[EBS 지식프라임] 심슨의 패러독스: 평균에 대한 착각

Main ideas / lessons

• Statistical thinking is powerful, but it can mislead if you rely on simple averages without considering how the data is structured.
• The video focuses on the illusion of the “average” and explains why this illusion can produce wrong conclusions.
• The key concept introduced is Simpson’s Paradox:
  • When you combine data into a single “average” (or aggregate statistic), you may see a trend that reverses the trend that exists within each separate subgroup.
• The lesson: To avoid being deceived by aggregate numbers, analyze data separately (e.g., by category such as “strong vs. weak” conditions, or by segments such as “high-end vs. low-end”).

Concepts illustrated with examples

1) Baseball batting average example (Simpson’s Paradox)

• Metric discussed: Batting average

  • Defined as: (number of hits) / (number of at-bats)
  • In other words, it treats performance as an average across all situations.

• Players compared:

  • Lee Seung-yeop (Korea’s representative power hitter)
  • Hong Gil-dong

• What looks surprising (the “trap”):

  • In a practice comparison:

    • Lee Seung-yeop: 15 hits
    • Hong Gil-dong: 15 at-bats (The subtitle wording is described as potentially garbled/incomplete; the intended point is that Hong’s aggregated numbers look better.)

  • The “average” comparison suggests Hong is better.

• Why the aggregated conclusion is misleading:

  • Their performance against strong pitchers differs from their performance against weak pitchers.
  • The number of times each player faced strong pitchers is unequal, creating a misleading aggregate when batting averages are combined.

• Core lesson from the example:

  • Separate the data by context:
    • Against strong pitchers: Lee’s batting average is 1 (per subtitle), Hong has no hits
    • Against weak pitchers: Lee has 8 hits, Hong’s batting is 0.600
  • So, within each subgroup, the “who’s better” conclusion differs from the combined average.

2) Apartment price example (aggregation problem)

• Observed claim (initial aggregate conclusion):

  • From three consecutive months after peaking last October:
    • Transaction volume decreases
    • Apartment prices fall

• Stated report figures (February 2007 article):

  • Compared to four months earlier:
    • Nationwide price per pyeong: -24.7%
    • Metropolitan area: -15.7%
    • Seoul: -12.2%

• Why this can be misleading (video’s argument):

  • The “drop” is explained as a statistical artifact from averaging across different types of apartments:
    • High-end apartments vs. low-end apartments were mixed to form an aggregate “average transaction price.”

• Mechanism described:

  • Real estate policies at the time targeted high-end apartments primarily.
  • Therefore, high-end transaction volume dropped sharply.
  • When you compute an average price by combining high-end and low-end segments, a major drop in high-end activity can pull the overall average down, even if the subgroup story is different.

• Core lesson applied:

  • If you reach a “strange” conclusion using aggregated averages, check whether subgroups (like high-end vs. low-end) behave differently.

3) Quick “average” misunderstanding (final depth/height lines)

Examples used to remind that averages are simple and can be misleading without context:

• Average staff lecture depth: 140 cm
• Average soldier height: 165 cm
• (Subtitle text ends mid-thought; the intent is to show how an average can lead to a questionable or incomplete inference if taken at face value.)

Methodology / instruction presented

• When analyzing statistics, do not rely solely on aggregate averages.
• Check whether the data should be split into meaningful subgroups, such as:
  • Performance by context (e.g., against strong vs. weak pitchers)
  • Performance by segment (e.g., high-end vs. low-end apartments)
• Analyze separately and then compare:
  • For each subgroup, determine which person/segment performs better.
  • Then compare the subgroup results rather than trusting the combined average.
• If the conclusion seems strange or contradictory:
  • You may be experiencing Simpson’s Paradox.
• Use averages as a starting point, not as final truth:
  • “Statistics do not lie,” but aggregates can create deceptive interpretations when subgroup conditions differ.

Sources / speakers featured

• British mathematician Simpson (Simpson’s Paradox namesake; mentioned as the source of the term)
• Lee Seung-yeop (baseball player used in the example)
• Hong Gil-dong (used as the comparison subject in the baseball example)
• A February 2007 article about apartment price changes (referenced; no specific author named)