8

Regression Analysis
In this chapter we seek answers to the question: What equation
fits my experimental data? The general terminology for this type
of activity is regression analysis. The reader may wish to Google to
find how this term came to be used.

We begin by looking at linear functions-functions that can be

recast asy = mx+b. We explore the use of chart trendlines and the
Excel functions SLOPE, INTERCEPT, TREND, FORECAST, and
LINE ST. Then we explore some nonlinear functions, again using
trendlines, and the Excel function LINEST and LOGEST. We
conclude by showing the use of Excel's Data Analysis tools. In
Chapter 12 we will see how Solver can be used for curve-fitting
problems, especially for problems where trendlines and Excel
function cannot be used.

Least-Squares Fitting Gauss is credited with developing the fundamentals of the basis
for least-squares analysis in 1795 at the age of 18. One speaks
about the line of best fit. In this instance, we will restrict
ourselves to linear fits. Let the experimental data consist of
pairs ofx- andy-values. We write the equation of the line of best
fit as J =mx + b, where J (read as ttyhat") is the predicted value.
The vertical displacement between the actualy-value and the
predicted J for a given x is called the residual. The least-
squares criterion requires that we adjust the constants m and
b such that the sum of the squares of the residuals, 'L.(Yi - Jlis
as small as possible. There are formulas for finding these
parameters, but we shall let Excel do the work.

Exercise 1: Trendline, Scenario: A physics student is tasked with finding the thermal
coefficient of resistance of a sample. Her experimental results
SLOPE, and are shown in Figure 8.1.
INTERCEPT
The textbook told her to work with Equation 1, where Ro is the
resistance at ODC, R, is the resistance at temperature t DC and a
is the required coefficient.
R, = RoO + at)
Of course, this can also be written as:
 Regression Analysis 131

which has the form of the well- known equation of a straightline

y= mx + b. The slope will be aR o and the intercept Ro; hence a
can be found from slope/intercept.

Figure 8.1
(a) Open a new workbook and on Sheetl enter the text and data
shown in columns A and B of Figure 8.1.

(b) Construct an XY chart using the first subtype-markers only.

Now we are ready to add the trendline. We could select the chart
and use Chart Tools / Layout / Trendline / Linear Trendline to
quickly add a trendline. This just adds the trendline; we want
more. The same steps, but ending with More Trendline Options,
R-squared gives a measure of will open the required dialog but we shall use the shortcut menu.
the goodness of the fit. In a
sense, it is a measure of how (c) Right click a marker on the chart and select Insert Trendline
much of the variability in the from the shortcut menu to open the Trendline dialog (see
y-values can be accounted for Figure 8.2).
by changes in the x-values.
(d) Clearly, we want a linear trendline, so make that selection.
Here it is 99/0; the rest may
For this demonstration also check the boxes to give us the
be attributed to experimental
equation of the best fit and the R-squared value. Our data
errors. starts at SoC but it will be interesting to have the trendline
start at O°C (then it will hit the y-axis), so in the Backwards
box of the Forecast group, enter a value of 5. Note that the
Trendline Equation box can be dragged around the chart.
132 A Guide to Microsoft Excel 2007 for Scientists and Engineers

We have used the equation y =

mx + b. Be aware that there
are other conventions. In the
UK it is Y = mx + c and m is
called the gradient.
Statisticians, and the authors
of some entries in Excel's
Help, like to use y = a + bx. So
that b may not be the b you
are thinking of when you flip
through a textbook or glance
at Help. Always check what
convention is being used.

Figure 8.2

Remember that we want both slope and intercept in order to

compute the coefficient a. The are three good reasons not to just
copy the trendline values into cells on the worksheet: (i) We may
not use enough significant figures (the Trendline equation can be
formatted to show more or less digits), (ii) should we make a
correction to the data used to make the plot, we may forget to
recopy the Trendline parameters, and (iii) it is an error-prone
operation. Rather, we shall get the parameters of the line of best
fit using functions.

(e) Enter the text in columns D and E.

(f) The formulas we need in column Fare:

F3: =SLOPECB4:Bll, A4:All)
F4: =INTERCEPTCB4:Bll, A4:All)
FS: =RSQCB4:Bll, A4:All)
F6: =F3/F4 to give a
 Regression Analysis 133

Note the syntax for the three regression formulas is

FUNCTION (y-value. x-values). Engineers and scientists
generally use x before y in this context, so be careful.

