Let's say you wanted to find out the relationship between the density of water and temperature. So, you measure the density of water at 10 C, 50 C and 90 C then fit a line through your data points. That's well and good, but you might run into problems using that line to predict the density of water at 101 C or -1 C.

If you stick with engineering you may hear about the concept of "inference space". Roughly speaking, that's the range of values for which we have experimental data. Using an empirical relationship outside of its inference space can lead to problems. In the example above, there are state changes just outside of the inference space which are not considered in the relationship. In other cases a variable may have a fairly constant value across the inference space (and therefore be ignored in the relationship) but not outside of that space.