Using nonlinear regression with an empirical model 

Using nonlinear regression with an empirical model 


If your goal is just to plot a smooth curve, without worrying about a model, you have several choices.
Splines created by Prism go through every point, so may wiggle too much, but smoothing splines work well.
Lowess curves follow the general trend of the data, but can be too jagged.
An alternative is to use nonlinear regression.
Nonlinear regression requires you pick a model, but you don't have to pay attention to the meaning of the model or the value of the parameters. Instead, you can pick a model empirically and judge it solely on the appearance of the curve. In this case, you are using nonlinear regression as a tool to create a smooth curve, and not as a method to analyze data.
If you use nonlinear regression in this way, you can experiment with any model you want. But first try fitting polynomial models, which are very general (and never give fitting problems due to poor initial values). If the curve strays too far from the trend of the data, pick a higher order model. If the polynomial curve wiggles too much, pick a lower order model.