" All models are wrong, some are useful "
I first came across this idea while watching an introductory lecture on macroeconomics. After some preliminaries and an explanation of modeling using the familiar equation y = mx + b, the professor quoted the statistician George Box: “All models are wrong, but some are useful.” That line struck me so much that I had to pause the video and think about it for a while. “All models are wrong, but some are useful.” It made perfect sense. After all, modeling is simply our attempt to create a function that fits the points on a graph, a way to make sense of scattered observations. To those who dislike math, think of it like this: You’re sitting in a restaurant and notice a couple, a neatly dressed man and a stunning woman in an eye-catching dress. Time passes, and suddenly you hear loud voices coming from their table. A girl’s name is mentioned, the woman stands up, slaps the man, and storms out. These are your data points, aka your observations. As humans, we naturally try to explain what we see, to connect the dots. So your brain forms a model: “She must have found out that he’s cheating on her, got angry, slapped him and left.” That explanation ( that story ) is your model. It’s the line that connects the dots and makes everything coherent. But remember: all models are wrong, though some are useful. Your model might be completely wrong. You only observed a small part of the situation. For instance, maybe the couple has a baby girl, and they were out for the first time in months. The father called his sister to babysit, but the mother dislikes her and disapproves of her influence on their daughter. When she found out the sister was babysitting, she got upset, slapped the man, and left. This second model also fits your observations, just like the first one, but it’s also wrong. You might say, “But surely one of them must be right?” Well, that depends on what you mean by right. We build models to explain the world around us, to make sense of things and, most importantly, to predict the future. But with your limited observations, you can’t really predict what will happen next. Even if you don’t care about prediction, your model is still uncertain, because you don’t truly know why she slapped him. There could be dozens of unseen causes, like work-stress or maybe hormonal problems, perhaps previouse fights that piled and lead to this .