Hi all,
is it possible to show the confidence curves in the new passing bablok option in fit y by x?
When I tried it the option was always greyed out. Maybe there is a work around? Asking for a customer.
thx, Jonas
Thank you both for the answers. Will forward the information next week.
Thank you for your response @peng_liu.
Computing confidence curves for Passing Bablok is an ongoing research topic for the development team. According to the developer, "once we figure out which method we should use, such as using bootstrap or the confidence interval of parameter’s estimations, then I will implement it JMP18.0."
It is intentionally disabled. I don't know there is a consensus on how the intervals should be calculated for this model. I cannot see whether they are unambiguously defined, in terms of traditional confidence interval of fitted values and prediction intervals of given explanatory variable values. This model is not defined on a probabilistic theory.
But that does not mean we cannot do anything. There is one thing, which I see people are doing is to use bootstrap, which is one way to define at least the confidence intervals of fitted values in some sense.
Here are the steps:
I am using "Method Comparison.jmp" among sample data sets.
Use this script to get one report:
Bivariate(
Y( :Method 1 ),
X( :Standard ),
Fit Passing Bablok( {Equality Line( 1 ), Line Color( {212, 73, 88} )} )
);
In the report, right click on estimate report, and choose Bootstrap. This should be a standard feature in JMP17. In the past, one needs Pro.
In the dialog, I recommend check the first box, and increase the number of samples if desired to achieve better precision. Then OK.
You get a table look like this. Intercept is the intercept, Standard is the slope.
Create a formula column, and my example is self-explained what the formula is.
Run Distribution on that column, and read off the quantiles to get the confidence interval for the fitted value at Standard = 100.
Because it is Passing-Bablok, the intervals should be defined in either directions: X and Y. Then one should define a reciprocal formula like the following, which is the fitted X value given Y. By such, run Distribution on that column, and the result is the confidence interval of fitted X given Y.
Just got the same question. Thank you all for the information. Waiting for the implementation in JMP 18.
