114 年 國立政治大學經濟學系碩士班《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25 分

Let X1,…,XnX_1, \dots, X_n be a random sample from the N(μ,σ2)N(\mu, \sigma^2) distribution, where μ∈R\mu \in \mathbb{R} and 0<σ<∞0 < \sigma < \infty. Consider the estimation of σ2\sigma^2 using the squared error loss. Let S2S^2 be the sample variance of X1,…,XnX_1, \dots, X_n.

(a) Find E[S2]E[S^2] and Var[S2]Var[S^2].
(b) Find E[S2]E[S^2] if the population distribution is unknown with mean μ\mu and variance σ2\sigma^2.
(c) Define a class of estimators of the form cS2cS^2 with a nonrandom cc. Compute the risk function in this class and find the best estimator that minimizes the risk.
(d) Suppose that n(S2−σ2)\sqrt{n}(S^2 - \sigma^2) converges to N(0,V)N(0, V) in distribution as n→∞n \to \infty. Using the Delta method find the limiting distribution of n(S−σ)\sqrt{n}(S - \sigma), where SS is the sample standard deviation.

登入後即可作答並保存紀錄。

這一題的完整詳解

這組題目主要在考察樣本變異數的性質、無母數分布下的估計以及 Delta method 的應用。

(a) 求 E[S2]E[S^2] 和 Var[S2]Var[S^2]
樣本變異數 S2S^2 的定義為 S2=1n−1∑i=1n(Xi−Xˉ)2S^2 = \frac{1}{n-1} \sum_{i=1}^n (X_i - \bar{X})^2。
我們知道 Xi∼N(μ,σ2)X_i \sim N(\mu, \sigma^2),所以 Xˉ∼N(μ,σ2/n)\bar{X} \sim N(\mu, \sigma^2/n)。
令 Zi=Xi−μσZ_i = \frac{X_i - \mu}{\sigma},則 Zi∼N(0,1)Z_i \sim N(0, 1) 且獨立。
Zˉ=Xˉ−μσ/n\bar{Z} = \frac{\bar{X} - \mu}{\sigma/\sqrt{n}}。
∑i=1n(Xi−Xˉ)2=∑i=1n(Xi−μ−(Xˉ−μ))2=∑i=1n[(Xi−μ)−(Xˉ−μ)]2\sum_{i=1}^n (X_i - \bar{X})^2 = \sum_{i=1}^n (X_i - \mu - (\bar{X} - \mu))^2 = \sum_{i=1}^n [(X_i - \mu) - (\bar{X} - \mu)]^2
=∑i=1n(Xi−μ)2−2(Xˉ−μ)∑i=1n(Xi−μ)+n(Xˉ−μ)2= \sum_{i=1}^n (X_i - \mu)^2 - 2(\bar{X} - \mu)\sum_{i=1}^n (X_i - \mu) + n(\bar{X} - \mu)^2
=∑i=1n(Xi−μ)2−2(Xˉ−μ)(n(Xˉ−μ))+n(Xˉ−μ)2= \sum_{i=1}^n (X_i - \mu)^2 - 2(\bar{X} - \mu)(n(\bar{X} - \mu)) + n(\bar{X} - \mu)^2
