CHAPTER 14

Model Validation

14.1 Introduction
Model validation is the most important step in the model-building process.
However, it is often neglected. Validating a mathematical model usually
consists of quoting the R~ statistic from the fit. Unfortunately, a high R2
value does not mean that the data actually fit well. If the model does not
fit the data well, this negates the purpose of building the model in the first
place.
There are many statistical tools that can be used for model validation.
This chapter covers the basic concepts for model validation, which include
residuals, normal distribution of random errors, missing terms in the func-
tional part of the model, and unnecessary terms in the model. The most
useful is graphical residual analysis. There are many types of plots of residu-
als that allow the model accuracy to be evaluated. There are also several
methods that are important to confirm the adequacy of graphical techniques.
To help interpret a borderline residual plot, a lack-of-fit test for assessing the
correctness of the functional part of the model can be used. The number of
plots that can be used for model validation is limited when the number of
parameters being estimated is relatively close to the size of the data set. This
occurs when there are designed experiments. In this case, residual plots are
often difficult to interpret because of the number of unknown parameters.

14.2 Residuals
The residuals from a fitted model are the differences of the responses at
each combination of variables, and the predicted response using the regres-
sion function. The definition of the residual for the ith observation in the
data set can be written as:
eij = Yii - xYij ( 14-1 )
with yij denoting the ith response in the data set and 'Jij represents the list
of explanatory variables, each set at the corresponding values found in the
ith observation in the data set. If a model is adequate, the residuals should
394 PEM Fuel Cell Modeling and Simulation Using MATLAB |

have no obvious patterns or systematic structure. The primary method of

determining whether the residuals have any particular pattern is by study-
ing the scatterplots. Scatterplots of the residuals are used to check the
assumption of constant standard deviation of random errors.

14.2.1 Drifts in the M e a s u r e m e n t Process

Drifts in the measurement process can be checked by creating a "run order"
or "run sequence" plot of the residuals. These are scatterplots where each
residual is plotted versus an index that indicates the order (in time) in
which the data were collected. This is useful when the data have been col-
lected in a randomized run order, or an order that is not increasing or
decreasing in any of the predictor variables. If the data are increasing or
decreasing with the predictor variables, then the drift in process may not
be separated from the functional relationship between the predictors and
the response. This is why randomization is encouraged in the design of
experiments 1,~.

14.2.2 I n d e p e n d e n t R a n d o m Errors
A lag plot of residuals helps to assess whether the random errors are inde-
pendent from one to the next. If the errors are independent, the estimate
of the error in the standard deviation will be biased, which leads to improper
inferences about the process. The lag plot works by plotting each residual
value versus the value of the successive residual. Due to the way that the
residuals are paired, there will be one less point than most other types of
residual plots.
There will be no pattern or structure in the lag plot if the errors are
independent. The points will appear randomly scattered across the plot,
and if there is a significant dependence between errors, there will be some
sort of deterministic pattern that is evident.
Example 14-1 shows the types of scatterplots used for determining
consistent standard deviation.

EXAMPLE 14-1: Plotting Residuals

A parametric model has been developed to predict the performance of
a PEM fuel cell over a range of operating currents and temperatures. The
parametric equation predicts activation overvoltage from a linear regres-
sion analysis. Table 14-1 shows the run order, experimental temperature,
experimental current, calculated experimental activation overpotential,
and predicted activation overpotential from the model given in Amphlett
et al. 3. (a) Calculate and plot the residuals versus the experimental
factors using scatterplots, (b) create the run order plot, and (c) create the
lag plot.
Using MATLAB to solve:
 Model Validation 395