(g) Save the workbook as Chap8.xlsx.

We see that the trendline values and the function results agree. In
the figure the function values are formatted to show only two
decimal places in keeping with the experimental data, butyou can
look at both the trendline equation and the function results to 15
decimals to compare them.

Exercise 2: In the previous exercise we have fitted data for temperatures in

the range 5 to 40°C in 5-degree intervals. How would we compute
Interpolation and the expected resistance of the sample at (i) 22°C, (ii) DoC, and (iii)
FORECAST 100°C? The first task is called interpolation (we want a value
within the known range); the others are extrapolation (we want
a value outside the known range). It is generally safe to
interpolate. Extrapolation is risky especially when the value lies
far from the know range. Many physical systems appear to behave
in a linear fashion over a short range but actually obey a more
complex law. A gas obeys the Ideal Gas Law at low pressure and
high temperatures but not under other conditions. Of course, the
second task is special, we are looking for the intercept-a value
If you copy a chart from one we already know!
worksheet to another, the
new chart will still be using
the data from the original
worksheet. But you can copy
an entire worksheet by
holding [Ctrl] and dragging the
tab of the source worksheet
to the right (or left) past the
next tab.

Figure 8.3

(a) Our completed worksheet will look like Figure 8.3. We can
copy much of it from Sheetl. Select from Sheetl Al :Fll, copy
it, and paste this to Al of Sheet2. Use Home / Editing / Clear
(looks like an eraser) / Clear All to remove D5:F7.

(b) Enter the text in D6:F6 and the values in D7:D9.

134 A Guide to Microsoft Exce/2007 for Scientists and Engineers

Ifwe know the parameters for the equation of the straight line, we
can find the value of y for any x with y = mx+ b. We do this in
E7:E9. In F7:F9 we use the FORECAST function to show that if this
is our only task we do not need to find the slope and intercept but
can have Excel do that "behind the scenes."

(c) Enter these formulas:

E7: =$F$3*D7+$F$4
You may wish to refer to Help F7: =FORECAST(D7, $B$4:$B$11, $A$4:$A$11).
to understand the syntax of Note that we have used some absolute references as we wish
the FORECAST function. to copy these formulas.

(d) CopyE7:F7 down torow9. Rememberthedouble-clickonthe

fill handle trick?

(e) You may wish to return to Sheetl and extend Trendline to

100°C and compare the chart and the values in E9 and F9.

(f) Save the worksheet.

Exercise 3: The LlNEST In this Exercise we use LINEST rather than SLOPE, INTERCEPT,
and RSQ to get the parameters for a linear fit. LINEST is more
Function flexible and can give more data, as we shall see in this and
subsequent Exercises.

In Figure 8.4 we have the results of a chemistry experiment to

measure the enthalpy of solution (i1H) of J-ascorbic acid at
various mole fractions (X).

Temporarily ignore E4:F8 and enter all text and values as shown
in Figure 8.4 onto Sheet3. Construct the chart.

(a) LINEST is an array function in that it returns more than one

value. With E4:F8 selected, type the formula
=LINEST(B4:B20,A4:A20,TRUE,TRUE) using [Ctrl] + [OShift]
+ [ .-J Ito commit it. Note the braces around the formula when
viewed in the formula bar.

(b) Save the workbook.

 Regression Analysis 135

Figure 8.4

When used with four arguments, LINEST returns the slope,

intercept, and R2 value as well as a number of other statistics that
we address in a later chapter. You will see that arguments one and
two are the y- and x-values as used with SLOPE. When the third
argument is TRUE or omitted, LINEST computes the intercept;
otherwise it sets the interceptto zero. The statistics in rows below
the fit parameter are not returned if the fourth argument is either
set to FALSE or is omitted. A two-argument formula such as
=LINEST(B4:B20, A4:A20) would just give the slope and
intercept.

Exercise 4: Fixed Occasionally, you want to get a fit with a fixed intercept. You may,
for example, want an intercept of zero or of some other value. If
Intercept you look at Figure 8.2, there is a setting Set Intercept where you
can specify the required intercept value. Getting a zero value with
LINEST is simple; you just enter FALSE for argument three.
Specifying a value such as S needs a "workaround." The linesy =
1.5x+5 and z = 1.5x are parallel. For a given x, the y-value equals
thez-value plus S. So if we subtractS fromeachy-value, we get the
z line and its intercept is O. Let's see how we implement that in
Excel.

