Empirical Software Engineering
As researchers investigate how software gets made, a new empire for empirical research opens up
Getting Good Programmers
Another holy grail of empirical software engineering research is finding a way to tell good programmers from bad. A widely quoted piece of folklore says that good programmers are 28 times better than average (or 40, or 100, or some other large number). As teacher and developer Steve McConnell discusses in Making Software, the precise quantification of this assertion may be suspect, given that the definition of “better” is elusive in any knowledge-based work. Yet everyone who has programmed for a living knows that there really are huge differences in productivity and capability among programmers. Other than hiring someone and watching her work for a couple of years, how can we spot the stars?
Jo Erskine Hannay at Simula Research Laboratory has addressed the issues of talent and expertise by breaking the question into parts: Can we actually define what it means to be a good software developer? Can we reliably determine that one developer is better than another? And if we can’t, should we surrender on those questions and focus instead on tools and techniques?
Hannay zeroes in on a field called individual differences research, in which individuals are classified by characteristics that separate one from another, such as personality. Although some pseudoscientific schemes used by human resources departments and dating websites have given personality testing a bad reputation, modern programs such as the five-factor model have solid scientific foundations (or as scientists might put it, we have construct validity for the concept of measuring personality). The five-factor model and similar protocols address dimensions that include extraversion, agreeableness, conscientiousness, emotional stability, and willingness and ability to learn from experience. We also know that programmers tend to have certain personality types and that they vary less in personality than the average population. Can we use findings like these to discover good programmers?
The short answer is no. Large meta-analyses and further studies by Hannay and others conclude that a programmer’s personality is not a strong predictor of performance. The people who swear by their beliefs about personality and programmer success have now been given reason to assess their position critically, along with methodological support for doing so.