You are on page 1of 11

Ramin Shamshiri STA6167, HW#1, Jan.24.

2008 Page 1

STA 6167, Section 1648, Fall 2007
Project #1
Due Thursday 1/24/08






RAMIN SHAMSHIRI
UFID#: 9021-3353































Ramin Shamshiri STA6167, HW#1, Jan.24.2008 Page 2

Part 1: Experiments on oysters

A study was conducted to measure the effects of several factors on the growth of oyster shells
the variables under study are:

Response: growth (mm in shell width)
Predictor 1: Food Concentration
Predictor 2: Flow speed

The authors fit 2 models:

Complete Model:
( ) ( ) Flow x Conc Flow Flow) ( conc) Food ( E(Width)
* 2 * * * *
4 3 2 1 0
| | | | | + + + + =

Source df SS
Regression 4 101.68
Residual 15 37.35
Total 19 139.03

Parameter Estimate Standard error
Intercept 0.96 n/a
Food conc 2.52 0.785
Flow 1.72 0.595
Flow
2
-0.10 0.064
Conc x flow -0.19 0.204

Reduced Model
Flow) ( conc) Food ( E(Width)
* *
2
* * * *
1 0
| | | + + =

Source df SS
Regression 2 93.33
Residual 17 45.70
Total 19 139.03

Parameter Estimate Standard error
Intercept 2.41 n/a
Food conc 1.98 0.549
Flow 0.67 0.144




Ramin Shamshiri STA6167, HW#1, Jan.24.2008 Page 3

- For each model, give the fitted value (prediction) when food conc=8 and flow=2.5
- Give the coefficient of determination for each model
- For the complete model, test H
0
:|
1
=|
2
=|
3
=|
4
=0
- for the complete model, Test H
0
: |
i
=0 (i=1,2,3,4)
- For the complete model, test H
0
:|
3
=|
4
=0

Solution:

For the complete model, we have:

Parameter Estimate Standard error t-stat
Intercept 0.96 n/a -
Food conc 2.52 0.785 2.52/0.785=3.21
Flow 1.72 0.595 1.72/0.595=2.89
Flow
2
-0.10 0.064 -0.10/0.064=-1.562
Conc x flow -0.19 0.204 -0.19/0.204=-0.931


Source df SS MS F
Regression p=4 SSR= 101.68 MSR=SSR/P =25.42 MSR/MSE=10.2
Residual n-(p+1)=15 SSE= 37.35 MSE=SSE/n-(p+1) =2.49
Total n-1=19 TSS= 139.03


( ) ( ) Flow x Conc Flow Flow) ( conc) Food ( E(Width)
* 2 * * * *
4 3 2 1 0
| | | | | + + + + =
|
0
=0.96
|
1
=2.52
|
2
=1.72
|
3
=-0.10
|
4
=-0.19
=> ( ) ( ) Flow x Conc 19 . 0 Flow 10 . 0 Flow) ( 72 . 1 conc) Food ( 52 . 2 96 . 0 E(Width)
2
+ + =
When conc=8 and flow=2.5 then we will have:
( ) ( ) 2.5 8 19 . 0 5 . 2 10 . 0 ) 5 . 2 ( 72 . 1 8) ( 52 . 2 96 . 0 E(Width)
2
+ + =
Width=20.995

Coefficient of determination R
2
for the complete model is:

2
=

=
101.68
139.03
= 0.731




Ramin Shamshiri STA6167, HW#1, Jan.24.2008 Page 4

For the reduced model we have:

Parameter Estimate Standard error t-stat p-value
Intercept 2.41 n/a
Food conc 1.98 0.549 1.98/0.549=3.6
Flow 0.67 0.144 0.67/0.144=4.65

Source df SS MS F
Regression P=2 SSR= 93.33 MSR=SSR/P =46.66 MSR/MSE=17.35
Residual n-(p+1)=17 SSE= 45.70 MSE=SSE/n-
(p+1)=2.68

