good fits - bad models ?

Enter six data points - e.g car speed at one second intervals[br]from beginning of braking until car has stopped, or average [br]temperature at two week intervals from September to December.[br][br]Fit a linear model to your data.[br][br]CHALLENGE - Deviation is defined as the square root[br]of the sum of the squares of the vertical distances to[br]to linear model function.[br][br]Why not use some measure that depends on the [br]perpendicular distances to the linear model function?[br][br][br]CHALLENGE - If we can ([i]almost[/i]) always fit a finite set of data with a[br]polynomial, why do we bother with statistical techniques like regression?[br]Why does it say [i]almost[/i]?

Information: good fits - bad models ?