115 年 國立臺灣大學商學研究所乙組《統計學(H)》
第 1 題
- The study uses Google search data and comparison search terms to test a behavioral hypothesis about goal initiation. Based on the research design and background information provided, which of the following statements is NOT correct?
A) Dieting commonly begins on psychologically salient temporal landmarks such as Mondays, the beginning of a new month, or the start of a new year.
B) "Diet" is an appropriate example of goal pursuit because it reflects a common, aspirational, self-regulatory behavior that individuals frequently intend to initiate at meaningful temporal landmarks.
C) The authors include search terms such as "news" and "weather" in their regression analysis primarily to increase the sample size and improve statistical power.
D) Google search activity for the term "diet" serves as a proxy for individuals' interest in initiating dieting behavior, even though it may not directly measure actual dieting outcomes.
E) Comparing "diet" searches with non-aspirational terms helps rule out the possibility that observed temporal patterns are driven by general Internet search behavior rather than goal-related motivation.
登入後即可作答並保存紀錄。
本題旨在考驗考生對研究設計與背景資訊的理解能力。題目要求選出一個「不正確」的敘述。
首先,我們來逐一檢視各選項:
A) 根據研究背景,人們在「新的開始」(例如新的一年、新的一個月、新的一週)等時間點更容易採取行動達成目標。這與「節日效應」(fresh start effect)的概念一致,而節日效應正是指人們在心理上重要的時間節點(temporal landmarks)更有動力追求目標。因此,將節食(dieting)視為在這些時間點開始是很合理的。此敘述正確。
B) 「節食」是一種常見的、有抱負的、自我調節的行為,人們經常會在重要的時間節點開始。這與研究中提到的「時間地標」和「抱負行為」(aspirational behavior)相符。因此,「節食」確實是個適當的目標追求例子。此敘述正確。
第 2 題
- Based on the regression results reported in the table, and interpreting all statements as referring to model-predicted
search volume while holding other regressors constant, which of the following statements is NOT correct?
A) The model implies that the predicted difference in search volume for "diet" between Monday and Tuesday is the
same as the predicted difference between Saturday and Sunday.
B) The model predicts that the search volume for "diet" tends to decline as the month progresses from its beginning
toward its end.
C) On average, the model predicts that daily search volume for "diet" decreases as the week progresses from Monday
to Sunday.
D) The regression results indicate a statistically significant difference in average "diet" search volume between the
beginning of the year and the middle of the year.
E) After controlling for month-of-year effects, temporal landmarks such as the start of a new week have no additional
association with goal-related search behavior.
登入後即可作答並保存紀錄。
本題要求考生根據提供的迴歸分析結果表格,判斷哪個敘述是「不正確」的。所有敘述都應在「其他迴歸變數固定」的條件下,解釋模型預測的搜尋量。
首先,我們需要理解表格中的迴歸係數。
「Days since the start of the week」:此變數代表星期幾(1=星期一,7=星期日)。
「Days since the start of the month」:此變數代表該月已過去的天數(1 到 31)。
「Months since the start of the year」:此變數代表該年已過去的月份數(1=一月,12=十二月)。
現在我們逐一檢視選項:
A) 關於「Diet」的搜尋量,星期一(Day 1)到星期二(Day 2)的差異,以及星期六(Day 6)到星期日(Day 7)的差異。
變數「Days since the start of the week」的係數為 。這表示搜尋量隨著星期幾的增加而「減少」。
星期一到星期二的差異:預測值變化為 ,其中 為包含此變數的函數。由於係數是 ,這個差異是 。
星期六到星期日的差異:預測值變化為 ,其中 為包含此變數的函數。這個差異是 。
因此,模型預測的搜尋量在星期一到星期二之間的差異,與星期六到星期日之間的差異是相同的(都是 )。此敘述正確。
B) 關於「Diet」的搜尋量,隨著月份的推進(從月初到月底)。
變數「Days since the start of the month」的係數為 。
第 3 題
- Based on the regression results reported in the table, and holding the other regressors constant, how many of the
following six statements are supported by the sign and statistical significance of the estimated coefficients for the
control search terms ("news" and "weather")?
