Data science: Spot Simpson's paradox

Two treatments were tried in several groups. Report success rates within each group and overall, and say whether the combined data reverse the group-level result.

What to build

A program that compares two treatments within groups and overall and detects Simpson's paradox.

Requirements

  • For each group, in order of first appearance, print GROUP: A X%, B Y%, with rates to one decimal place.
  • Then print overall: A X%, B Y%.
  • Finally print paradox: yes if one treatment has the strictly higher rate in every group while the other has the strictly higher overall rate; otherwise paradox: no.

Input

Line 1: n. Then n lines GROUP,TREATMENT,SUCCESSES,TOTAL, where TREATMENT is A or B; every group has exactly one A line and one B line.

Output

One line per group, then two lines.

Constraints

1 to 20 groups; 1 ≤ TOTAL ≤ 100000; 0 ≤ SUCCESSES ≤ TOTAL.

Example

Input:

4
small,A,81,87
small,B,234,270
large,A,192,263
large,B,55,80

Output:

small: A 93.1%, B 86.7%
large: A 73.0%, B 68.8%
overall: A 78.0%, B 82.6%
paradox: yes

Why it works

Overall rates weight each group by how many cases it has, so a treatment given mostly to hard cases can win in every group and still lose overall; checking both levels exposes it.

Examples

Example 1 · Input
4
small,A,81,87
small,B,234,270
large,A,192,263
large,B,55,80
Expected output
small: A 93.1%, B 86.7%
large: A 73.0%, B 68.8%
overall: A 78.0%, B 82.6%
paradox: yes
Example 2 · Input
4
g1,A,8,10
g1,B,6,10
g2,A,5,10
g2,B,3,10
Expected output
g1: A 80.0%, B 60.0%
g2: A 50.0%, B 30.0%
overall: A 65.0%, B 45.0%
paradox: no

Prompts inside input() are ignored when checking, so input("n: ") is fine.

Hints

Hint 1

Keep, for each group, the successes and totals of A and B; also keep running overall totals.

Hint 2

Remember the groups in the order they first appear.

Hint 3

The paradox needs the same treatment strictly better in every group and the other strictly better overall.

Solution

Show one correct solution (try it yourself first)
n = int(input())
groups, order = {}, []
overall = {"A": [0, 0], "B": [0, 0]}
for _ in range(n):
    g, t, s, total = input().split(",")
    if g not in groups:
        groups[g] = {}
        order.append(g)
    groups[g][t] = (int(s), int(total))
    overall[t][0] += int(s)
    overall[t][1] += int(total)


def rate(pair):
    return 100 * pair[0] / pair[1]


a_wins = b_wins = 0
for g in order:
    ra, rb = rate(groups[g]["A"]), rate(groups[g]["B"])
    print(f"{g}: A {ra:.1f}%, B {rb:.1f}%")
    a_wins += ra > rb
    b_wins += rb > ra
oa, ob = rate(overall["A"]), rate(overall["B"])
print(f"overall: A {oa:.1f}%, B {ob:.1f}%")
paradox = (a_wins == len(order) and ob > oa) or (b_wins == len(order) and oa > ob)
print("paradox:", "yes" if paradox else "no")

Lesson: Relationships between variables →