Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

RLH very big

I run CBC result using CBC/HB sawtooth software. In first I run I get result
RLH
63
65
62
64
71
68

In second trial I get this result
RLH
0.077321964
0.065412867
0.073315509
0.07220652
0.145737896
0.086991154

It make me confused what is the problem? Please help
asked May 29, 2015 by Retno Jati Sahari

2 Answers

+1 vote
Hi Retno,

Which programs are you using for the two trials? I assume you are switching between CBC/HB and something else (perhaps HB within SSI Web, MBC, ACBC, etc.).

RLH stands for Root Likelihood, and is a measure of model fit derived from the maximum likelihood utility estimation procedure. The RLH fit statistic is a probability that is naturally bounded to the range 0 and 1, where 0 indicates that the model is perfectly non-predictive and 1 indicates that the model perfectly fits the data.

In many of our analysis packages, including the standalone CBC/HB program, the RLH is rescaled as RLH * 1000 to avoid having to store and manipulate fractional data. This practice originated when our programs were written for DOS, and decimal data was a bit of a nuisance. The "RLH" therefore ranges from 0 to 1000. (Other programs, such as CVA and ACA, implement this practice for r^2 values as well.)

Some of our newer analysis packages represent RLH on the more formally correct 0-to-1 bounded scale. It appears that your second data set uses the more formally correct scale, while the first data set uses RLH x 1000. To compare the two, you'd simply divide the first by 1000, or multiply the second by 1000.

--Aaron Hill
answered May 29, 2015 by Aaron Hill Gold Sawtooth Software, Inc. (10,095 points)
Such a thoughful, succinct, and useful response, Aaron!  Is there an acceptable range of RLH figures?
Chuck
0 votes
I'm sorry, but it's very hard to know without more information what has produced these two sets of numbers you are displaying in this post.  I suggest you contact our technical support team at support@sawtoothsoftware.com or by phoning +1 801 477 4700.
answered May 29, 2015 by Bryan Orme Platinum Sawtooth Software, Inc. (162,290 points)
...