- The predicted search volume for "news" significantly decreases as the week progresses from Monday to Sunday.
- The predicted search volume for "news" significantly decreases as the month progresses from the beginning to the end of
the month. - The predicted search volume for "news" significantly decreases as the year progresses from the beginning to the end of
the year. - The predicted search volume for "weather" significantly decreases as the week progresses from Monday to Sunday.
- The predicted search volume for "weather" significantly decreases as the month progresses from the beginning to the end of
the month. - The predicted search volume for "weather" significantly decreases as the year progresses from the beginning to the end of
the year.
A) None of the statements are supported.
B) Only one of the statements is supported.
C) Two of the statements are supported.
D) Three of the statements are supported.
E) More than three statements are supported.
登入後即可作答並保存紀錄。
本題要求根據表格中「news」和「weather」的迴歸係數,判斷有多少個敘述是正確的。這裡我們需要關注這些係數的符號(sign)和統計顯著性(statistical significance)。
首先,我們來檢視「news」搜尋詞的迴歸結果:
-
"The predicted search volume for 'news' significantly decreases as the week progresses from Monday to Sunday."
- 變數: "Days since the start of the week"
- 係數:
- 顯著性: * (p < .05)
- 符號: 負
- 結論: 係數為負且顯著,表示搜尋量隨著星期幾增加而顯著下降。此敘述正確。
-
"The predicted search volume for 'news' significantly decreases as the month progresses from the beginning to the end of the month."
- 變數: "Days since the start of the month"
- 係數:
- 顯著性: * (p < .05)
- 符號: 負
- 結論: 係數為負且顯著,表示搜尋量隨著該月已過天數增加而顯著下降。此敘述正確。
-
"The predicted search volume for 'news' significantly decreases as the year progresses from the beginning to the end of the year."
- 變數: "Months since the start of the year"
- 係數:
- 顯著性: 無標示(與 p > .05 相關)
- 符號: 負
- 結論: 係數為負,但非統計顯著。因此,不能說搜尋量「顯著」下降。此敘述不正確。
第 4 題
- Based on the regression results reported in the table and the background information provided, which of the following
charts is most likely to represent the time series pattern of the daily Google search volume for the term "diet" over the
period January 1, 2012 to December 31, 2014?
A)
100
75
50
25
Jan 1, 2012
Dec 2, 2012
Nov 3, 2013
Oct 5, 2014
B)
100
75
50
25
Jan 1, 2012
Dec 2, 2012
Nov 3, 2013
Oct 5, 2014
C)
100
75
50
25
Jan 1, 2012
Dec 2, 2012
Nov 3, 2013
Oct 5, 2014
D)
100
75
50
25
Jan 1, 2012
Dec 2, 2012
Nov 3, 2013
Oct 5, 2014
E) It is not possible to rule out any of the charts, because the regression analysis is based on data only through June 30,
2012, and no evidence is available on search behavior beyond that period.
登入後即可作答並保存紀錄。
核心觀念
本卷第 1–4 題共用同一份研究背景:研究以 Google 搜尋量檢驗「新開始效應」(fresh start effect)——人們傾向在具有心理意義的時間地標開始追求目標,例如新的一週、新的一個月、新年開始節食。依這個假說與迴歸設計,「diet」的搜尋量在週初、月初、年初較高,並隨著一週、一個月、一年往後推進而下降。
解題方法
把「年初最高、之後在一年中逐漸下降、下一個新年再跳升」的型態對到四張圖:
- 圖 A 在 2012 年 1 月 1 日從最高點開始,全年大致緩降;接近 2013 年、2014 年元旦時突然跳升回高點,之後又逐漸下滑,呈現每年一次的鋸齒狀。
- 圖 C 雖然也有年度週期,但高峰出現在年中(春夏之際),而且上升、下降都很平滑,相位與「新年開始節食」不符。
選項分析
第 5 題
- Based on the background information and the figure described above, which of the following statements is correct?
A) A larger proportion of respondents report that obedience is especially important for children than report that
independence is especially important.
