Learner can identify misleading visuals and can name and apply correlation≠causation, survivorship bias and Simpson's paradox.
How charts lie and how reasoning goes wrong
How charts lie and how reasoning goes wrong
A technically correct number can still mislead if the chart that presents it is poorly designed. The most common chart-honesty problems are: (1) a bar chart whose y-axis does not start at zero — this makes small differences look enormous; (2) a truncated or stretched scale that exaggerates or hides variation; (3) missing axis labels or units, so the reader cannot tell what is actually being measured. Rules of thumb: always start bar chart axes at zero, always label both axes with the variable name and unit, and always give your chart a clear title.
Beyond chart design, three thinking traps cause analysts to draw wrong conclusions from correct data. Correlation ≠ causation: two variables can move together because a third hidden factor (a confounder) drives both. Ice-cream sales and drowning rates both rise in summer — the confounder is hot weather, not ice cream. Seeing a correlation never proves that one variable causes the other.
Survivorship bias: you analyse only the cases that survived a selection process and miss the ones that did not. The classic example is Abraham Wald's WWII study of returning bombers. Returning planes had bullet holes everywhere except the engines. The Army wanted to add armour where the holes were — but Wald pointed out that the planes that were hit in the engines never made it back. The data was biased toward survivors. Armour should go where the survivors were NOT hit.
Simpson's paradox: an aggregate trend can reverse completely when you look inside subgroups. A university might show a higher overall acceptance rate for Group A than Group B, yet within every individual department Group B has a higher rate. This happens because Group A applied mostly to easy-to-enter departments and Group B applied mostly to competitive ones. The aggregate hides the composition. Always ask: could a subgroup structure be reversing this trend?
Lesson notes
How charts lie and how reasoning goes wrong
A technically correct number can still mislead if the chart that presents it is poorly designed. The most common chart-honesty problems are: (1) a bar chart whose y-axis does not start at zero — this makes small differences look enormous; (2) a truncated or stretched scale that exaggerates or hides variation; (3) missing axis labels or units, so the reader cannot tell what is actually being measured. Rules of thumb: always start bar chart axes at zero, always label both axes with the variable name and unit, and always give your chart a clear title.
Beyond chart design, three thinking traps cause analysts to draw wrong conclusions from correct data. Correlation ≠ causation: two variables can move together because a third hidden factor (a confounder) drives both. Ice-cream sales and drowning rates both rise in summer — the confounder is hot weather, not ice cream. Seeing a correlation never proves that one variable causes the other.
Survivorship bias: you analyse only the cases that survived a selection process and miss the ones that did not. The classic example is Abraham Wald's WWII study of returning bombers. Returning planes had bullet holes everywhere except the engines. The Army wanted to add armour where the holes were — but Wald pointed out that the planes that were hit in the engines never made it back. The data was biased toward survivors. Armour should go where the survivors were NOT hit.
Simpson's paradox: an aggregate trend can reverse completely when you look inside subgroups. A university might show a higher overall acceptance rate for Group A than Group B, yet within every individual department Group B has a higher rate. This happens because Group A applied mostly to easy-to-enter departments and Group B applied mostly to competitive ones. The aggregate hides the composition. Always ask: could a subgroup structure be reversing this trend?