Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Linear Fit
evaporation coefficient (mm2/s) = 0.0692424 + 0.0038288*air velocity (cm/s)
Summary of Fit
RSquare
RSquare Adj
Root Mean Square Error
Mean of Response
Observations (or Sum Wgts)
0.905317
0.893481
0.159052
0.835
10
Analysis of Variance
Source
DF Sum of Squares
Model
1
1.9350694
Error
8
0.2023806
C. Total
9
2.1374500
Parameter Estimates
Term
Intercept
air velocity (cm/s)
Estimate
0.0692424
0.0038288
Mean Square
1.93507
0.02530
Std Error
0.100974
0.000438
F Ratio
76.4923
Prob > F
<.0001*
t Ratio
0.69
8.75
Prob>|t|
0.5123
<.0001*
Parameterr estimatess:
For a signiificant relaationship w
we test
H0 : 1 0
H1 : 1 0
and the reesult of thiss test is give
en by the p
pvalue <0.0
0001 in the
e Prob>|t| ccolumn th
his test is
done heree at the defaault level off 0.05
The gradieent is signifficant i.e. non zero we have aa significant relationshhip betwee
en X and Y
Four requ
uirements for fitting
g a regresssion modell:
Testing th
he residua
als
1. zero m
mean
2. indepeendent
3. normaally distribu
uted
4. equal vvariance
Constructt the resid
duals which
h are storeed in data file
Go to red trian
ngle next to
o Linear Fitt under scaatterplot
In tthe popup
p window ch
hoose Savee Predicted
d and Save R
Residuals
Thee data tablee has been augmented
d:
new column
n: Analyze Distribution (i.e., univariate aanalysis like
e in
Anaalyse this n
Desscriptive Sttats chapte
er 2)
uantile plott
Normal qu
In fact theere is a featu
ure which g
gives an ou
utput analo
ogous to the
e MINITAB output in tthe text on
page 357 ((exercise 1
11.24)
Graph the
e residualss
Go to red trian
ngle next to
o Linear Fitt under scaatterplot
In tthe popup
p window ch
hoose Plot Residuals
Thee followingg Diagnostics Plots aree produced
d:
Ressidual Norm
mal Quantile Plot as above indiicates nonnormality
Ressidual by R
Row Plot v
visualises rresiduals in
n row orderr used forr assessing
ind
dependencee (randomlly scattered
d around th
he zero line
e) in this ccase there seems to
be no pattern
n between tthe dots so independeence is acce
epted anoother view o
of this is
ollows usin
ng the Graph menu option where
e we can jooin the dotss
obttained as fo
Gra
aph Over lay Plot red triangle
r
e Y optio ns conne
ect points
graphs the ressiduals in o
order of row
w entry
Ressiduals by P
Predicted P
Plot assessees whetherr we have e
equal (consstant) varia
ance which
app
pears to bee true.
Acttual by Predicted Plott:
Thiis is similar to the regreession plot eexcept that it contains more
m inform
mation. Thee degree to
whiich the dotss fit along th
he solid linee is another indication of
o the goodnness of fit of
o the
model the acctual respon
nse values arre close to the
t predicted response vvalues.
n to Fit Y by
b X and wiill try a quaadratic mo
odel.
We return
Polynomial Fiit Degree=2
evaporation co
oefficient (mm2//s) = 0.0308049 + 0.0038288*a
air velocity (cm/s
s) + 2.9119e-6*(air velocity (cm
m/s)-200)^2
Fit
Summary of F
RSquare
RSquare Adj
Root Mean Square Error
onse
Mean of Respo
Observations ((or Sum Wgts)
Analysis of Va
ariance
Source
DF
2
Model
7
Error
9
C. Total
Parameter Estimates
Term
Intercept
m/s)
air velocity (cm
(air velocity (cm
m/s)-200)^2
0 .910679
0 .885159
0 .165149
0.835
10
Sum off Squares
1.9465308
1
0.1909192
0
2.1374500
2
Mean Square
0.973265
0.027274
Std Error
0.12045
0.000455
4.492e-6
E
Estimate
0.0
0308049
0.0
0038288
2..9119e-6
F Ratio
35.6845
Prob > F
0.0002*
o
t Ratio
0.266
8.422
0.655
Residual by
y Predicted Plot
Acttual by Pred
dicted Plot
Residual by
y Row Plot
Res
sidual Norm
mal Quantile Plot
Prob>|t|
0.8
8055
<.0
0001*
0.5
5375
he data welll (visually))
Thee curve seeems to fit th
Nottice the R2 value has rrisen margiinally to 91
1.1% a slig
ghtly betterr fit but o
only
maarginal.
Thee residual p
plots for co
onditions foor regressio
on fitting are better iin that the residuals
aree clearly clo
oser to norm
mality
How
wever, the coefficientt for the qu
uadratic terrm in the ne
ew formulaa is actually
y not
significant (pvalue of 0..5375)
of lack of siignificancee of the quadratic coeffficient, we will
So, on balancee, because o
ugh its resiiduals were
e not norm
mal, they we
ere
acccept the linear model even thou
sym
mmetrical. And the da
ata set is on
nly of size 1
10
Exercise 7
7.32
Consider tthe scatterp
plot
There is no way a lin
ne fit could be justified
d the curv
vature is so pronounceed.
2
(If you waant to do the analysis, it gives an R of 18.4%
%
R2 is 70.5% and the Fratio is significant pvalue < 0
0.05 all off which inddicate this q
quadratic
is a good m
model fit.
Interesting is that the single power term ffor location
n has a coeffficient whiich is not siignificant
best modell probably h
has a form of Y = b0 + b2X2
at the 5% level: pvalue of 0.0749. So the b
Fit Y by X does not haave the abiility to deall with this ssubtlety, bu
ut the Fit M
Model platfo
orm, which
h
will be meet later, doees.
Condition
ns for Fitting a Regre
ession Mod
del
Residual by Predicte
ed Plot
Shows a reasonable
e fit of the
actual daata points w
with
predictedd around th
he solid
line.
One posssible outlier.
Reasonabbly random
mly
scatteredd around th
he zero
line no clear patte
ern we
can assum
me independence.
Residual Normal Qua
antile Plot
Reasonabble indication of
normalitty clearly there is
one outliier.
Moments report show
ws mean of residuals = 0
hich is Ana
alyze Dis
stribution Residuals
s)
(See reporrt below wh