Total n-1= 19 TSS= 139.03

Flow) ( conc) Food ( E(Width)
* *
2
* * * *
1 0
| | | + + =
|
0
=2.41
|
1
=1.98
|
2
=0.67
=> Flow) ( 67 . 0 conc) Food ( 98 . 1 41 . 2 E(Width) + + =
when food conc=8 and flow=2.5
) 5 . 2 ( 67 . 0 8) ( 98 . 1 41 . 2 E(Width) + + =
Width= 19.925

Coefficient of determination R
2
for the complete model is:

2
=

=
93.33
139.03
= 0.671


Ramin Shamshiri STA6167, HW#1, Jan.24.2008 Page 5

For the complete model, test H
0
:|
1
=|
2
=|
3
=|
4
=0 versus H
A
: Not all |
1
=0, we have:

= 10.2
R.R.: F
obs
F
, P, n-(p+1)

P-value: P (FF
obs
)
= 0.05 p=4 n-(p+1)=15
From appendix table, F
, P, n-(p+1)
= F
0.05, 4, 15
=3.056
Since Fobs> F
0.05, 4, 15
We will have a P-value less than 0.05, thus rejecting the null hypothesis.
We conclude that there is not enough evidence that |
1
=|
2
=|
3
=|
4
=0

For the complete model, Test H
0
: |
i
=0 (i=1,2,3,4)
T.S.:


R.R.: |t
obs
|t
/2, n-(p+1)

P-value: 2p (t|t
obs
|)
= 0.05
n-(p+1)=15

We will have four sets of test as below:

H
0
: |
1
=0
H
A
: |
1
0
T.S.:

=

1

1
= 3.21
t
/2, n-(p+1)
= t
0.025, 15
=2.131<3.21
Since |t
obs
|t
/2, n-(p+1)
it means that the corresponding P-value of this test is less than the
significant level, 0.025, thus we reject the null hypothesis and conclude that there is not enough
evidence that |
1
=0

H
0
: |
2
=0
H
A
: |
2
0
T.S.:

=

2

2
= 2.89
|2.89|<3.21 => Reject H
0


H
0
: |
3
=0
H
A
: |
3
0
T.S.:

=

3

3
= 1.562<3.21
|-1.562|<3.21 => Reject H
0


H
0
: |
4
=0
H
A
: |
4
0
T.S.:

=

4

4
= 0.931<3.21
|-0.931|<3.21 => Reject H
0

Ramin Shamshiri STA6167, HW#1, Jan.24.2008 Page 6

For the complete model, test H
0
:|
3
=|
4
=0
Our alternative hypothesis will be:
H
A
:|
3
AND/OR |
4
0
Complete Model:
( ) ( ) Flow x Conc Flow Flow) ( conc) Food ( E(Width)
* 2 * * * *
4 3 2 1 0
| | | | | + + + + =

Reduced Model:
Flow) ( conc) Food ( E(Width)
* *
2
* * * *
1 0
| | | + + =

SSR
c
=101.68
SSE
c
=37.35
SSR
r
=93.33
n=20 p=4 g= 2

TS:

=
(

)/()

/[(+1)]
=
(101.6893.33)/(42)
37.35/20(4+1)
=
4.175
2.49
= 1.676
R.R.: F
obs
F
0.05, 4, 15
= 3.056
P-value: P (FF
obs
)
Decision: DO NOT reject the Null hypothesis. There are evidence to show that |
3
=|
4
=0

















