I mention R-squared in this video, which is a good predictor of fit for a linear (y=mx+b) model. However, we are using an exponential model. JMP reports RMSE (reduced mean square error). Values closer to 0 are best. The lowest RMSE I achieved was 4.05 or so, which indicates the model is the best fit compared to other models I use. You'll notice the previous iterations of the curve I made have a higher RMSE, indicating they are less good of a fit than the final values I settled on. The multiple iterations allows me to compare the models made by looking at RMSE. Further iterations with locked in values can help determine better RMSE fits of the nonlinear equation, which is what we want. The lowest achievable RMSE is generally the best fit within the specified exponential model. Other models may provide a better RMSE (like perhaps a logarithmic model). However, the logarithmic shape does not accurately represent the biological response happening here, so I would not include such a model even if it had a better RMSE. RMSE is similar to R-squared in that it is a representation of variance of the data from the curve or model made. However, RMSE is a better representation of this variance for non-linear models, which is what I ended up reporting in my publications.
This video is life saving. If I have the data of soil nitrogen mineral concentrations (Nm) from an incubation study of given time, t. Can I estimate parameters N2, k1 and K2 using a a two parallel pool, zero order plus first order kinetic model (Nmin = k1t + N2[1 − exp(−k2t)]) on this software. Which option do I choose for this particular equation?
Hi Daniel, sorry I didnt see your comment until now. You should be able to write that equation yourself in the equation editor window. When you double click the column, you can "set formula". In that case, you would write each of the components of your equation as a "reference" to the data in the other columns. Then, you should be able to "fit" the model iteratively.
I mention R-squared in this video, which is a good predictor of fit for a linear (y=mx+b) model. However, we are using an exponential model. JMP reports RMSE (reduced mean square error). Values closer to 0 are best. The lowest RMSE I achieved was 4.05 or so, which indicates the model is the best fit compared to other models I use. You'll notice the previous iterations of the curve I made have a higher RMSE, indicating they are less good of a fit than the final values I settled on. The multiple iterations allows me to compare the models made by looking at RMSE. Further iterations with locked in values can help determine better RMSE fits of the nonlinear equation, which is what we want. The lowest achievable RMSE is generally the best fit within the specified exponential model. Other models may provide a better RMSE (like perhaps a logarithmic model). However, the logarithmic shape does not accurately represent the biological response happening here, so I would not include such a model even if it had a better RMSE.
RMSE is similar to R-squared in that it is a representation of variance of the data from the curve or model made. However, RMSE is a better representation of this variance for non-linear models, which is what I ended up reporting in my publications.
This video is life saving.
If I have the data of soil nitrogen mineral concentrations (Nm) from an incubation study of given time, t. Can I estimate parameters N2, k1 and K2 using a a two parallel pool, zero order plus first order kinetic model (Nmin = k1t + N2[1 − exp(−k2t)]) on this software. Which option do I choose for this particular equation?
Hi Daniel, sorry I didnt see your comment until now. You should be able to write that equation yourself in the equation editor window. When you double click the column, you can "set formula". In that case, you would write each of the components of your equation as a "reference" to the data in the other columns. Then, you should be able to "fit" the model iteratively.