H & H Servicco Corp. Sampling Plans Site Map  

Main Areas: Home Best Practices Software Forums
Software Area:  
Acceptance sampling software ] Audit sampling software ] How to Buy Software ] Specifications & Support ]

Specifications & Support Information:

Validation of Software
Validation of TSS Plans
Customer Service
Windows Compatibility
Year 2000
Off-Site Information

Software Validation

visitors=Hit Counter

This area describes the process that H & H Servicco Corp. used to validate the programs: ASP, TP105, TP414, TP781.

Sections:

  1. Validation of the algorithm code.
  2. How program results compare to standards.
  3. Robustness to extreme input data.
  4. Warning messages and disallowed input.
  5. Product release testing.
  6. Usage experience of the software.
  7. Year 2000.

1  Validation of the algorithm code:

Each program is based on the calculations of a national sampling standard or a published article. That reference was then used to validate the calculation by reproducing the results of the standard.

1A) Sampling Standards:

bulletAttribute (binomial and poisson) - Mil-std-105d.
bulletVariables (percent nonconforming) - Mil-Std-414, K-method.
The TSS method has veen validated by simulation.
bulletReliability (MTBF). Mil-Std-781

1B) Books, Articles, and Statistical Tables:

bulletBiometrica Tables for Statisticians. Volume I.
bulletEngineering Handbook AMPC 706-109 - "Tables of the Cumulative Binomial Distribution", US Army Materials Command, June 1972.
bulletArticle: "A Nomograph of the Binomial Distribution" by Harry R. Larson, in Industrial Quality Control, Dec, 1966. An example from this artical is the default data of TP105. I recommend this article to anyone desiring to sharpen his/her comprehension of binomial sampling.
bulletQuality Control and Industrial Statistics, Acheson J. Duncan. (calculations for variables percent nonconforming plans)
bulletStatistics, an Introduction, Rickmers and Todd. (calculations for variables mean plans)
bulletReliability in Engineering Design, K.C. Kapur and L.R. Lamberson.
bulletSampling with Confidence, Dr. Donald J. Wheeler, Quality magazine, August 1984. This article contains statistical methods and example result for attribute confidence limits. (calculations for attribute confidence limits in ASP)

2  How program results compare to standards:

You can verify compliance to a standard yourself by selecting a sampling plan from the standard and entering it into the program as decision rule input (choose the "D" method of input). For example, from Mil-Std-105, choose n,Ac. Or, from Mil-Std-414, choose n,K. This will reproduce the OC Curves that are published in the standard.

Special Note: If you choose the plan for variables (K-method) from Mil-Std-414, TP414 will reproduce the OC curve exactly. If you choose it from ANSI/ASQC Z1.9, there will be a slight difference because of a known error in the graphs in Z1.9. TP414 gives the correct result.

3  Robustness to extreme input data:

For each data input type, extreme values were input to assure robustness of the program code. Subroutines that calculate probability distributions were compared to published tables, covering the range of the tables: binomial, poisson, normal, students t, exponential.

Warning messages and disallowed input:

The programs notify the user when he/she inputs values that will produce inaccurate results. Depending on the situation, there may be either warning messages or prohibited input. Most values have this type of internal check.

bulletWarnings Generated about Input: You can observe how the programs handle this by trying to enter a value for alpha or beta risk that is less than 0.000001 (1 part per million).
bulletWarnings Generated about Output: Also in ASP, the table of sample size will flag those values that may violate accuracy limitations. Footnotes to the table explains these flags to the user.

5  Product release testing:

Each program is thoroughly tested before commercialization, using a sequential sampling plan developed with TP781. The testing strategy used is described on the reliability growth page.


Usage experience of the software:

The first commercial program was TP414 sold in 1986. There has not been even one numerical error reported by users -- which is a testimonial to the efficacy of acceptance sampling.

Year 2000:

The only date related variable in the programs are the system date that prints at the bottom of each printed (but not onscreen) report. This line is intended to document the program that produced the report, the version number, and the date. Software with file dates (.exe file) prior to 7/2/1997, will not print that documentation line after 1999. Program files dated after 7/2/1997 will not be effected by Y2K.