---
id: "concept-lasso-regression-workforce"
type: "concept"
source_timestamps: ["\\\"§ Rely on Data", "Not Intuition\\\"", "§ A Playbook to Customize Scheduling"]
tags: ["statistical-methods", "data-science", "predictive-modeling"]
related: ["concept-scheduling-quality-dimensions", "action-mine-workforce-data", "prereq-advanced-analytical-capability", "concept-operational-noise"]
definition: "A statistical method used to isolate the most critical variables predicting turnover from hundreds of potential scheduling metrics."
sources: ["tail1"]
sourceVaultSlug: "hbr-seg-tail1"
originDay: 1
articleStem: "hbr-tail-111-service-worker-churn"
sourceUrl: "https://hbr.org/2026/03/the-solution-to-service-worker-churn"
sourceTitle: "The Solution to Service-Worker Churn"
---
# LASSO Regression in Workforce Analytics

**LASSO** (Least Absolute Shrinkage and Selection Operator) is the advanced statistical method the authors used to cut through hundreds of potential scheduling variables — **166 in their study** — and isolate the few that actually predict employee turnover.

LASSO works as a **"truth detector"**: it penalizes less-important variables, effectively shrinking their coefficients toward zero and leaving only the minimal set of true predictors. This is precisely what lets analysts separate structural scheduling problems from [[concept-operational-noise|operational noise]].

Crucially, the authors ran LASSO **separately for each company, each state, and each worker group** (part-time, full-time, tenured, new) — proving that the drivers of retention are deeply dependent on local context rather than universal rules. The output of this modeling is the [[concept-scheduling-quality-dimensions|five dimensions of scheduling quality]], and it is the analytical engine behind step one of the [[framework-customized-scheduling-playbook|playbook]] and the recommendation to [[action-mine-workforce-data|mine existing workforce data]]. Running it requires the [[prereq-advanced-analytical-capability|advanced analytical capability]] noted in the prerequisites.

**Enrichment note:** The published article confirms this description almost verbatim, quoting the authors that they used LASSO regression "designed to cut through hundreds of potential variables and isolate the few that matter most." Data-science critics do caution that alternative models (random forests, gradient boosting) may capture nonlinearities LASSO misses, and that variable selection is not causal proof — experimentally validated interventions (see [[action-ab-test-schedules]]) remain necessary.

> **Definition:** A statistical method used to isolate the most critical variables predicting turnover from hundreds of potential scheduling metrics.