(a) On Sheet4, we will make a worksheet like Figure 8.5. Enter

the values and text in columns A, B, and C, and the text in
columns E and F.
136 A Guide to Microsoft Exce/2007 for Scientists and Engineers

Figure 8.5
(b) Create the chart. Add the trendlines giving the y-line an
intercept of a and the z-line and intercept of S. Delete the
trendline entries in the legend. Edit the second trendline
equation to show z= ... rather thany= ...

(c) In E6:F6 the LINEST formula for they-line is

=LINEST(B5:B12,A5:A12,FALSE).

(d) In Ell:Fll we wantto usey-values that have S subtracted

from them using =LINEST(C5:C12-5,A5:A12,FALSE).

(e) Butthe a in Fll is misleading. In Ell use

=INDEX(LINEST(C5:C12-5,A5:A12,FALSE),1,1) which
extracts from the LINEST array the item in row 1, column 1.
Since we want the very first element, we could use
LINEST(C5:C12-5,A5) butthe longer formula shows how any
element can be extracted. In F12 enter the value 5 since this
is the fixed value.

(f) Save the workbook.

Exercise 5: A Looking at Figure 8.2, one can see that Excel can do more than just
linear trendlines. How about LINEST? Can that cope with other
Polynomial Fit than linear functions? We will look at a polynomial fit

Scenario: An engineer has measured the temperature of an

extruder machine's die at various settings of the screw revolution
speed. The results are shown in Figure 8.6. He would like an
empirical formula to summarize the data.

(a) On SheetS, enter the text and values shown in rows 1 through
6 of Figure 8.6. If you enter formatted text in two cells
(B6:C6), select the range and drag the fill handle; Excel will
automatically complete the rest of the test.
 Regression Analysis 137

(b) Make an XY chart with only markers. Experiment with

trendlines with polynomials of order 2 (quadratic), 3 (cubic),
and so on. Stop when all the points are fairly close to the line;
in this case a fourth order (quartic) gives a reasonable fit
Remember that with six data points, a fifth-order function
will give a perfect fit There is, however, no justification in
using this, as the leading coefficient is very small.

Figure 8.6

Now we will have LINEST generate the same coefficients. The

At the time of writing, Excel naive way is to insert rows with values of the screw speed raised
2007 had a bug that caused it to the powers of 2,3, and 4. But there is a better way.
to occasionally drop the
leading coefficient in the (c) To hold the coefficients of x 4, x 3, x 2, x, and b (or x''), we need
trendline polynomial equation. five cells. Select B7:F7 and enter the array formula
Microsoft expects this wiII be =LINEST(B4:G4,B3:G3"'{1;2;3;4}). Commit it with
fixed in service pack 2. Make [Ctrll+[ -0 Shift1+[ -.J I.
sure you have downloaded it.
(d) Save the worksheet

NOTE: In the formula used in step (c) we have an array of

constants: {1;2;3;4}, the elements of which are separated by semi-
colons. This is because our x-values are in columns. Had we made
a vertical table, the array would have used commas to separate
elements: {1,2,3,4}.

Exercise 6: A A simple model for the growth of bacteria predicts that if the
initial population is No, the population N, at time twill be given by
Logarithmic Fit the following equation, in which B is the reproduction rate.
(LOG EST)
N, = No exp(Bt)
138 A Guide to Microsoft Exce/2007 for Scientists and Engineers

We can linearize (which means to give it the formy = mx + b) by

taking natural logs on both sides:

Before computers, the normal practice was to convert equations

to linear form since fitting to a straight line is relatively simple.

Figure 8.7

For this demonstration, we will find the fitting parameters of some

exponential data both with and without linearization.

