In regression and curve fitting, you can add additional, often spurious parameters, and get a closer fit. A better fit does not mean you have a better model, it

In regression and curve fitting, you can add additional, often spurious parameters, and get a closer fit. A better fit does not mean you have a better model, it usually means the opposite, that you've overfitted the model, and that it's completely useless for forecasting and predictive purposes. This is why in machine learning, researchers are careful to separate training and testing data. This is just physicists roasting each other over the same issue.