
In the Anchor Experiment (on the first day of class), we found an effect size of around 15 and a p-value of approximately zero, leading us to reject the no that there is no causal link between the value on the card (X = 11 or X = 74) and the percentage guess. In the potential outcomes framework, which populations does this conclusion generalize to?

\[ \widehat{ATE} = \hat{\bar{Y}}_1 - \hat{\bar{Y}}_0 \]
designrbi R package# A tibble: 74 × 2
d y
<fct> <dbl>
1 11 16
2 11 5
3 73 20
4 73 40
5 11 73
6 73 70
7 11 6.5
8 73 12
9 11 15
10 11 70
# ℹ 64 more rows
# A tibble: 74 × 4
d y Y11 Y73
<fct> <dbl> <dbl> <dbl>
1 11 16 16 NA
2 11 5 5 NA
3 73 20 NA 20
4 73 40 NA 40
5 11 73 73 NA
6 73 70 NA 70
7 11 6.5 6.5 NA
8 73 12 NA 12
9 11 15 15 NA
10 11 70 70 NA
# ℹ 64 more rows
# A tibble: 74 × 4
d y Y11 Y73
<fct> <dbl> <dbl> <dbl>
1 11 16 16 31.5
2 11 5 5 20.5
3 73 20 4.53 20
4 73 40 24.5 40
5 11 73 73 88.5
6 73 70 54.5 70
7 11 6.5 6.5 22.0
8 73 12 -3.47 12
9 11 15 15 30.5
10 11 70 70 85.5
# ℹ 64 more rows
# A tibble: 5 × 2
replicate ate
<int> <dbl>
1 1 16.1
2 2 15.6
3 3 20.6
4 4 14.5
5 5 17.0

# A tibble: 5,000 × 2
replicate ate
<int> <dbl>
1 1 8.25
2 2 10.9
3 3 13.0
4 4 9.84
5 5 22.9
6 6 11.6
7 7 24.0
8 8 14.2
9 9 15.9
10 10 10.8
# ℹ 4,990 more rows

2.5% 97.5%
6.75478 24.21553