108 年 國立中央大學企業管理學系碩士班丙組《統計學》

📄 試題原卷 免費註冊後即可對照原始考卷 PDF免費註冊

第 1 題

Suppose there are 3 random variables, X1,X2,X3X_1, X_2, X_3, and their expected values are 2, 0, and -1, respectively. We also know their variances are 2, 1, and 2, respectively. The covariance between X1X_1 and X2X_2 is -1, and the covariance between X2X_2 and X3X_3 is 1, while X1X_1 and X3X_3 are independent. Try to find the following:
(a) The expected value and the variance of W1=X1+X2+X3W_1 = X_1+X_2+X_3? (5 pts)
(b) The expected value and the variance of W2=2X1−X3W_2 = 2X_1-X_3? (5 pts)
(c) The covariance between W1W_1 and W2W_2. (5 pts)

登入後即可作答並保存紀錄。

這一題的完整詳解

本題主要在考驗對隨機變數期望值、變異數以及共變異數的基本性質的理解與應用。

已知:
E(X1)=2E(X_1) = 2, Var(X1)=2Var(X_1) = 2
E(X2)=0E(X_2) = 0, Var(X2)=1Var(X_2) = 1
E(X3)=−1E(X_3) = -1, Var(X3)=2Var(X_3) = 2
Cov(X1,X2)=−1Cov(X_1, X_2) = -1
Cov(X2,X3)=1Cov(X_2, X_3) = 1
X1X_1 and X3X_3 are independent, which implies Cov(X1,X3)=0Cov(X_1, X_3) = 0.

(a) 求 W1=X1+X2+X3W_1 = X_1+X_2+X_3 的期望值與變異數。
期望值性質:E(aX+bY+cZ)=aE(X)+bE(Y)+cE(Z)E(aX + bY + cZ) = aE(X) + bE(Y) + cE(Z)
E(W1)=E(X1+X2+X3)=E(X1)+E(X2)+E(X3)E(W_1) = E(X_1+X_2+X_3) = E(X_1) + E(X_2) + E(X_3)
E(W1)=2+0+(−1)=1E(W_1) = 2 + 0 + (-1) = 1

變異數性質:Var(aX+bY+cZ)=a2Var(X)+b2Var(Y)+c2Var(Z)+2abCov(X,Y)+2acCov(X,Z)+2bcCov(Y,Z)Var(aX + bY + cZ) = a^2Var(X) + b^2Var(Y) + c^2Var(Z) + 2abCov(X,Y) + 2acCov(X,Z) + 2bcCov(Y,Z)
Var(W1)=Var(X1+X2+X3)Var(W_1) = Var(X_1+X_2+X_3)
Var(W1)=Var(X1)+Var(X2)+Var(X3)+2Cov(X1,X2)+2Cov(X1,X3)+2Cov(X2,X3)Var(W_1) = Var(X_1) + Var(X_2) + Var(X_3) + 2Cov(X_1, X_2) + 2Cov(X_1, X_3) + 2Cov(X_2, X_3)
Var(W1)=2+1+2+2(−1)+2(0)+2(1)Var(W_1) = 2 + 1 + 2 + 2(-1) + 2(0) + 2(1)
Var(W1)=2+1+2−2+0+2=5Var(W_1) = 2 + 1 + 2 - 2 + 0 + 2 = 5

【答案】(a) E(W1)=1E(W_1) = 1, Var(W1)=5Var(W_1) = 5

(b) 求 W2=2X1−X3W_2 = 2X_1-X_3 的期望值與變異數。
期望值:
E(W2)=E(2X1−X3)=2E(X1)−E(X3)E(W_2) = E(2X_1-X_3) = 2E(X_1) - E(X_3)
E(W2)=2(2)−(−1)=4+1=5E(W_2) = 2(2) - (-1) = 4 + 1 = 5

變異數:
Var(W2)=Var(2X1−X3)=Var(2X1)+Var(−X3)+2Cov(2X1,−X3)Var(W_2) = Var(2X_1-X_3) = Var(2X_1) + Var(-X_3) + 2Cov(2X_1, -X_3)

