# A tibble: 1 × 2
Female Age
<dbl> <dbl>
1 0.55 38
3/ 無作為化比較試験
関西大学総合情報学部
内生性の存在 \(\rightarrow\) ATE推定値の信頼性\(\downarrow\)
例) やる気のある学生だけがソンさんの講義を履修した場合
内生性は因果推論の敵! どうすれば…?
\(\downarrow\)
無作為割当 (Random Assignment)無作為割当 (Random Assignment)
コインを投げ、表( \(H\) )なら統制群、裏( \(T\) )なら処置群に割当
| ID | Female | Age | ID | Female | Age | |
|---|---|---|---|---|---|---|
| 1 | 1 | 31 | 11 | 0 | 38 | |
| 2 | 1 | 41 | 12 | 1 | 29 | |
| 3 | 0 | 31 | 13 | 0 | 21 | |
| 4 | 1 | 46 | 14 | 0 | 26 | |
| 5 | 1 | 37 | 15 | 1 | 36 | |
| 6 | 1 | 37 | 16 | 1 | 40 | |
| 7 | 0 | 30 | 17 | 0 | 50 | |
| 8 | 1 | 46 | 18 | 0 | 42 | |
| 9 | 1 | 56 | 19 | 0 | 29 | |
| 10 | 0 | 47 | 20 | 1 | 47 |
コイン投げの結果
| ID | Female | Age | Coin | ID | Female | Age | Coin | |
|---|---|---|---|---|---|---|---|---|
| 1 | 1 | 31 | H | 11 | 0 | 38 | H | |
| 2 | 1 | 41 | T | 12 | 1 | 29 | T | |
| 3 | 0 | 31 | T | 13 | 0 | 21 | H | |
| 4 | 1 | 46 | T | 14 | 0 | 26 | T | |
| 5 | 1 | 37 | H | 15 | 1 | 36 | H | |
| 6 | 1 | 37 | H | 16 | 1 | 40 | T | |
| 7 | 0 | 30 | H | 17 | 0 | 50 | T | |
| 8 | 1 | 46 | T | 18 | 0 | 42 | T | |
| 9 | 1 | 56 | H | 19 | 0 | 29 | H | |
| 10 | 0 | 47 | H | 20 | 1 | 47 | H |
統制群と処置群が比較的同質的なグループに
集団として処置群と統制群は、母集団とほぼ同質
| 女性の割合 | 平均年齢 | |
|---|---|---|
| 母集団(\(n=20\)) | 55.0% | 38.0歳 |
| 統制群(\(n=11\)) | 54.5% | 37.2歳 |
| 処置群(\(n=9\)) | 55.6% | 39.0歳 |
無作為割当は均質な複数のグループを作る手法







Randomized Controlled Trial(RCT)
\[ \mbox{Income} = \beta_0 + \beta_1 \cdot \mbox{Quant} + \varepsilon \]


Hyde (2015) による分類
実際の社会を舞台に行う実験
Ito, Koichiro, Takanori Ida, and Makoto Tanaka. 2018. “Moral Suasion and Economic Incentives: Field Experimental Evidence from Energy Demand,” American Economic Journal: Economic Policy, 10 (1): 240-267.
Gerber, Alan S., Donald P. Green, and Christopher W. Larimer. 2010. “An Experiment Testing the Relative Effectiveness of Encouraging Voter Participation by Inducing Feelings of Pride or Shame,” Political Behavior, 32: 409-422.
人為的に作られた環境内で行う実験
Blais, André, Simon Labbé-St-Vincent, Laslier Jean-François, Nicolas Sauger, and Karine Van der Straeten. 2011. “Strategic Vote Choice in One-Round and Two-Round Elections: An Experimental Study,” Political Research Quarterly, 64(3): 637–645.
Mueller, Pam A. and Daniel M. Oppenheimer. 2014. “The Pen Is Mightier Than the Keyboard: Advantages of Longhand Over Laptop Note Taking,” Psychological Science, 25(6): 1159-1168.
世論調査に実験を埋め込む方法
Asaba, Yuki, Kyu S Hahn, Seulgi Jang, Tetsuro Kobayashi, and Atsushi Tago. 2020. “38 seconds above the 38th parallel: how short video clips produced by the US military can promote alignment despite antagonism between Japan and Korea,” International Relations of the Asia-Pacific, 20(2): 253–273.
Song, Jaehyun, Takeshi Iida, Yuriko Takahashi, and Jesús Tovar. 2022. “Buying Votes across Borders? A List Experiment on Mexican Immigrants in the US,” Canadian Journal of Political Science, 55 (4): 852-872.
Now we are going to show you four activities that some people may experience during the electoral campaign. After you read all four, just answer HOW MANY activities you experienced during the last electoral campaign. (We do NOT want to know which ones, just how many.)
Bertrand, Marianne, and Sendhil Mullainathan. 2004. “Are Emily and Greg More Employable Than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination,” American Economic Review, 94(4): 991-1013.
処置変数: 人種 ( \(\in \{\text{black}, \text{white}\}\) )
結果変数: 連絡の有無 ( \(\in \{0, 1\}\) )

