Science Whats in a name? Television adverts are more successful if they show the product first and brand name later, according to research examining which creative factors are most likely to be effective. Liam Kay-McClean reports A dvertising, depending on who you speak to, can be seen as an art or a science. On the one hand, it is intended to stir emotions, draw in consumers and maintain a loyal fanbase. On the other, there is the need to generate repeated sales, encourage unconscious purchasing decisions and appeal to behavioural traits. One recent study, led by researchers from the University of Otago in New Zealand and EhrenbergBass Institute for Marketing Science in Australia, sought to examine how advertising could be made more effective, and the factors affecting its likelihood of success. The resulting paper, Finding creative drivers of advertising effectiveness with modern data analysis, makes several suggestions for improving advertising effectiveness and the means by which researchers analyse campaigns. The research set out to address two issues with previous studies of advertising effectiveness, the first being the way that the relationships between creative variables and various advertising outcomes had been determined often using ordinary least squares or logistic regression models. The models assume that the effect of explanatory variables is independent of the effect of other variables, according to the paper. The second issue is that most studies on advertising effectiveness lack replication of analysis, as they apply a single analytical method and do not use a hold-out or alternative data set for validation. This means that an alternative methodology or algorithm could lead to the results of an experiment not being replicated, according to the authors. To address this, the research team applied a modern data analysis paradigm to a data set originally reported by Hartnett, Kennedy et al (2016) that examined creative variables from 312 television advertisements against their short-term sales using an ordinal logistic regression model. The data for the 2016 study was provided by a consumer packaged goods company, and the adverts were aired and measured in five developed markets for more than 60 brands. The advertisements were accompanied by a commercially validated measure of sales effectiveness using single-source data. In the 2023 research, researchers re-analysed the same ads using traditional, artificial intelligence and machine-learning models, to identify which creative variables contributed the most to sales performance. John Williams, director of the Bachelor of entrepreneurship at the University of Otago and one of the report authors, says it was important to use a variety of methods in the research. Every statistical method, model or algorithm has pros and cons, and assumptions that are often not met in practice and where the impact of those assumptions being violated are rarely known and difficult or impossible to quantify. If we use several methods equally and they all give the same results, then we have more 44 Impact ISSUE 43 2023_pp44-45 Science.indd 44 18/09/2023 12:16