions
are the most popular models to fit continuous bounded data. Further, these models
have many features in common and in a practical situation one question of interest
is how to select the most adequate model (between the Beta and Kumaraswamy
distributions) to fit a certain continuous bounded data set. Up-to-date, there is no
preference in practical applications to favor the Beta against the Kumaraswamy
model.

SUMMARY AND IMPLICATIONS

This endeavor tries to expand the seminal work JR (1964) applying their same
numerical example to explore four different applications of the CVP model.
Results of profit (z) using the four different distributions (Normal, Lognormal,
Beta-PERT, and Kumaraswamy) are presented in Table (1). This study
demonstrates how the deterministic model is being transformed into different
stochastic models. The dissimilarity in means and other location parameters
(mode, median) and variations in risk measurements (e.g., standard deviations,
coefficient of variations, skewness, and kurtosis) are to be taken into consideration
when making current (using historical records) or future (budgeting) decisions.
The extreme simplicity of the basic deterministic CVP model enables a clearer
perception of the elements added by generalizing the model to stochastic ones.
When independency between inputs (Q, P, V, C, F) is suspect then the CVP model
may not produce the factual reality of risk, hence a flawed decision is made, as the
case with the Normal distribution. Correlation between these variables should be
examined initially using the most common and simple method of Pearson
correlation. Thus, managers must overcome of these pitfalls and appreciate a
deeper understanding of dependency issue needed to model the real worlds facets
of uncertainty.
Another issue is the skewness and symmetry of the variables in the data when
fitted to the wrong distribution. Minor abnormality could be tolerated, but

20
Journal of Business and Accounting
significant deviation from normalcy may exhibits significant skewness or kurtosis
that clearly indicate that data need to be fitted with the proper distribution of
adjusted (e.g., transformation though log) to fit normalcy. And if there is
asymmetry, are those extreme values equally befalling, or is one more likely
occurring than the other. Inaccuracies in elicited values from experts are likely to
be larger for lower than upper bounds, because of availability bias that essentially
arises from managers judgment based on available information. Like all fallible
experts expectations, even knowledgeable managers tend to underestimate the
pessimistic case leading to miscalculation by a wide margin. Drawing on prior
managerial experience and know-how to predict projected results as to elicit
benchmarks (costs, profits, sales, and production levels) helps capture uncertainty
inherent in the decision-making process.
Distribution-fitting software are alternative approaches that use various
quantitative measures (e.g., estimation and goodness-of-fit) to rank the suitability
of among distributions for the data. Fitting and simulating variables and
application of risk in the workplace are common now in both government and
businesses. The application of Beta PERT and Kumaraswamy Distributions
explored here highlights the fact that they can be applied without much statistical
rigor for analysis of uncertainty given the advancement in software technologies.
They are pedagogical tools that were expensive or unavailable in the past, the study
of risk and familiarity with common distributions should be part of the textbook
material in several human behavioral sciences including business. Although
students as well as mangers may use the surrogate (deterministic model) and forgo
the more intricate analytical applications, they must not, however, be relied upon
to the exclusion of more sophisticated methods whenever circumstances warrant
and resources permit their execution.

Table (1) Profits (Z)

Distrib- (Z) Median Mode CV(z)
2(z) Skewness Kurtosis
ution (000) (000) (000) %
4.644*101
Normal $450 $450 $450 1 7.21 0.154 0.0328

Log 50.17*101
$450 $328 $120.6 0 1.31 6.185 110.94
Normal
1.982*101
Beta-PERT $443 $396 $400 1 111.3 0.045 - 0.543

Kumara-
$ 284 $281 $281 2.12*1011 1.62 0.0527 - 0.578
swamy

21
Said

REFERENCES

AICPA, Functional Competencies, Core Competency Framework & Educational

Competency Assessment, website: Last viewed, 2/15/2015
http://www.aicpa.org/interestareas/accountingeducation/resources/pages/cor
ecompetency.aspx
Banker, R, Byzalov, D and Plehn-Dujowich, J (2014) Demand Uncertainty and
Cost Behavior, The Accounting Review, Vol. 89, No. 3, pp. 839-865
Basu, N and Conrad, E (1994) Cost-Volume-Profit Analysis: Uses and
Complexities in a Bank,The Journal of Bank Cost & Management
Accounting , vol. 7, No. 2 , p 58
Berge, J (2006) The Case for Objective Bayesian Analysis, Bayesian Analysis,
vol. 1, No 3, pp. 385-402
Bhimani, A, Horngren, C, Datar, S and Foster, G (2008) Management and Cost
Accounting, 4th edition, Pearson Education Ltd, Upper Saddle River, New
Jersey, NJ
Buzby, S (1974) Extending the Applicability of Probabilistic Management
Planning and Control Models, The Accounting Review, vol. 49, No 1, pp.
42-49
Davis, R (2008) Teaching Project Simulation in Excel Using Beta-PERT
Distributions, Informs - Transactions on Education, vol. 8, No. 3, pp. 139
148
Farnum, R and Stanton, W (1987) Some results concerning the estimation of
beta distribution parameters in PERT, Journal of Operation Research
Society, vol. 38, pp. 287-290
Ferrara, W, Hayya, J, and Nachman, D (1972) Normalcy of Profit in the
Jaedicke-Robichek Model, The Accounting Review, Vol. 47, No. 2, pp.
299-307
Fienberg, S (2006) Does it Make Sense to be an 'Objective Bayesian'?
(Comment on Articles by Berger and by Goldstein)." Bayesian Analysis,
vol. 1, No 3, 429-432
Garg, M (2009) On Generalized Order Statistics from Kumaraswamy
Distribution, Tamsui Oxford Journal of Mathematical Sciences, vol. 25, No
2, pp. 153-166
Garrison, R, Noreen, E and Brewer, P (2011) Managerial Accounting, 14th
Edition, McGraw-Hill Education, New York, NY
Golenko-Ginzburg, D (1988) On the Distribution of Activity Time in PERT,
The Journal of the Operational Research Society, Vol. 39, No. 8, pp. 767-
771
Greer, W (1970) Capital Budgeting Analysis with the Timing of Events
Uncertain, The Accounting Review, Vol. 45, No. 1, pp. 103-114
Guidry, F, Horrigan, J, and Craycraft, C (1998) CVP Analysis: A New Look,
Journal of Managerial Issues, vol. 10, No. 1, pp. 74-85

