SED navigation bar go to SED homepage go to NIST home page SED Home Page SED Contacts SED Projects SED Products and Publications Search SED Pages

RECIPE : REGression Confidence Intervals for PErcentiles

    Version 1.0, April 1995
    Mark Vangel
    vangel@nmr.mgh.harvard.edu

    NOTE: Mark Vangel developed this code while he was employeed by the NIST Statistical Engineering Division. Mark is no longer at NIST, but he can be reached at the above e-mail address. Questions regarding downloading and installation of Recipe can be directed to alan.heckert@nist.gov. However, Alan Heckert does not support Recipe other than providing assistance with downloading and installing the software. If you have technical questions about RECIPE, you may contact Mark directly.

    PLEASE NOTE:
      If you obtained this software by anonymous ftp, please send me an e-mail message and let me know. If you do this, and provide information on how I can contact you, then I will be able to keep you informed about new versions or bug fixes!


Overview:

    RECIPE is a FORTRAN program for determining one-sded tolerance limits (in particular, A- and B-basis material property values) for regression models in the presence of between-batch variability. The users guide for this program is

      Vangel, M. G. (1995), ``A Users Guide to RECIPE: A FORTRAN Program for Determining Regression Basis Values''.

      Vangel, M. G. (1996), ``Design Allowables From Regression Models Using Data From Several Batches'', Proceedings of the 12th ASTM Symposium on Composite Materials: Testing and Design, to appear.

      Vangel, M. G. (1996), ``One-Sided Tolerance Limits for Mixed Models With Two Components of Variance'', submitted, Technometrics.

      Vangel, M. G. (1996), ``ANOVA Estimates of Variance Components for `Partially-Balanced' Mixed Models'', submitted, Journal of Statistical Planning and Inference.

      Mil-HDBK-17, Volume 1, Chapter 8, ``Statistical Methods for Material Basis Values'' (currently a draft under review)

    The on-line version of the user's guide is in PDF (Portable Document Format) and requires Adobe Acrobat to be loaded to allow viewing.

    You only need the Acrobat reader. This reader is free (downloadable from the above Web site) and supported for most common computer systems (IBM PC Windows 3.1/Windows 95/Window NT, MacIntosh, most common Unix workstations). For best results with Netscape, it is recommended that you install the reader as a plug-in rather than a helper application. This is done automatically for the PC installation. Other software may also be used to view PDF files (e.g., the latest version of ghostview), although the performance is typically not as seamless as the Adobe Acrobat reader.

    Latex and Postscript versions of the user's guide are distributed with the software. The technical documents are available from the author.

    The program RECIPE can be used to calculate one-sided tolerance limits (or, equivalently, one-sided confidence limits on quantiles) from an arbitrary regression model with or without a random batch effect.


Files distributed:

For convenience, Unix users can download a single compressed tar file. After downloading, enter the following command to uncompress and unpack the file:

    uncompress recipe.tar.Z | tar -xvf

For convenience, PC users can download the source and data files via a single self-extracting zip file. After downloading, enter the following command to unpack the file:

    recipe.exe
The software distributed consists of the following files:
    FORTRAN Source
    recipe.f Main program
    simcov.f Utility for determining actual confidence levels
    simpvt.f Utility for simulating a pivotal quantity; useful for highly unbalanced data
    subs.for FORTRAN subroutines
Since many users may not have a 32-bit Fortran compiler for Microsoft Windows, the executables for the Recipe program, Simcov program, and Simpvt program are provided. These were created using the Intel version 9.0 Fortran compiler for Windows 2000/XP. Since these were created as console applications, you can run them under DOS or under a DOS prompt window using Windows 2000/XP. These executables were created using the default array sizes.

NOTE: The Windows executables were update 2/1/2008. The previous executables were built using a much older compiler for Windows NT/95/98. If you still have one of these older operating systems, contact alan.heckert@nist.gov to obtain these older versions.


Example files:

    Five example datasets are distributed with the software in the following files:

    Sample Data Files
    ex1.dat Example 1: Simple random sample
    ex2.dat Example 2: One-way ANOVA
    ex3.dat Example 3: Regression with a single batch
    ex4.dat Example 4: Regression with multiple batches
    ex5.dat Example 5: Basis values using data from multiple sources
    ex5a.dat Example 5a: Unbalanced example to illustrate SIMPVT and SIMCOV (see appendix of user's manual)

    Sample Output
    ex1.out Example 1: Simple random sample
    ex2.out Example 2: One-way ANOVA
    ex3.out Example 3: Regression with a single batch
    ex4.out Example 4: Regression with multiple batches
    ex5.out Example 5: Basis values using data from multiple sources
    ex5a.crt Critical value file from SIMPVT for example ex5a.dat (See appendix to the user's manual)

    These examples are discussed in the users guide.


Problem size limitations:

    As distributed, the program can be used for up to MTOT=100 total observations, at up to MLVL=100 fixed levels, with data from up to MBCH=25 batches, using regression models with up to MPAR=10 fixed parameters, and with the tolerance limit being evaluated at up to MPTS=100 points on the regression surface. These maximum values can be changed by editing the `parameter' statement at the beginning of the `recipe.f' file. After this, you must recompile and relink the program in order to create a new executable file.


Incorporation into Dataplot:

    The RECIPE code has now been incorporated into the Dataplot software program. Dataplot is general purpose program for graphical data analysis.




Date created: 6/5/2001
Last updated: 2/1/2008
Please email comments on this WWW page to alan.heckert@nist.gov.