Ramin Shamshiri STA6167, HW#1, Jan.24.2008 Page 7

Part 2: Ballistic Tests on various layers of cloth panels
A study was conducted to measure the effect of the number of layers of panels in cloth fabric
and the velocity needed for half of a ballistic discharge to penetrate the fabric, separately for 3
types of bullets (Rounded, sharp, fsp).
The model fit is:
( ) ( ) ( ) FSP x Layers Sharp x Layers F Sharp Layers
5 4 3 2 1 0
2
50
| | | | | | + + + + + = SP V E
- Give the fitted equation
- Test whether regressions differ among bullet types (H
0
: |
2
= |
3
=|
4
= |
5
=0)
- Test whether the layers effect is the same for each bullet type (H
0
: |
4
= |
5
=0)
- Obtain the influence statistics.
- Do any observations appear to have undue influence on regression coefficients (Dfbetas), own
fitted values (dffits), or appear to be outliers (studentized residuals). Give the rules for extreme
cases for each measure

Solution:
From the SAS output, we have:

Number of Observations Read 25
Number of Observations Used 25

Standard
Parameter Estimate Error t Value Pr > |t|

Intercept 3.643316892 0.83614504 4.36 0.0003
layers 0.854688870 0.03371662 25.35 <.0001
sharp 0.768657468 1.18248765 0.65 0.5235
fsp 0.498872798 1.13634386 0.44 0.6656
layers*sharp 0.149624096 0.04768250 3.14 0.0054
layers*fsp 0.136967041 0.04670056 2.93 0.0085

The fitted equation will be:

( ) ( ) ( ) FSP x Layers Sharp x Layers F Sharp Layers
5 4 3 2 1 0
2
50
| | | | | | + + + + + = SP V E
( ) ( ) ( ) FSP x Layers 136 . 0 Sharp x Layers 149 . 0 F 498 . 0 Sharp 768 . 0 Layers 854 . 0 643 . 3
2
50
+ + + + + = SP V E





Test whether regressions differ among bullet types (H
0
: |
2
= |
3
=|
4
= |
5
=0) versus H
A
: Not all
|i=0, From the SAS output we have:
Ramin Shamshiri STA6167, HW#1, Jan.24.2008 Page 8


The GLM Procedure

Dependent Variable: v502
Sum of
Source DF Squares Mean Square F Value Pr > F
Model 5 3737.170151 747.434030 502.86 <.0001
Error 19 28.241217 1.486380
Corrected Total 24 3765.411368

R-Square Coeff Var Root MSE v502 Mean
0.992500 5.094881 1.219172 23.92935

= 502.86
R.R.: F
obs
F
, P, n-(p+1)

P-value: P (FF
obs
)
= 0.05 p=3 n-(p+1)=25-(3+1)=21
From appendix table, F
, P, n-(p+1)
= F
0.05, 3, 21
=3.072
Since Fobs>> F
0.05, 3, 21
We will have a P-value less than 0.05, thus rejecting the null hypothesis.
We conclude that there is not enough evidence that |
1
=|
2
=|
3
=|
4
=0


The GLM Procedure
Dependent Variable: v502
Sum of
Source DF Squares Mean Square F Value Pr > F
Model 3 3718.973407 1239.657802 560.59 <.0001
Error 21 46.437962 2.211332
Corrected Total 24 3765.411368


R-Square Coeff Var Root MSE v502 Mean
0.987667 6.214355 1.487055 23.92935


Standard
Parameter Estimate Error t Value Pr > |t|

Intercept 1.587992667 0.72365169 2.19 0.0396
layers 0.951410010 0.02339989 40.66 <.0001
sharp 3.948169500 0.74352732 5.31 <.0001
fsp 3.368058580 0.72297879 4.66 0.0001

Ramin Shamshiri STA6167, HW#1, Jan.24.2008 Page 9

The GLM Procedure

Dependent Variable: v502

Sum of
Source DF Squares Mean Square F Value Pr > F
Model 1 3645.445455 3645.445455 698.91 <.0001
Error 23 119.965913 5.215909
Corrected Total 24 3765.411368


R-Square Coeff Var Root MSE v502 Mean
0.968140 9.544081 2.283837 23.92935


Standard
Parameter Estimate Error t Value Pr > |t|

Intercept 4.106509697 0.87798784 4.68 0.0001
layers 0.949369698 0.03591080 26.44 <.0001


Obtain the influence statistics.