B) Because the data are nationally representative and span many countries and years, the observed relationship
between family income and parental values can be interpreted as causal.
C) The figure provides evidence of an association between family income and the probability that respondents
emphasize independence or obedience as important child qualities.
D) Each respondent is restricted to selecting only one child quality in the survey.
E) None of the statements above can be concluded based on the information provided.
登入後即可作答並保存紀錄。
核心觀念
本題考查「關聯不等於因果」及圖表中條件機率的解讀。題目將「重視獨立」與「重視服從」分別記為二元指標,並比較不同家庭所得組別中,受訪者重視這兩項特質的比例。
圖中橫軸是家庭所得組別,縱軸是重視該特質的機率。若各所得組別的機率隨所得改變,表示家庭所得與該特質的重視程度存在關聯;但僅憑觀察資料,不能推論家庭所得造成了這種變化。
解題方法
先比較兩張圖的趨勢:家庭所得提高時,重視獨立的機率整體上升;重視服從的機率整體下降。因此,圖形顯示家庭所得與受訪者強調的子女特質之間有關聯,符合選項 C。
接著檢查其他選項是否能由圖中資訊推出。圖呈現各所得組別內的機率,沒有提供各所得組別的人數比例,因此不能直接比較全體受訪者中重視獨立與重視服從的總比例。
選項分析
第 6 題
- Suppose the researcher conducts a correlation analysis between each outcome variable (independence and obedience)
and family income. Based on the patterns shown in the figure, which of the following results is most plausible?
A) The correlation coefficient between independence and family income is significantly positive, and the correlation
coefficient between obedience and family income is significantly positive.
B) The correlation coefficient between independence and family income is significantly positive, and the correlation
coefficient between obedience and family income is significantly negative.
C) The correlation coefficient between independence and family income is significantly negative, and the correlation
coefficient between obedience and family income is significantly positive.
D) The correlation coefficient between independence and family income is not significantly different from zero, while
the correlation coefficient between obedience and family income is significantly negative.
E) A correlation analysis cannot be conducted because the outcome variables are binary indicators.
登入後即可作答並保存紀錄。
本題要求根據圖表中家庭收入與親子特質(獨立性、服從性)的關係模式,判斷哪個相關分析結果最為合理。
首先,我們回顧圖表中的模式:
- Independence (獨立性):隨著家庭收入 (Family income) 的增加,強調獨立性的機率呈現 上升趨勢。
- Obedience (服從性):隨著家庭收入 (Family income) 的增加,強調服從性的機率呈現 下降趨勢。
相關係數(Correlation coefficient)衡量兩個變數之間線性關係的強度和方向。
- 正相關係數表示一個變數增加時,另一個變數也傾向於增加。
- 負相關係數表示一個變數增加時,另一個變數傾向於減少。
- 顯著的相關係數表示觀察到的關係不太可能是由隨機抽樣誤差造成的。
現在我們來評估各選項:
A) "The correlation coefficient between independence and family income is significantly positive, and the correlation coefficient between obedience and family income is significantly positive."
- 圖表顯示獨立性與收入是正相關趨勢,這部分合理。
- 但服從性與收入是負相關趨勢,此處聲稱服從性與收入是「顯著正相關」,與圖表模式相反。此選項不合理。
B) "The correlation coefficient between independence and family income is significantly positive, and the correlation coefficient between obedience and family income is significantly negative."
- 獨立性與家庭收入:圖表顯示隨著收入增加,獨立性機率上升,這符合「顯著正相關」。
- 服從性與家庭收入:圖表顯示隨著收入增加,服從性機率下降,這符合「顯著負相關」。
- 這個選項的描述與圖表顯示的模式完全一致。此選項最為合理。
第 7 題
- Suppose the researcher estimates separate logistic regression models for each outcome variable—independence and
obedience—using family income as the explanatory variable. Based on the patterns shown in the figure, which of the
following results is most plausible?
A) The estimated coefficient on family income is positive and statistically significant in both the independence and
obedience regressions.
B) The estimated coefficient on family income is positive and statistically significant in the independence regression,
and negative and statistically significant in the obedience regression.