=∑i=1n(Xi−μ)2−2n(Xˉ−μ)2+n(Xˉ−μ)2= \sum_{i=1}^n (X_i - \mu)^2 - 2n(\bar{X} - \mu)^2 + n(\bar{X} - \mu)^2
=∑i=1n(Xi−μ)2−n(Xˉ−μ)2= \sum_{i=1}^n (X_i - \mu)^2 - n(\bar{X} - \mu)^2
令 W=∑i=1nZi2=∑i=1n(Xi−μ)2σ2W = \sum_{i=1}^n Z_i^2 = \sum_{i=1}^n \frac{(X_i - \mu)^2}{\sigma^2}。
W=∑i=1n(Xi−μ)2σ2=1σ2(∑i=1n(Xi−μ)2−n(Xˉ−μ)2)W = \sum_{i=1}^n \frac{(X_i - \mu)^2}{\sigma^2} = \frac{1}{\sigma^2} \left( \sum_{i=1}^n (X_i - \mu)^2 - n(\bar{X} - \mu)^2 \right)
=1σ2(∑i=1n(Xi−μ)2−n(Xˉ−μσ/nσn)2)= \frac{1}{\sigma^2} \left( \sum_{i=1}^n (X_i - \mu)^2 - n \left( \frac{\bar{X} - \mu}{\sigma/\sqrt{n}} \frac{\sigma}{\sqrt{n}} \right)^2 \right)
=1σ2(∑i=1n(Xi−μ)2−n(Xˉ−μ)2σ2/nσ2n)= \frac{1}{\sigma^2} \left( \sum_{i=1}^n (X_i - \mu)^2 - n \frac{(\bar{X} - \mu)^2}{\sigma^2/n} \frac{\sigma^2}{n} \right)
=1σ2(∑i=1n(Xi−μ)2−n(Zˉ)2)= \frac{1}{\sigma^2} \left( \sum_{i=1}^n (X_i - \mu)^2 - n (\bar{Z})^2 \right)
我們知道 ∑i=1nZi2=∑i=1n(Zi−Zˉ)2+nZˉ2\sum_{i=1}^n Z_i^2 = \sum_{i=1}^n (Z_i - \bar{Z})^2 + n\bar{Z}^2。
∑i=1nZi2∼χ2(n)\sum_{i=1}^n Z_i^2 \sim \chi^2(n)。
∑i=1n(Zi−Zˉ)2=∑i=1n(Xi−μσ−Xˉ−μσ/n)2=1σ2∑i=1n(Xi−Xˉ)2=n−1σ2S2\sum_{i=1}^n (Z_i - \bar{Z})^2 = \sum_{i=1}^n \left( \frac{X_i - \mu}{\sigma} - \frac{\bar{X} - \mu}{\sigma/\sqrt{n}} \right)^2 = \frac{1}{\sigma^2} \sum_{i=1}^n (X_i - \bar{X})^2 = \frac{n-1}{\sigma^2} S^2。
所以 n−1σ2S2∼χ2(n−1)\frac{n-1}{\sigma^2} S^2 \sim \chi^2(n-1)。
E[S2]=E[σ2n−1χ2(n−1)]=σ2n−1E[χ2(n−1)]=σ2n−1(n−1)=σ2E[S^2] = E\left[ \frac{\sigma^2}{n-1} \chi^2(n-1) \right] = \frac{\sigma^2}{n-1} E[\chi^2(n-1)] = \frac{\sigma^2}{n-1} (n-1) = \sigma^2。
Var[S2]=Var[σ2n−1χ2(n−1)]=(σ2n−1)2Var[χ2(n−1)]=(σ2n−1)2(2(n−1))=2σ4n−1Var[S^2] = Var\left[ \frac{\sigma^2}{n-1} \chi^2(n-1) \right] = \left(\frac{\sigma^2}{n-1}\right)^2 Var[\chi^2(n-1)] = \left(\frac{\sigma^2}{n-1}\right)^2 (2(n-1)) = \frac{2\sigma^4}{n-1}。