Parameter Estimates

Parameter Standard Variance
Variable DF Estimate Error t Value Pr > |t| Inflation

Intercept 1 1.58799 0.72365 2.19 0.0396 0
layers 1 0.95141 0.02340 40.66 <.0001 1.00151
sharp 1 3.94817 0.74353 5.31 <.0001 1.36000
fsp 1 3.36806 0.72298 4.66 0.0001 1.36151


The REG Procedure
Model: MODEL1
Dependent Variable: v502

Number of Observations Read 25
Number of Observations Used 25


Analysis of Variance

Sum of Mean
Source DF Squares Square F Value Pr > F

Model 3 3718.97341 1239.65780 560.59 <.0001
Error 21 46.43796 2.21133
Corrected Total 24 3765.41137


Root MSE 1.48705 R-Square 0.9877
Dependent Mean 23.92935 Adj R-Sq 0.9859
Coeff Var 6.21435

Do any observations appear to have undue influence on regression coefficients (Dfbetas), own fitted
values (dffits), or appear to be outliers (studentized residuals). Give the rules for extreme cases for each
measure?
Model: MODEL1
Dependent Variable: v502
Output Statistics

Dependent Predicted Std Error Std Error Student
Obs Variable Value Mean Predict Residual Residual Residual
Ramin Shamshiri STA6167, HW#1, Jan.24.2008 Page 10


1 4.5412 3.4908 0.6923 1.0503 1.316 0.798
2 8.7261 7.2965 0.6354 1.4297 1.344 1.063
3 16.8757 13.9563 0.5601 2.9193 1.378 2.119
4 17.7915 19.6648 0.5284 -1.8733 1.390 -1.348
5 27.0400 25.3732 0.5330 1.6668 1.388 1.201
6 28.6118 30.1303 0.5642 -1.5185 1.376 -1.104
7 32.6155 34.8873 0.6164 -2.2718 1.353 -1.679
8 38.2419 39.6444 0.6848 -1.4025 1.320 -1.063
9 7.0809 7.4390 0.6923 -0.3581 1.316 -0.272
10 10.8175 11.2446 0.6354 -0.4271 1.344 -0.318
11 16.5080 17.9045 0.5601 -1.3965 1.378 -1.014
12 22.0618 23.6130 0.5284 -1.5511 1.390 -1.116
13 30.3050 29.3214 0.5330 0.9836 1.388 0.709
14 35.7245 34.0785 0.5642 1.6461 1.376 1.196
15 38.4400 38.8355 0.6164 -0.3955 1.353 -0.292
16 45.0912 43.5926 0.6848 1.4987 1.320 1.135
17 5.6074 6.8589 0.6538 -1.2514 1.336 -0.937
18 9.4004 9.7131 0.6104 -0.3127 1.356 -0.231
19 15.3194 14.4702 0.5504 0.8492 1.381 0.615
20 18.9747 19.2272 0.5105 -0.2525 1.397 -0.181
21 23.5128 23.9843 0.4957 -0.4715 1.402 -0.336
22 27.5205 28.7413 0.5081 -1.2208 1.398 -0.874
23 34.5391 33.4984 0.5459 1.0408 1.383 0.752
24 38.1306 38.2554 0.6044 -0.1248 1.359 -0.0918
25 44.7561 43.0125 0.6781 1.7436 1.323 1.318
Output Statistics