TABLE 14-1
Experimental and Calculated Results for Example 14-1
Run Temperature Current Experimental Activation Predicted Activation
Order (K) (Amps) Overvoltage (V) Overvoltage (V)
1 358 2.72 -0.2717 -0.2647
2 328 6.66 -0.4017 -0.4014
3 343 6.66 -0.3522 -0.35
4 358 6.66 -0.3038 -0.3025
5 343 6.66 -0.3341 -0.3283
6 343 6.66 -0.3 756 -0.3775
7 328 6.66 -0.3727 -0.3747
8 343 2.72 -0.322 -0.3188
9 343 6.66 -0.3492 -0.35
10 343 6.66 -0.3472 -0.35
11 328 2.72 -0.3141 -0.3193
12 328 6.66 -0.352 -0.3541
13 343 6.66 -0.3482 -0.35
14 358 6.66 -0.3473 -0.353 7
15 358 6.66 -0.325 -0.3252
16 358 6.66 -0.3218 -0.3252
17 343 16.33 -0.4075 -0.4083
18 343 16.33 -0.3902 -0.3868
19 343 6.66 -0.3492 -0.35
20 343 6.66 -0.3 788 -0.3775
21 358 16.33 -0.3834 -0.386
22 343 2.72 -0.2969 -0.292
23 343 6.66 -0.3249 -0.3283
24 343 2.72 -0.2868 -0.292
25 343 16.33 -0.4453 -0.4379
26 343 6.66 -0.3502 -0.35
27 328 6.66 -0.3793 -0.3747
28 343 16.33 -0.4062 -0.4083
396 PEM Fuel Cell Modeling and Simulation Using MATLAB |
 Model Validation 397

X 10 -3
8

0
0 0

-1 "T

'
I 0 'I 0
' 0 '
' ' 0

m
t~
- - - @ - - I '
I
I
ei
-I . . . . . .

'I 0 'I
I I _ ~ _ _ _
"0
|m
t~ 0 I I
I
I
I
I
I I I I
, 0 , , , 0
I I I I
-2 1. . . . ~,q~ -- - -t 1" t-
I t,,D / I I
I I I I
I ~ I I I
I ~ I I I
I I I I
-4 I I I I
0 I
I
~
~
I
I
I
I
I
I
I I I I
I ~ I I I
-6 ......... I........
I
~ -I . . . . . ~.
I
.~
I
.I.
I
L . . . . . . .
I
-I
I
I I I I I I I
I I I I I I 0 I
I I I 0 I I I I
II I ,I ; ,I ; ,
-8 I

325 330 335 340 345 350 355 360

Temperature (K)

FIGURE 14-1. Scatterplot of temperature versus residuals.

Figures 14-1 through 14-4 show the residuals versus the experimental
factors, the run order plot, and the lag plot for Example 14-2.
For Figures 14-1 and 14-2, the range of the residuals in these figures
looks essentially constant across the levels of the predictor variables, which
are temperature, and current. There are points randomly scattered above
and below y = 0 line. This suggests that the standard deviation of the
398 PEM Fuel Cell Modeling and Simulation Using MATLAB |

X 10 "3
........ ;- -; T T
; -;- ;
I I I I I I I
I I I I I I I
I I i
I I I I I
I I ~ I I I I I
. . . . . . . . I. . . . . . . -4 . . . . . . . 4. . . . . . . . 4- . . . . . . . I- . . . . . . . I. . . . . . . . . . . . . .
I I I
I I I
I I I
I I I
I I I
J . . . . . . . 1 . . . . . . . /- . . . . . . .

0
0 o

m
t _ . . . . . .

O
:3
. . . . . . . .]
T . . . . . . . T . . . . . . . . . . . . . . . I. . . . . . . I
iim
I I I I
I I I I
I I I I

IZ, 0 I I I I
I I I I
-2 - D . . . . T . . . . . . . T
. . . . . . . r . . . . . . . I. . . . . . . "Ii
I I
I I I I
I I I I
0 I I I I
O
I I I I
-4 . . . . . . . . I. . . . . . .
I
-I . . . . . . .
I
-I- .
I
. . . . . . t-
I
. . . . . . . I-
I
-I

0 '
I , 0
I
I
I
I
I I
I I I I I
I I _~il'IP~i_ I I I
-6 . . . . . . . . I. . . . . . . -.I . . . . . . 4.. .
I
.. . . . . . .. . . .4-
I
. . . .
--Ii . . . . . . . 41 . . . . . . .
I I I I I I I I
I I I I I I I I
I I I I I I I O I
,
I ', ,, I , I
-8 I

2 4 6 8 10 12 14 16 18
Current(A)

FIGURE 14-2. Scatterplot of current versus residuals.

random error is the same for the responses observed at each temperature
and current. These scatter plots indicate that the parametric model is most
likely a good fit to the experimental data.