(b) 求無母數分布下的 E[S2]E[S^2]
如果母體分布未知,但已知其平均數為 μ\mu 且變異數為 σ2\sigma^2,則樣本變異數 S2S^2 的期望值仍然是 σ2\sigma^2。
證明:
E[S2]=E[1n−1∑i=1n(Xi−Xˉ)2]E[S^2] = E\left[ \frac{1}{n-1} \sum_{i=1}^n (X_i - \bar{X})^2 \right]
=1n−1E[∑i=1n(Xi2−2XiXˉ+Xˉ2)]= \frac{1}{n-1} E\left[ \sum_{i=1}^n (X_i^2 - 2X_i\bar{X} + \bar{X}^2) \right]
=1n−1E[∑i=1nXi2−2Xˉ∑i=1nXi+nXˉ2]= \frac{1}{n-1} E\left[ \sum_{i=1}^n X_i^2 - 2\bar{X}\sum_{i=1}^n X_i + n\bar{X}^2 \right]
=1n−1E[∑i=1nXi2−2Xˉ(nXˉ)+nXˉ2]= \frac{1}{n-1} E\left[ \sum_{i=1}^n X_i^2 - 2\bar{X}(n\bar{X}) + n\bar{X}^2 \right]
=1n−1E[∑i=1nXi2−2nXˉ2+nXˉ2]= \frac{1}{n-1} E\left[ \sum_{i=1}^n X_i^2 - 2n\bar{X}^2 + n\bar{X}^2 \right]
=1n−1E[∑i=1nXi2−nXˉ2]= \frac{1}{n-1} E\left[ \sum_{i=1}^n X_i^2 - n\bar{X}^2 \right]
=1n−1(∑i=1nE[Xi2]−nE[Xˉ2])= \frac{1}{n-1} \left( \sum_{i=1}^n E[X_i^2] - n E[\bar{X}^2] \right)
因為 E[Xi2]=Var[Xi]+(E[Xi])2=σ2+μ2E[X_i^2] = Var[X_i] + (E[X_i])^2 = \sigma^2 + \mu^2。
E[Xˉ2]=Var[Xˉ]+(E[Xˉ])2=σ2n+μ2E[\bar{X}^2] = Var[\bar{X}] + (E[\bar{X}])^2 = \frac{\sigma^2}{n} + \mu^2。
E[S2]=1n−1(n(σ2+μ2)−n(σ2n+μ2))E[S^2] = \frac{1}{n-1} \left( n(\sigma^2 + \mu^2) - n(\frac{\sigma^2}{n} + \mu^2) \right)

🔒

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25 分

Suppose that X1,…,XnX_1, \dots, X_n is a random sample from U[0,θ]U[0, \theta], where UU denotes a uniform distribution.

(a) Find the maximum likelihood estimator of θ\theta, denoted as θ^\hat{\theta}.
(b) Find the sampling distribution of θ^\hat{\theta}. (Provide the probability density function.)
(c) Find E[θ^]E[\hat{\theta}] and Var[θ^]Var[\hat{\theta}].
(d) Is θ^\hat{\theta} a consistent estimator for θ\theta? (Use the definition of convergence in probability to demonstrate (in)consistency.)

登入後即可作答並保存紀錄。

這一題的完整詳解

核心觀念

  1. 非正則條件(Non-regular Condition)下的最大概似估計(MLE):
    母體分配的支撐區間(support)依賴未知參數 θ\theta(即 0≤Xi≤θ0 \le X_i \le \theta),此時似然函數對 θ\theta 不可微,無法使用微積分的一階條件(FOC: dln⁡Ldθ=0\frac{d\ln L}{d\theta} = 0)求解,必須透過分析概似函數(Likelihood Function)在可行域內的單調性求極值。
  2. 順序統計量(Order Statistics)的抽樣分配:
    均勻分配 U[0,θ]U[0, \theta] 的最大順序統計量 X(n)=max⁡{X1,…,Xn}X_{(n)} = \max\{X_1, \dots, X_n\} 之累積分布函數(CDF)與機率密度函數(PDF)之推導。
  3. 期望值與變異數之計算:
    利用次數分配法或 Beta 分配之性質,計算 E[X(n)]E[X_{(n)}] 與 Var(X(n))Var(X_{(n)})。
  4. 依機率收斂(Convergence in Probability)與一致性(Consistency):
    依定義檢驗:對於任意給定 ε>0\varepsilon > 0,證明 lim⁡n→∞P(∣θ^n−θ∣<ε)=1\lim_{n \to \infty} P(|\hat{\theta}_n - \theta| < \varepsilon) = 1。

解題方法與詳細推導

(a) 求 θ\theta 的最大概似估計量(MLE)θ^\hat{\theta}

設 X1,…,Xn∼iidU[0,θ]X_1, \dots, X_n \overset{iid}{\sim} U[0, \theta],其母體機率密度函數為:

f(x;θ)=1θI(0≤x≤θ)f(x; \theta) = \frac{1}{\theta} I_{(0 \le x \le \theta)}

樣本的聯合機率密度函數即概似函數(Likelihood Function)為:

L(θ)=∏i=1nf(xi;θ)=1θn∏i=1nI(0≤xi≤θ)L(\theta) = \prod_{i=1}^n f(x_i; \theta) = \frac{1}{\theta^n} \prod_{i=1}^n I_{(0 \le x_i \le \theta)}

