Fitted lines are useful as first order approximations, but it’s often useful to look a little deeper. In the process of exploring a paper’s time trends, I found an unexpected data processing issue.

Trying to adapt the warming stripes graphic with latitude bands, with a detour to import NetCDF data on an equally-spaced global grid.

Like many programmers, I’ve had fun writing code to explore optimal guessing strategies for the online game Wordle. How many guesses are needed for any starting word?

How reproducible are studies with shared data? I collect one more data point in a quest for the answer, and make a side exploration into a linear regression conundrum.

Five alternatives to a broken axis scale for data visualization’s wide scale problem, and a broken line chart tip.

Any investigation into small R² values leads to a different finding about extrapolation.

A study of gender balance in Star Trek dialog leads to an adventure in data cleaning and a smoothing revelation.

Where a more realistic (skeuomorphic) data encoding is less accurate to read.

A study of a map of wine consumption reveals many underlying choices, both cosmetic and substantive.

When a linear fit doesn’t look right, should we use it anyway? Let’s not — here are some alternatives.