14.3 N o r m a l D i s t r i b u t i o n of N o r m a l R a n d o m Errors

14.3.1 Histograms and N o r m a l Probability Plots

In order to determine the normal distribution of normal random errors,
there are two plots that can be used: the histogram and the normal prob-
ability plot. These plots can be used to verify if the random error in the
process has been obtained from a normal distribution. When making a deci-
sion about the model and process, the normality assumption is needed to
obtain more information about the error rates. If the random errors are not
from a normal distribution, incorrect decisions will be made more or less
often than the stated confidence levels indicate.
 Model Validation 399

x 10.3

I
F
I
~I
I I I I
I I I I
I I I I
-I- 4- -4
I
I I R I I
,0
I I
I ~ I
I
I
I
I I I I
_1 J. L J

0 0
()
0
w -0-0 O-0 v

0 0 0
"o 3" "1
..(
we

rr, 0 ' ' 0

I
-2 . . . . . 0 - - - ' l
I

0 0 I
I
-4 4"
I I ~
"4
I
I I I R ~ I
I I I ~ I
I I I

-6 6 I

I
I

I
I

I
I I I
0 I
I
I
I f~
I
I

-8
0 5 10 15 20 25 30
Run Order

FIGURE 14-3. Run order plot for Example 14-1.

The normal probability plot is a plot of the sorted values of the

residuals versus the associated theoretical values from the standard normal
distribution. Unlike most residual scatterplots, a random scatter of points
does not validate that the assumption being checked was met. If the random
errors are normally distributed, the plotted points will lie close to a straight,
diagonal line. If points deviate significantly from the line, the random errors
are probably not randomly distributed, and the data have some outliers in
them.
If the normal probability plot suggests that the errors come from a
non-normal distribution, then a histogram can be used to determine what
the distribution looks like. If the histogram is bell-shaped, the conclusions
of the normal probability plots are correct. It is important to note that
information about the distribution of random errors from the process can
only be obtained if the functional part of the model is correctly specified,
the standard deviation is constant, there is no drift in the process, and the
random errors are not dependent on the r u n 4's.
400 PEM Fuel Cell Modeling and Simulation Using MATLAB |

X 10 "3
. . . . . . . . I. . . . . . .
I -I ....... "I ....... 7 ....... T ....... ,r- ....... I
I I I I I I I
I I I I I I I
I I I llml I I I
I I I ~ I I I
. . . . . . . . I. . . . . . . -1 . . . . . . . "1 . . . . . . . "1' . . . . . . . t . . . . . . . I- . . . . . . . I-- . . . . . . .
I / I l /

0
I

I
I
I
I
I
I
I
I
I
I
b I
I
I
I
I
I
I
I I I I I I I
. . . . . . . . I. . . . . . . 4 . . . . . . . 4 . . . . . . . 4 . . . . . . . 4- . . . . . . . I- . . . . . . . I. . . . . . . .
I I I I I

, , 0 , , 0 ,
I I I I I
I I I I I
0 I I ( ~ I I
_ _ L~Jk _ _ _ I . . . . . . . I . . . . . . . . . . . . . . L . . . . . . .
V
'-0
i

0 0 0
Q
"o
ii u Q
in I

IZ: 0 '
I 0
I I
-2
I v I
I I
,
I
,
I
~
I I
. . . . . . . . l. . . . . . . -I . . . . . . . "I' . . . . . . . "I" . . . . . . . 1"- . . . . . . I- . . . . . . . I. . . . . . .
I I I I 0 I I
I I I I ~ I I
I I I I I ~ I I
I
I
I
I
I
I
I
I l:i '
I 'I
-6 . . . . . . . . 11. . . . . . . . . . . -~--
. . -4I . . . . . . . . . . 4.. . . . . . . . . . I I--
I . . . . . . . Ii. . . . . . . .
I I I I I I I
I I I (~1 I I I
I I I bll I I I

-8 I 1 I , 1 I ,'
-8 -6 -4 -2 0 2 4 6 8
Residuals x lo .3

FIGURE 14-4. Lag plot for Example 14-1.

E X A M P L E 14-2: Creating a H i s t o g r a m and N o r m a l

Probability Plot
Using the residuals from Example 14-1, create a normal probability plot
and histogram to determine if the data are normally distributed. Another
method of presenting the error data is through the use of a box plot. Use
a box plot to view the residuals from Example 14-1.

Using MATLAB to solve:

 Model Validation 401

Figures 14-5 through 14-7 show the normal probability plot, histo-
gram, and box plot for Example 14-2.