C) The estimated coefficient on family income is negative and statistically significant in the independence regression,
and positive and statistically significant in the obedience regression.
D) The estimated coefficient on family income is negative and statistically significant in both the independence and
obedience regressions.
E) Logistic regression is inappropriate in this context because family income is measured on an ordinal scale.
登入後即可作答並保存紀錄。
本題要求根據圖表中家庭收入與親子特質(獨立性、服從性)的關係模式,判斷使用羅吉斯迴歸(logistic regression)分析時,家庭收入作為解釋變數的估計係數最可能為何。
首先,我們需要理解羅吉斯迴歸中係數的意義:
- 在羅吉斯迴歸中,我們預測的是某個結果發生的機率(或 log-odds)。
- 如果一個解釋變數的係數是正的且統計顯著,表示該變數的增加會顯著增加結果發生的機率(或 log-odds)。
- 如果一個解釋變數的係數是負的且統計顯著,表示該變數的增加會顯著降低結果發生的機率(或 log-odds)。
家庭收入 (Family income) 是自變數,親子特質(Independence, Obedience)是依變數。依變數是二元的。
我們再次檢視圖表中的模式:
- Independence (獨立性):隨著家庭收入 (Family income) 的增加,強調獨立性的機率 上升。
- Obedience (服從性):隨著家庭收入 (Family income) 的增加,強調服從性的機率 下降。
現在我們來評估各選項:
A) "The estimated coefficient on family income is positive and statistically significant in both the independence and obedience regressions."
- 獨立性:收入增加,機率上升,係數應為正且顯著。這部分合理。
- 服從性:收入增加,機率下降,係數應為負且顯著。此處聲稱係數為正且顯著,與圖表模式相反。此選項不合理。
B) "The estimated coefficient on family income is positive and statistically significant in the independence regression, and negative and statistically significant in the obedience regression."
第 8 題
- In the control condition, 62.5% of buyers purchased unattractive produce and 56.0% purchased attractive produce. In
the "ugly" label condition, 81.6% of buyers purchased unattractive produce and 26.5% purchased attractive produce.
(Percentages do not sum to 100% because some buyers purchased both types of produce.) Which of the following
statement is correct?
A) A logistic regression can be used to examine whether labeling unattractive produce as "ugly" (versus the control
label) affects buyers' likelihood of purchasing unattractive produce.
B) A linear regression can be used to examine whether labeling unattractive produce as "ugly" (versus the control
label) affects buyers' likelihood of purchasing unattractive produce.
C) The reported percentages alone are sufficient to conclude that labeling unattractive produce as "ugly" caused a
change in purchasing behavior.
D) Labeling unattractive produce as "ugly" (versus the control label) significantly increased buyers' likelihood of
purchasing attractive produce.
E) Labeling unattractive produce as "ugly" (versus the control label) significantly decreased buyers' likelihood of
purchasing unattractive produce.
登入後即可作答並保存紀錄。
核心觀念
本題考查二元應變數的統計建模(Binary Outcome Modeling)與統計推論的基本原則:
- 二元邏輯斯迴歸(Binary Logistic Regression):當依變數(Dependent Variable)為二元分類變數(例如:買家「有購買」或「未購買」醜蔬果,以 表示)時,一般線性迴歸(OLS)會違反誤差項常態性、同質變異數假設,且預測機率可能超出 區間。因此,應採用邏輯斯迴歸,透過 Logit 連結函數 檢驗自變數(標籤處理組 vs. 對照組)對購買機率的影響與勝算比(Odds Ratio)。
- 敘述統計 vs. 統計推論(Descriptive vs. Inferential Statistics):僅憑樣本百分比的差異(敘述統計),在未提供樣本數(Sample Size)、標準誤(Standard Error)或檢定統計量(-value)的情況下,無法直接宣稱具有統計顯著性(Statistical Significance)或確立因果關係。
解題方法
- 確認應變數性質:
探討標籤("ugly" 標籤組 vs. 對照組)是否會影響買家「是否購買不美觀蔬果(unattractive produce)」。買家的決策為二元結果(二分變數:, ),適合以二元邏輯斯迴歸(Logistic Regression)進行檢定。 - 檢視統計推論前提:
題幹僅給出購買比例(Control: 62.5% vs. "ugly": 81.6%),未提供樣本數 與假說檢定結果,因此不能直接宣稱「顯著增加/減少(significantly increased/decreased)」或單憑比例即確立因果結論。
選項分析
第 9 題
- Based on the figure showing spending as the dependent variable, which of the following analytical approaches is most
appropriate for analyzing the data?