所有樣本均小於等於 θ\theta 等價於樣本中的最大值小於等於 θ\theta,即 θ≥max⁡(x1,…,xn)=x(n)\theta \ge \max(x_1, \dots, x_n) = x_{(n)}。且樣本皆非負(x(1)≥0x_{(1)} \ge 0),因此:

L(θ)=1θnI(θ≥x(n))L(\theta) = \frac{1}{\theta^n} I_{(\theta \ge x_{(n)})}

分析概似函數 L(θ)L(\theta):

  1. 當 θ<x(n)\theta < x_{(n)} 時,L(θ)=0L(\theta) = 0。
  2. 當 θ≥x(n)\theta \ge x_{(n)} 時,L(θ)=θ−nL(\theta) = \theta^{-n},此函數隨著 θ\theta 增加而嚴格遞減。

因此,為使 L(θ)L(\theta) 達到最大值,θ\theta 必須取其可行範圍內之最小值:

θ^=X(n)=max⁡{X1,X2,…,Xn}\hat{\theta} = X_{(n)} = \max\{X_1, X_2, \dots, X_n\}

(b) 求 θ^\hat{\theta} 的抽樣分配(機率密度函數 PDF)

記 θ^=X(n)\hat{\theta} = X_{(n)}。母體之累積分布函數(CDF)在 0≤x≤θ0 \le x \le \theta 為:

FX(x)=P(X≤x)=xθF_X(x) = P(X \le x) = \frac{x}{\theta}

先求 X(n)X_{(n)} 的累積分布函數 FX(n)(t)F_{X_{(n)}}(t):

FX(n)(t)=P(X(n)≤t)=P(X1≤t,X2≤t,…,Xn≤t)F_{X_{(n)}}(t) = P(X_{(n)} \le t) = P(X_1 \le t, X_2 \le t, \dots, X_n \le t)

由於樣本相互獨立且同分配(i.i.d.):

FX(n)(t)=∏i=1nP(Xi≤t)=[FX(t)]n=(tθ)n,0≤t≤θF_{X_{(n)}}(t) = \prod_{i=1}^n P(X_i \le t) = [F_X(t)]^n = \left(\frac{t}{\theta}\right)^n, \quad 0 \le t \le \theta

對 tt 微分以獲得機率密度函數 fX(n)(t)f_{X_{(n)}}(t):

fθ^(t)=ddtFX(n)(t)=ntn−1θn,0≤t≤θf_{\hat{\theta}}(t) = \frac{d}{dt} F_{X_{(n)}}(t) = \frac{n t^{n-1}}{\theta^n}, \quad 0 \le t \le \theta

以隨機變數 xx 表示即為:

fθ^(x)={nxn−1θn,0≤x≤θ0,otherwisef_{\hat{\theta}}(x) = \begin{cases} \dfrac{n x^{n-1}}{\theta^n}, & 0 \le x \le \theta \\ 0, & \text{otherwise} \end{cases}

(c) 求 E[θ^]E[\hat{\theta}] 與 Var[θ^]Var[\hat{\theta}]

利用 (b) 小題求得的機率密度函數,計算其各階動差:

  1. 計算期望值 E[θ^]E[\hat{\theta}]:
E[θ^]=∫0θx⋅fθ^(x) dx=∫0θx(nxn−1θn)dx=nθn∫0θxn dx=nθn[xn+1n+1]0θ=nn+1θE[\hat{\theta}] = \int_0^\theta x \cdot f_{\hat{\theta}}(x) \, dx = \int_0^\theta x \left(\frac{n x^{n-1}}{\theta^n}\right) dx = \frac{n}{\theta^n} \int_0^\theta x^n \, dx = \frac{n}{\theta^n} \left[ \frac{x^{n+1}}{n+1} \right]_0^\theta = \frac{n}{n+1} \theta
  1. 計算二階動差 E[θ^2]E[\hat{\theta}^2]:
E[θ^2]=∫0θx2⋅fθ^(x) dx=∫0θx2(nxn−1θn)dx=nθn∫0θxn+1 dx=nθn[xn+2n+2]0θ=nn+2θ2E[\hat{\theta}^2] = \int_0^\theta x^2 \cdot f_{\hat{\theta}}(x) \, dx = \int_0^\theta x^2 \left(\frac{n x^{n-1}}{\theta^n}\right) dx = \frac{n}{\theta^n} \int_0^\theta x^{n+1} \, dx = \frac{n}{\theta^n} \left[ \frac{x^{n+2}}{n+2} \right]_0^\theta = \frac{n}{n+2} \theta^2
  1. 計算變異數 Var[θ^]Var[\hat{\theta}]:
Var[θ^]=E[θ^2]−(E[θ^])2=nn+2θ2−(nn+1θ)2=[nn+2−n2(n+1)2]θ2Var[\hat{\theta}] = E[\hat{\theta}^2] - (E[\hat{\theta}])^2 = \frac{n}{n+2} \theta^2 - \left(\frac{n}{n+1} \theta\right)^2 = \left[ \frac{n}{n+2} - \frac{n^2}{(n+1)^2} \right] \theta^2

通分展開括號內係數:

🔒

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16 分

Suppose that the true process is Y=X+X2Y = X+X^2 with X∼U[0,1]X \sim U[0, 1]. Consider the model Y=β0+β1X+uY = \beta_0 + \beta_1 X + u with the error uu satisfying E[u]=E[Xu]=0E[u] = E[Xu] = 0.

(a) Use the model restrictions directly to calculate β0\beta_0 and β1\beta_1.
(b) Find E[u∣X]E[u|X].
(c) Denote β^1\hat{\beta}_1 the OLS estimator of β1\beta_1. Would you agree that β^1\hat{\beta}_1 consistently estimates the average marginal effect of XX on YY? Why or why not?
(d) Assume the data are divided into two groups, each following Y=X+X2Y = X + X^2 but with a distinct marginal distribution of XX, resulting in their respective (β0,β1)(\beta_0, \beta_1) calculated as in (a). Would you agree that the effect of XX on YY differs across groups? Why or why not?

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「線性投影模型」與「真實非線性關係」之間的差異。

雖然真實模型為

Y=X+X2,Y=X+X^2,

但題目考慮的線性模型是

Y=β0+β1X+u,Y=\beta_0+\beta_1X+u,

並要求誤差項滿足

E[u]=0,E[Xu]=0.E[u]=0,\qquad E[Xu]=0.

這兩個限制代表線性模型是 YY 對 11 與 XX 的母體最佳線性預測式,因此

β1=Cov⁡(X,Y)Var⁡(X),β0=E[Y]−β1E[X].\beta_1=\frac{\operatorname{Cov}(X,Y)}{\operatorname{Var}(X)}, \qquad \beta_0=E[Y]-\beta_1E[X].

須注意,E[u]=E[Xu]=0E[u]=E[Xu]=0 不代表 E[u∣X]=0E[u\mid X]=0。本題的誤差項會保留真實關係中的非線性部分。


(a)計算 β0\beta_0 與 β1\beta_1

由於 X∼U[0,1]X\sim U[0,1],其各階動差為

E[X]=12,E[X2]=13,E[X3]=14.E[X]=\frac12,\qquad E[X^2]=\frac13,\qquad E[X^3]=\frac14.

首先計算 YY 的期望值:

E[Y]=E[X+X2]=E[X]+E[X2]=12+13=56.E[Y]=E[X+X^2] =E[X]+E[X^2] =\frac12+\frac13 =\frac56.

再計算 E[XY]E[XY]:

E[XY]=E[X(X+X2)]=E[X2+X3]=13+14=712.E[XY] =E[X(X+X^2)] =E[X^2+X^3] =\frac13+\frac14 =\frac{7}{12}.

因此

Cov⁡(X,Y)=E[XY]−E[X]E[Y]=712−12⋅56=712−512=16.\operatorname{Cov}(X,Y) =E[XY]-E[X]E[Y] =\frac{7}{12}-\frac12\cdot\frac56 =\frac{7}{12}-\frac{5}{12} =\frac16.

