If the 1st differences of consecutive y-values are constant or very nearly constant then a linear model will probably fit well. If the ratios of consecutive y-values are constant or very nearly constant then an exponential model will likely work well.