14.4 Missing Terms in the Functional Part of the Model

Residual plots are the most valuable tool for assessing whether variables
are missing in the functional part of the model. However, if the results are
402 PEM Fuel Cell Modeling and Simulation Using M A T L A B |

Normal Probability Plot

0.99 ....... -I" I t ....... -I ,
I I I I

0.98 . . . . . . .L . . . . . . . . I. . . . . . . . J. . . . . . . . _I 'L_ . . . . ---: :-*-r

I
I I I I |

0.95
I
I
I
,
I'-
.__,
I

I
.....
=l = S*-I
I
~'_ _1
I I h ~'- I
I I * I
0.90 . . . . . . . . I. . . . . . . . . _ _ ~ _ ~'_'_"]"_ _ _,_ _ _

,,iu,* I

0.75

m
m
I
~
9
|-.= 0.50 . . . . . . . . I. . . . . . . .
I
.... L- . . . . . . . -L

I
(D I
I ,/~ ",
I
I
0.25 . . . . . . . +
I
'1-
I
-
t ........ ', ~ ,
I I I I I I
I I I I I I
I I I I
I "~.,J I I I I I

0.10 J..- - - ~-~'- - -I . . . . . . . . L . . . . . . . ..1

I
I- . . . . . . . -L . . . . . . . .
I
I----~
I
I I I I
I I I I
0.05 _ _ _§ 97 s - ~ "I 1"
I
!
I
"I"
I
I----~
I
I I I I
# I I I I I
0.02 - ~ . . . . . . 4-
I
4- . . . . . . . -4 I- . . . . . . . 4 I----~
I I I I III--

0.01 ~- ,I ', .,I __

-6 -4 -2 0 2 4 6
Quantities from Standard Normal Distribution x 10.3

FIGURE 14-5. Normal probability plot for Example 14-2.

nebulous, it may be helpful to use statistical tests for the hypothesis of the
model. One may wonder if it may be more useful to jump directly to the
statistical tests (since they are more quantitative), however, residual plots
provide the best overall feedback of the model fit. These quantitative tests
are termed "lack-of-fit" tests, and there are many illustrated in any statis-
tics textbook.
The most commonly used strategy is to compare the amount of vari-
ation in the residuals with an estimate of the random variation in the model
is to use an additional data set. If the random variation is similar, then it
can be assumed that no terms are missing from the model. If the random
variation from the model is larger than the random variation from the
independent data set, then terms may be missing or misspecified in the
functional part of the model.
Comparing the variation between experimental and model data sets
is very useful, however, there are many instances where a replicate meas-
urement is not available. If this is the case, the lack-of-fit statistics can be
 Model Validation 403

0.9

0.8

0.7

0.6
,. 4
c
0.5
o
O3
0.4

0.3

0.2

0.1

-0.01 -0.008 -0.006 -0.004 -0.002 0 0.002 0.004 0.006 0.008 0.01
Residuals

FIGURE 14-6. Histogram for Example 14-2.

calculated by partitioning the residual standard deviation into two inde-

pendent estimators of the random variation in the process.
One estimator depends upon the model, and the means of the repli-
cated sets of data (Crm), and the other estimator is a standard deviation of
the variation observed in each set of replicated measurements (rrr). The
squares of these two estimators are often called "mean square for lack-of-
fit." The model estimator can be calculated by6:

O'm = I ni~
I
(nu 1- p) .= n,(yij - Yij )2 (14-2)

where p is the number of unknown parameters in the model, n is the

sample size of the data set used to fit the model, nu is the number of com-
binations of predictor variable levels, and ni is the number of replicated
observations at the ith combination of predictor variable levels.
If the model is a good fit, the value of the function would be a good
estimate of the mean value of response for every combination of predictor
variable values. If the function provides good estimates of the mean response
404 PEM Fuel Cell Modeling and Simulation Using MATLAB |

X 10 .3

if}

i .