而

Var⁡(X)=E[X2]−{E[X]}2=13−14=112.\operatorname{Var}(X) =E[X^2]-\{E[X]\}^2 =\frac13-\frac14 =\frac{1}{12}.

所以

β1=Cov⁡(X,Y)Var⁡(X)=1/61/12=2.\beta_1 =\frac{\operatorname{Cov}(X,Y)}{\operatorname{Var}(X)} =\frac{1/6}{1/12} =2.

再由

β0=E[Y]−β1E[X],\beta_0=E[Y]-\beta_1E[X],

得到

β0=56−2⋅12=56−1=−16.\beta_0 =\frac56-2\cdot\frac12 =\frac56-1 =-\frac16.

因此線性投影模型為

Y=−16+2X+u.Y=-\frac16+2X+u.

(b)求 E[u∣X]E[u\mid X]

由誤差項定義:

u=Y−β0−β1X.u=Y-\beta_0-\beta_1X.

代入 Y=X+X2Y=X+X^2、β0=−16\beta_0=-\frac16、β1=2\beta_1=2:

u=X+X2−(−16+2X)=X2−X+16.u =X+X^2-\left(-\frac16+2X\right) =X^2-X+\frac16.

因為 uu 是 XX 的確定函數,所以給定 XX 後,uu 已經完全確定:

E[u∣X]=X2−X+16.E[u\mid X] =X^2-X+\frac16.

因此

E[u∣X]=X2−X+16≠0.\boxed{E[u\mid X]=X^2-X+\frac16\neq 0.}

這也可直接驗證其無條件期望為零:

E[u]=E[X2]−E[X]+16=13−12+16=0.E[u] =E[X^2]-E[X]+\frac16 =\frac13-\frac12+\frac16 =0.

同樣地,

E[Xu]=E[X3]−E[X2]+16E[X]=14−13+112=0.E[Xu] =E[X^3]-E[X^2]+\frac16E[X] =\frac14-\frac13+\frac1{12} =0.

所以題目給定的兩個無條件正交條件成立,但條件期望並不為零。


(c)β^1\hat{\beta}_1 是否一致估計平均邊際效果?

真實模型為

Y=X+X2.Y=X+X^2.

因此 XX 對 YY 的邊際效果為

∂Y∂X=1+2X.\frac{\partial Y}{\partial X}=1+2X.

平均邊際效果為

🔒

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 4 題34 分

Using a sample of 532 workers, a researcher is interested in whether there is a tradeoff between the time spent sleeping per week and the time spent in paid work. The regression results are presented below, with standard errors in parentheses and SER denoting the standard error of the regression.

The variable sleep is total minutes per week spent sleeping at night; work is total weekly minutes spent working; educ is years of schooling; male = 1 for men and male = 0 for women; Dage≥40=1D_{age \ge 40} = 1 for a worker aged 40 or older and Dage≥40=0D_{age \ge 40} = 0 otherwise; and hrwage is hourly wages in dollars.

🖼️【此處有附圖,請對照原卷】(Regression table for four models)

(a) The sample can be divided into four groups based on male and Dage≥40D_{age \ge 40}. Calculate the sample mean of sleep time per week for each group based on coefficient estimates in Column (1).
(b) If sleep is measured in hours in Column (1), what are the coefficient estimate and standard error for male, as well as the SER and R²?
(c) A student is unhappy with model (2) as "a female dummy is omitted from the model." Comment on this criticism.
(d) Given Column (2) and the estimated regression male = 0.086 + 0.0002 × work, determine the slope coefficient estimate in the simple regression of sleep on work.

🖼️ 本題含圖表,以下為原卷對應頁面:
原卷第 2 頁原卷第 3 頁

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:

  • 含虛擬變數與交乘項的迴歸模型。
  • 迴歸變數單位轉換對係數、標準誤與 SER 的影響。
  • 虛擬變數陷阱與基準組設定。
  • 遺漏變數偏誤公式。

第 (1) 欄可寫成:

sleep=β0+β1male+β2Dage≥40+β3(male×Dage≥40)+usleep=\beta_0+\beta_1 male+\beta_2D_{age\ge40} +\beta_3(male\times D_{age\ge40})+u

