In the world of analytics there are a number of artificial barriers we put in place to separate disciplines. Biostatistics to the right, pure theoretical statistics to the left, econometrics in the back corner, Bayesians center stage, oh and marketing research can wait in the hall. Not only are these implicit value judgments we put upon the work done in each field, they become mental walls that researchers put upon themselves. Despite the interconnected nature of information today, we often turn only to the tried and true methods of our disciplines while ignoring the insights of our fellow researchers across the aisle. Once we broaden our vision, we quickly notice the methods other disciplines have to offer that directly address the needs of our industry.
Two Camps of Marketing Research, Two Camps of Medical Research
For the sake of our discussion let’s say that marketing research can be separated into two camps: econometric modeling and customer modeling. Any aggregate level models will fall under the umbrella of econometric modeling and any individual level models will fall under the umbrella of customer modeling. Therefore econometric modeling encompasses price / promotion, media mix and forecasting, while customer modeling encompasses multivariate testing, attribution modeling and engagement scoring.
For each of these marketing research camps there is a corresponding camp in medical research. Epidemiology, the study of the distribution and determinants of health-related issues, easily corresponds to econometric modeling. Both examine the impact of factors on groups and seek predictive information from observational data. Clinical studies, clinical trials and other forms of patient-based research easily correspond to customer modeling. Both examine the impact of factors on individuals and experimental designs whenever practical.
What We Can Learn From Epidemiology and Clinical Studies
Case control studies are a cornerstone of epidemiological research. Individuals with a condition are matched to similar individuals then further analysis is used to determine the differences that exist between these groups. This use of observational rather than randomized data bodes well for the marketing researcher. Often, we do not have the luxury to institute large-scale tests and need to make recommendations off existing data. When given the opportunity to create experimental designs, such as market level media tests, the matching algorithms developed for cancer studies provide excellent results when needing to determine which DMA is most like Omaha, for example.
Cohort studies follow at-risk individuals through time. Cohort groups are created based upon their exposure to the risk factors of interest, such as smokers and non-smokers, and are then tracked over time to monitor for differences in outcomes. Website analysts can track registered consumers, create cohorts based on a number of factors collected at time of registration and analyze the groups over time for their relative likelihood of conversion/upgrade.
Case-series studies follow individuals over time and use Poisson regression to study the rate at which disease occurs pre and post exposure to the factor of interest. The most common example in the medical research world is examining the side effects of vaccinations or medication. A parallel in the marketing research world would be examining consumers’ conversion rates pre and post exposure to CRM efforts.
Survival Analysis studies the time elapsed until death for test and control groups. This type of analysis has made major inroads into other arenas of research under the title of duration analysis. Survival Analysis has numerous applications in marketing research ranging from CRM and loyalty programs to web site engagement analysis.
Toxicity studies determine the amount a substance that is required to cause harm to an organism. The concept of acute toxicity could be translated into depth of discount analysis for marketing researchers.
Stopping rules are a standard practice in phase III clinical trials. An interim analysis is performed on the data to determine whether a study should continue or come to an early end. This methodology is notably different from the process used in MVT testing software to determine unsuccessful elements that should be dropped from a testing matrix. In the MVT setting, elements that reach negative statistical significance are routinely dropped from the continuing test to maximize the number of observations for the remaining elements. Stopping rules bring an end to a study both in the face of efficacy or futility. Whether choosing a Haybittle-Peto boundary, a Pocock boundary, an O’Brien-Fleming, or some variant utilizing the power family method, your study can come to an abbreviated end once substantial evidence is gathered. Stopping rules such as these have somewhat integrated themselves into operations research and industrial experimental design but have yet to take root in marketing research despite that fact that they are a natural fit with focus groups and new initiative testing.
Outbreak investigations use a number of steps outlined by the Center for Disease control. Researchers study the existence of an outbreak, define who is infected, map the spread of an outbreak, hypothesize the root cause and the develop prevention systems. Marketing researchers can utilize the same techniques used to examine the epidemic curve to measure the spread of viral marketing or social media.
As you can see from this handful of examples, there are many parallels between the analytics of medical research and marketing research. If we take the time, we will find approaches we can borrow from disciplines such as engineering, geology, actuarial science, sociology, etc. By becoming aware of the methods other disciplines utilize, we are provided a number of tools to solve the Next Question.
Via | Three minds by Jonathan Prantner, Director of Advanced Analytics at Organic
