Fit Normal() - Differences of parameter estimates standard error and confidence intervals between versions (JMP 16 vs JMP 18.2)Hi,
I noticed some numerical differences between JMP versions, and I was wondering what has changed in the computation of the standard errors and confidence intervals of the parameters estimates of Fit Normal() from JMP 16 to the next versions?
It would be great if you can specify any difference in formulas and methods between the two versions.
The following output has been produced by using the same data and running the following chunk of code in JMP 16.0 (left) and JMP 18.2 (right).
Names Default To Here( 1 );
dt = Open( "$SAMPLE_DATA/Big Class.jmp" );
obj = dt << Distribution( Column( :Weight ) );
obj << Fit Normal;
Thanks in advance to anyone who replies!
Basic Data Analysis and ModelingRe: Fit Normal() - Differences of parameter estimates standard error and confidence intervals between versions (JMP 16 vs JMP 18.2)Hi,
Could you highlight which parameters do not match between the two outputs you shared? I scanned the table for differences, but I could not detect any.
Thanks,
TS
Re: Fit Normal() - Differences of parameter estimates standard error and confidence intervals between versions (JMP 16 vs JMP 18.2)Hi,
Thanks for your response.
The standard error of the dispersion parameter is different:
JMP16.0 = 0.5369693
JMP18.2 = 2.530138
the confidence interval for the location is different:
JMP16.0 = 98.119695 - 111.8803
JMP18.2 = 97.899497 - 112.1005
the confidence interval for the dispersion is different:
JMP16.0 = 21.173986 - 23.293976
JMP18.2 = 18.186907 - 28.50799
Re: Fit Normal() - Differences of parameter estimates standard error and confidence intervals between versions (JMP 16 vs JMP 18.2)Are the data tables identical? Are the data types identical? I don't have JMP 16 to test anymore, but perhaps attaching the actual data table would help (It looks like you got the data table from JMP files). You may need to contact support@jmp.com
Re: Fit Normal() - Differences of parameter estimates standard error and confidence intervals between versions (JMP 16 vs JMP 18.2)Thanks but as I specified initially the data is the same (and can be tested with any data). Even when saving the data and performing the analysis on the same saved data table column (for any continuous column) in the 2 JMP versions the results are different and I did not find, so far, anything in the documentation that could explain this variation in Fit Normal(). Interestingly, the CI and SE seem to match for other distributions (i.e. Lognormal).
Re: Fit Normal() - Differences of parameter estimates standard error and confidence intervals between versions (JMP 16 vs JMP 18.2)I ran a Fit Normal on the same data on JMP 17, 18 and 19. Each version returned the same results for the Fitted Normal Distribution
I was running 16.2 so there might be differences.
Re: Fit Normal() - Differences of parameter estimates standard error and confidence intervals between versions (JMP 16 vs JMP 18.2)Thanks for checking and sharing, I also ran the analysis in JMP 19 and the results were the same as 18.2. So, I assume something changed in Fit Normal() from JMP 16 to JMP 16.2.
Re: Fit Normal() - Differences of parameter estimates standard error and confidence intervals between versions (JMP 16 vs JMP 18.2)This is all I could find related to the issue. "The confidence interval calculations for the normal fit have been updated."
Re: Fit Normal() - Differences of parameter estimates standard error and confidence intervals between versions (JMP 16 vs JMP 18.2)The issues stemmed from the fact that the sigma for the hessian entry in the Standard Error term was off by a factor of 1/sigma, so the variance estimate was off by a factor of sigma. Additionally the Confidence Intervals were different in JMP 16.0 vs JMP 16.1 because 16.0 used the z-distribution, while 16.1 (and up) uses the t-distribution. This better reflects current best-practice sine the t-distribution reflects the distribution of the mean for a normally distributed population and because the sample estimate for sigma is usually unknown (given we are collecting data to estimate the population)
