Model, model, everywhere, but what problems could it bear?

Abstract

We, I mean colleagues in academia, like to develop models and want to get them published. In fact, a large proportion of the submissions to QREI contain newly developed models. In this editorial, I would like to share with you what are the key elements that we look for in screening/reviewing these submissions. I hope this will serve as a guide for young colleagues in preparing their research work for submission. The key purpose of a model is to solve a real problem; preferably a significant, recurring problem that is experienced by people in a certain sector of the industry. For example, profile monitoring was motivated by process control in semiconductor industry to ensure uniformity across a piece of wafer. In the other words, the context from which the model was conceived must be well articulated. This will address the concern that “…while the paper contains interesting theory development but will there ever be an application” or simply put, the classic case of “hammer looking for nails”. This has been well-expounded in an editorial piece contributed by Professor Brombacher, one of the Co-Editor-in-Chief, in 2015 that commented on “Conceptual studies”. The next consideration is that the model should be implementable. Very often, we encounter papers that presented a model, followed by a numerical example in which the values of the unknown parameters were given. Unless there is an obvious means of obtaining these values, the research work is deemed to be incomplete. As the “test of the pudding is in the eating”, the real test of a model is how does it work in the face of data arising from the problem (i.e. when it is applied to the actual scenarios). The parameters should be estimated from some real-life data or pseudo data if the associated data are confidential. This will also help to identify the data field and the preferred or correct data collection scheme for the model parameters to be estimatable. Sometimes, from the data structure, one could provide additional insight on the ease of implementation of the proposed model in relation to other competing models. Moreover, the associated sampling error and the variability of the estimated parameter could also be quantified. A classic example on this implementation issue is the Phase 1 problem for SPC. For a model to be adopted, it should outperform other competing models that address the same or similar problem. Inevitably, a comprehensive comparison is expected and the strength of the model in some aspects (not necessarily all aspects) should be clearly illustrated. In particular, the circumstances in which the model works best against others should be convincingly put across. The weaknesses of the model should also be discussed. The above is a summary of three key ingredients in preparing a model-centric paper. It is not meant to be a standard or template. Any other ways of presenting the relevance and utilities of a model to solving practical problems are welcome.