Source: https://toddschiller.com/artifacts/belief-phylogenetics/index.html

[← toddschiller.com](https://toddschiller.com/)

# Belief phylogenetics

A toy clustering of public-facing thinkers by the "genome" of beliefs they publicly
profess. Each person is a presence/absence vector over a fixed trait set; the tree is
[UPGMA](https://en.wikipedia.org/wiki/UPGMA)
(Unweighted Pair Group Method with Arithmetic mean) over
[Hamming distance](https://en.wikipedia.org/wiki/Hamming_distance). Switch the
dataset below to swap communities. Categories drive the cell colors; pattern overlays flag
hitchhiking traits and crossover individuals.

Dataset:

Wellness & podcast circuit

AI thought leaders

Climate & energy commentators

## What you're looking at

**The data model.**
Each thinker is a presence/absence vector over a fixed trait set. Traits are grouped
into categories (shown above the heatmap) that color the cells. Strength of belief,
recency, and context are all flattened to a single bit.

**The distance metric.**
Pairwise distances are
[Hamming distances](https://en.wikipedia.org/wiki/Hamming_distance): count
the traits where two people differ, weight each trait equally. No normalization or IDF
reweighting. The PCA panel below is a separate view, not folded into the tree's
distance.

**The clustering.**
The
[dendrogram](https://en.wikipedia.org/wiki/Phylogenetic_tree)
is built by
[UPGMA](https://en.wikipedia.org/wiki/UPGMA), an agglomerative
average-linkage algorithm. It runs client-side on page load over the inline data;
editing the inline arrays in the page source would recompute the tree.

**Pattern overlays.**
"Hitchhiking" is a loose nod to
[genetic hitchhiking](https://en.wikipedia.org/wiki/Genetic_hitchhiking);
strictly, the pattern is
[linkage](https://en.wikipedia.org/wiki/Linkage_disequilibrium): traits
that co-occur often enough that one predicts the others. Whether selection is doing
the dragging here is left to the reader. "Crossover" cribs from
[chromosomal crossover](https://en.wikipedia.org/wiki/Chromosomal_crossover): individuals who recombine traits across otherwise linked clusters. Each bundle
below is anchored on a single individual in this slice; the overlays are layered on
top of the tree, not computed from the data.

## Tree and trait genome

Hover or focus a row · click a cluster pill to highlight

Hover a cell to see what trait it represents.

Highlight cluster:
Clear highlight

Analysis ▾

## Principal components

Same data, projected onto the top two axes of variation

[PCA](https://en.wikipedia.org/wiki/Principal_component_analysis)
collapses the trait matrix to the directions of largest variance. PC1 is the strongest
axis of belief divergence; PC2 is the next-strongest orthogonal direction. Hitchhiking
clusters appear as labeled centroids over the people who carry at least half their
traits. Solid amber dots are full carriers (every trait in the bundle); hollow amber
dots are partial carriers (at least half, but not all). Clusters with only one carrier
skip the centroid label, since it would sit on the anchor's dot. Click a cluster pill
above to emphasize its members; click a centroid to do the same.

## Pattern overlays

### Hitchhiking clusters

Trait bundles that travel together when they appear at all. Click a bundle to
highlight its traits in the heatmap above.

### Crossover individuals

Unexpected recombinations: people who pair traits from clusters that otherwise
rarely mix.

Reading the tree