A) An independent-samples T test comparing average spending between the "ugly" label condition and the control
condition.
B) A paired t test comparing spending on unattractive produce and attractive produce, ignoring label condition.
C) A linear regression of spending on a single dummy variable indicating whether the produce is labeled "ugly,"
without accounting for produce appearance.
D) A mixed-effects (or mixed-design) regression/ANOVA of spending, with label condition ("ugly" vs. control) as a
between-subjects factor and produce appearance (unattractive vs. attractive) as a within-subject factor.
E) No formal statistical analysis is appropriate because some buyers purchased both attractive and unattractive
produce.
登入後即可作答並保存紀錄。
本題要求根據圖表,判斷最適合分析「購買支出」(spending)作為依變數的數據的統計方法。圖表顯示了「Mean Spending by Buyers (CAD)」在四種組合下的情況:
- Control Label, Spending on Unattractive Produce
- Control Label, Spending on Attractive Produce
- "Ugly" Label, Spending on Unattractive Produce
- "Ugly" Label, Spending on Attractive Produce
這是一個典型的實驗設計,其中有兩個因子(factor):
- Label Condition (標籤條件):分為 "Control" 和 "Ugly" 兩個水平。這是受試者之間(between-subjects)的因子,因為一個消費者被分配到 control 組或 ugly 組。
- Produce Appearance (農產品外觀):分為 "Unattractive" 和 "Attractive" 兩個水平。這是受試者內部(within-subject)的因子,因為同一批消費者(在某個時間段內)被觀察了在 unattractive 和 attractive 產品上的支出。
圖表中的數據是「Mean Spending by Buyers」。
現在我們逐一檢視選項:
A) "An independent-samples T test comparing average spending between the 'ugly' label condition and the control condition."
- 這只比較了兩個組別(ugly vs. control)的平均支出,但忽略了農產品外觀(unattractive vs. attractive)這個重要的因子。因此,它沒有考慮到兩因子之間的交互作用,也不夠全面。
B) "A paired t test comparing spending on unattractive produce and attractive produce, ignoring label condition."
- 這比較了同一批消費者在 unattractive 和 attractive 產品上的支出,但忽略了標籤條件(ugly vs. control)這個因子。因此,它也無法捕捉到實驗的全部結構。
C) "A linear regression of spending on a single dummy variable indicating whether the produce is labeled 'ugly,' without accounting for produce
第 10 題
- Based on the figure showing mean spending by buyers, which of the following statements is correct? (Interpret dollar
amounts as sample means, and statistical significance as indicated in the figure.)
A) In the control condition, buyers spent on average 3.41 on attractive produce.
B) In the "ugly" label condition, buyers spent on average 1.78 on attractive produce.
C) The figure indicates that the main effect of label condition ("ugly" vs. control) on total spending is not statistically
significant.
D) The figure indicates a statistically significant interaction effect between label condition ("ugly" vs. control) and
produce appearance (attractive vs. unattractive).
E) All of the statements above are correct.
登入後即可作答並保存紀錄。
核心觀念
本題考查 因子設計中的:
- 兩個主效果:標籤條件、農產品外觀。
- 交互作用效果:標籤條件是否改變外觀對消費金額的影響。
- 各實驗條件下的樣本平均數與圖示的統計顯著性。
令消費金額為 ,標籤條件為「控制組」與「醜標籤組」,外觀為「不具吸引力」與「具吸引力」。
解題方法
依圖中四組平均數:
| 標籤條件 | 不具吸引力 | 具吸引力 |
|---|---|---|
| 控制組 | ||
| 醜標籤組 |
控制組中,具吸引力與不具吸引力產品的平均消費差異為
醜標籤組中,具吸引力與不具吸引力產品的平均消費差異為
兩組的外觀效果方向相反,交互作用對比為
因此,標籤條件會改變外觀對消費金額的影響;依圖中的顯著性標記,此交互作用達統計顯著。
第 11 題
- A scatter diagram reveals a strong positive linear relationship between two variables. Which of the following
numerical techniques will not give us more detailed information about this relationship?