Cook's Hat Diag Cov
Obs -2-1 0 1 2 D RStudent H Ratio DFFITS

1 | |* | 0.044 0.7910 0.2168 1.3720 0.4161
2 | |** | 0.063 1.0669 0.1826 1.1917 0.5042
3 | |**** | 0.186 2.3326 0.1419 0.5410 0.9484
4 | **| | 0.066 -1.3760 0.1263 0.9688 -0.5231
5 | |** | 0.053 1.2141 0.1285 1.0494 0.4662
6 | **| | 0.051 -1.1097 0.1440 1.1180 -0.4551
7 | ***| | 0.146 -1.7607 0.1718 0.8247 -0.8019
8 | **| | 0.076 -1.0660 0.2121 1.2367 -0.5530
9 | | | 0.005 -0.2660 0.2168 1.5301 -0.1399
10 | | | 0.006 -0.3108 0.1826 1.4586 -0.1469
11 | **| | 0.042 -1.0145 0.1419 1.1589 -0.4125

Dependent Variable: v502
Output Statistics

Cook's Hat Diag Cov
Obs -2-1 0 1 2 D RStudent H Ratio DFFITS

12 | **| | 0.045 -1.1228 0.1263 1.0894 -0.4268
13 | |* | 0.019 0.6999 0.1285 1.2660 0.2687
14 | |** | 0.060 1.2095 0.1440 1.0706 0.4960
15 | | | 0.004 -0.2858 0.1718 1.4439 -0.1302
16 | |** | 0.087 1.1436 0.2121 1.1973 0.5933
17 | *| | 0.053 -0.9342 0.1933 1.2702 -0.4573
18 | | | 0.003 -0.2254 0.1685 1.4470 -0.1014
19 | |* | 0.015 0.6054 0.1370 1.3098 0.2412
20 | | | 0.001 -0.1765 0.1179 1.3694 -0.0645
21 | | | 0.004 -0.3291 0.1111 1.3382 -0.1163
22 | *| | 0.025 -0.8684 0.1168 1.1868 -0.3157
23 | |* | 0.022 0.7444 0.1348 1.2594 0.2938
24 | | | 0.000 -0.0896 0.1652 1.4537 -0.0399
25 | |** | 0.114 1.3425 0.2080 1.0868 0.6879


Output Statistics

-------------------DFBETAS-------------------
Obs Intercept layers sharp fsp
Ramin Shamshiri STA6167, HW#1, Jan.24.2008 Page 11


1 0.4156 -0.2707 -0.2234 -0.2388
2 0.4977 -0.2832 -0.2950 -0.3128
3 0.8714 -0.3269 -0.6295 -0.6583
4 -0.4139 0.0521 0.3680 0.3802
5 0.2813 0.0767 -0.3251 -0.3318
6 -0.1946 -0.1651 0.2999 0.3029
7 -0.2093 -0.4186 0.4837 0.4835
8 -0.0650 -0.3543 0.3002 0.2970
9 -0.0626 0.0910 -0.0751 0.0030
10 -0.0567 0.0825 -0.0859 0.0027
11 -0.0977 0.1422 -0.2738 0.0047
12 -0.0292 0.0425 -0.3003 0.0014
13 -0.0304 0.0442 0.1874 0.0015
14 -0.1237 0.1800 0.3268 0.0060
15 0.0467 -0.0679 -0.0785 -0.0023
16 -0.2612 0.3801 0.3221 0.0126
17 -0.2049 0.2982 0.0000 -0.2278
18 -0.0407 0.0592 0.0000 -0.0545
19 0.0720 -0.1048 0.0000 0.1454
20 -0.0106 0.0154 0.0000 -0.0424
21 -0.0008 0.0012 0.0000 -0.0797
22 0.0477 -0.0695 0.0000 -0.2135

The REG Procedure
Model: MODEL1
Dependent Variable: v502

Output Statistics

-------------------DFBETAS-------------------
Obs Intercept layers sharp fsp

23 -0.0846 0.1231 0.0000 0.1870
24 0.0157 -0.0228 0.0000 -0.0232
25 -0.3226 0.4695 0.0000 0.3604


Sum of Residuals 0
Sum of Squared Residuals 46.43796
Predicted Residual SS (PRESS) 65.72896

You might also like