>
-2

I
-4
I
-6
I
,,, !
-8 I
1
Column Number

FIGURE 14-7. Box plot for Example 14-2.

at the ith combination, then Crm should be close in value to cr~ and should
also be a good estimate of cr. If the model is missing any important terms,
or any of the terms are correctly specified, then the function will provide
a poor estimate of the mean response for some combination of predictors,
and Crmwill probably be greater than cry.
The model-dependent estimator can be calculated usingT:

I 1 n~n, )~
= 2(Yij--Yij (14-3)
G ( n - n u ) .= i=l

Since Crr depends only on the data and not on the functional part of the
model, this indicates that cr~ will be a good estimator of rx, regardless of
whether the model is a complete description of the process. Typically, if
O'm > Or, then one or more parts of the model may be missing or improperly

specified. Due to random error in the model, sometimes Crmwill be greater

than cr~even when the model is accurate. To ensure that the model hypoth-
 Model Validation 405

esis is not rejected by accident, it is necessary to understand how much

greater rr~ can possibly be. This will ensure that the hypothesis is only
rejected when am is greater than rrr. A ratio that can be used when the model
fits the data is8:

L = (14-4)

The probability of rejecting the hypothesis is controlled by the prob-

ability distribution that describes the behavior of the statistic, L. One
method of defining the cut-off value is using the value of L when it is
greater than the upper-tail cut-off value from the F distribution. This allows
a quantitative method of determining when am is greater than rr~.
The probability specified by the cut-off value from the F distribution
is called the "significance level" of the test. The most commonly used
significance value is a = .05, which means that the hypothesis of an ade-
quate model will only be rejected in 5 % of tests for which the model really
is adequate. The cut-off values can be calculated using the F distribution
described in most statistics textbooks.

14.5 Unnecessary Terms in the Model

Sometimes models fit the data very well, but there are additional unneces-
sary terms. These models are said to "overfit" the data. Since the param-
eters for any unnecessary terms in the model usually have values near zero,
it may seem harmless to leave them in the model. However, if there are
many extra terms in the model, there could be occurrences where the error
from the model may be larger than necessary, and this may affect conclu-
sions drawn from the data.
Overfitting often occurs when developing purely empirical models for
experimental data, with little understanding of the total and random vari-
ation in the data. This happens when regression methods fit the data set
instead of using functions to describe the structure in the data. There are
models that are sometimes are made to fit very complex patterns, but these
may actually be finding structure in the noise if the model is analyzed
carefully.
To determine if a model has too many terms, statistical tests can also
be used. The tests for overfitting of data is one area in which statistical
tests are more effective than residual plots. In this case, individual tests for
each parameter in the model are used rather than a single test. The test
statistics for testing whether or not each parameter is zero is typically based
on T distribution. Each parameter estimated in the model is measured in
terms of how many standard deviations it is from its hypothesized value
of zero. If the parameter's estimated value is close enough to the hypoth-
406 PEM Fuel Cell Modeling and Simulation Using MATLAB |

esized value, then any additional deviation can be attributed to random

error, and the hypothesis that the parameter's true value is not zero is
accepted. However, if the parameter's estimated value is so far away from
the hypothesized value that the deviation cannot be plausibly explained by
random error, the hypothesis that the true value of the parameter is zero
is rejected.
The test statistic for each of these tests is simply the estimated param-
eter value divided by its estimated standard deviation:

T = fl~ (14-5)
aa

Equation 14-5 provides a measure of the distance between the estimated

and hypothesized values of the parameter in standard deviations. Since the
random errors are normally distributed, and the value of the parameter is
zero, the test statistic has a Student's T distribution with n - p degrees of
freedom. Therefore, the cut-off values from the T distribution can be used
to determine the amount of variable that is due to random error. These
tests should each be used with cut-off values with a significance level of
a/2 since these tests are generally used to simultaneously test whether or
not a parameter value is greater than or less than zero. This will ensure
that the hypothesis of each parameter equals zero will be rejected by chance
with probability a.

EXAMPLE 14-3: Creating ANOVA Tables and

Confidence Intervals
Using the residual data given in Example 14-1, create the ANOVA table
for the activation voltage, temperature, and pressure. Find the confi-
dence interval for error mean at a 95 % confidence level.

Using MATLAB to solve:

 Model Validation 407

Source Sum Sq. d.f. Mean Sq. F Prob>F

X1 0_00696 2 0.00348 II_4S O_ 0 0 0 4

X2 0.03096 2 0.01548 50.93 0
Error 0.00699 23 0.0003
Total 0.04324 Z7

FIGURE 14-8. ANOVA for temperature and current for Example 14-3.

