Hypothesis Formation Assistant

Hypothesis Formation: Hybrid Work and the Perceived-vs-Objective Productivity Gap at Meridian

1. Framing

Phenomenon. In the August 2025 survey, employees in hybrid divisions (Support and Engineering, 3 remote days per week) rated their productivity change over the prior six months at a mean of 5.8/7. Employees in fully in-office divisions (Sales and Finance) rated theirs at 4.6/7. Objective throughput for Feb–Jul 2025 is reported as nearly identical after seasonal adjustment. The core puzzle is therefore a divergence between perceived and measured productivity, not simply a productivity effect.

Research gap. Published post-2020 work shows more consistent gains in self-reported outcomes than in objective output. There is little evidence on why these two measures diverge, and none at Meridian. The useful question is whether hybrid work shifts how people appraise their work (through autonomy or lower demands) without shifting output, or whether the gap comes from division differences and measurement design.

Assumptions to make explicit.

1. Comparable objective metrics. Tickets closed (Support/Engineering) and CRM activities (Sales/Finance) are assumed to measure "productivity" on a common scale. They do not. If you z-score within role and roles are nested in divisions, each group's mean is roughly 0 by construction. In that case the "nearly identical" objective result is partly an artifact of standardization, not a finding. Only change from baseline is comparable across divisions.
2. Change item vs. level metric. The survey item asks about change ("Compared to six months ago…"), while the objective comparison is currently a level (Feb–Jul throughput). The perceived gap must be compared to objective change (2025 H1 vs. 2024 H2), not to objective level. Otherwise the two measures answer different questions.
3. Policy equals behavior. Assignment to "hybrid" is assumed to equal actual remote days. Because Tue/Wed/Thu attendance was optional, actual remote frequency likely varies from about 3 to 5 days. Badge or VPN data should be used to measure it.

Critical design constraint

Treatment is perfectly confounded with division. Every Support/Engineering employee is hybrid and every Sales/Finance employee is in-office. Three consequences follow:

• The between-group comparison cannot separate the effect of hybrid work from division culture, job design, management, or events specific to the division during the window.
• Your proposed control "compare within Support/Engineering vs. within Sales/Finance separately" gives no treatment contrast, because there is no policy variation inside either group.
• Direct manager is nested within division and is therefore collinear with treatment. It can be used as a random effect or clustering level, but not as a control that removes confounding.

Credible leverage must come from (a) pre/post change (difference-in-differences), and (b) within-hybrid variation in actual remote days (dose-response).

2. Variables

Recommendation on the "composite productivity index": do not merge self-report and objective output into one index. The phenomenon is their divergence, and averaging them hides it. Model DV-a and DV-b separately and treat DV-c as its own outcome.

3. Hypotheses

H1 — Perception–output divergence (primary)

• H₀: After adjusting for baseline, tenure, and team size, the gap between perceived and objective productivity change (DV-c) does not differ between hybrid and in-office employees.
• H₁: Hybrid employees show a larger positive gap (perceived change exceeds objective change) than in-office employees.
• Directional. Both the observed data and SDT's claim that autonomy raises perceived competence independently of output predict the sign in advance. A one-tailed framing is justified only if you pre-register it before recomputing the corrected DV-b.
• Caveat: Because of the division confound, support for H1 shows association, not a causal effect of hybrid work.

H2 — Dose-response within hybrid divisions (causal leverage)

• H₀: Among Support/Engineering employees, actual remote days per week are unrelated to perceived productivity change, controlling for baseline output, tenure, role, and commute time.
• H₁: More actual remote days predict higher perceived productivity change. This association is weak or absent for objective productivity change.
• Directional. JD-R and SDT both predict a positive slope. This test holds division constant, so it is the strongest available check on whether remote work itself matters rather than division membership.
• Caveat: Remote frequency is self-selected here. For example, more motivated or less socially connected employees may stay home more. Include baseline output and, if available, prior engagement as covariates.

