Misleading Graphical Displays of Data

The selection of class width in a histogram can significantly alter the perceived shape of a data distribution.
In the case of President Obama’s $2009$ approval ratings (referenced in Exercise $19$ of Homework $3.4$ ), a smaller class width of $2$ percentage points reveals a bimodal distribution, indicating high ratings in the first half of the year and low ratings in the second half.
A larger class width of $5$ percentage points is misleading because it suggests a unimodal distribution, obscuring the actual drop in approval ratings.

Starting the vertical axis at $0$ (e.g., $0\%$ or $0$ thousand) emphasizes the absolute magnitude of values and de-emphasizes differences or increases.
Starting the vertical axis at a nonzero value (e.g., $35\%$ or $38$ thousand) emphasizes the relative differences or changes between data points.
Entities often choose an axis start to support a specific narrative: Northwestern University might use a zero start to de-emphasize tuition increases, whereas Facebook might use a nonzero start to highlight its lead over Snapchat.
If a company like Snap wanted to show their percentage is close to Facebook’s, they would display a graph starting at $0\%$ to make the bar heights appear similar.

Choosing a nonzero starting point for the vertical axis often allows for smaller scale increments.
This increased granularity improves the ability to estimate specific values accurately, such as determining that approximately $38\%$ of Instagram users visited the platform multiple times daily in $2018$ .