The dangers of categorical thinking

Categorical thinking (thinking about a problem as relationships between coarse, distinct groups) needs to be valid and useful to be of any value. The article highlights 4 common pitfalls of categorical thinking, which might prevent that: compression, amplification, discrimination and fossilization.

Reading outline

  • We tend to think in categories: distinct entities which we map over a messy reality. We do that consciously and unconsciously.
    • For example. Hearing a difference between how we say ta and da comes down not to the movement within our mouths (which is identical in both cases), but down to the attack of the sound, the ‘voice onset time’ – if that time is less than 40 milliseconds, we hear ta, if it’s more, then we hear da.
    • Our brain is constantly categorizing the messy reality around us into distinct groups, allowing us to make quick decisions.
  • The way we categorize is often context-specific.
  • For categorization to have value, it must fulfill two criteria: the categories must be valid, and they must be useful.
    • Valid, means that the categorization must not be arbitrary. You can’t just divide a homogenous group into arbitrary parts.
    • Useful, i.e. there must be a difference in the categories that you care about.
      • Example: it’s useful to differentiate snakes from sticks, because that will help you survive.
  • In business (or generally in any organization) we often use categories, which are neither useful, nor valid.
    • Example: 80% of Fortune 500 companies include the Meyers-Briggs type indicator into their HR decision making.
      • The 93 questions of the test only have two possible responses to them. The problem is that these questions demand complex, continual assessment. For example, few people will be able to answer confidently if they make their decisions based on facts or intuition (let alone objectively), yet you have to choose in the Meyers-Briggs test.
      • The output categories of the test are ‘extravert’ or ‘introvert’, ‘judger’ or ‘perceiver’ – categories that are not valid, according to the authors.
      • Even though some categorizations (like the Meyers-Briggs type indicator) have incredible staying power, and continue to be employed by a lot of organizations, because they create a powerful illusions (see below).
  • Categorical thinking can be dangerous through: compression, amplification, discrimination, and fossilization.
    • Compression eliminates any in-category variance between members, which might lead to crucial information loss. {#compression}
      • The ‘average woman’ contest of 1945 by a newspaper in Cleveland offers a stark example, how prototypical categorization can be wrong.
        • A newspaper in Cleveland ran a contest in 1945 to find the anatomically prototypical woman.
        • Not long before, a study had determined the average values for a variety of anatomical measurements, and the paper’s editors used those measurements to define their prototype.
        • A total of 3,864 women submitted their measurements.
        • None of them were close to the average on every dimension.
      • Another good example in a similar vein is a study by Foroni and Rothbart where perception of body shapes before and after categorization was studied.
        • The participants were shown silhouettes of people twice: once without categorization, and once categorized from ‘anorexic’ to ‘obese’.
        • The participants tended to assume more similarity between categorized silhouettes (up to assumptions about similar lifestyles).
      • Segmentation studies often show how categorization can be applied carelessly in a business setting.
        • Segmentation is the most common tool used by marketing departments, where the goal is to separate customers into categories and then identify target customers.
        • A target customer is a category of customers that deserves special attention and strategic focus.
        • Typically conducted by asking customers a serious of questions about their behavior, desires, and demographic characteristics.
        • Then a clustering algorithm then divides respondents into groups according to similarities in how they answered.
        • Marketers rarely evaluate whether the clusters are valid.
        • Rather they move on to profiling, personas, etc.
        • This is how categories like ‘minivan moms’ are born, even though not everyone in those categories might have a minivan or children.
      • Another good example of categorization misuse leading to compression is (of course) the financial industry.
        • During early 2000s it was enough for a company to rebrand itself, putting a ‘dot com’ in its name for the investors to pump cash into its stock.
        • When a company joins a major index, such as the S&P500, its price starts to move more closely with the index, even though nothing about the company has actually changed.
    • Amplification means differences across category boundaries can be exaggerated. {#amplification}
      • Members of opposing political parties tend to overestimate the extremity of each other’s views.
      • In firms, teams might struggle to work together when there’s a strong perception of difference between organizational categories, i.e. ‘techies’ not understanding business or ‘MBAs’ not understanding technology.
      • Decision-making can suffer, when decisions are based on a cut-off or threshold based on categories, specifically due to amplification: the cut-off might be in reality arbitrary, when the difference between categories isn’t great.
      • A/B testing might not be a rigorous tool, since the differences between category ‘A’ case and category ‘B’ case might occur just through random chance or an unknown variable in the problem domain, making the distinction between these two categories meaningless.
    • Discrimination occurs when to much weight/investment/attention is given to only some categories. {#discrimination}
      • Highly targeted ads usually have negligible impact on ROI or probability of purchase (as explained in this 2018 study) meaning focusing only on ‘key segments’ might be detrimental to your business.
      • Net promoter score has a problem which might lead you to overestimate the amount of detractor clients.
        • Customers are asked (on a scale of 1 to 10) how likely they are to recommend the product to their friends and family.
        • Answers are clustered into three categories: detractors (0-6), passives (7-8), and promoters (9-10).
        • The net promoter score is then calculated as:
          NPS=promotersdetractorspromoters+passives+detractors\text{NPS}=\frac{\text{promoters}-\text{detractors}}{\text{promoters}+\text{passives}+\text{detractors}}
        • ==Notice how the score 6 is way closer to 7, rather than 0, but still lands in the detractor category.==
        • Another problem the categorization of net promoter scores is the blindness to passives.
          • Case A: There are 50 promoters, 0 passives and 50 detractors.
          • Case B: There are 0 promoters, 100 passives and 0 detractors.
          • In both cases the NPS is the same (0), but the two citations are very-very different: in one case your product is polarizing, while in the other irrelevant.
    • Fossilization happens, when established categories are taken as gospel and not reviewed. {#fossilization}
      • People tend to forget that categories are not how things are, but rather how someone chose to organize the world.
      • “The difficulty lies not in new ideas, but in escaping the old ones” — John Maynard Keynes
      • Example: the decline of the Schwinn Bicyle Company
        • Dominated the market in the 1950s focusing on the youth market
        • Failed to recognize that there is a market for adults, which grew substantially up to 1970s. {#schwinn-decline}
        • Consumers chose European and Japanese bicycles instead.
        • Lead to a slow and painful decline of the company due to a fossilized view from decades of success selling bikes to kids.
      • Innovation can be hampered if you stick to your old categories too much.
        • Example: in a study from the University of Toronto in 2016 people were asked to build an alien with legos.
          • One group was given a random assortment of bricks.
          • Another group was given bricks grouped into different categories.
          • A third group needed to rate the creativity of both groups. The group with the random assortment of bricks won.
  • You limit the dangers of categorical thinking by: increasing awareness, analyze the data, audit your decision criteria, regular ‘defossilization’.
    • Increase awareness
      • Make sure to remember that categories are simplifications
      • As such, they might invite biases
    • Develop capabilities to analyze data continuously
      • Continuous analytics are key in reviewing and recalibrating established categories, like client personas, for example.
    • Audit decision criteria
      • Any categorical decision criteria needs to be challenged and refreshed to continue to make sense.
    • Schedule regular ‘defossilization’ meetings
      • Hold regular meetings at which you crutinize your most basic beliefs about what is happening in your industry and how the world is.