A) Coefficient of correlation
B) Coefficient of variation
C) Coefficient of determination
D) All of these choices help us describe this relationship
E) None of these choices helps us describe this relationship
登入後即可作答並保存紀錄。
本題探討在散佈圖顯示兩個變數之間存在強烈的正向線性關係時,哪些數值技術「無法」提供更多關於此關係的詳細資訊。
首先,我們理解題目所指的「強烈的正向線性關係」。這意味著散佈圖上的點大致沿著一條從左下到右上的直線分佈,且點與直線的距離很近。
接著,我們分析各選項提供的數值技術:
A) Coefficient of correlation (相關係數):
- 相關係數(例如 Pearson correlation coefficient, )直接衡量兩個變數之間線性關係的強度和方向。
- 對於一個強烈的正向線性關係,相關係數會接近 +1。
- 了解相關係數的值(例如 0.9 或 0.95)能提供關於關係強度的精確數值資訊。因此,它提供了關於此關係的詳細資訊。
B) Coefficient of variation (變異係數):
- 變異係數(CV)是標準差(Standard Deviation, )與平均值(Mean, )的比值,通常表示為 。
- 變異係數主要用於衡量一個數據集的相對變異程度,即變異數相對於平均值的比例。它常用於比較不同平均值水平但變異程度相似的數據集,或者反之。
第 12 題
- Which of the following is true regarding the sampling distribution of the mean for a large sample size? Assume the
population distribution is not normal.
A) It has the same shape, mean, and standard deviation as the population.
B) It has the same mean and standard deviation as the population, but a different shape.
C) It has the same shape and mean as the population, but a different standard deviation.
D) It has the same mean as the population, but a different shape and standard deviation.
E) It has a different shape, mean, and standard deviation compared to the population.
登入後即可作答並保存紀錄。
本題考查中央極限定理(Central Limit Theorem, CLT)的應用。題目假設母體分佈(population distribution)不是常態分佈,但樣本數(sample size)足夠大(large sample size)。我們需要判斷樣本平均數的抽樣分佈(sampling distribution of the mean)的特性。
中央極限定理的核心內容是:
- 平均值 (Mean):無論母體分佈為何,當樣本數足夠大時,樣本平均數的抽樣分佈的平均值會趨近於母體平均值 ()。即 。
- 標準差 (Standard Deviation):樣本平均數的抽樣分佈的標準差(稱為標準誤, standard error)是母體標準差 () 除以樣本數的平方根 ()。即 。當 夠大時,這個標準誤會趨於一個固定值,且這個值通常與母體標準差 不同(除非 )。
- 形狀 (Shape):無論母體分佈為何,當樣本數足夠大時,樣本平均數的抽樣分佈會趨近於常態分佈(Normal Distribution)。
現在我們根據 CLT 的結論來分析選項:
A) "It has the same shape, mean, and standard deviation as the population."
