How the data works

Methodology

Statwren data comes from a network of consumer publications covering home design, fashion, travel, food, autos, RVs, personal finance, outdoors, waterfront living, homemaking and science. Readers leave signals in the normal course of reading. This page explains what those signals are, how we turn them into numbers and where the numbers stop being trustworthy.

The instruments

Reactions

A strip of emoji under images (love, fire, inspired, thumbs-down and others). One tap per image per session.

Saves and keeps

A save button on images and articles. A “keep” is a save or a heart: the reader wants to come back to it.

Head-to-head

Two images side by side; the reader picks one. Win rates are reported only above a minimum number of appearances.

Polls

Short in-article questions with fixed options, written to map onto the same attributes we tag in images, so stated preference can be checked against behaviour.

Reasons

After a thumbs-down or a love, readers can tap a reason (“too frumpy”, “looks dated”, “would buy”). These explain the verdict rather than repeat it.

Tagging what people reacted to

Every image in a study is tagged against a fixed schema for its category: material, colour, style, era, light, silhouette, hem and so on. Tags are assigned by a vision model and stored with a confidence score; low-confidence tags are excluded. The schema is fixed before the field window opens, so we cannot invent a category after seeing the result.

From signals to an index

Most reports use an index where 100 is average for the room, garment or article in question. An index of 130 means 30% more of the signal than average; 70 means 30% less. Views are sampled 1-in-10 and scaled, and every index is published with a 95% confidence interval. Values whose interval crosses 100 are reported as “not separable from average”, and values below a minimum sample are not reported at all.

The fair comparison

Some articles draw more engagement than others, whatever is in them. Where that matters, we compare each image only with the other images in the same article. The reports show both the first-pass number and the fair one, so you can see how much the comparison changes.

Who the panel is

The panel is the readership of our publications: mostly adults in the United States, skewing over 40, arriving from search, social platforms and news apps. It is large and it behaves naturally, but it is not a national probability sample. Each report states its own audience and, where we can measure it, the age profile.

Privacy

Signals are anonymous. We do not collect names or contact details with reactions, and nothing in a report or dataset identifies a person. Datasets contain aggregated counts and public image and article URLs only.

Reproducibility

Every report ships with its dataset and a data dictionary. If a number in the report cannot be reproduced from the files, tell us at hello@statwren.com and we will correct it.