🔒

後續完整解題步驟與【答案】

免費註冊,享三天全站完整詳解閱覽。

免費註冊

第 2 題

To test a particular coin fair or not, (i.e. p=Pr("head")=0.5p = Pr(\text{"head"}) = 0.5), we decide to flip the coin nn times. Let XX denote the number of heads appearing in nn flips.
(a) Please provide a M.L.E. of pp; (5 points)
(b) Let n=4n = 4, try to use XX to be the test statistics to develop a test on H0:p=0.5H_0: p = 0.5
(1) Show the distribution of XX. (5 points)
(2) Let's set the significant level, α\alpha, to be 0.15, show the rejection rule. (5 points)
(3) Based on your rejection rule in (b-2), what is the testing power on Ha:p=0.3H_a: p = 0.3? (5 points)
(4) If only one head appears in 4 flips, what is your conclusion? (5 points)
(5) Please develop a 85% confidence interval of pp. (5 points)
(c) If n=50n = 50, let's use Y=X/nY = X/n to be the test statistics to test on H0:p=0.5H_0: p = 0.5
(1) Since n=50n = 50, we can invoke the Central Limit Theory, show the approximate distribution of YY. (5 points)
(2) Let's set the significant level, α\alpha, to be 0.15, show the rejection rule. In other words, what is the critical number of heads appearing to reject H0H_0. (5 points)
(3) Based on your rejection rule in (c-2), what is the testing power on Ha:p=0.3H_a: p = 0.3? (5 points)
(4) Please develop a 85% confidence interval of pp. (5 points)

登入後即可作答並保存紀錄。

這一題的完整詳解

核心觀念

本題考查:

  • 二項分配:X∼Binomial⁡(n,p)X\sim\operatorname{Binomial}(n,p)。
  • 最大概似估計法(M.L.E.)。
  • 雙尾假設檢定:H0:p=0.5H_0:p=0.5 對 Ha:p≠0.5H_a:p\ne0.5。
  • 顯著水準 α\alpha、檢定力(power)。
  • 中央極限定理(Central Limit Theorem)。
  • 二項母體比例的信賴區間。

「硬幣是否公平」表示只要 pp 偏離 0.50.5,無論偏大或偏小都應拒絕虛無假設,因此採用雙尾檢定。


(a) pp 的最大概似估計量

令觀察到 nn 次擲幣中有 xx 次正面,則

P(X=x)=(nx)px(1−p)n−x.P(X=x)=\binom{n}{x}p^x(1-p)^{n-x}.

概似函數為

L(p)=(nx)px(1−p)n−x.L(p)=\binom{n}{x}p^x(1-p)^{n-x}.

取對數概似函數:

ℓ(p)=ln⁡(nx)+xln⁡p+(n−x)ln⁡(1−p).\ell(p)=\ln\binom{n}{x}+x\ln p+(n-x)\ln(1-p).

對 pp 微分並令其等於 00:

dℓ(p)dp=xp−n−x1−p=0.\frac{d\ell(p)}{dp} = \frac{x}{p}-\frac{n-x}{1-p}=0.

因此

x(1−p)=(n−x)p,x(1-p)=(n-x)p, x=np.x=np.

所以最大概似估計量為

p^=Xn.\boxed{\hat p=\frac{X}{n}}.

(b) n=4n=4,以 XX 為檢定統計量

(b-1) XX 的分配

因為每次擲幣只有正面或反面,且各次擲幣相互獨立,所以

X∼Binomial⁡(4,p).X\sim\operatorname{Binomial}(4,p).

在虛無假設 H0:p=0.5H_0:p=0.5 下,

P(X=x)=(4x)(0.5)x(0.5)4−x=(4x)(12)4.P(X=x)=\binom{4}{x}(0.5)^x(0.5)^{4-x} =\binom{4}{x}\left(\frac12\right)^4.

因此:

xxPH0(X=x)P_{H_0}(X=x)
00116=0.0625\dfrac{1}{16}=0.0625
11416=0.25\dfrac{4}{16}=0.25
22616=0.375\dfrac{6}{16}=0.375
33416=0.25\dfrac{4}{16}=0.25
44116=0.0625\dfrac{1}{16}=0.0625

分配具有對稱性,最能支持「硬幣不公平」的結果是 X=0X=0 或 X=4X=4,因為四次全部為反面或全部為正面。


(b-2) 顯著水準 α=0.15\alpha=0.15 下的拒絕域

由於是雙尾檢定,先選擇兩端最極端的結果:

PH0(X=0 或 X=4)=116+116=18=0.125.P_{H_0}(X=0\text{ 或 }X=4) = \frac{1}{16}+\frac{1}{16} = \frac18 = 0.125.

