Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit

Semester 1 Assessment, 2022

MAST30025 Linear Statistical Models

Question 1 (14 marks)

[3 marks] Let A be a symmetric matrix with eigenvalues all either 0 or 1. Show that A is idempotent.

[3 marks] Let B be a symmetric matrix with positive eigenvalues. Show that B is positive definite.

(c) [3 marks] Let y be a random vector with E[y] = µ and Var y = V , and let C be a matrix. Show directly that Var Cy = CVCT .

[2 marks] Let D be a matrix. Show directly that r(Dc D) = r(D).

[3 marks] Find a conditional inverse of

E =

' 3   3   333   3333 ' .

Question 2 (11 marks)

Let

y = y(y)扌(|) ( MVN 2(1) , 1(3)   2(1) ,

and

A = 21 11 ,    V = 1(3)   2(1) .

(a) [2 marks] Calculate E[yT Ay].

[2 marks] Let z = y 一 E[y]. Describe the distribution of zT V-|z, where V = Var y.

[3 marks] Let x = (y|, y , y| 一 y )T . Describe the distribution of x.

[4 marks] Find a constant a such that

ay|

y| 3y

( t| .

Question 3 (16 marks)

For this question, we study the relationship between the compressible strength  (MPA) of a concrete slump and the following seven ingredients:

❼ Cement: cement (kg/m )

Slag: blast furnace slag (kg/m )

❼ Flyash: fly ash (kg/m )

❼ Water: water (kg/m )

❼ SP: superplasticizer (kg/m )

Coarse: coarse sand aggregate (kg/m )

❼ Fine: fine sand aggregate (kg/m )

>  slump  <- readRDS("slump.Rds")

>  head(slump,  n  =  5)

Cement 1 273 2 163 3 162 4 162 5 154

Slag

82

149

148

148

112

Flyash

105

191

191

190

144

Water SP Coarse

210 9

180 12

179 16

179 19

220 10

904

843

840

838

923

Fine 680 746 743 741 658

MPA 34.99 41.14 41.81 42.08 26.82

>  y  =  slump$MPA

>  X1  =  as.matrix(cbind(1,  slump$SP))

>  round(solve(t(X1)%*%X1),6)

[,1]            [,2]

[1,]    0.100417  -0.010622

[2,]  -0.010622    0.001244

>  t(X1)%*%y

[,1]

[1,]    3712.06

[2,]  31615.27

>  sum(y^2)

[1]  140047.1

>  sum((y  - mean(y))^2)

[1]  6266.664

>  fullmodel  <- lm(MPA ~ ., data = slump)

>  summary(fullmodel)

Call:

lm(formula  =  MPA  ~  .,  data  =  slump)

Residuals:

Min            1Q   Median           3Q         Max

-5.8411  -1.7063  -0.2831    1.2986    7.9424

Coefficients:

Estimate  Std.  Error  t  value  Pr(>|t|)

(Intercept) 139.78150 Cement 0.06141 Slag -0.02971 Flyash 0.05053 Water -0.23270

SP 0.10315

Coarse -0.05562

Fine -0.03908

71.10128 0.02282 0.03176 0.02316 0.07166 0.13459 0.02744 0.02882

1.966 2.691 -0.935 2.182 -3.247 0.766 -2.027 -1.356

0.05222 . 0.00842 ** 0.35200 0.03159 * 0.00161 ** 0.44532 0.04546 * 0.17833

---

Signif.  codes:    0  ‘***’  0.001  ‘**’  0.01  ‘*’  0.05  ‘.’  0.1  ‘  ’  1

Residual  standard  error:  2.609  on  95  degrees  of  freedom                Multiple  R-squared:    0.8968,               Adjusted  R-squared:    0.8892 F-statistic:      118  on  7  and  95  DF,   p-value:  < 2.2e-16

>  qvec  =  c(0.025,  0.05,  0.1,  0.9,  0.95,  0.975)

>  qchisq(qvec,  df  =  95)

[1]    69.92487    73.51984    77.81843  113.03769  118.75161  123.85797 >  qt(qvec,  df  =  95)

[1]  -1.985251  -1.661052  -1.290527    1.290527    1.661052    1.985251 >  qf(0.95,  5:10,  95)

[1]  2.310225  2.195548  2.107506  2.037370  1.979923  1.931838

(a) [2 marks] For the model MPA = βb + β|SP + ε, calculate the least squares estimate b.

[2 marks] For the model MPA = βb + β|SP + ε, calculate SSRes .

[2 marks] For the full model, find a 95% confidence interval for βin.

[3 marks] For the full model, test the hypothesis Hb : βslag  = 0 at the 5% significance level using an F-test.   State the F-statistic, the p-value, and your conclusion in the context of the study.

[4 marks] For the full model, complete the following ANOVA table:

Variation SS

Regression

Residual

Total


[3 marks] For the full model, test Hb : β廿。m。nt = βslag = βlyash = βwatr = β廿oars = βin = 0 (i.e., all parameters except the intercept and βsP ), with significance level α = 0.05. (Hint: Use your results from part (b).)

Question 4 (11 marks)

Consider the generalised least squares estimator

b = (XT V-|X)-|XT V-|y

for a full rank linear model

y = Xβ + ε ,

where X is an n × p matrix with rank p, ε ( MVN (0, V), and V is known.

[2 marks] Show directly that E[b] = β .

[3 marks] Calculate Var b.


[3 marks] For design variables x* with response y* , find a 100(1 一 α)% confidence interval

for E[y*] = (x* )T β .

[3 marks] For the null hypothesis Hb : β = 0, find the distribution of the test statistic bT XT V-|Xb, and the rejection region for significance level α .