Statistical Reliability Modelling

July 1, 2016

---

What is Statistical Reliability ?

In software systems, reliability means how far the system is able to deliver the required functionality with the best performance. It aims to predict through various techniques whether a system is dependable and robust enough to sustain through the extreme phases during its entire lifetime.
The process of statistical reliability measures the performance of a system through test scores and based on the scores it evaluates the degree of reliability of the product. The test is run repeatedly to ascertain the output of the software system. If the results produced are same all across then the system is a reliable one, but if it doesn't then the possible causes of error should be rectified.

Estimating reliability is a way to ascertain whether the difference is due to errors in measurement or variability in test scores. There are practically four methods of estimating test reliability.

• Test retest reliability method :Checks for consistency between the test scores of one administration to another. This activity involves administering testing tasks assigned to a group of individuals, re-administering and then finally cross check the two results.
• Parallel forms method :This method involves creation of two forms that shall be evaluated in parallel. This is done by creating a large set of questions and randomly divide them into two sets. The correlation that exists between the two sets of question is an estimate of reliability. But one issue with this approach is that one may not be able to generate sufficient number of questions for the same construct. It also assumes that the randomly divided halves are parallel, which may not be true sometimes.
• Internal consistency

  This aims to check for consistency of results across the different items within the same test. It is the estimation of how well the different items work in collaboration with one another within the same system, and produce the same result.

  Internal consistency includes various types of methods -

  • Average Inter-item correlation : This is used to group the items with each other in a way that they all form some sort of correlation. The average/mean is calculated for all these correlations.
  • Split- half method : This method treats the two halves as alternate options for measurement. It comprises of splitting the test into halves. Then the items belonging to each half is compared with one another.
• Cronbach's AlphaThis approach relies on the estimation technique where the split half's are further split into further halves. Each set of half's are computed and re-computed until we have finished with all possible estimates of reliability.

Conclusion :

Different reliability models have their own unique way of assessing reliability parameter of a software system, each of them having their own advantages and disadvantages. Inter rater is considered to be the best among other methods. Each method offers a quantifiable way in which we can estimate the reliability of a software system. It is efficient in making an estimate about the system's performance. The term statistic is used because we decide a system's capability on the basis of the scores produced by each technique.