Skip to main content
HR & People

Why Comparing This Year's Engagement Score to Last Year's Isn't as Simple as It Looks

An organization's workforce composition changes every year through hiring and attrition, which means a year-over-year engagement score comparison is often comparing two meaningfully different populations, not tracking the same group over time.

Key Takeaways
  • An organization's workforce composition changes every year through hiring and voluntary attrition, meaning a year-over-year engagement score comparison often reflects two meaningfully different underlying populations, not the same group tracked consistently over time
  • A rising aggregate engagement score can result from genuine improvement among continuously employed staff, from disengaged employees leaving and being replaced by more satisfied new hires, or from some combination of both, and the aggregate number alone can't distinguish between these explanations
  • This is directly related to the survivorship bias problem discussed elsewhere in attrition modeling, applied here specifically to the interpretation of aggregate engagement trend data over time
  • Cohort-based analysis, tracking specifically continuously employed staff separately from newly hired staff, is what actually distinguishes genuine engagement improvement from population turnover effects

An organization's aggregate employee engagement score rises from one year to the next, presented as evidence that recent culture or management initiatives have genuinely improved engagement — a conclusion the aggregate number alone can't actually support, since the underlying population contributing to each year's score has changed through ordinary hiring and voluntary attrition, meaning the comparison may be tracking two meaningfully different groups of people rather than genuine improvement within a consistent group over time.

Why workforce composition changes complicate a simple year-over-year comparison

Employees who were disengaged and dissatisfied in the earlier survey period are statistically more likely to have voluntarily left the organization by the time the later survey is conducted, and new hires joining during the intervening period bring their own separate, independent engagement levels, typically starting from a more positive baseline during an initial period before any accumulated dissatisfaction has had time to develop — meaning the later survey's respondent population can differ substantially from the earlier one, entirely independent of whether engagement genuinely improved among people who remained employed throughout the entire period.

Why this is directly related to the survivorship bias problem discussed in attrition modeling

The same underlying logic behind survivorship bias in attrition analysis applies directly here — the population contributing to an engagement score at any given point already reflects who chose to stay through that point, systematically excluding people who left specifically because of low engagement, meaning aggregate engagement trend data is subject to exactly the same kind of population-composition distortion that makes naive attrition analysis unreliable without a proper comparison group.

Why the aggregate number alone can't distinguish between genuine improvement and population turnover

A rising aggregate engagement score is consistent with several genuinely different underlying explanations — real improvement in engagement among continuously employed staff, replacement of disengaged departed employees with more naturally satisfied new hires, or some combination of both — and the single aggregate number, examined in isolation, provides no way to distinguish which of these explanations, or what mix of them, actually produced the observed change.

What actually distinguishes genuine improvement from population turnover effects

Cohort-based analysis, specifically tracking the engagement scores of employees who were continuously employed across both survey periods separately from newly hired employees who joined during the intervening period, directly reveals whether engagement genuinely improved among the consistent, continuously employed group — a genuine improvement specifically within this consistently tracked cohort is considerably stronger evidence of a real, causal improvement than an aggregate score that could equally reflect nothing more than ordinary workforce turnover.

Why this distinction matters directly for evaluating whether an engagement initiative actually worked

An organization crediting a specific engagement initiative for a rising aggregate score, without checking whether the improvement holds specifically within the continuously employed cohort, risks crediting an initiative for an improvement that actually reflects nothing more than ordinary employee turnover replacing less satisfied departed staff with more naturally satisfied new hires, entirely independent of whatever the initiative itself actually accomplished.

What this means for interpreting year-over-year engagement survey trends

  • Analyze engagement trends specifically within a continuously employed cohort, not just as an aggregate score across the full, changing workforce population
  • Be specifically skeptical of a rising aggregate engagement score as sole evidence that a specific initiative meaningfully improved engagement
  • Track new hire engagement separately from continuously employed staff engagement, since these reflect genuinely different underlying populations
  • Recognize this as directly related to the same survivorship bias concern relevant to attrition modeling generally

An aggregate engagement score's year-over-year comparison is only as meaningful as the population it's actually comparing — and given how much a workforce's composition changes through ordinary hiring and attrition, cohort-based analysis, not the aggregate trend line alone, is what actually reveals whether genuine improvement occurred.

engagement survey year over year comparisonworkforce composition changecohort analysis HR metricspeople analyticsattrition survivorship in engagement data