Validation of code usually takes two forms:
1. If I write code where all I’m using is existing methods that a part of a computer package and all I’m generating is output from those package components someone else has to look at the code and agree that the approach is the correct one.
2. If I’m writing code where I actually have to create something new – say write a subroutine that does some kind of test not a part of the current package then to validate that someone else has to independently write code using the same package, run the same data and arrive at the same answer.
As for data – the focus is on issues of changes to the data. Most of this has to do with tools used to create, import, and export data. For example, Excel would not be an acceptable format for a validated data set because it is far too easy to make irreversible mistakes in data set construction and manipulation (e.g. thinking you were sorting the entire page when you actually sorted a single column).
Where I work data and code validation are a requirement on many projects. As a result we never use interactive anything in our work. Everything is written in code so programs like SAS, R, and Splus are the packages used in our analysis.
.
