Difference in Differences in Marketing Performance Measurement

Abstract Marketing performance measurement is important in retail companies, it is critical for the budget planning and adjustment of marketing strategies and tactics over time. Compared to digital online marketing, such as social media advertisers where tagging techniques [1] can be applied to track the lifecycle of the consumption and the customer’s behaviours, traditional TV and radio advertising will be harder to evaluate at the individual level. Instead, common approach in this scenario is to study the relationship between ads spend and sales. This paper presents a casual inference methodology – “Difference in Differences”(DID) on the application of TV marketing measurement. Instead of building direct relationship between ads and sales, we attempt to create an experimental research design and study the differential effect on a ‘treatment group’ versus a ‘control group’. In order to study the performance of advertising in a city, the treatment group is sales performance during and after TV advertisings and the control group is the statistical model that mimics the sales of the same city without ads influence. It is proved that the model can significantly predict the sales of city in nature by considering its past sales, the neighbourhood districts and weather conditions. And a case study in industrial data shows that DID methodology can effectively evaluate the TV advertising results on regions.