Random Thoughts on Marketing Research, Statistics and Data Science (July 3, 2018)

My opinions. Nothing more. Why spend money on marketing research that is faster, cheaper…and cr*ppy? Wouldn’t it make more sense to trust our gut than untrustworthy data? Why make a bad decision more quickly? A penny saved is a penny earned.

It would be not be a huge exaggeration to say chaos and confusion reign in data science. To begin with, the term itself implies many things and is interpreted in a multitude of ways.

Data scientist and data science are often lumped together in such a way that many job descriptions are for unicorns, instead. They list an assortment of IT skills, numerous programing languages, and experience with statistics, machine learning and software development – AI, especially. Not to mention the soft skills.

Sorry, this describes a team, not an individual. Moreover, data management is not data analysis. Data analysis is not just descriptive statistics and visualization. Pattern recognition is not data analysis either, just an early step in the analytics process. Last, but not least, insights are not descriptive findings. It takes a human to turn data into insights.

Meanwhile, around the globe, millions of talented, ambitious young people have been convinced that coding paves the road to riches…

Some marketing researchers may wonder “What has all this got to do with me?” Unfortunately, quite a lot. One reason to worry is that many in the data science community essentially see all data as the same, apart from volume, velocity and structure. Their focus is on data management or simple predictive analytics.

From that perspective, marketing research data is much the same as any other data, and knowledge of marketing or research is not needed in order to mine “insights” from it. This is a threat to MR and marketing I feel should receive more attention.

We humans debate whether something is this or that endlessly and sometimes, tragically, even go to war over it. We often forget that neither this or that actually exist – they are verbal labels.

Take Myers-Briggs, for instance – wouldn’t it be better just to look at dimension scores instead of 16 personality types that don’t actually exist? Another example, is NHST. Though now a pariah in the eyes of many statisticians, Null Hypothesis Significance Testing recognizes this ingrained need of ours to categorize, usually into…this or that. We are far less comfortable with probabilities, a human foible easy for statisticians to lose sight of because of the way we’re trained to think.

Personally, I’d rather look at distributions of effect size estimates than asterisks. But, if my education and career path had taken a different course, asterisks would most likely have been my choice.

Oops…this or that…

IMO, branding is now more important than ever.

Unfortunately, fundamental marketing knowledge and commonsense is being lost, perhaps in part squeezed out by a near obsession with programming and tech.

I recall one fellow, an American with an MBA, defining marketing as “selling to a lot of people.” Sorry, but it’s a bit more than that… Another example is a claim that the appearance of food and its packaging affects its taste perceptions (true) and than we’ve only learned this recently through neuroscience (baloney). Kinda think that’s something babies figure out sans fMRI. Yet another is that, once again, it’s being re-discovered that humans make decisions that aren’t always as tidy as mathematical formulas.

If marketers aren’t taken seriously at C-Level, small wonder why.

If you want to extract some general information from a vast amount of social media or other text data very quickly, automated text mining (now often called “AI”) can do this for you.

However, if instead you want in-depth analysis and true insights, a skilled and experienced qualitative researcher is your best resource until Artificial General Intelligence arrives.

That might be a long wait. Text Analytics: A Primer is a short interview with Professor Bing Liu, a noted authority on text mining, and provides an overview of the current strengths and limitations of text mining.

Arrange a Conversation