Empirical results can be important to show relations that work, but are not necessarily explainable. In my field of functional size measurement of software Albrecht discovered an empirical relation between design elements and build effort in the late 70's. This became the empirical FPA method for functional size measurement. About twenty years later the COSMIC method was designed from scientifically proven software engineering principles. I believe that the empirical results on the use of FPA have been a great help to develop a scientifically built method.

In other fields of Software Engineering there are much more examples I'm sure.

As we are attesting various methodologies to new paradigms of software engineering , from this fact we need empirical software engineering to have a proof of which methodology can be applied on which case study , so it gives us some indications of strength and weaknesses of the software development process.

A good source for information the impact of software engineering research upon software engineering practice is the so-called "Impact Project". This project also considers the role of empirical findings.. Examples cases for impacts in the areas of middleware, inspections, configuration management, and programming language design are systematically analyzed. The website for the project can be found here:

http://www.sigsoft.org/impact/publications.htm

Empirical results are needed to validate new concepts. In an applied discipline like software engineering, this is important. Otherwise, we will not be able to distinguish from validated proposals from interesting but yet unvalidated proposals.

The "Impact Project" mentioned by Jürgen seems not be so active anymore. The last publication added was in 2008. It looks like the ACM is not actively supporting this anymore. The publications on the website are still a valuable source though.

Agree with Anurag. Without any evidence, improvements risk to be done as "best guess" and "trail and error" approach. Moreover, errors made once may be repeated.

So, empirical research can provide some insights and evidence. However, the point is that results are made public in a reproducible manner. Unfortunately, reporting is often poor (especially in SPI): no detailed context description, single case argumentation, incomplete research settings...

IMO, a major challenge is the definition of useful and general metrics (and corresponding measurement procedures). The problem is (IMO) that everybody collects somewhat different data - just to prove novelty to get a paper accepted (not necessarily the authors' fault). This makes, among others, comparability and transfer hard. We lack replication studies!

Over the last decade, it has become clear that software engineering is fundamentally an empirical discipline: Software development practices and technologies must be investigated by empirical means in order to be understood, evaluated, and deployed in proper contexts. The main basis for development in any software discipline is empirical verification of knowledge. Therefore, I think it is essential in software development field