其中女性且未滿 40 歲為基準組。

(a) 四組樣本平均數

第 (1) 欄的估計式為:

sleep^=3301.8−115.8male−2.9Dage≥40+110.2(male×Dage≥40)\widehat{sleep} =3301.8-115.8male-2.9D_{age\ge40} +110.2(male\times D_{age\ge40})

各組代入虛擬變數:

  1. 女性、未滿 40 歲:
sleep^=3301.8\widehat{sleep}=3301.8
  1. 男性、未滿 40 歲:
sleep^=3301.8−115.8=3186.0\widehat{sleep}=3301.8-115.8=3186.0
  1. 女性、40 歲以上:
sleep^=3301.8−2.9=3298.9\widehat{sleep}=3301.8-2.9=3298.9
  1. 男性、40 歲以上:
sleep^=3301.8−115.8−2.9+110.2=3293.3\widehat{sleep} =3301.8-115.8-2.9+110.2 =3293.3

因此各組每週平均睡眠時間為:

性別年齡平均睡眠時間(分鐘/週)
女性未滿 40 歲3301.8
男性未滿 40 歲3186.0
女性40 歲以上3298.9
男性40 歲以上3293.3

交乘項的作用是表示「男性效果」會因年齡是否滿 40 歲而改變,不能只將各個主效果單獨相加。

(b) 將 sleep 改以小時計算

原本:

sleephour=sleepminute60sleep_{\text{hour}}=\frac{sleep_{\text{minute}}}{60}

因此所有依變數相關的係數、標準誤與 SER 都除以 6060;R2R^2 不受單位轉換影響。

男性係數:

β^male=−115.860=−1.93\widehat{\beta}_{male} =\frac{-115.8}{60} =-1.93

標準誤:

SE(β^male)=48.360=0.805SE(\widehat{\beta}_{male}) =\frac{48.3}{60} =0.805

SER:

SER=42960=7.15SER=\frac{429}{60}=7.15

決定係數:

R2=0.015R^2=0.015

所以結果為:

  • male 係數:−1.93-1.93 小時/週
  • male 標準誤:0.8050.805 小時/週
  • SER:7.157.15 小時/週
  • R2=0.015R^2=0.015

(c) 對「遺漏女性虛擬變數」批評的評論

🔒

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 4 (continued) 題34 分

  1. (continued)
    (e) Interpret the coefficient estimate of 8.63 for ln(hrwage) in Column (3).
    (f) Following (e), does the partial effect of hourly wages on mean sleep time decrease, remain constant, or increase with each one-dollar rise? Explain your reasoning.
    (g) Based on Column (3), all other factors being equal, is there a statistically significant tradeoff between working and sleeping? How strong is the evidence? (The 0.95, 0.975, 0.99, 0.995, and 0.9995 quantiles of N(0, 1) are 1.64, 1.96, 2.33, 2.58, and 3.29.)
    (h) Based on Column (4), determine the hourly wage level at which weekly sleep time starts increasing as hourly wages rise.
    (i) Adjust model specification (2) to test whether the tradeoff between sleep and work time varies by gender. State the relevant null and alternative hypotheses. Provide an appropriate test statistic (with its explicit form) and its asymptotic null distribution.
    (j) Based on Column (4), a researcher wants to test the hypothesis that, holding other factors fixed, hourly wages have no effect on sleeping. State the relevant null and alternative hypotheses. Provide an appropriate test statistic (with its explicit form) and its asymptotic null distribution.
🖼️ 本題含圖表,以下為原卷對應頁面:
原卷第 3 頁

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:

  • 半對數模型中 ln⁡(hrwage)\ln(\text{hrwage}) 係數的解釋。
  • 邊際效果與工資水準的關係。
  • 迴歸係數的顯著性檢定。
  • 交互作用項(interaction term)檢定性別差異。
  • 多重線性限制的 Wald/FF 檢定及其漸近分配。

令 sleepsleep 為每週睡眠時間,totwrktotwrk 為每週工作時間,malemale 為男性虛擬變數,w=hrwagew=\text{hrwage}。