\(\Rightarrow\) 内生性がある限り、因果効果の識別は困難
\(\Rightarrow\) ケースによって政策的含意が変わる。
| 白人の名前 | 黒人の名前 | |
|---|---|---|
| Female | 76.42% | 77.45% |
| HighQuality | 50.23% | 50.23% |
| Call Rate | 9.65% | 6.45% |
| 計 (人) | 2435 | 2435 |
無作為割当が行われているか否かを確認

標準化差分を使用
連続変数
\[ \text{SB}_{T-C} = 100 \cdot \frac{\bar{X}_T - \bar{X}_C}{\sqrt{0.5 \cdot (s_T^2 + s_C^2)}} \]
二値変数
\[ \text{SB}_{T-C} = 100 \cdot \frac{\bar{X}_T - \bar{X}_C}{\sqrt{0.5 \cdot (\bar{X}_T(1-\bar{X}_T) + \bar{X}_C(1-\bar{X}_C))}} \]
方法1: グループ間の結果変数の差分の検定 (\(t\)検定)
方法2: 単回帰分析 (線形 or ロジスティックス/プロビット)
| Covriates | Est. | S.E. |
|---|---|---|
| Intercept | 0.064 | 0.006 |
| Race: White | 0.032 | 0.008 |
| Covriates | Est. | S.E. |
|---|---|---|
| Intercept | -2.675 | 0.083 |
| Race: White | 0.438 | 0.107 |
| Covriates | Est. | S.E. |
|---|---|---|
| Intercept | -1.518 | 0.039 |
| Race: White | 0.217 | 0.053 |
無作為割当のおかげですべての変数が互いに独立
| Covriates | Est. | S.E. |
|---|---|---|
| Intercept | 0.057 | 0.007 |
| Race: White | 0.032 | 0.08 |
| Female | 0.007 | 0.009 |
| Military | -0.027 | 0.014 |
| Education | -0.002 | 0.005 |
| High Quality | 0.019 | 0.008 |
因果効果が下位グループによって異なる場合
intro_data2.csv)方法1: 男女に分けてATEを推定
| 統制群 | 処置群 | ATE | \(t\) | \(p\) | |
|---|---|---|---|---|---|
| 男性のみ | 0.611 | 1.561 | 0.951 | -7.521 | < 0.001 |
| 女性のみ | 0.493 | 2.480 | 1.987 | -15.573 | < 0.001 |
| 全体 | 0.551 | 2.057 | 1.506 | -15.945 | < 0.001 |
男性のみ
Welch Two Sample t-test
data: Outcome by Treatment
t = -7.5211, df = 235.95, p-value = 1.132e-12
alternative hypothesis: true difference in means between group 0 and group 1 is not equal to 0
95 percent confidence interval:
-1.1996845 -0.7016501
sample estimates:
mean in group 0 mean in group 1
0.6105137 1.5611810
女性のみ
Welch Two Sample t-test
data: Outcome by Treatment
t = -15.573, df = 259.72, p-value < 2.2e-16
alternative hypothesis: true difference in means between group 0 and group 1 is not equal to 0
95 percent confidence interval:
-2.238053 -1.735599
sample estimates:
mean in group 0 mean in group 1
0.4931905 2.4800169
全体
Welch Two Sample t-test
data: Outcome by Treatment
t = -15.945, df = 494.24, p-value < 2.2e-16
alternative hypothesis: true difference in means between group 0 and group 1 is not equal to 0
95 percent confidence interval:
-1.692061 -1.320817
sample estimates:
mean in group 0 mean in group 1
0.5509135 2.0573524
方法2: 性別と処置有無の交差項を投入した重回帰分析
| (1) | |
|---|---|
| (Intercept) | 0.611 (0.091) |
| Treatment | 0.951 (0.131) |
| Female | -0.117 (0.127) |
| Treatment × Female | 1.036 (0.180) |
| Num.Obs. | 500 |
| R2 Adj. | 0.398 |
| F | 110.905 |
\[ \begin{align} \hat{y} = & \beta_0 + \beta_1 \mbox{Treatment} + \beta_2 \mbox{Female} + \\ & \beta_3 \mbox{Treatment} \cdot \mbox{Female} \\ = & \beta_0 + (\beta_1 + \beta_3 \mbox{Female}) \mbox{Treatment} + \beta_2 \mbox{Female}. \end{align} \]
Stable Unit Treatment Value Assumption
非干渉性: 他人の処置・統制有無が処置効果に影響を与えないこと
| Aさんが統制群 | Aさんが処置群 | |
|---|---|---|
| Bさんが統制群 | 0 | 10 |
| Bさんが処置群 | 15 | 20 |
| Aさんが統制群 | Aさんが処置群 | |
|---|---|---|
| Bさんが統制群 | 5 | 10 |
| Bさんが処置群 | 15 | 20 |
処置の無分散性: 同じグループに属する対象は同じ処置を受けること
二重盲検法(Double Blind Test):ある被験者がどのような処置を受けているかについて研究者と被験者両方において不明な状態で実験を行うこと
二重盲検法を使えば以下の問題点に対処することが可能
第4、5、6回はRの実習