How data analysts experience cognitive biases and how to recognise them: Part 2

21 February 2018

Cognitive biases: We all have them and we regularly encounter them in everyday life. But how do data analysts actually experience these biases? And how can you recognise them? In this series we will discuss cognitive biases in the context of analysts. Each blog picks out 3 biases from 20 categories.

In this blog, there are two biases from the “Too Much Information” spectrum: the Availability heuristic and the Contrast effect, and one bias from the “Not Enough Meaning” spectrum: the Clustering illusion.

The Availability heuristic

A heuristic is a mental shortcut that your brain takes when you have to make a decision. The Availability heuristic says that the thoughts you have recently had available to you will be selected again more quickly. Example: You are working in the open-plan-office and have to decide what analysis to apply to the data. Beforehand you have just received an e-mail about a seminar called “New techniques in random forest modelling”. Somewhere in the office two of your colleagues are talking about pruning the trees in the garden. You are now more likely to select decision trees as your technique. Not because this technique is particularly appropriate for your data: purely because this technique is “available” to you. Consequently you assume there must be some reason why this technique came to mind (because, of course, you were busy working hard and not surreptitiously eavesdropping on your colleagues or reading your e-mail).

The Contrast effect

The Contrast effect says that when you are presented with two things simultaneously or in quick succession, you will compare them. Even if these things have absolutely nothing to do with each other. Example: You are hard at work and have built a model that only has a hit rate of 47% – i.e. your model works less well than when you go on random selections. The second model you make has a hit rate of 61% and so you feel much better! You compare 61% with the 47% hit rate and therefore you feel your model is better than you would if you had not had that comparative information. So the next time you are satisfied with your re-make of a 3D pie chart: Ask yourself whether it really is a good graph or whether it is the contrast that makes you satisfied.

The Clustering illusion

For analysts, the clustering illusion is not only relevant for cluster analysis (what’s in a name), it is also relevant for factor analysis, correlations, regression analyses and more. This illusion means you see clusters or correlation in small, random samples. In other words: You think the data isn’t randomly distributed, even though it is. This bias occurs due to the fact that people like to have correlation and predictability and therefore try to find it everywhere. So, next time you are sure you have observed a pattern between the number of phone calls to helpdesk per day and the number of the day in the Chinese calendar… Think again :-).

Also read the other parts:

Part 1: Self-relevance effect, Confirmation bias and the Bias blind spot.

Part 3: Essentialism, the Positivity Effect and the Appeal to Probability Fallacy.

Part 4: Curse of Knowledge, the Hindsight Bias and the Hard-Easy Effect.

Part 5: Sunk Cost Effect, the IKEA effect and the Google Effect.

Part 6: Ambiguity Bias, Appeal to Novelty and False-Consensus Effect.

Contact

Do you want to know more about this subject? Please contact us on +31 (0)33 258 28 30 or info@cmotions.nl.