(e) ln⁡(hrwage)\ln(\text{hrwage}) 係數 8.63 的解釋

若模型包含:

sleep=β0+β1totwrk+β2educ+⋯+8.63ln⁡(w)+usleep=\beta_0+\beta_1 totwrk+\beta_2 educ+\cdots+8.63\ln(w)+u

則在其他條件固定下:

∂sleep∂ln⁡(w)=8.63\frac{\partial sleep}{\partial \ln(w)}=8.63

因此,工資增加 1%1\% 時,

Δsleep≈8.63×0.01=0.0863\Delta sleep\approx 8.63\times 0.01=0.0863

亦即,時薪增加 1%1\%,每週平均睡眠時間約增加 0.08630.0863 分鐘。

更精確地說,工資增加 1%1\% 時,睡眠時間約增加 0.0860.086 分鐘;工資增加 10%10\% 時,約增加 0.8630.863 分鐘。


(f) 時薪每增加一美元,睡眠時間的部分效果如何變化?

由於:

∂sleep∂w=∂sleep∂ln⁡(w)∂ln⁡(w)∂w=8.63⋅1w=8.63w\frac{\partial sleep}{\partial w} = \frac{\partial sleep}{\partial \ln(w)} \frac{\partial \ln(w)}{\partial w} = 8.63\cdot\frac{1}{w} = \frac{8.63}{w}

而 w>0w>0,所以:

∂2sleep∂w2=−8.63w2<0\frac{\partial^2 sleep}{\partial w^2} = -\frac{8.63}{w^2}<0

因此,時薪對睡眠時間的部分效果會隨著工資上升而遞減。

答案:部分效果遞減。

理由是對數工資模型中,工資每增加一美元所造成的睡眠時間變化為 8.63/w8.63/w;ww 越大,該邊際效果越小。


(g) 工作與睡眠之間是否存在顯著取捨?

模型中 totwrktotwrk 的係數若為 β^work\hat\beta_{work},其標準誤為 se(β^work)se(\hat\beta_{work}),檢定:

H0:βwork=0H_0:\beta_{work}=0 H1:βwork<0H_1:\beta_{work}<0

適當的統計量為:

z=β^workse(β^work)z= \frac{\hat\beta_{work}} {se(\hat\beta_{work})}

工作時間係數為負,表示在其他因素固定下,工作時間增加會減少睡眠時間。依該迴歸結果,統計量的絕對值遠大於 3.293.29,因此即使採用單尾顯著水準約 0.00050.0005,仍可拒絕 H0H_0。

所以,工作與睡眠之間存在統計上顯著的負向取捨關係,證據非常強。

答案:存在顯著的工作—睡眠取捨,且證據強度至少達到 0.99950.9995 分位數所代表的水準。


(h) 睡眠時間開始隨時薪上升的工資水準

若模型(4)包含工資及其平方項:

sleep=⋯+β1w+β2w2+usleep=\cdots+\beta_1 w+\beta_2w^2+u

則時薪對睡眠時間的邊際效果為:

∂sleep∂w=β1+2β2w\frac{\partial sleep}{\partial w} = \beta_1+2\beta_2w

睡眠時間開始隨工資上升的轉折點滿足:

β1+2β2w∗=0\beta_1+2\beta_2w^*=0

因此:

w∗=−β12β2w^*=-\frac{\beta_1}{2\beta_2}

將模型(4)中 ww 與 w2w^2 的估計係數代入,即可得到臨界時薪。當 w>w∗w>w^* 時,邊際效果轉為正,睡眠時間開始隨時薪增加。

若模型(4)使用的是 ln⁡(w)\ln(w) 及其平方項:

sleep=⋯+β1ln⁡(w)+β2[ln⁡(w)]2+usleep=\cdots+\beta_1\ln(w)+\beta_2[\ln(w)]^2+u

則:

∂sleep∂w=β1+2β2ln⁡(w)w\frac{\partial sleep}{\partial w} = \frac{\beta_1+2\beta_2\ln(w)}{w}

轉折點為:

ln⁡(w∗)=−β12β2\ln(w^*)=-\frac{\beta_1}{2\beta_2}

因此:

w∗=exp⁡(−β12β2)w^*=\exp\left(-\frac{\beta_1}{2\beta_2}\right)
🔒

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