Teaching Replication in Quantitative Empirical Economics
Abstract
In empirical economics, a twofold lack of incentives leads to chronic problems with replicability: For authors of empirical studies providing replicable material is not awarded in the same way as publishing new irreplicable studies is. Neither is authoring replication studies.
We offer a strategy to set incentives for replicability and replication. By integrating replication studies in the education of young scholars, we raise the awareness for the importance of replicability among the next generation of researchers and ensure that a big number of scientists get incentives to write replication studies: credit points and the prospect of publications at least of working papers already during their time as students.
By raising the number of researchers involved in replication and by providing an infrastructure for sharing their information, on the one hand we help to lower the amount of work researchers need to put into making their studies replicable. On the other hand, we facilitate the dissemination of insights derived from replication studies. This as a side effect imposes a significant threat of detection of irreplicable research, following the cases of recently introduced wiki projects for the revelation of plagiarism. In contrast to previous efforts like the report on the American Economic Review Data Availability Compliance Project, with our project we build the basis for the first replicable review paper on reblicability as we give account of which studies were tested and which results were found in each case. After exploring several dozen studies published in highly ranked journals, we have not yet determined a single case where we see replicability is fully ensured. We identified two main problems: First, not all published results can be obtained from the replication material provided. Second, information about how the used data were obtained from the raw data is hardly ever sufficient.
For our investigation, we gave seminars at several faculties. We set up a wiki project for documenting the results of our replications as well as those found in the literature. In our database, we provide information about more than 1500 empirical studies, especially with regards to the availability of material for their replication.
We invite for discussion to develop standards for how to make research replicable and how to write replication studies. For this we provide information about existing projects that facilitate the sharing of material for empirical econometric research.
Do you think the six journals your paper focuses are the top journals of econometrics in the world? If so, why?
Haiyun Zhao
We chose the American Economic Review because it was the leading force in introducing an online archive for data and code for replication and it is among the top journals that publish applied econometric studies. The Journal of Political Economy followed the same policy and is also among the very top journals. The four American Economic Journals were introduced by the American Economic Association because they saw that so many good papers were submitted to the American Economic Review and underwent some editorial process but not all of them could be published and not all of them had the quality to be published in similarly renowned journals so they offered the authors the option to choose the option to undergo a quicker review process and get published in these field journals that are good but not among the very top. We included these journals because they have the same replication policy, so material is easily available. Many studies published there are a bit more accessible to students. And it is a good opportunity to check if there is a difference in replicability between journals that are among the very top and those that are still very good but not quite as well established.
Certainly to speak of top journals in econometrics we would also need journals that publish theoretical econometric studies but that was not our focus. Another journal that is important in the field of replication as it has an online archive and an own replication section is the Journal of Applied Econometrics. We however do not suggest our students to replicate studies from that journal for two reasons. Since this journal focuses on advances in econometrics many of the studies published there are quite complicated for many students. In addition, their policy is not to ask for the original code because they think for a true replication the code should be written again. I see good reason for this. For seminars however students cannot be expected to write code that often took researchers many months to write for a publication. Also in case of deviations from published results it could require quite some effort to find out about the reasons and it would often require contact to the original authors and you never know how fast they will reply. In our seminars students also often get in touch with original authors but the way we do it they do not depend on them, otherwise it would not be feasible for seminars.