Figure 8.7 shows the population of a bacteria colony at various

times. We wish to estimate No and B. For a change of pace, the
reader is asked to develop this worksheet without detailed
instructions. Row 5 has the linearized data, which we fit with
LINEST in row 9. Row 13 also uses LINEST, but we do the
logarithm transformation within the formula. Also in row 9 we use
LOGEST to do the fit in a most straightforward way. We also check
out functions by using a chart with two trendlines whose labels we
have edited to remove the default x and y text The formulas
required are:

BS: =LN(B4)
A9:B9: =LINEST(B5:F5,B3: F3)
C9: =EXP(B9)
A13:B13: =LINEST(LN(B4:F4),B3: F3)
C13: =EXP(B13)
E9:F9: =LOGEST(B4: F4,B3: F3)
G9: =LN(E9)

The LOGEST function (which can also return fitting statistics) fits
 Regression Analysis 139

our data to y = km" while the exponential trendline fits it to Y =

kExp(ax). It is left to the reader to show that a = Ln(m).

As expected, the results of the three methods are in agreement.

Remember to save the workbook when done.

The TREND and The parameters in the LINEST and LOGEST arrays can, of course,
be used to find the values in the trendlines or to interpolate or
GROWTH Functions extrapolate. However, this is more readily done with TREND (for
LINEST fits) and GROWTH (for LOGEST fits). The syntaxes for
these functions are:
TREND (known_ y' s,known_x' s,new_x's,const)
GROWTH (knowny' s,known_x's,new_x's,const)

Figure 8.8

Figure 8.8 shows the use ofTREND to compute the fitted values in
the problem posed in Exercise 6 and the use of GROWTH for the
problem in Exercise 7. The formulas, which refer to different
worksheets, are:

P3:U3: =TREND(P2:U2,Pl:Ul) (in the top diagram)

P3:T3 =GROWTH(P2:T2,Pl:Tl) (the lower diagram)

The two, of course, must be entered as array formulas.

Residuals Recall that we have defined residuals as the difference between

the actual and the predicted values in a curve-fitting problem. If
the residuals are the result of normal experimental errors, we
Computing and plotting
would expect them to be distributed randomly above and below
residuals can sometimes the x = a line. If, on the other hand, the residuals display an
reveal otherwise hidden observable trend, then one should question the fit.
information.
140 A Guide to Microsoft Excel 2007 for Scientists and Engineers

In the example shown in Figure 8.9 (albeit a very contrived set of

data), a linear fitseems very appropriate, but the residuals appear
to follow a parabolic rule. Two trendlines were added to the
original chart; both give R2 as 1, but the results from REQ (linear
fit) and LINEST (using =INDEX(LINEST(y, x"{1,2}, TRUE,
TRUE),3,1) for a quadratic fit) show very slightly different values.
The meaningfulness of the x2 coefficient (called by some a lurking
variable) will depend very much on the circumstance of the
experiment.

Figure 8.9

Exercise 7: Slope and In this Exercise we see how to compute the slope of a polynomial
and how to display a tangent line on a chart Suppose we find the
Tangent slope m ata pointx!J'Yo' then the tangent is the line that obeys Yo =
mxo + b. Hence, b = Yo - mxo and we can find a second point on the
tangent usingy = m[x-xoJ+Yo'

The next table gives the formulas needed to compute

approximations to the first and second derivatives of tabulated
data. The central formula is generally more accurate but requires
that we have a point each side of the point of interest.

Order Forward Backward Central

First dy Yl - Yo dy Yo - Y-l dy Yl - Y-l
= = =
dx h dx h dx 2h
Second d 2y Y2 -2Yl + Yo d 2y Yo - 2Y_l + Y- 2 d 2y Yl -2yo + Y-l
2= = 2=
dx h2 dx 2 h2 dx h2
 Regression Analysis 141

Figure 8.10

Figure 8.11

(a) On Sheet8 of Chap8.xlsx, enter all the text shown in Figure

Data Validation is the best 8.10. Enter the values shown in A3:B13.
way to prevent users from
entering inappropriate data in (b) The formula to compute the slope using the Central
cells. We have barely touched Difference method in CS is =(B6-B4)/(2*(A5-A4)), and this
on all it can do. The reader is must be copied down to row 12.
encouraged to experiment.
(c) Enter a number in the range 2 to 9 in G3. We will use this as
an index to the x-values. With the central difference method
we cannot use points 1 or 10, so we need to prevent users
entering invalid data here. Select G3 and use Data / Data Tools
/ Data Validation to open and complete the dialog box shown
in Figure 8.11.

