Garbage in Garbage out

Notes

Claim

The idea that the quality of a model (an assumption about the connection between a variable and an outcome) is only as good as its input data.

Explanation

Why it matters

Therefore, when we design a system (or a model) it is not enough to think about the different components interact with each other, but also how this system interacts with the world.

Examples

For example, when testing a drug effectiveness, if you only include healthy young people for the group who took the drug, and elderly prone-to-sickness people who didn't take the drug, it is not surprising that you might find differences between the groups, creating a false positive result about the drug's effectiveness Biased inputs produce biased conclusions.

For example, habits are examples of automated systems that rarely care about the inputs. We can have good habits such as exercising and eating healthy, but also bad habits such as being prone to anger or mindlessly doom scrolling. Both are examples of the same system, yet one takes inputs into considerations.

For example, having only healthy food at the house almost guarantees that we will eat healthy (at least on average), no matter what our cooking/eating habits are like. Environmental design

Supporters

Fortunately, this also means that it's true the other way around. Pruning the inputs such that only good inputs remain, is very effective in making sure that the outcome will be positive, regardless of the process.

Opposers

Similarly, if our decision making is biased, for example due to Overestimating abilities blinds us to genuine risks, We prefer immediate benefits over delayed ones, Preferring personal benefits blinds us to others' perspectives, it is unsurprising that we will make bad decisions, even if it seems that we have checked all the relevant boxes. The process itself is not enough, we need to verify our intuitions and our Perspective..

Open questions

Visual

Overview

🔼Topic:: Selective Evidence ◀Origin:: 🔗Link::