Previous: Why do I get

Up: Frequently Asked Questions

Next: Why is the program

How do I get this program to run faster?

A variety of ${\mathfrak{E}I}$ options can significantly increase the speed of the program. In the order in which we suggest you try them, these are,

1. If you have many observations, you can use a random subset in the first stage (by setting _EselRnd to the fraction of observations to include). Setting this global still results in all observations still being included in the second stage, the result being that estimates will still be available for all observations. If a large $n$ is the issue, then this is the best way to speed estimation without much cost.

2. You can use the asymptotic normal approximation, and eliminate the importance sampling refinement, by setting _EisFac$=-1$ or to go even faster use _EisFac$=-2$ to exclude estimation uncertainty.

3. Choose good starting values (by setting _Estval). For example, if you have many similar analyses to run, do a typical analysis and use those numbers as starting values for the others. If you have a guess as to the values of $\phi$ on the untruncated scale, you can use the following command (documented only in eirepar.src) to set the starting values: e.g., _Estval=eireparinv(0|0|-1|.4|.1).

4. Buying enough RAM so you do not need to take advantage of the virtual memory feature of Gauss is helpful. Or, just buy a really fast computer!

5. If you are willing to live with lower precision, you can draw fewer simulations by setting _Esims to a number smaller than the default (100).

6. You can set _EdirTol to a larger number. The default is 0.0001; you can try 0.001 or larger. You could also reduce the maximum number of iterations with _EmaxIter.

7. If you are running many analyses, run one analysis to make sure that the globals are properly defined and then set _Echeck=0 to turn off global checking. This will only save a small bit of time and so is only helpful if you are running many analyses. (If you turn off checking, and you define a global incorrectly, ${\mathfrak{E}I}$ will not give a pretty error message.)

8. For EI2, set _ei2_m$=-1$.