Often what we think are valuable insights from marketing analytics are often skewed by common biases – resulting in flawed marketing analytic reports. The problem is that every marketing analyst has an interpretation and opinion about consumer data. Yet research has shown that when presented with a set of facts most people will interpret the facts in a way that bolsters their respective opinions. One study showed that when people were presented with a set of facts they used their interpretations (of those facts) as an opportunity to bolster or rationalize their existing partisan views and opinions. These divergent views, opinions and interpretations often lead to faulty conclusions and erroneous marketing decisions.
See Also – Video: Sources of bias: How data goes bad
The paradox is that consumer behavior generates a litany of data about consumer preferences and attitudes. Yet the (raw) data is meaningless and requires that a presumably biased human give meaning to the numbers, draw inferences from them, and define their meaning through his/her interpretation.
Some common cognitive and data biases include:
- Confirmation Bias: occurring when there’s an intentional or unintentional desire to prove a hypothesis, assumption, or opinion.
- Selection Bias: occurring when the population sampled is non-random and does not represent the actual population. This is common in survey research.
- Data Outliers: these are extreme data values that are significantly outside the range of normal values or the pattern of normal distribution. These values can significantly skew data.
- Simpson’s Paradox – this can occur when a trend indicated in distinct groups of data can reverse when these groups of data are combined. This is often the explanation for seemingly successful marketing campaigns proving to be unsuccessful.
- Overfitting/Under-fitting data models: Overfitting involves models that are overly complex and includes noise. Under-fitting involves models that are over-simplified.
- Confounding Variables: Occurs when a perceived relationship between two variables is proven partially false or entirely false because a confounding third variable has been omitted or overlooked.
- Non-Normality: Occurs when some statistical tests, e.g. a t-test, assume that a normal distribution of the data exists. However, if the data is not normally distributed the results may be biased or misleading
Is there a way to interpret this mountain of evidence about consumers and their behavior in a way that is not tainted by interpretation? Is it possible to objectively evaluate this data to find meaning in consumer behavior and solve persistent, perplexing consumer data problems?
The solution to bias-free interpretation lie in a confluence of improved research methodology and big data analytics. Marketers need to carefully evaluate where their data comes from, the methodology used to gather and analyze that data, and what cognitive biases they might bring to its interpretation (https://hbr.org/2013/04/the-hidden-biases-in-big-data/). Implementing best practices in data collection and data management can help generate relatively clean raw data while improving the universe of existing data. Applying marketing analytics approaches and computation analysis to cleaner data will model patterns, trends, and associations about human behavior and interactions (Google.com) that are closer to actual consumer behavior.
Anthony Hawkins is a graduate student in the Masters of Digital Marketing Analytics program @Aurora University. He is a digital marketing and analytics professional passionate about the natural and organic foods market.