Quality Metrics - Ryan Cook
^Table Of Contents^ Naming Conventions >>
Definition of quality

    Before we can measure quality, we need to understand and define what quality is exactly. Quality is commonly interpreted as something that can be judged but not measured and as indefinable because quality means many different things to many different people. Such a vague understanding of the concept quality obviously could not be helpful to professionals attempting to quantify it. Therefore, the workable definition of quality was created as the conformance of a product to the customers’ requirements.

^Top^

Measurement theory

    Scientific advancement is made through the observation of data resulting in new theories and affirmation or refutation of these theories based upon new data, thus making measurement of that data crucial to scientific progress.

    The first component of a measurement is it scale, the four levels of scales are nominal scale, ordinal scale, interval scale, and ratio scale.

    Nominal scale classifies or categorizes the attribute being measured. An example of this would be the classifications male and female. The nominal scale doesn’t allow for comparisons to be made or mathematical operations to be preformed. So the expression male > female or male < female are invalid. As well, it is impossible to subtract a female from a male. The key requirement of the nominal scale is that each data item can fit into only one category.

    Ordinal scale is vary similar to the nominal scale except categories can be ordered and comparisons can be made. An example of this would be customer satisfaction surveys which often require an answer of completely dissatisfied, somewhat dissatisfied, neutral, somewhat satisfied, or completely satisfied. In such a case, there is a natural order to the scale unlike nominal scales. However like nominal scales mathematical operations still can’t be performed.

    Interval scale indicates exact differences between measurement points unlike the previous two scales. Only the mathematical operations add and subtract can be applied to an interval scale because the zero point on the scale has been arbitrarily decided upon. For example the temperature 20 degrees is hotter than 10 degrees but 20 degrees may not be twice as hot as 10 degrees. The key requirement of the interval scale is that units be clearly defined.

    Ratio scale is the highest level of measurement much the same as interval scales except zero is a non-arbitrary point allowing mathematical operations such as division. Examples of ratio scale measurements are weight and length.

    The next important component is the quality of measurements taken. The two facets of measurement quality are reliability and validity. First reliability, which simply means how repeatable the results are and low reliability indicates that many random errors are occurring during measurement. Second, the concept of validity, which means the metric measures what it intended to measure. Errors in validity, called systematic errors, are usually experienced because not all factors were taken into account when the metric was designed.

    The next component of measurement is causality. The cause and effect relationship has three requirements:
  1. cause precedes effect in time or by logic.
  2. shown cause and effect relationship
  3. direct constant relationship
    When the relationship between variables can be graphed and a line of best fit can be found, however be careful of extreme and inaccurate data values. When defining a casual relationship be sure to watch for the following spurious relationships.

    CASE 1
        Z causes changes in X
        Z causes changes in Y
        X is perceived to causes changes in Y or vice versa and Z is overlooked

    CASE 2
        X causes changes in Z
        Z causes changes in Y
        X is perceived to causes changes in Y and Z is overlooked

^Top^

Many metrics

    Most metrics in software quality assurance fall under one of two categories, product metrics and process metrics.

    Product metrics are for describing characteristics of product such as it’s size, complexity, features, and performance. Several common product metrics are mean time to failure, defect density, customer problem, and customer satisfaction metrics.
  • Mean time to failure metric is, to put it plainly, the average time the product runs before experiencing a crash, which is important for systems like air traffic control that are required to have no more than a few seconds of down time in a full year.
  • Defect density metric refers to number of imperfections per:
    • Lines of code
    • Function definitions
    • Lines on input screens
  • Customer problem metric is a measure of problems customers have encountered with the product over the total usage of the product. This metric takes into account that multiple instances of the product can be used at the same time, which effectively multiplies the length of time the product has been in operation by the number of product licenses.
  • Customer satisfaction metric is generally a survey asking customers to rate their satisfaction with the product and/or it’s features on a five-point scale.

Process metrics are strictly for evaluating & improving the effectiveness of development and maintenance processes. Some common process metrics are defect arrival pattern, backlog management index, fix quality, and fix response time metrics.
  • Defect arrival pattern metric is a combination of the rate at which defects are reported and the growth of these rates through out the process being measured.
  • Backlog management index metric measures the effectiveness of the process by comparing the number of problems that arrive to the number of fixes during a specified period.
  • Fix quality metric measures the number of defective fixes compared to the number of successful fixes. A fix is defective when either it doesn’t repair the problem reported or it repairs the problem but introduces new defects into the product.
  • Fix response time metric refers to the time between when a problem is reported and the time a fixed is issued.

^Top^

^Table Of Contents^ Naming Conventions >>