H3 — Mechanism: autonomy vs. demand reduction

• H₀: Perceived autonomy does not mediate the relationship between remote frequency and perceived productivity change, and commute time does not moderate it.
• H₁ (SDT pathway): Perceived autonomy partially mediates the positive relationship between remote days and perceived productivity change, but does not mediate any effect on objective change.
• H₁ (JD-R pathway, competing): The remote-days effect on perceived productivity is stronger for employees with longer commutes (positive interaction) and is associated with fewer reported interruptions.
• Directional for each pathway. The two pathways are not mutually exclusive. Report the relative size of each rather than declaring one the winner.
• Caveat: Mediator and outcome come from the same survey at the same time point. This gives a weak basis for causal mediation claims (common-method variance, reverse causation: people who feel productive may also report feeling autonomous). Label the result "consistent with mediation," not "mediation demonstrated."

Supplementary — objective output equivalence

• H₀ (equivalence framing): The difference in objective productivity change between groups lies outside ±δ, where δ is a smallest effect of interest (e.g., 0.2 SD) that you set in advance with leadership.
• H₁: The difference lies within ±δ, meaning practically no difference.
• Non-directional (two one-sided tests). Prior evidence on objective output is mixed, and the operational claim is "nearly identical." A failed standard t-test cannot establish "no difference," but an equivalence test can.

4. Stress-test

None of the hypotheses fails outright. H1 fails as a causal claim under the current design, and H3's causal interpretation is weak.

5. Recommended next steps

1. Fix the objective comparison first. Recompute DV-b as within-person change from 2024 H2 to 2025 H1. Then check whether "nearly identical" still holds. If it does not, the framing of the puzzle changes.
2. Difference-in-differences on objective output. Model (Hybrid × Post) on weekly output, with person fixed effects and week fixed effects. Before interpreting the result, plot and test pre-trends across 2024 H2.
   • Supports H₁ of equivalence: the DiD estimate and its 90% CI fall inside ±δ.
   • Refutes: the CI excludes the equivalence bounds.
3. Look for a perceived-productivity baseline. Search for any 2024 engagement item on productivity or effectiveness. If one exists, run a DiD on self-report too. That would greatly strengthen H1.
4. Test H2 within hybrid divisions. Run a multilevel regression of DV-a (and separately DV-b) on actual remote days, with manager random intercepts.
   • Supports H₁: a positive, CI-excluding-zero slope for DV-a together with a null slope for DV-b.
5. Test H3 with bootstrapped indirect effects (remote days → autonomy → DV-a) and a remote days × commute interaction. Pre-register both.
6. Pre-register δ, the direction of each hypothesis, and the covariate sets before analysis. This matters because the survey data have already been seen.

Primary rival explanation to rule out: division-level differences, not hybrid work. Sales and Finance may have faced events unrelated to remote policy during Feb–Jul 2025, such as quota changes, fiscal close, or client churn, which would depress their felt productivity. Their jobs may also be inherently less autonomous. Ways to probe this:

• Check whether Sales/Finance ratings on survey items unrelated to remote work (e.g., pay satisfaction, role clarity) are also depressed. A general division-level depression points to confounding rather than a hybrid effect.
• Run the H2 dose-response test. A within-hybrid slope cannot be explained by division membership.

Falsifiability & Confounds

• What would falsify the "hybrid raises perceived productivity" account: no dose-response slope within Support/Engineering, and the between-group gap shrinks to ≈0 once baseline and division-general dissatisfaction are adjusted for.
• Confounds the design cannot remove: division, job type, and manager are fully collinear with policy, and assignment was chosen by leadership based on operational needs. Matching on tenure or team size across divisions does not address this.
• Measurement confounds:
  • Self-report vs. system logs have different error structures.
  • The survey item is a change item, while the objective metric is currently a level.
  • Ticket counts and CRM counts measure different constructs.
• Response biases:
  • Strategic responding: hybrid employees have an incentive to report success so the pilot continues. Consider an anonymous item asking about preference to keep hybrid, and test whether it predicts DV-a beyond the other variables.
  • Nonresponse: 27% did not respond. Compare responders and non-responders on baseline output.
• Also check:
  • Regression to the mean, if divisions entered the pilot with unusually low or high 2024 H2 performance.
  • Novelty effects: the survey came about seven months into a new perk, and self-report gains may fade. A follow-up survey in early 2026 would test durability.
• Honest ceiling: with this design, the strongest defensible conclusion is "evidence consistent with (or inconsistent with) a hybrid effect on perceived productivity, operating plausibly through autonomy and/or reduced demands." A clean causal estimate would need staggered or randomized rollout within a division in a future phase. If leadership plans to expand the pilot, that is the design to recommend.