Are models easier to understand than code? An empirical study on comprehension of entity-relationship (ER) models vs. structured query language (SQL) code

Abstract

Models in Software Engineering are considered as abstract representations of software systems. Models highlight relevant details for a certain purpose, whereas irrelevant ones are hidden. Models are supposed to make system comprehension easier by reducing complexity. Therefore, models should play a key role in education, since they would ease the students' learning process. Although these statements are widely accepted, to the best of our knowledge, there is no empirical evidence that supports these hypotheses (beyond practitioners' personal experience). This article aims to contribute to fill thisgap by performing an empirical study on how well students understand entity-relationship database models as compared to structured query language (SQL) code. Several ER models and their corresponding SQL code (more specifically, the data definition language (DDL) statements required to create such models) were shown to a heterogeneous group of students, who answered different questions about the database systems represented by these artifacts. Then, we analysed the correctness of the answers to check whether the ER models really improved students' comprehension.

Keywords:

• entity-relationship models
• SQL
• empirical research
• software models
• abstraction

Acknowledgements

We thank the companies (Nuclenor, Predictia, Semicrol and Suomitech 6) which helped us by providing different case studies that served as guide to prepare the material for the experiments. We also thank all the students who participated in the experiments. We really appreciate their extra effort, collaboration and commitment. Finally, we express our thanks to our colleagues Carlos Blanco and Daniel Sadornil, who gave a hand whenever it was required.

Notes

1.  http://www.nuclenor.org/

2.  http://www.predictia.es/

3.  http://www.semicrol.es

4.  http://www.suomitech.com/

5. We would like to point out we refer to the median instead of the mean because as thedistribution of the data is unknown we prefer to refer to a robust measure of centrality.

6. All acknowledgements are given in alphabetical ordering.