Professional Documents
Culture Documents
5a)
Pearson Correlation Coefficients, N = 16
Prob > |r| under H0: Rho=0
y
x1
x2
1.00000
0.89239
<.0001
0.39458
0.1304
x1
0.89239
<.0001
1.00000
0.00000
1.0000
x2
0.39458
0.1304
0.00000
1.0000
1.00000
x2
x2
x1
You see moderate correlation from the table, which is good because it means that we have the right predictors. By
looking at scatter plot, it seems that data is evenly scattered.
6.5b)
Parameter Estimates
Variable DF
Parameter Standard
Estimate
Error t Value Pr > |t|
Intercept
37.65000
2.99610
12.57 <.0001
x1
4.42500
0.30112
14.70 <.0001
x2
4.37500
0.67332
6.50 <.0001
The estimated regression line is y = 37.65 + 4.42 X1 + 4.375 X2. Beta1 indicates that as moisture content increases, the
degree of brand liking increases about 4.425 on average, given fixed amount of sweetness.
6.5c)
Boxplot of residual is below,
Residual
-2
-4
4
x1
10
2.0
2.5
3.0
x2
3.5
4.0
6.5e)
Heteroscedasticity Test
Equation Test
BreuschPagan
0.00
<.0001 1, x1, x2
I have tried this, but for some reason, the code below did not work. if p-value is lower than .01, we reject null hypothesis,
which means that it does not have a constant variance.
proc model data=brand;
parms x1 x2;
y = z0 + x1 *z1+ x2*z2 ;
fit y /breusch=(1 z1 z2);
run;
6.5f)
Analysis of Variance
Source
DF
Sum of
Squares
Model
1872.70000
936.35000
Error
13
94.30000
7.25385
Lack of Fit
37.30000
7.46000
Pure Error
57.00000
7.12500
15
1967.00000
Corrected Total
Mean
Square F Value
Pr > F
129.08
<.0001
1.05
0.4530
Null hypothesis is that there is no lack of linear fit. Alternative hypothesis is that there is lack of linear fit. As we see, pvalue is very high, so we fail to reject the null hypothesis, which means that y and x1 and x2 has a linear fit. P-value
is .453.
6.6a)
Parameter Estimates
Parameter Standard
Variable DF Estimate
Error t Value Pr > |t|
Intercept
37.65000
2.99610
12.57 <.0001
x1
4.42500
0.30112
14.70 <.0001
x2
4.37500
0.67332
6.50 <.0001
Null hypothesis is that slopes are zero. Alternative hypothesis is that slopes are not zero. As we see, p-value is low, so we
reject the null hypothesis, which means that y and x1 and x2 are not zero and beta1 and beta2 are significant.
6.6b)
Analysis of Variance
Source
DF
Sum of
Squares
Mean
Square F Value
Model
1872.70000
936.35000
Error
13
94.30000
7.25385
Corrected Total
15
1967.00000
129.08
Pr > F
<.0001
0.9521
0.9447
Root MSE
Dependent Mean
3.29455
Coeff Var
R^2 is equal to .9521. it means that the model explains 95 percent variability of the response data around its mean.
6.7b)
Yes, It is the same. Refer to page 224 in the textbook.
6.8)
Parameter Estimates
Parameter Standard
Variable DF Estimate
Error t Value Pr > |t|
95% Confidence
Limits
Intercept
37.65000
2.99610
12.57 <.0001
31.17731
44.12269
x1
4.42500
0.30112
14.70 <.0001
3.77447
5.07553
x2
4.37500
0.67332
6.50 <.0001
2.92037
5.82963
7.3a)
Sum of
Squares Mean Square F Value Pr > F
Source
DF
Model
1872.700000
936.350000
Error
13
94.300000
7.253846
Corrected Total
15
1967.000000
Parameter
Standard
Error t Value Pr > |t|
Estimate
129.08 <.0001
37.65000000
2.99610324
x1
4.42500000
0.30111971
14.70 <.0001
3.77447043
5.07552957
x2
4.37500000
0.67332413
6.50 <.0001
2.92037165
5.82962835
Intercept
44.12268753
7.3b)
Test 1 Results for Dependent Variable y
Source
Numerator
Denominator
DF
Mean
Square F Value Pr > F
306.25000
13
7.25385
42.22 <.0001
p-value is very small so we cannot discard the variable (keep in mind that our sample size is a bit small). P-value is less
than .0001