因為

0.125≤0.15,0.125\le 0.15,

所以非隨機化檢定的拒絕域為

R={X=0,4}.\boxed{R=\{X=0,4\}}.

亦即:

若 X=0 或 X=4, 則拒絕 H0。\boxed{\text{若 }X=0\text{ 或 }X=4,\text{ 則拒絕 }H_0。}

此檢定的實際顯著水準為 0.1250.125,小於名目顯著水準 0.150.15。若再納入 X=1X=1 或 X=3X=3,拒絕機率會增加為

0.125+0.25+0.25=0.625,0.125+0.25+0.25=0.625,

超過 0.150.15,因此不能納入。


(b-3) 在 Ha:p=0.3H_a:p=0.3 下的檢定力

檢定力定義為在替代假設成立時拒絕 H0H_0 的機率。

本題拒絕域為 X=0X=0 或 X=4X=4,所以在 p=0.3p=0.3 下:

Power=Pp=0.3(X=0 或 X=4)=Pp=0.3(X=0)+Pp=0.3(X=4)=(40)(0.3)0(0.7)4+(44)(0.3)4(0.7)0=(0.7)4+(0.3)4=0.2401+0.0081=0.2482.\begin{aligned} \text{Power} &=P_{p=0.3}(X=0\text{ 或 }X=4)\\ &=P_{p=0.3}(X=0)+P_{p=0.3}(X=4)\\ &=\binom40(0.3)^0(0.7)^4 +\binom44(0.3)^4(0.7)^0\\ &=(0.7)^4+(0.3)^4\\ &=0.2401+0.0081\\ &=\boxed{0.2482}. \end{aligned}

因此檢定力為

24.82%.\boxed{24.82\%}.

(b-4) 若四次擲幣只有一次正面

此時

X=1.X=1.

由於拒絕域為 {0,4}\{0,4\},而 X=1X=1 不在拒絕域內,因此

不拒絕 H0:p=0.5。\boxed{\text{不拒絕 }H_0:p=0.5。}

統計上的結論是:在此顯著水準下,沒有足夠證據認定硬幣不公平。

必須注意,「不拒絕 H0H_0」不等於證明硬幣一定公平,只表示目前的樣本證據不足以支持硬幣不公平。


(b-5) pp 的 85% 信賴區間

承接上一小題,觀察值為 x=1x=1。由於樣本數只有 44,採用二項分配的 Clopper–Pearson 精確信賴區間較合適。

信賴水準為 85%85\%,所以

α=0.15,α2=0.075.\alpha=0.15,\qquad \frac{\alpha}{2}=0.075.

當 x=1,n=4x=1,n=4 時,精確信賴區間為

[B−1(0.075;1,4),B−1(0.925;2,3)],\left[ B^{-1}(0.075;1,4), B^{-1}(0.925;2,3) \right],

其中 B−1(q;a,b)B^{-1}(q;a,b) 表示 Beta 分配的第 qq 分位數。

下限可由

1−(1−pL)4=0.0751-(1-p_L)^4=0.075

求得:

pL=1−(0.925)1/4≈0.0193.p_L=1-(0.925)^{1/4}\approx0.0193.

上限由

PpU(X≤1)=0.075P_{p_U}(X\le1)=0.075

求得,數值約為

pU≈0.7117.p_U\approx0.7117.

因此,85% 精確信賴區間為

0.0193≤p≤0.7117.\boxed{0.0193\le p\le0.7117}.

這個區間相當寬,原因是樣本數僅有 44 次,估計比例的不確定性很高。


(c) n=50n=50,以 Y=X/nY=X/n 為檢定統計量

令

Y=Xn=X50.Y=\frac{X}{n}=\frac{X}{50}.

YY 是樣本中正面出現的比例。

(c-1) YY 的近似分配

因為

X∼Binomial⁡(n,p),X\sim\operatorname{Binomial}(n,p),

所以

E(X)=np,Var⁡(X)=np(1−p).E(X)=np,\qquad \operatorname{Var}(X)=np(1-p).