- 形狀:抽樣分佈趨近常態,母體分佈非常態。形狀不同。
- 平均值:抽樣分佈平均值等於母體平均值。這部分正確。
- 標準差:抽樣分佈標準差是 ,母體標準差是 。除非 ,否則標準差不同。
- 因此,此選項錯誤。
B) "It has the same mean and standard deviation as the population, but a different s
第 13 題
- In a one-way ANOVA, error variability is computed as the sum of the squared errors, SSE, for all values of the
response variable. This variability is the:
A) total variation
B) within-treatment variation
C) covariation
D) between-treatment variation
E) None of the choices above
登入後即可作答並保存紀錄。
本題考查單因子變異數分析(One-way ANOVA)中各類變異數的定義。題目中提到「error variability is computed as the sum of the squared errors, SSE」。SSE 代表 Sum of Squared Errors,也就是殘差平方和。
在單因子 ANOVA 中,總變異(Total Variation, SST)被分解為組間變異(Between-treatment Variation, SSB 或 SSG)和組內變異(Within-treatment Variation, SSE 或 SSW)。
- SST (Total Sum of Squares):所有觀測值與總平均值之間差異的平方和。衡量了整個數據集的總變異。
- SSB (Sum of Squares Between Groups):各組平均值與總平均值之間差異的平方和,乘以各組樣本數。
第 14 題
- If we are testing for the difference between the means of two independent populations with equal variances, samples
of n₁ = 15 and n₂ = 15 are taken, then the number of degrees of freedom is equal to
A) 13
B) 14
C) 15
D) 28
E) 29
登入後即可作答並保存紀錄。
本題考查獨立樣本 t 檢定(Independent samples t-test)中,當兩母體變異數相等時的自由度(degrees of freedom, df)計算方法。
當我們檢定兩個獨立樣本均值的差異,並且假設兩個母體變異數相等 () 時,我們通常採用合併變異數(pooled variance)的方法。合併變異數的計算公式為:
第 15 題
- A candidate can earn the Chartered Financial Analyst (CFA) designation only after passing three annual
examinations: CFA I, II, and III. Candidates who pass an examination are eligible to take the examination for the
next level in the following year. The pass rates for levels I, II, and III are 57%, 73%, and 85%, respectively. Suppose
that 3,000 candidates take the level I examination, 2,500 take the level II examination, and 2,000 take the level III
examination.
Now, consider a randomly selected candidate who is going to take an examination. If the candidate manages to pass
the examination, what is the probability that he or she is now effectively a CFA already? Choose the closest answer.
A) 0.220
B) 0.228
C) 0.243
D) 0.325
E) 0.348
登入後即可作答並保存紀錄。
核心觀念
本題考查條件機率與加權計算。
候選人通過哪一級考試的機率,不能只看三個通過率的平均值,必須先考慮各級考試的應試人數。題目給定:
- CFA I: 人,通過率
- CFA II: 人,通過率
- CFA III: 人,通過率
只有通過 CFA III 後,候選人才算「effectively a CFA already」。因此要求的是:
解題方法
先計算各級考試的預期通過人數:
在所有通過考試的候選人中,總人數為:
其中,通過 CFA III、因而已具備 CFA 身分者有 人。因此:
取最接近的選項:
選項分析
第 16 題
- A recent election study showed that 45% of registered American voters are Democrats, 42% are Republicans, and
13% are others. If two registered voters are selected at random, provided that they are either Democrats or
Republicans, what is the probability that they have the same party affiliation? Choose the closest answer.
A) 0.495
B) 0.498
C) 0.501
D) 0.504
E) 0.507
登入後即可作答並保存紀錄。
本題考驗條件機率和組合的應用。題目要求計算在一個前提條件下,兩個隨機選取的選民具有相同政黨歸屬的機率。
首先,我們整理給定的資訊:
- Democrats (D): 45%
- Republicans (R): 42%
- Others (O): 13%
總計:45% + 42% + 13% = 100%
前提條件是:「provided that they are either Democrats or Republicans」。這意味著我們只考慮 D 或 R 的選民,而排除 O。
在排除 O 之後,D 和 R 的比例需要重新歸一化(normalize)。
新的比例:
- Democrats (D):
- Republicans (R):
(檢查:)
現在,我們從這些 D 或 R 的選民中隨機選取兩位。
我們要計算他們具有相同政黨歸屬的機率。這意味著:
- 選出的兩位都是 Democrats。
- 選出的兩位都是 Republicans。
我們假設選取是「有放回」或「獨立」進行的,因為樣本量很大,這種近似是合理的。如果是非放回抽樣,則需要用組合來計算。題目中「selected at random」通常暗示獨立抽樣或足夠大的總體。我們採用獨立抽樣的假設。
情況 1:兩位都是 Democrats。
機率 =
機率 =
情況 2:兩位都是 Republicans。
機率 =
機率 =
第 17 題
- What statistical test do we use to test the validity of a regression model?
