vault backup: 2025-08-07 15:12:16

This commit is contained in:
2025-08-07 15:12:16 -04:00
parent 8cc5b7ca70
commit baca4a962d
8 changed files with 239 additions and 129 deletions
+50 -119
View File
@@ -12,6 +12,14 @@ by Douglas W. Hubbard
## Key Takeaways
### Definition of Risk
As it is most commonly understood,
risk _always_ implies a negative impact.
For boolean cases,
risk can be represented as a vector of **probability** and **loss**.
### Qualitative Metrics Must Be Avoided
Qualitative risk analysis
@@ -44,6 +52,8 @@ It describes the process of "calibration"
by which people can be trained to compensate for this bias
and make predictions far more accurately.
See [[calibration-questions]] for more.
Experts tend to be good at creating heuristics,
but do not apply them consistently in practice.
@@ -55,125 +65,8 @@ Chapter 13 introduces the [Brier Score](https://en.wikipedia.org/wiki/Brier_scor
as a method of evaluating the performance of an estimator,
evaluated as the mean squared error of their forecasts.
#### The Difficulty of Calibration
##### Boolean Examples
> The melting point of tin is higher than the melting point of aluminum.
> In English, the word “quality” is more frequently used that the word “speed”.
reductive (used more frequently where?)
> Any male pig is referred to as a hog.
reductive (referred to by whom?)
> Californias giant sequoia trees are named for an early 19th century leader of the Cherokee Indians.
reductive
> The Model T was the first car produced by Henry Ford.
reductive (Henry Ford didn't produce cars)
> When rolling 2 dice, a roll of 7 is more likely than a 3.
facile
> No one has ever been reported to have been hit by any object that fell from space.
reductive (reported by whom?)
> Sir Christopher Wren was a British anthropologist.
> Pakistan does not border Russia.
unnecessary negative form, otherwise good.
> The Navy won the first Army-Navy football game.
should specify the official event name, otherwise good.
> The paperback version of the book “The Da Vinci Code”, as of July 2007, still ranks in the top 500 bestselling books on Amazon.
obtuse phrasing, dated topic, otherwise good
> Italian has more words than any other language.
reductive (what is a word? what dialect?)
> The month of August is named after a Greek god.
borderline facile, reductive
> The deepest ocean trench is deeper than the Grand Canyon.
facile
> Abraham Lincoln was the first president born in a log cabin.
deceptive phrasing
> As of July of 2007, more people search Google for  “Harry Potter” than “Hillary Clinton” (according to GoogleTrends).
obtuse phrasing, dated topic, otherwise good
> The population of Alabama is higher than the population of Arizona.
borderline facile, deceptive phrasing
> No category 5 hurricane hit the US in 2004.
> The UK is among the top 10 largest economies in the world (by GDP).
> The movie Forest Gump has grossed more to date than E.T. The Extra Terrestrial.
obtuse phrasing, dated topic, otherwise good
##### Interval Examples
> What percentage of bronze is typically made of copper?
reductive
> How many countries have at least one McDonalds?
As of when?
> How many employees did eBay have in the first quarter of 2006
> What was the population of Miami (within the city limits, not the entire metropolitan area) in 1990?
> How many casualties did the French suffer in the Battle of Waterloo?
> What is the range in miles of a Minuteman Missile?
> What is the percentage of IT jobs in the US were unfilled in 1997?
> The Supremes (with Diana Ross) song “Stop! In the Name of Love” was how long? (minutes, seconds)
> How many undergraduates attended Cambridge in 1990?
> If you could jump 50 feet straight up into the air, how many seconds would you be airborne before you landed?
> How many gallons are in a bushel (they are both measures of volume)?
> How many sovereign rulers has England had in the last thousand years?
> If the air temperature was 5 degrees below zero (Fahrenheit) and the wind speed was 15 mph, what would the temperature adjusted for wind-chill be?
> Average cost of testing in software development is what percentage of total project costs?
> On average, if a software development project was projected to take 17 months, it actually takes how many months?
> How many meters tall is the Sears Tower?
> How many gold medals did Jesse Owens win at the 1936 Berlin Olympics?
> In 2005, the average combined MPG for all US cars and light trucks on the road was how much?
> The average house in the United States uses how many gallons of water per day?
> What was the average price in the United States of a house sold in 2001?
##### Writing Good Calibration Questions
A good calibration question should not feel like it could be a "trick" question.
Definitions/terminology are *always* contentious,
questions based on them always feel deceptive.
Interval "questions" should describe the quantity
rather than phrase it as a question.
##### Strategy for Answering Calibration Questions
Confidence should never be less than probability of picking randomly
(50% for true)
> [!example] p. 198
> Models based on expert opinion consistently outperform the same experts.
### Luck Looks Like Skill
@@ -189,8 +82,25 @@ to overvalue competence and undervalue luck
in the role of achieving improbable accomplishments
as the "Red Baron effect".
How many success stories are simply cases of winning a coin flipping tournament?
### Qualitative Labels are Problematic
> [!example] p. 170
> Experts do not agree on the bounds of terms expressing probability.
> "Likely" vs. "Very Likely"
> [!example] p. 182
> risk matrix type bucketing tends to inflate the significance of small risks.
### There's Always Enough Data
> [!quote] Voltaire
> Perfect is the enemy of good.
> [!quote] Jon Von Neumann
> The truth is much too complicated to allow anything but approximations.
Hubbard challenges the popular rebuttal
that any industry is so niche that
data sufficient for quantitative models
@@ -200,6 +110,27 @@ does not exist.
> ...the belief that unless two things are identical in every way,
> nothing learned from one can be applied to the other.
### Value of Information
* Expected Value of Information (EVI)
* Expected Opportunity Loss (EOL)
$$
\text{EVI} = \text{EOL} - \text{EOL}|I
$$
EOL translates well to continuous probabilities.
### Single Point Estimates are Problematic
> [!example] p. 232
> Hubbard describes a case in the oil industry
> where decent estimating is simplified to the point of serious error
> (collapsing distributions to a single point for "accounting purposes")
> leading to the widespread underestimating of Earth's oil reserves.
The case closely mirrors construction estimating.
## Mentioned Topics and Abbreviations
* Analytic Hierarchy Process (AHP)