The Privilege Embedded in your Unit of Analysis

How much water per bouquet? If we watered them all using the average required per bouquet, we’d over water one and underwater one. What’s the problem: we’re using the denominator of bouquets instead of flowers.

Defining your denominator is as important as defining your numerator.

When we’re working with descriptive data questions — trying to use statistics to get a picture of what’s happening — we’re often trying to describe the ‘typical’ or ‘average’ rate of something. Rates have denominators, or units of analysis, or what I call the “after per”. Low-birthweight babies per capita. Libraries per neighborhood. Meals per week.

We spend so much time in data science talking about and defining the numerator -the first part — of these rates. What do we consider low birthweight? Are stillbirths and miscarriages being included? Is low birthweight based on a benchmark for full-term births? These are common and important conversations happening across all kinds of data projects.

If I just eat crackers accompanied by spoonfuls of jam before flopping down into bed, does that count as a “meal”? According to me, yes. My spouse and I debate this numerator frequently.