因此

E(Y)=E(Xn)=p,E(Y)=E\left(\frac Xn\right)=p,

且

Var⁡(Y)=Var⁡(X)n2=p(1−p)n.\operatorname{Var}(Y) = \frac{\operatorname{Var}(X)}{n^2} = \frac{p(1-p)}{n}.
🔒

後續完整解題步驟與【答案】

免費註冊,享三天全站完整詳解閱覽。

免費註冊

第 3 題

An experiment was performed to investigate the effectiveness of three materials. Five samples of each material were tested and the failure times (in seconds) are shown below:

| Material | \multicolumn{5}{|c|}{Failure Time (seconds)} |
|---|---|---|---|---|---|
| A | 52 | 50 | 47 | 65 | 46 |
| B | 44 | 58 | 48 | 60 | 64 |
| C | 70 | 95 | 88 | 86 | 89 |

Assume that the variances for three materials are the same, and the failure times are normally distributed.
(a) Use α=0.05\alpha=0.05, do all three materials have the same effect on the mean failure time? (10 pts)
(b) Please develop the 95% confidence intervals of the mean failure times for three materials separately. Construct a graphical display of these three confidence intervals. (10 pts)
(c) When we do pairwise comparison, at α=0.05\alpha=0.05, does there exist any significant difference on mean failure time between any pair of materials? (15 pts)

登入後即可作答並保存紀錄。

這一題的完整詳解

本題主要考驗單因子變異數分析 (One-way ANOVA) 的應用,包括整體檢定、個別信賴區間以及事後檢定 (pairwise comparison)。

首先,整理資料並計算基本統計量:
樣本數 nA=nB=nC=5n_A=n_B=n_C=5。總樣本數 N=3×5=15N = 3 \times 5 = 15。
各組平均數:
xˉA=(52+50+47+65+46)/5=260/5=52\bar{x}_A = (52+50+47+65+46)/5 = 260/5 = 52
xˉB=(44+58+48+60+64)/5=274/5=54.8\bar{x}_B = (44+58+48+60+64)/5 = 274/5 = 54.8
xˉC=(70+95+88+86+89)/5=428/5=85.6\bar{x}_C = (70+95+88+86+89)/5 = 428/5 = 85.6

各組樣本變異數:
sA2=15−1∑i=15(xA,i−xˉA)2=14[(52−52)2+(50−52)2+(47−52)2+(65−52)2+(46−52)2]s_A^2 = \frac{1}{5-1} \sum_{i=1}^5 (x_{A,i} - \bar{x}_A)^2 = \frac{1}{4} [(52-52)^2 + (50-52)^2 + (47-52)^2 + (65-52)^2 + (46-52)^2]
sA2=14[02+(−2)2+(−5)2+132+(−6)2]=14[0+4+25+169+36]=2344=58.5s_A^2 = \frac{1}{4} [0^2 + (-2)^2 + (-5)^2 + 13^2 + (-6)^2] = \frac{1}{4} [0 + 4 + 25 + 169 + 36] = \frac{234}{4} = 58.5

sB2=15−1∑i=15(xB,i−xˉB)2=14[(44−54.8)2+(58−54.8)2+(48−54.8)2+(60−54.8)2+(64−54.8)2]s_B^2 = \frac{1}{5-1} \sum_{i=1}^5 (x_{B,i} - \bar{x}_B)^2 = \frac{1}{4} [(44-54.8)^2 + (58-54.8)^2 + (48-54.8)^2 + (60-54.8)^2 + (64-54.8)^2]
sB2=14[(−10.8)2+(3.2)2+(−6.8)2+(5.2)2+(9.2)2]s_B^2 = \frac{1}{4} [(-10.8)^2 + (3.2)^2 + (-6.8)^2 + (5.2)^2 + (9.2)^2]
sB2=14[116.64+10.24+46.24+27.04+84.64]=284.84=71.2s_B^2 = \frac{1}{4} [116.64 + 10.24 + 46.24 + 27.04 + 84.64] = \frac{284.8}{4} = 71.2

