Measuring one thing to get information about some other, unmeasurable thing.
Good proxies have a high level of correlation to the thing they are being used to measure-by-proxy.
Bad proxies have a low correlation to the thing they are a proxy for. They can also lead to Rule Beating.
Proxies are subject to confounding variables.
Proxy measures can be either Leading Measures or Lagging Measures.
Example Proxy Measures
- Tylenol consumption as a proxy for overall pain in your life.
- Confounding variable: you share your bottle of Tylenol without others, who will introduce Noise into the signal.
- Infant mortality rate as a proxy for a countries level of health care
- GDP as a proxy for the average standard of living in a country
- Proxy gap: assumes a relatively even distribution of wealth. If the entire GDP is earned by one person and everyone else in the country makes nothing.
- Dates or arguments as proxy measures for how well a relationship is going