how do you get an r2 and AICc value for a custom nonlinear regression model? How do you adjust the curve for built in regression models?

[#1]

I don't know what equation best fits the line. I think that the curve for some of my data looks like a power function. Since I don't know the exponent, I used the custom model for nonlinear regression. I need an R2 and AICc values to compare different models. I created a column using the power equation and then went to analyze to modeling to nonlinear and put in my y values and then put in the power equation column into the predictor column. I then pushed go  to calculate the parameters. However, I can't get an R2 or AICc value or a p value. I also have used the built in logistic 3p regression fit, but there are some outliers that are skewing the curve. I would like to be able to adjust the curve to better fit my data. How do I do this? Please help me. I have read all the jmp help sections on nonlinear regression and even watched some youtube videos, but I still can't figure it out. If anyone knows how to do this, please let me know,

[#2]

---

@des2870 wrote:

I don't know what equation best fits the line. I think that the curve for some of my data looks like a power function. Since I don't know the exponent, I used the custom model for nonlinear regression. I need an R2 and AICc values to compare different models. I created a column using the power equation and then went to analyze to modeling to nonlinear and put in my y values and then put in the power equation column into the predictor column. I then pushed go  to calculate the parameters. However, I can't get an R2 or AICc value or a p value. I also have used the built in logistic 3p regression fit, but there are some outliers that are skewing the curve. I would like to be able to adjust the curve to better fit my data. How do I do this? Please help me. I have read all the jmp help sections on nonlinear regression and even watched some youtube videos, but I still can't figure it out. If anyone knows how to do this, please let me know,

Five years later I have the same problem. Is there anybody who can help in this matter?

---

[#3]

First of all, what version of JMP are you currently using? JMP Pro 14? JMP 14?

Second of all, have you tried using Fit Curve? (I think so but I am not sure.) This platform is easier than Nonlinear for many common models. It provides a model comparison report to help you select the best model based on fit statistics and information criteria.

(Note - R square is not recommended for model selection.)

Third, you are correct that R square is not available. That intentional omission is because that sample statistic assumes that the SS Error and the SS Model sum to SS Total, but that is not true for nonlinear case. You are also correct that AICc is not provided by Nonlinear but you can compute the AICc for the continuous response as n*Log( SSE/n ) + 2*k + (2*k^2 + 2*k)/(n-k-1) where k is the number of model parameters and n is the number of observations.

Fourth, if the outliers are not data errors, then you might try using the Weight analysis role. For example, this way can be used to perform a weighted regression with the reciprocal variance weight.

[#4]

Thanks Mark for your response,

1. I am using JMP Pro 14

2. I have tried to use Fit Curve, hovewer there is no a power function which I prefer in my analysis.

3. I need AICc for model selection among models based on power function with various independent variables. R2 I use occasionally as additional information describing the fitness of regression model (using fit Y by X). However I did not know that it is incorrect here (now I know why it is not available).

I have computed AICc according to your proposition from previous discussion (https://community.jmp.com/t5/Discussions/AIC-for-self-created-nonlinear-model/td-p/86335), however I don't know how to implement it into JSL. It is pretty annoying to compute it manually for each of my models set.

4. However, this solution doesn't give me AICc :(

[#5]

My earlier post is equivalent but more concise, so I would use the previous formula.

I am sorry that you are annoyed. Perhaps this data table will help with calculating the AICc for your models:

[#6]

Thank you for your help Mark. Is it possible to add an option to automatically obtain AICc (and BIC) in "nonlinears" to the next JMP versions?

[#7]

See the JMP Wish List in the JMP Community menu bar. You can search to see if someone has already suggested it and add your vote or start a new request.