From a few years working in modeling and simulation I came up with the following (I’ll call these Ellen’s Postulates):
1. Any model containing more than 5 variables will have limited utility. (I might flex on the number, based on following points).
2. Models that rely on variables with a high percentage of binary choices (1 vs. 2; A vs. B; Male vs. Female) are more likely to produce accurate predictions than models that rely on variables with a broad set of choices and choice ranges.
3. Models designed to predict future states for situations in which there are an unknowable number of factors or variables can offer, at best, crude speculations on what can be expected.
4. In these complex situations, the limitations on what is unknown and unknowable, and the interactions between and among known/unknown/unknowable variables is effectively impossible to quantify or comprehend.
***
Weather can be somewhat reliably predicted (within a short term window — typically 5 days). With each additional day, likelihood of prediction declines.
To wit, weather forecasts beyond five days don’t have much value, when you factor out the main fixed aspects of what season and what climate zone you are in.
Climate forecasts gets very complex and therefore unreliable in a hurry IMO.
“All models are wrong; some are useful.”
From a few years working in modeling and simulation I came up with the following (I’ll call these Ellen’s Postulates):
1. Any model containing more than 5 variables will have limited utility. (I might flex on the number, based on following points).
2. Models that rely on variables with a high percentage of binary choices (1 vs. 2; A vs. B; Male vs. Female) are more likely to produce accurate predictions than models that rely on variables with a broad set of choices and choice ranges.
3. Models designed to predict future states for situations in which there are an unknowable number of factors or variables can offer, at best, crude speculations on what can be expected.
4. In these complex situations, the limitations on what is unknown and unknowable, and the interactions between and among known/unknown/unknowable variables is effectively impossible to quantify or comprehend.
***
Weather can be somewhat reliably predicted (within a short term window — typically 5 days). With each additional day, likelihood of prediction declines.
To wit, weather forecasts beyond five days don’t have much value, when you factor out the main fixed aspects of what season and what climate zone you are in.
Climate forecasts gets very complex and therefore unreliable in a hurry IMO.