You can only control for weight and smoking if you have good data, which is to say that at least a significant portion of the weight and smoking behaviours are

You can only control for weight and smoking if you have good data, which is to say that at least a significant portion of the weight and smoking behaviours are not causally linked to the exercise behaviour. If the data is not sufficiently diverse, no amount of statistical power will overcome the garbage-in garbage-out nature of regression.