As an undergraduate majoring in Economics and Political Science, I spend majority of my time studying and working on models that have been developed to explain our reality such as Game Theory. A quote by a famous statistician, George Box, has stuck with me since freshman year.
"All models are wrong, but some are more useful than others." - George Box
In many of my courses, we are introduced to a plethora of models that are used to explain our complex world and professors always attach critical assumptions. These models that we use are already too complicated and difficult to grasp but the truth is that models are just merely nothing but a simplified framework.
So what is the purpose of models if they, at the end of the day, fail to explain the entirety of the world we live in with 100% accuracy?
That may be true, but consider this. Let's say I am making a bet with my friend on whether the next coin flip will be a heads or tails. Basic probability will state that the chance is 50/50. Knowing this, my decision to pick heads or tails will be nothing but "impulse". However, what if the coin was asymmetric and did not have an even curvature? What if the ground had some influence on the way the coin would land? I could take these parameters and do an experiment and ultimately create a model that could predict the probability the coin will land on heads or tails. In the best case, the model may predict heads with a 99.9% chance. But even if the model predicted the coin will land on heads with a 51% chance, that would still be more helpful than the naive forecast of a 50/50.
Yes, the model is imperfect. All models are. But, good models can lead to good forecasts. Good forecasts can lead to good decisions.
So what is the purpose of models if they, at the end of the day, fail to explain the entirety of the world we live in with 100% accuracy?
That may be true, but consider this. Let's say I am making a bet with my friend on whether the next coin flip will be a heads or tails. Basic probability will state that the chance is 50/50. Knowing this, my decision to pick heads or tails will be nothing but "impulse". However, what if the coin was asymmetric and did not have an even curvature? What if the ground had some influence on the way the coin would land? I could take these parameters and do an experiment and ultimately create a model that could predict the probability the coin will land on heads or tails. In the best case, the model may predict heads with a 99.9% chance. But even if the model predicted the coin will land on heads with a 51% chance, that would still be more helpful than the naive forecast of a 50/50.
Yes, the model is imperfect. All models are. But, good models can lead to good forecasts. Good forecasts can lead to good decisions.
If the model (coin flip) predicted heads with 51% chance, it is indeed better than native 50/50. However, it would be equally fair state that there will be 49% of getting tails, according to the model. If that is a fair statement, it is also fair to state that "bad models can lead to bad forecasts. Bad models can lead to bad decision" (?)
ReplyDeleteAbsolutely true. The concluding statement is symmetrical. It would ultimately come down to what the "decision boundary" is, which is subject to the individual(s). That threshold can be much higher dependent upon how risk averse one would be in evaluating the model.
Delete