(d) Cells El0 and Fl.O hold our xo,Yo data pair; this is the painton
the curve where we want the tangent Cells Ell and Fll hold
the second painton the tangent. In Gl0:Hl0 we compute the
slope and intercept values of the tangent line.
El0: =INDEX($A$4:$A$13,$G$3)
142 A Guide to Microsoft Excel 2007 for Scientists and Engineers

rro. =INDEX($B$4:$B$13,$G$3)
Gl0: =INDEX($C$4:$C$13,$G$3)
Hl0: =FlO-ElO*GlO
Ell: =INDEX($A$4:$A$13,$G$3+1)
Fll: =GlO*Ell+HlO

(e) Make an XY chart with the data in A3:B13. Using the

Copy-&-Paste Special method from Exercise 11 in Chapter 7,
add the points defined by El0:Fll as a second series. We
want only the first point to be visible, so click on the second
point twice and format it to have no marker line or fill.

(f) Add a linear trendline to the new series with appropriate

forward and backward projections to make the tangent line
visible.

(g) Save the workbook.

Exercise 8: The Excel has a feature called the Analysis Toolpak, which has a
variety of tools that enable the user to generate results without
Analysis Toolpak using formulas and formatting. In this exercise we will see the use
of the Regression Tool by repeating the problem set out in
Exercise 3 for comparison purposes.

(a) Copy Al:B20 from Sheet3 to Al in Sheet9.

(b) Use the command Data / Analysis / Data Analysis and from the
resulting dialog select Regression, which opens the dialog
shown in Figure 8.12.

(c) The x range is B3:B20, and they range is A3:A20. Ensure you
have checked the Labels box. A suitable output range for our
purposes is ES, but you will note that you could output to a
new worksheet or workbook. Check the box Line Fit Plots to
generate a chart. Click the OK button.

If you compare the results in F21 and F22, shown in Figure 8.13,
you will see that the slope and intercept are the same as were
generated with LINEST in Exercise 3. You will also see that the
statistics are in agreement. None of this is surprising as the Tool
uses the LINEST function.
 Regression Analysis 143

Figure 8.12

Figure 8.13

There are two major drawbacks to using this Tool. The user has no
control over the positioning of the various resulting values and,
like all Data Analysis Tools, the results are static. This means that
if you make a change in the input data you must remember to
rerun the Tool.

Artthe time ofwriting, the Regression Tool has a small bug in that
it produces a Column chart when an XY chart is required. The
user should right click and change the chart type.
144 A Guide to Microsoft Exce/2007 for Scientists and Engineers

Problems 1. *What function best fits the datal in the following table?

Galaxy Distance Radial Velocity

(Megaparsec1 (km/s)
Virgo 15 1200
lwww.astro.indiana.edu/
Perseus 71 5400
catyp/activities/hubble.doc Coma 83 6600
Hercules 150 10500
Ursa Major I 313 15600
Leo 337 19500
Corona Borealis 347 21600
Gemini 402 23400
Bootes 650 39300
Ursa Major II 653 40200
Hydra 831 60600

2. *A chemical engineer is studying the rate at which compound

X reacts under certain conditions. The following table gives
the percentages of X remaining after measured times. Fit the
data to (1-X] = Exp(-kt) to determine kusing (a) a graphical
method, and (b) a single cell with a LINEST or LOGEST
formula.

t (sec) X (percent)
200 0.18
400 0.29
600 0.42
800 0.51

3. *In Problem 3 of Chapter 4 we used numerical differentiation

formulas to find dijdt for same tabulated data. Another
approach is to use LINESTto get the polynomial coefficients;
then from f(x) we can find the coefficients off'(x). Compare
the results from each method.

4. The solution to Problem 9 of Chapter 2 consisted of a table

giving the amount of solute rna remaining in the water after
extraction n.
(i) Plotthis data and add an exponential trendline in the form
rna = 5exp(-An).
(ii) Fit the data using the LOGEST function to get parameter
Band 5.
 Regression Analysis 145

(iii) Clearly, 5 results from the fact we started with 5g. How
are A and B related to each other and to the data in the
experiment?
(iv) Do a mathematical analysis of the experiment to explain
the exponential fit

5. Fit the data below" to the equation N = apk by: (i) making a
2 W. L Friend and A. B. plot and adding a power trendline; (ii) plotting Ln(N) against
Metzner, American Institute Ln(P) and adding a linear trendline; and using LINE ST.
of Chemical Engineering Ensure you understand the relationship between the various
Journal 4, 393, 1958. fitting parameters. Note that you can plot P vs. N and give
both axes a logarithmic scale to get a straight line, but this
does not help with regression analysis.

