This video delves into the statistical likelihood of an accumulation of unlikely events, prompted by the observation of multiple AfD candidates dying before an election. The speaker critically examines common misconceptions in probability calculation, using the example of a specific car license plate to illustrate how precisely defining the event is crucial. He differentiates between purely statistical and game-theoretical thinking, where patterns might suggest intentional influence rather than mere chance. Employing a simplified "urn model" (binomial/Poisson distribution) and conservative estimates for death probabilities, the analysis concludes that while the observed cluster of 4 deaths is statistically striking (less than 1% probability under certain conditions) and highly significant, it remains within a range that cannot be entirely dismissed as implausible by chance alone. However, 7 deaths under the same assumptions would be extremely improbable. The video recommends precise actuarial and forensic investigations for such cases and highlights the "side bias" that influences immediate interpretations, urging for abstract thinking to avoid preconceptions.
Summarizing the Likelihood of Unlikely Events: The Case of AfD Candidate Deaths
0:00:01 The Core Problem: The video addresses the statistical question of how likely it is for unlikely events to accumulate, using the specific example of 4 (or 7, depending on the count) AfD candidates dying directly before an election.
0:00:40 Misconceptions in Probability: Initial calculations circulating on social media, claiming the event was "one in millions," are identified as likely incorrect due to a flawed understanding of how to calculate such probabilities.
0:01:41 Illustrative Example: License Plate Probability: The speaker uses the probability of a specific car license plate (e.g., FMW4879) appearing in front of a door to demonstrate common errors.
0:02:15 Specific Calculation: For a theoretical 3.456 billion possible license plates, the chance of that exact one is extremely low (0.00289%).
0:03:46 Defining the Event: This calculation is correct but answers the wrong question; typically, we're interested in any license plate, or a plate with certain properties, not one specific combination.
0:04:32 Key Takeaway: There are so many unlikely events that the probability of an improbable event occurring purely by chance can be quite high. Base probabilities (e.g., "F" for Frankfurt) and assigned identifiers also matter.
0:06:03 Statistical vs. Game-Theoretical Thinking: The video introduces the distinction between pure chance and situations where patterns or combinations might convey information or imply external influence (e.g., a car with "FHR3" from a radio station building). The speaker even subtly demonstrates this by choosing "3456" in his own calculation.
0:09:14 False Positives and Bias: The risk of perceiving intent when it's pure coincidence (false positives) is discussed, citing the example of a company with an exact $1 million turnover triggering a tax audit due to its "unlikely" exactness.
0:10:10 Precisely Defining the "Event" of Death: For the AfD case, it's crucial to define what "event" is being investigated (e.g., exactly these four, at least four, only from party A, compared to total population/other parties, excess mortality).
0:13:58 The Research Question: The focus shifts to whether AfD candidates exhibit an excess mortality rate compared to a comparable population, considering age and health factors.
0:14:20 Modeling with Urns and Balls: A simplified statistical model (binomial distribution, approximated by Poisson for rare events) is introduced, where "red balls" represent deaths and "white balls" represent survival.
0:16:51 Estimating Individual Death Probability: A "worst-case" annual death probability of 6% is estimated for older (approx. 75 years, with pre-existing conditions) candidates, acknowledging this is an overestimate for younger individuals.
0:17:49 Initial Plausibility Check (6% Annual Death Rate, 100 People):
0:18:38 At least 4 deaths: The probability is surprisingly high, around 85%.
0:18:54 At least 7 deaths: Probability is significantly lower but still "not completely out of this world."
0:19:12 Monte Carlo Simulation: The theoretical calculations are validated through a Monte Carlo simulation, confirming their stability and accuracy.
0:20:06 Refined Probabilities for the Specific Case (0.6% Monthly Death Rate, 100 People): Considering younger candidates and the short timeframe (1 month, scaling down 6% annual to 0.6% monthly):
0:21:01 At least 4 deaths: Probability is approximately 0.34%. This is considered "significant" (below 1% significance level) but not "insane" or "completely absurd."
0:21:37 At least 7 deaths: Probability drops strongly non-linearly to less than 0.001%, becoming "incredibly rare" and "very unlikely."
0:22:00 Visualizing Probability Curves: Graphs show how the probability of observing at least 4 or 7 deaths changes with the individual death probability, highlighting the "relevant area" for the actual case at lower probabilities.
0:24:10 Analogies for Understanding: A class reunion of 75-year-olds: 4 deaths in a year is plausible, but 4 deaths in the month before the reunion is "strange," and 7 deaths in the month before would "not sound good," suggesting non-randomness.
0:25:55 Importance of Population Size: The number of individuals included in the analysis (e.g., 100 candidates vs. 300 including reserves/relatives) greatly impacts the calculated probabilities.
0:28:07 Actuarial Expertise: Hire an actuary to perform more precise calculations using detailed life tables and data to refine a priori probabilities.
0:29:01 Forensic Investigations: Conduct forensic analysis to look for further clues, autopsies, or external influences (e.g., shared exposures).
0:29:30 Analogy to Tax Audits: Just as a tax office would immediately audit a 1% statistical anomaly, human lives warrant thorough investigation.
0:30:17 Common Explanations and Their Implications:
0:30:28 Older Candidates: Being older is a significant factor contributing to a higher death probability, making the observation (for 4 deaths) more plausible. If younger candidates were involved, it would be much more alarming.
0:31:05 Increased Attacks/Stress: The argument that AfD candidates face more attacks or stress is noted as a non-statistical argument that could point to either increased external influence or natural stress, highlighting the complexity.
0:31:48 The Problem of "Side Bias": The video emphasizes the reflex reactions of different groups (e.g., "conspiracy theory" vs. "stinks to high heaven") and suggests using abstract thought experiments (e.g., "what if it were a different party/country?") to combat bias.
0:33:40 Final Conclusion: The event is "unlikely, but still within a plausible range." It's clear that "it is not clear" whether it's pure chance or something else, but it warrants thorough investigation.