GraphPad Curve Fitting Guide

Troubleshooting nonlinear regression

Troubleshooting nonlinear regression

Previous topic Next topic No expanding text in this topic  

Troubleshooting nonlinear regression

Previous topic Next topic JavaScript is required for expanding text JavaScript is required for the print function Mail us feedback on this topic!  

The #1 tip for troubleshooting

Nonlinear regression starts with initial values for each parameter. If these initial values are very far from the correct values, nonlinear regression might go astray. To check this, go to theDiagnostics tab of the nonlinear regression dialog and choose the option at the top of the dialog: Don't fit a curve. Instead plot the curve defined by the initial values of the parameters.

If the curve defined by the initial values does not generally follow the shape of the data, and go near the data points, you should change the initial values (Initial Values tab of nonlinear regression) until they do.

If Prism reports an error code

Prism reports error codes as a single phrase that appears over the column of results, for example "Bad initial values" or "Impossible weights". Learn more about each error message.

Common problems in nonlinear regression

Here is a short list of potential curve fitting problems with suggested solutions.

The equation simply does not describe the data.

Try a different equation.

The initial values are too far from their correct values.        

Enter different initial (estimated) values for the parameters. If you entered your own equation, check that you entered sensible  rules for generating initial values.

The range of X values is too narrow to define the curve completely.

If possible, collect more data. Otherwise, hold one of the parameters to a constant value.

You have not collected enough data in a critical range of X values.

Collect more data in the important regions.

Your data are very scattered and don't really define a curve.

Try to collect less scattered data. If you are combining several experiments, consider normalizing the data for each experiment to an internal control.

The equation includes more than one component, but your data don't follow a multicomponent model.

Use a simpler equation.

Your numbers are too large or too small.

If your X or Y values are huge, divide by a constant to change the units. Avoid values greater than, say,  100,000.

If your X or Y values are tiny, multiply by a constant change the units. Avoid values less than about 0.00001.

Changing the units is unlikely to solve the problem, but it is worth a try.

You've set a parameter to an inappropriate constant value.

Check that you haven't made a simple mistake like setting a maximum plateau to 1.0 when it should be  100, or a Hill slope to +1.0 when it should be -1.0.

Consult an analysis check list

Prism has three different analysis checklists for nonlinear regression. Use the one that matches your goal.

Analysis checklist: Fitting a model

Analysis checklist: Comparing nonlinear fits

Analysis checklist: Interpolating from a standard curve