P N P N P N
0.46 24.80 10.00 84.50 55.00 195.00
0.53 26.50 17.70 115.00 58.50 193.00
0.63 28.50 18.60 115.00 70.30 189.00
0.74 30.00 25.30 150.00 93.00 245.00
3.00 58.40 31.60 127.00 95.00 245.00
4.20 60.30 32.00 140.00 185.00 315.00
5.00 70.70 37.00 165.00 340.00 380.00
5.60 69.00 41.00 170.00 590.00 480.00

6. A chemist measured the partial pressure of a decomposing

gas at various times; see the following table. Make an
appropriate chart to show that this data follows the
equation Ln(po I p) = kt where Po is the pressure attime t =
O. What value of k is reported by the trendline. Can you get
the same result with a LINEST formula?

7. *The heat of vaporization of a liquid (L1HJ may be found by

measuring the liquid's vapor pressure at various
temperatures and applying the Clausius-Clapeyron equation,
which chemists like to write as:

In[P2) = f.J{v [~_~)

r. R 1; 1;
146 A Guide to Microsoft Excel 2007 for Scientists and Engineers

3 H. F. Stimson, Journal of A plot of liT against liP where T is measured in Kelvin and
Research of the National P in torr will give a straight line with a slope -!JHJR where R
Bureau of Standards, 73A, has the value 8.3145 J mole" K 1. From the following data"
493,1969. find ~Hv for water.

8. Make a plot of the following data and add two trendlines, one
quadratic and the other cubic. Format the cubic trendline as
a dotted line. Hint: Remember the Selection group in Chart
Tools / Format.

9. *A sociological study in 1976 tested the hypothesis that the

larger the city the more rushed were the inhabitants. Google
with Pace of Life for more details. The table that follows lists
some results. Which model best fits the data; (i) a power
model V=kpG, or (ii) a logarithmic model V= mLn(P)+c?

Location Population V (ft/sec)

Brno, Czechoslovakia 341,948 4.81
Prague, Czechoslovakia 1,092,759 5.88
Corte, Corsica 5,491 3.31
Bastia, France 49,375 4.90
Munich, Germany 1,340,000 5.62
Psychro,Crete 365 2.76
Itea, Greece 2,500 2.27
Iraklion, Greece 78,200 3.85
Athens, Greece 867,023 5.21
Safed, Israel 14,000 3.70
Dimona, Israel 23,700 3.27
Netanya, Israel 70,700 4.31
Jerusalem, Israel 304,500 4.42
New Haven, U.S.A. 138,000 4.39
Brooklyn, U.S.A. 2,602,000 5.05
Applied
4 F. E. Croxton et 01.,
10. *The following data" records observation of the number of
General Statistics, Prentice-
chirps per 20 seconds of crickets as a function of
Hall, Englewood Cliffs, N J, temperature. What relationship do you find? A search of the
1967 (page 390). Internet found two comments that if you count the chirps in
 Regression Analysis 147

15 seconds and add 40 (one said 37) you get a good estimate
of the temperature in of. Does this data agree with these
comments?

T COF) 46 49 51 52 54 56 57 58 59 60
chirps 40 50 55 63 72 70 77 73 90 93
T COF) 61 62 63 64 66 67 68 71 71 72
chirps 96 88 99 110 113 120 127 137 137 132

11. The table that follows shows the results of an enzyme kinetics
experiment. The quantity V is the velocity of the reaction,
while [5] is the concentration of the substrate S. Ideally, this
data should be fitted to the Michaelis- Menten equation to find
K. Traditionally, biochemists linearize the M-M equation to
give the Lineweaver-Burke equation and then plot l/V
against 1/[5]. What value of K is obtained using a trendline
and using LINEST? We revisitthis problem in Chapter 12 and
use Solver to make a direct fit.

Michaelis-Menten Eqn: v = VmaJS]

[S]+K
1 K 1 1
Lineweaver-Burke Eqn: -=----+--
V Vmax [S] Vmax

[S] (mM) V (mM/sec)

8.33 3.62E-06
5.55 3.39E-06
2.77 2.7SE-06
1.38 1.99E-06
0.83 1.49E-06