22
Journal of Business and Accounting
Herreras-Velasco, J, Herreras-Pleguezuelo, R. and van Dorp, J., 2011)
Revisiting the PERT mean and variance, European Journal of Operational
Research, Vol. 210, pp. 448451,
Hilliard, E, and Leitch, R (1975) Cost-volume-profit analysis under uncertainty:
A log normal approach, The Accounting Review, vol. L, No. 1, pp. 69-80
Hillier, F and Lieberman, G (2009)Introduction to Operations Research, 9the
edition, McGraw-Hill Higher Education, New York, NY
Horngren, C and Foster, G (2010) Cost Accounting: A Managerial Emphasis,
Prentice Hall, Inc. Upper Saddle River, NJ
Jaedicke, R and Robichek, A (1964) Cost-volume-profit analysis under
conditions of uncertainty The Accounting Review, vol. 39, No 4, pp. 917-
926
Jones, M (2009) Kumaraswamys Distribution: A Beta Type Distribution with
Some Tractability Advantages, Statistical Methodology, vol. 6, No 1, pp.
70-81
Johnson, G and Simik, S (1971) Multiproduct C-V-P Analysis Under
Uncertainty, Journal of Accounting Research, vol. 9, No. 2, pp. 278-286
Kottas, J and Lau, H (1978) A general approach to stochastic management
planning models: An overview, The Accounting Review, vol. 53, No. 2, pp.
389-401
Kottas, J and Lau, H (1978) Direct Simulation in Stochastic CVP Analysis, The
Accounting Review, vol. 53, No. 3, pp. 698-707
Kottas, J and Lau, H (1978) On the accuracy of normalcy approximation in
stochastic C-V-P Analysis: A comment, The Accounting Review, vol. 53,
No. 1, pp. 247-251
Kumaraswamy, P (1980) A generalized probability density function for double-
bounded random processes, Journal of Hydrology, vol. 46, No 1-2, pp. 79
88
Liao, M (1975) Model sampling: A stochastic cost-volume-profit analysis, The
Accounting Review, vol. 50, No. 4, pp. 780-790
Lau, AH and Lau, H (1976) CVP analysis under uncertainty - A log normal
approach: A comment, The Accounting Review, Vol. 51, No. 1, pp. 163-
167
Madgett, A (1998) Some Uses for Distribution-Fitting Software in Teaching
Statistic, The American Statistician, Vol. 52, No. 3, pp. 253-256
Malcolm DG, Roseboom C, Clark C, and Fazar, W (1959) Application of a
technique for research and development program evaluation, Operations
Research, vol. 7, No. 5, 646-649
Nadarajah, S (2008) On the Distribution of Kumaraswamy, Journal of
Hydrology, vol. 348, 34, pp. 568-569
Nolan, D and Duncan, L (2009) Approaches to Broadening the Statistics
Curricula, In Shelley, C, Yore, L and Hand, B (Eds.), Quality Research in
Literacy and Science Education, Chapter 6, Page 1,
<http://www.stat.berkeley.edu/~statcur/Papers/ComputingCurriculumPostio
nPaper.pdf>

23
Said

Last reviewed 2/23/2015

Olsson, U (2005) Confidence Intervals for the Mean of a Log-Normal
Distribution, Journal of Statistics Education, vol. 13, No 1, pp
www.amstat.org/publications/jse/v13n1/olsson.html> last viewed 2/15/2015
Regnier, E (2005) Activity Completion Times in PERTand Scheduling Network
Simulation, DRMI Newsletter Part II, Issue 12 www.nps.navy.mil/drmi/
April 8, 2005, www.nps.navy.mil/drmi/> last viewed 2/15/2015
Shih, W (1979) A general decision model for cost-volume-profit analysis under
uncertainty, The Accounting Review, vol. 54, No 4, 687-706
Vose, D. (1996) Quantitative Risk Analysis: A Guide to Monte Carlo
Simulation Modelling. John Wiley & Sons, New York, NY
Ware, R and Lad, F (2003) The influence of ratios and combined ratios on the
distribution of the product of two independent Gaussian random variables,
University of Canterbury,
http://www.math.canterbury.ac.nz/research/ucdms2003n15.pdf, last viewed
2/23/2015
Warren, C, Reeve, J, and Duchac J (2011) ACCOUNTING, 24th Edition,
Cengage Learning, Inc., Boston, MA
Yunker, J and Yunker, P (2003) Stochastic CVP analysis as a gateway to
decision-making under uncertainty, Journal of Accounting Education, vol.
21, No 4, pp. 339-365
Zimmerman, H (2013) Accounting for Decision Making and Control, 8th Eighth,
McGraw-Hill Higher Education, New York, NY

24
Copyright of Journal of Business & Accounting is the property of American Society of
Business & Behavioral Sciences and its content may not be copied or emailed to multiple
sites or posted to a listserv without the copyright holder's express written permission.
However, users may print, download, or email articles for individual use.