sC2=15−1∑i=15(xC,i−xˉC)2=14[(70−85.6)2+(95−85.6)2+(88−85.6)2+(86−85.6)2+(89−85.6)2]s_C^2 = \frac{1}{5-1} \sum_{i=1}^5 (x_{C,i} - \bar{x}_C)^2 = \frac{1}{4} [(70-85.6)^2 + (95-85.6)^2 + (88-85.6)^2 + (86-85.6)^2 + (89-85.6)^2]
sC2=14[(−15.6)2+(9.4)2+(2.4)2+(0.4)2+(3.4)2]s_C^2 = \frac{1}{4} [(-15.6)^2 + (9.4)^2 + (2.4)^2 + (0.4)^2 + (3.4)^2]
sC2=14[243.36+88.36+5.76+0.16+11.56]=349.24=87.3s_C^2 = \frac{1}{4} [243.36 + 88.36 + 5.76 + 0.16 + 11.56] = \frac{349.2}{4} = 87.3

根據題目假設,三個材料的變異數相同,這意味著我們應該使用合併變異數 (pooled variance)。
合併變異數 sp2s_p^2 的計算:
sp2=∑j=1k(nj−1)sj2∑j=1k(nj−1)=(5−1)sA2+(5−1)sB2+(5−1)sC2(5−1)+(5−1)+(5−1)s_p^2 = \frac{\sum_{j=1}^k (n_j-1)s_j^2}{\sum_{j=1}^k (n_j-1)} = \frac{(5-1)s_A^2 + (5-1)s_B^2 + (5-1)s_C^2}{(5-1)+(5-1)+(5-1)}
sp2=4×58.5+4×71.2+4×87.34+4+4=4×(58.5+71.2+87.3)12=58.5+71.2+87.33s_p^2 = \frac{4 \times 58.5 + 4 \times 71.2 + 4 \times 87.3}{4+4+4} = \frac{4 \times (58.5 + 71.2 + 87.3)}{12} = \frac{58.5 + 71.2 + 87.3}{3}
sp2=2173≈72.333s_p^2 = \frac{217}{3} \approx 72.333

總平均數 xˉtotal=(260+274+428)/15=962/15≈64.133\bar{x}_{total} = (260 + 274 + 428) / 15 = 962 / 15 \approx 64.133。

(a) α=0.05\alpha=0.05,檢定所有三個材料是否對平均失效時間有相同影響。
這是單因子變異數分析的整體檢定。
虛無假設 H0H_0: μA=μB=μC\mu_A = \mu_B = \mu_C (所有材料的平均失效時間相同)。
對立假設 HaH_a: 至少有一個材料的平均失效時間不同。
檢定統計量為 F 統計量:F=MSTMSEF = \frac{MST}{MSE}。
其中 MST (Mean Square Treatment) 是組間均方,MSE (Mean Square Error) 是組內均方 (或稱殘差均方)。
SST=∑j=1knj(xˉj−xˉtotal)2SST = \sum_{j=1}^k n_j (\bar{x}_j - \bar{x}_{total})^2 (總平方和)
SSTA=5×(52−64.133)2≈5×(−12.133)2≈5×147.2=736SST_A = 5 \times (52 - 64.133)^2 \approx 5 \times (-12.133)^2 \approx 5 \times 147.2 = 736
SSTB=5×(54.8−64.133)2≈5×(−9.333)2≈5×87.1=435.5SST_B = 5 \times (54.8 - 64.133)^2 \approx 5 \times (-9.333)^2 \approx 5 \times 87.1 = 435.5
SSTC=5×(85.6−64.133)2≈5×(21.467)2≈5×460.8=2304SST_C = 5 \times (85.6 - 64.133)^2 \approx 5 \times (21.467)^2 \approx 5 \times 460.8 = 2304
SST=SSTA+SSTB+SSTC=736+435.5+2304=3475.5SST = SST_A + SST_B + SST_C = 736 + 435.5 + 2304 = 3475.5
MST = SST/(k−1)SST / (k-1),其中 k=3k=3 是組數。
MST = 3475.5/(3−1)=3475.5/2=1737.753475.5 / (3-1) = 3475.5 / 2 = 1737.75

MSE = sp2=217/3≈72.333s_p^2 = 217/3 \approx 72.333 (這是組內變異數的估計)。

🔒

後續完整解題步驟與【答案】

免費註冊,享三天全站完整詳解閱覽。

免費註冊

其他考古題