Recent software developments are reviewed from the vantage point of reproducible econometric research. We argue that the emergence of new tools, particularly in the open-source community, have greatly eased the burden of documenting and archiving both empirical and simulation work in econometrics. Some of these tools are highlighted in the discussion of three small replication exercises. / Series: Research Report Series / Department of Statistics and Mathematics

<p>With the general engineering practices being followed for the development of scientific software, scientists are seemingly able to simulate real world problems successfully and generate accurate numerical results. However, scientific software is rarely presented in such a way that an external reviewer would feel comfortable in certifying that the software is fit for its intended use. The documentation of the software development - Requirements, Design and Implementation, is not being given the importance it deserves. Often, the requirements are improperly and insufficiently recorded, which make the design decisions difficult. Similarly, incomplete documentation of design decisions and numerical algorithms make the implementation difficult. Lack of traceability between the requirements, design and the code leads to problems with building confidence in the results.</p> <p>To study the problems faced during certification, a case study was performed on a legacy software used by a nuclear power generating company in the 1980's for safety analysis in a nuclear reactor. Unlike many other scientific codes of that time, the nuclear power generating company included a full theory manual with their code. Although the theory manual was very helpful, the documentation and development approach still needed significant updating. During the case study, 27 issues were found with the documentation of the theory manual, 2 opportunities to update the design and 6 programming style issues were found in the original FORTRAN code. This shows room for improvement in the documentation techniques in the development of scientific software based on a physical model.</p> <p>This thesis provides a solution to the certification problem, by introducing software engineering methodologies in the documentation of the scientific software. This work proposes a new template for the Software Requirements Specification (SRS) to clearly and sufficiently state the functional and the non-functional requirements, while satisfying the desired qualities for a good SRS. Furthermore, the proposed template acts as a checklist and helps in systematically and adequately developing the requirements document. For developing the design and implementation, this thesis introduces Literate Programming (LP) as an alternative to traditional structured programming. Literate Programming documents the numerical algorithms, logic behind the development and the code together in the same document, the Literate Programmer's Manual (LPM). The LPM is developed in connection with the SRS. The explicit traceability between the theory, numerical algorithms and implementation (code), simplifies the process of verification and the associated certification.</p> / Master of Applied Science (MASc)

Documentation is an essential activity in software development, for source code as well as modelling artefacts. Typically, documentation is created and maintained manually which leads to inconsistencies as documented artefacts like source code or models evolve during development. Existing approaches like literate/elucidative programming or literate modelling address these problems by deriving documentation from software development artefacts or vice versa. However, these approaches restrict themselves to a certain kind of artefact and to a certain phase of the software development life-cycle. In this paper, we propose elucidative development as a generalisation of these approaches supporting heterogeneous kinds of artefacts as well as the analysis, design and implementation phases of the software development life-cycle. Elucidative development allows for linking source code and model artefacts into documentation and thus, maintains and updates their presentation semi-automatically. We present DEFT as an integrated development environment for elucidative development. We show, how DEFT can be applied to language specifications like the UML specification and help to avoid inconsistencies caused by maintenance and evolution of such a specification.

Documentation is an essential activity in software development, for source code as well as modelling artefacts. Typically, documentation is created and maintained manually which leads to inconsistencies as documented artefacts like source code or models evolve during development. Existing approaches like literate/elucidative programming or literate modelling address these problems by deriving documentation from software development artefacts or vice versa. However, these approaches restrict themselves to a certain kind of artefact and to a certain phase of the software development life-cycle. In this paper, we propose elucidative development as a generalisation of these approaches supporting heterogeneous kinds of artefacts as well as the analysis, design and implementation phases of the software development life-cycle. Elucidative development allows for linking source code and model artefacts into documentation and thus, maintains and updates their presentation semi-automatically. We present DEFT as an integrated development environment for elucidative development. We show, how DEFT can be applied to language specifications like the UML specification and help to avoid inconsistencies caused by maintenance and evolution of such a specification.