peopleanalyst

Field Craft

The ways a dashboard lies

A working set of methodology pieces on how people-analytics data misleads — and how to read it honestly. Each takes one trap that turns a confident chart into a wrong decision, names it, and shows the discipline that defuses it. No products required: this is the craft, free to use.

Read them as a set — they're the recurring ways the same mistake shows up — or grab the one that matches the chart in front of you.

  1. 01

    Methodology · causal inference

    Correlation Isn't a Driver

    The "top drivers of engagement" slide is a correlation wearing a causal word. A driver is a claim about what happens when you change something — and you can't get that from a regression on survey data, however good the chart looks.

    Read →

  2. 02

    Methodology · benchmarking

    The Benchmark Trap

    A raw benchmark compares you to a population that differs on everything, then hands you the gap as if it were about the one thing you care about. The percentile is mostly a measurement of who you stood next to.

    Read →

  3. 03

    Methodology · small-sample inference · draft

    The Law of Small Numbers

    A team of seven swings on a single response, and the dashboard paints the wobble red and hands it to a manager as a verdict. The smallest cells make the most dramatic charts and carry the least information.

    Read →

  4. 04

    Methodology · survivorship & selection

    What the Exit Data Can't See

    Exit surveys are a damage map of the planes that came back. The intelligence you need — the people still at their desks, quietly looking — is exactly what they structurally can't contain.

    Read →

  5. 05

    Methodology · multiple comparisons

    Significance on Demand

    Slice an engagement survey enough ways and something always crosses p < .05. That's not a discovery — it's arithmetic. The threshold was built for one test, not five hundred.

    Read →

  6. 06

    Methodology · base rates & prediction

    The Base Rate Problem

    A hiring test can be 85% accurate and wrong about most of the people it flags. Accuracy is a property of the test; usefulness is a property of the test and how rare the thing it's hunting actually is.

    Read →

These also live in the main magazinefeed — they're about the field, not the firm. For how we think about building products, see the Principles.