HogQL: Segment Analysis
This query segments experiment results by tenant size, revealing Simpson's Paradox effects masked in aggregate data.
-- Segment activation rate by tenant size
SELECT
properties.$feature/onboarding_test as variant,
person.properties.tenant_size as segment,
count(distinct person_id) as users,
count(distinct CASE WHEN event = 'activation'
THEN person_id END)
* 100.0 / count(distinct person_id)
as activation_rate
FROM events
WHERE event IN ('$pageview', 'activation')
GROUP BY variant, segment
| variant | segment | users | activation_rate |
|---|
control | SMB | 4,280 | 12.4% |
control | Enterprise | 1,840 | 18.7% |
variant-b | SMB | 4,310 | 14.1% |
variant-b | Enterprise | 1,790 | 17.9% |
Insight: Variant-B lifts SMB activation by 1.7 percentage points but slightly lowers Enterprise activation. The aggregate view would mask this divergence. Segment-level analysis reveals where the variant actually wins.
productquant.dev