Figure 14-8 illustrates the ANOVA table with the temperature and
current as factors.
The confidence limits printed in the MATLAB workspace from
Example 14-3 are as follows:
Sample mean = -3.5714e-006
Confidence interval for sample mean at 95% confidence l e v e l -
-0.0014334 <= Sample mean <= 0.0014262
In Figure 14-8, the model F-values of 50.93 and 11.45 indicate that the
model terms are significant. There is a 0.0% and 0.04% chance that a model
F-Value this large can be due to noise. When the values of Prob > F are less
than 0.05, this typically indicates that the model terms are significant. If
the values are greater than 0.100, this indicates that the model terms are
not significant. If there are many insignificant model terms, model reduc-
tion may improve the model.

Chapter Summary
Fuel cell validation is the most important step in the model-building
process. However, little attention is usually given to this step. A fast
method for analyzing the validity of a model is to look at plots of residuals
versus the experimental factors, run plots, and lag plots. These plots give
a good feel for how accurately a model fits the experimental data, and
how dependable it is. Selecting various statistical techniques, or using a
combination of them, will tell the user if there are any unnecessary por-
tions of the model, or will help determine the amount of noise. Some of
the techniques that are useful in comparing experimental and calculated
408 PEM Fuel Cell Modeling and Simulation Using MATLAB |

results are histograms, normal probability plots, T and F distributions,

analysis of variance, and confidence levels. If residual scatterplots are used
with one or more of the tests listed above, there will be substantial evidence
that a model is a good fit to the experimental data.

Problems
9 Perform a regression analysis for the data in Example 14-1.
9 Determine the T and F distribution tests for the data in Example 14-1.
9 How well do you think that the calculated data in Example 14-1 fit the
experimental data?
9 Perform a lack-of-fit test for the data in Example 14-1.

Endnotes
[1] NIST/SEMATECH e-Handbook of Statistical Methods, http://www.itl.nist
.gov/div898/handbook/. Date created 06/01/2003. Last updated 07/18/2006.
[2] Montgomery, D.C. Design and Analysis of Experiments. 5th ed. 2001. New
York: John Wiley & Sons.
[3] Amphlett, J.C., R.M. Baumert, R.F. Mann, B.A. Peppley, P.R. Roberge, and T.J.
Harris. Performance modeling of the Ballard Marck IV solid polymer electrolyte
fuel cell. J. Electrochem. Soc. Vol. 142, No. 1, January 1995.
[4] NIST/SEMATECH e-Handbook of Statistical Methods.
[5] Montgomery, Design and Analysis of Experiments.
[6] NIST/SEMATECH e-Handbook of Statistical Methods.
[7] Ibid.
[8] Ibid.

Bibliography
Barbir, F. 2005. PEM Fuel Cells: Theory and Practice. Burlington, MA: Elsevier
Academic Press.
Lu, G.Q., and C.Y. Wang. Development of micro direct methanol fuel cells for high
power applications. J. Power Sources. Vol. 144, 2005, pp. 141-145.
O'Hayre, R., S.-W. Cha, W. Colella, and F.B. Prinz. 2006. Fuel Cell Fundamentals.
New York: John Wiley & Sons.
Pekula, N., K. Heller, P.A. Chuang, A. Turhan, M.M. Mench, J.S. Brenzier, and K. Unlu.
Study of water distribution and transport in a polymer electrolyte fuel cell using
neutron imaging. Nucl. Instrum. Methods Phys. Res. A. Vol. 542, 2005, pp. 134-141.
Raposa, G. Performing AC impedance spectroscopy measurements on fuel cells.
Fuel Cell Magazine. February/March 2003.
Smith, M., K. Cooper, D. Johnson, and L. Scribner. Comparison of fuel cell electro-
lyte resistance measurement techniques. Fuel Cell Magazine. April/May 2005.
Turhan, A., K. Heller, J.S. Brenizer, and M.M Mench. Quantification of liquid water
accumulation and distribution in a polymer electrolyte fuel cell using neutron
imaging. J. Power Sources. Vol. 160, 2006, pp. 1195-1203.
The U.S. Fuel Cell Council's Joint Hydrogen Quality Task Force. November 29,
2004. Primer on Fuel Cell Component Testing: Primer for Generating Test Plans.
Document No. USFCC 04-003. Available at: www.usfcc.com.