Methodology
How we measure
what we publish.
This page documents the operational controls behind every Love Pulse Labs publication. For the broader picture of what we study and why, see Our Research.
Methodology White Paper, v1
The formal Love Pulse Labs methodology white paper covers our measurement model, construct validity work, signal-classification taxonomy, K-anonymity rationale, and consent architecture. It will be released through the Open Science Framework once peer review of the draft is complete. Until then, the operational controls below are in force and reflect the production system.
Operational controls in force today
The protections below are enforced in code, not in policy. They run on every export, every aggregation, every publication. They are independent of any legal structure and apply to all participants.
K-anonymity floor of 50
No aggregate publication or licensed dataset is derived from fewer than 50 couples. The threshold is enforced in code, verified in tests, and stamped on every published report.
Both-partner Tier 2 consent
Relational signals require both partners in a couple to have affirmatively opted into Tier 2 anonymized research at the time of computation. One-partner consent leaks information about the non-consenting partner. We do not accept that trade.
Five-way signal classification
Every research signal is classified as behavioral, linguistic, self-report, relational, or outcome. Classification is enforced in code: any new signal that ships unclassified fails our test suite. This makes the audit trail of what data crosses the research boundary append-only and inspectable.
Append-only audit trail
Every export from the product into the research arm writes an immutable audit row. The row records function, timestamp, cohort size, K-anonymity check result, the user and couple identifiers that contributed, and any refusal reasons. The table is never updated, only appended.
Revocation cascade
When a participant revokes Tier 2 consent, our cascade finds every export row referencing them and flags it for review. Aggregate cells whose underlying cohort drops below 50 couples after exclusion are zeroed in subsequent reports. Withdrawal is immediate and complete.
No re-identification, no resale
Aggregate datasets licensed to qualified third parties are governed by separate agreements that prohibit re-identification attempts and downstream resale of underlying records. We do not sell personal information.
How we publish
The principles that govern what gets published and how, from hypothesis to retraction.
Pre-registration of hypotheses
Where possible, we register hypotheses before analysis. This protects against the silent garden of forking paths that makes findings unreliable.
Confidence intervals, not point estimates
Every reported effect comes with its uncertainty. A finding without confidence intervals is a finding designed to be misread.
Limitations published with findings
Every report disclosing what we found also discloses what we cannot conclude from this dataset. Sample bias, observational design constraints, attrition, all of it.
Retraction is a separate, visible act
Reports cannot be silently revised. If a finding needs correction or retraction, the action writes a new immutable row pointing back at the original. The record of what we said, when, and why we changed our mind, stays public.
Related surfaces
- Our Research for the higher-level view of our process and areas of study.
- Data Ethics for the three-tier consent architecture and participant rights.
- Lexicon for definitions of every term we use in publications.
- Publications for current and upcoming quarterly reports.
Collaborating on the white paper
We are actively seeking academic collaborators for peer review of the methodology white paper, particularly researchers with expertise in dyadic measurement, consent architecture for behavioral data, and longitudinal relationship research.
Get In Touch