A) F-test
B) z-test
C) t-test
D) chi-square test
E) binomial test
登入後即可作答並保存紀錄。
本題考查迴歸模型有效性檢定的統計方法。
迴歸模型(特別是多元迴歸)的有效性通常是指模型是否整體上比一個只包含截距項的簡單模型(null model)更能解釋依變數的變異。這通常通過檢定迴歸係數的整體顯著性來實現。
在線性迴歸中,檢定所有斜率係數(不包括截距項)是否同時為零的統計檢定是 F-test。
- 虛無假設 : (所有斜率係數皆為零)。
- 對立假設 : 至少有一個 。
這個 F-test 的統計量是基於迴歸平方和 (SSR) 和殘差平方和 (SSE) 計算的:
其中 是自變數的數量, 是樣本數。
第 18 題
- Chlorofluorocarbons (CFCs) damage the ozone layer. A city is planning to ban the use of CFCs in vehicle air
conditioners. Suppose there are 5.8 million vehicles in the city, and you guess 20% of them use CFCs in air
conditioners. How many vehicles will you need to sample to estimate the number of vehicles using CFCs with a
margin of error of 0.3 million vehicles, with 95% confidence? Choose the closest answer.
A) 200
B) 210
C) 220
D) 230
E) 300
登入後即可作答並保存紀錄。
核心觀念
本題考查「母體比例的估計」與「樣本數決定」。
令:
- 母體總數: 百萬輛
- 使用 CFCs 的比例:
- 不使用 CFCs 的比例:
- 信賴水準:
- 估計車輛總數的誤差上限: 百萬輛
母體比例的估計誤差為
95% 信賴水準的標準常態臨界值為
估計比例所需樣本數公式為
其中 為預估比例,。
解題方法
題目要估計的是「使用 CFCs 的車輛數」,而不是直接估計比例。因此先將車輛數的誤差 百萬輛換算為比例誤差:
代入樣本數公式:
分別計算:
因此
第 19 題
- You are fitting a multiple linear regression model. Which of the following is always true about the adjusted R² of
this model?
A) It will increase when a predictor is added.
B) It will increase when a predictor is added and if the added predictor's p-value is < 0.05.
C) It will increase when a predictor is added and if the added predictor is not correlated with all existing predictors.
D) It will increase whenever R² increases.
E) It increases if the t-statistic for the added predictor satisfies |t| > 1.
登入後即可作答並保存紀錄。
核心觀念
本題考查新增解釋變數後,決定係數 與調整後決定係數 的變化。
若樣本數為 ,模型中有 個解釋變數,則
新增一個解釋變數時,普通 不會下降,但 只有在新增變數帶來的解釋力足夠大時才會上升。
解題方法
設原模型有 個解釋變數,新增一個變數後,殘差平方和由 降為 。
新增變數後調整後決定係數上升的條件為
整理得
左側即為新增變數的部分 統計量。當只新增一個變數時,
因此
所以調整後 的精確判準是新增變數的 統計量滿足 。
選項分析
A) It will increase when a predictor is added.
錯誤。新增變數一定使普通 不下降,但調整後 會因自由度懲罰而下降、維持不變或上升。
B) It will increase when a predictor is added and if the added predictor's p-value is < 0.05.
第 20 題
- The difference between a sample mean and the population mean is called:
A) selection bias
B) survival error
C) sampling error
D) non-sampling error
E) nonresponse error
登入後即可作答並保存紀錄。
核心觀念
抽樣誤差(sampling error)是樣本統計量與其對應母體參數之間的差異。題目中的樣本平均數為 ,母體平均數為 ,兩者之差可寫為:
這個差異來自只觀察母體中的部分個體,而非整個母體。
解題方法
辨認題目中的兩個量:sample mean 是樣本統計量 ;population mean 是母體參數 。樣本統計量與母體參數之間的差異,稱為抽樣誤差,因此選項 C 正確。
選項分析
- A) selection bias(選擇偏誤):樣本的選取方式造成特定群體被過度或不足代表,屬於系統性偏誤,不是題目所述差異的名稱。