A survey of Microsoft software engineers had respondents answer a series of questions about software development. The number of respondents was 564, which represents a 22 percent response rate. There was most agreement for the proposition that code reviews improve software quality. There was least agreement for the proposition that code quality depends on which programming language is used. Some respondents thought that programming skill matters more than language while others thought that fewer mistakes are made using statically typed languages. The survey also asked about how opinions were formed. The boxplot of Figure 1 makes clear that personal experience was the most influential factor while findings from empirical research was in fifth position. Several examples are discussed in which the beliefs of respondents are not aligned with research findings. For example, respondents believed that coding standards help improve software quality, but there is no evidence to support this.

Two geographically distributed Microsoft projects were studied in detail. Millions of changes to code were analyzed to determine the numbers of defect repairs associated with each file. Regression modeling, which took account of confounding factors, found that both projects were essentially comparable in terms of code quality. Members of one of the projects, however, had previously disagreed with the proposition that being geographically distributed did not affect code quality.

A convincing argument is made for improved dissemination of research findings and for software engineers to base their decisions on such findings. This paper is very strongly recommended to the software engineering community.