we-solve (beta)

Econ-533 · economics · Economics

Long-run effects of remote and hybrid work

LLM status Open unverified

Statement

Original atlas ID: L7 (entry 27). Classification: Editorial long-horizon policy, interference, and equilibrium problem.

Original atlas question: How does remote work affect productivity, promotion, geography, sorting, innovation and family formation?

Definitions and assumptions

Fix a baseline worker cohort, teams, a horizon $H$, and arrangements $r\in\{o,h,f\}$ denoting fully onsite, specified hybrid, and fully remote work. A treatment protocol includes hours, equipment, management, eligibility, and office access. Let $\mathbf r$ be the vector of team assignments and $e_j(\mathbf r)$ an announced exposure mapping for team $j$. Potential outcomes $Q_{jt}(\mathbf r)$ measure team output on a prespecified quality-adjusted scale; $Y_{it}(\mathbf r)$, $P_{it}(\mathbf r)$, and $F_{it}(\mathbf r)$ measure earnings, promotion, and family outcomes for baseline workers, including leavers. Do not assume no interference unless justified; communication and mentoring naturally connect workers.

Formal research question

For two specified allocation policies $\pi_1,\pi_0$ over assignment vectors, identify \[ \tau_H^V=\mathbb E_{\mathbf r\sim\pi_1} \left[\sum_{t=0}^{H}\beta^t V_t(\mathbf r)\right] -\mathbb E_{\mathbf r\sim\pi_0} \left[\sum_{t=0}^{H}\beta^t V_t(\mathbf r)\right], \] where $\beta\in(0,1]$ is fixed and $V$ is separately chosen as output, worker earnings, promotion probability, innovation, or another defined outcome. Characterize heterogeneity by pretreatment type and spillover exposure. For economy-wide adoption, construct a separate model $M$ of firm entry, worker sorting, residence, commuting, and office/housing prices, and bound its equilibrium outcome difference under the corresponding policies.

Interpretation and scope

The finite-cohort experimental estimand and the economy-wide equilibrium effect are different objects. Measuring workers only while they remain in their original firm excludes a potentially central retention and sorting pathway. Innovation proxies require a definition and validation; counts of messages or patents are not automatically productivity. Randomizing teams rather than individuals can identify some spillovers, but only exposure combinations with design support are identified. Families and residential moves may respond over longer horizons than available records, so their effects require explicitly stated follow-up and missing-data assumptions. A statistically insignificant productivity contrast does not demonstrate equality: prespecified equivalence margins and uncertainty intervals are needed for that conclusion.

Source audit and status

[S1] is a firm-specific experiment; [S2] and [S3] document broader work-from-home patterns. Added [S4] provides a 2024 randomized hybrid-work trial, so short-run performance and retention effects are already studied. The editorial frontier is transport, organizational spillovers, and long-horizon equilibrium outcomes. The atlas's tractability label is not a mathematical claim, and no universal open status is established.

References

    },leftmargin=*]
  1. Nicholas Bloom, James Liang, John Roberts, and Zhichun Jenny Ying (2015). Does Working from Home Work? Evidence from a Chinese Experiment. The Quarterly Journal of Economics 130(1), 165--218. https://academic.oup.com/qje/article-abstract/130/1/165/2337855. Locator: Primary publisher, working-paper, or author record: bibliographic information and abstract; full-text theorem verification is not claimed. Role: Randomized work-from-home evidence in a particular firm and occupation.
  2. Alexander Bick, Adam Blandin, and Karel Mertens (2023). Work from Home before and after the COVID-19 Outbreak. American Economic Journal: Macroeconomics 15(4). https://doi.org/10.1257/mac.20210061. Locator: Primary publisher, working-paper, or author record: bibliographic information and abstract; full-text theorem verification is not claimed. Role: Measurement of work-from-home patterns; descriptive evidence.
  3. José María Barrero, Nicholas Bloom, and Steven J. Davis (2023). The Evolution of Work from Home. Journal of Economic Perspectives 37(4), 23--49. https://doi.org/10.1257/jep.37.4.23. Locator: Primary publisher, working-paper, or author record: bibliographic information and abstract; full-text theorem verification is not claimed. Role: Review and measurement of remote-work arrangements.
  4. Nicholas Bloom, Ruobing Han, and James Liang (2024). Hybrid Working from Home Improves Retention without Damaging Performance. Nature 630, 920--925. https://doi.org/10.1038/s41586-024-07500-2. Locator: Primary publisher, working-paper, or author record: bibliographic information and abstract; full-text theorem verification is not claimed. Role: More recent randomized hybrid-work evidence; limits a blanket claim that hybrid performance is unstudied.

Common conventions: Source mathematical conventions

Unless an entry states otherwise, agent and firm sets are finite. State and action spaces are measurable, strategies and outcome maps are measurable, probability laws are specified, and expectations exist for the targets under discussion. Infinite-horizon discount factors lie in $(0,1)$ and discounted objectives are finite. Norms, tolerances and complexity input sizes must be specified for computational comparisons. Constants or primitives that appear locally are inputs, not empirical facts asserted by this catalogue.

Identification. A statistical model $m\in\mathcal M$ implies an observable law $P_m$ and target $\theta(m)$. At law $P$, its identified set is \[ \Theta(P;\mathcal M)=\{\theta(m):m\in\mathcal M,\ P_m=P\}. \] Point identification holds when this set is a singleton, or equivalently when $P_{m_1}=P_{m_2}$ implies $\theta(m_1)=\theta(m_2)$ for all compatible models. Sharp bounds mean bounds attained, or approached when the boundary is not attained, by compatible models. Writing this set is a definition, not a solution: the substantive task is to establish informative restrictions and derive or estimate the set. An unrestricted-confounding model may give only trivial bounds.

Inference. Identification, estimability and valid uncertainty quantification are separate questions. Fix $\alpha\in(0,1)$. A confidence procedure $C_n$ over a declared law class $\mathcal P$ has asymptotic uniform coverage at level $1-\alpha$ when \[ \liminf_{n\to\infty}\inf_{P\in\mathcal P}\Pr_P\{\theta(P)\in C_n\}\geq1-\alpha. \] For set-identified targets the coverage event must be defined separately, for example coverage of the full identified set. If an entry uses identification-indexed models instead of uniquely indexed laws, the same coverage convention is applied over those models. Pointwise limits do not establish uniform validity, and asymptotic validity does not guarantee finite-sample precision. Sample sizes, dependence structures and weak-identification sequences are part of each application.

Counterfactuals. $Y_i(a)$ is a potential outcome under a specified intervention, including its timing, implementation and versions. It is not $Y_i$ conditional on voluntarily choosing $a$. A full intervention vector permits interference; shorthand scalar treatment notation does not silently impose no interference. Assignment, selection, attrition, anticipation and measurement must be specified. A claim about an instrument's local effect is not automatically a population policy effect.

Economic equilibrium. A counterfactual model includes best responses, constraints, feasibility, market clearing, state transitions and expectations consistent with the stated information. When several equilibria exist, either a declared selector is an input or the target is set-valued. Comparisons across policy regimes must use the stated selection convention. Reduced-form relationships are not presumed invariant after an equilibrium-changing intervention.

Optimization and welfare. Optimization is over the stated feasible set. If attainment is not established, $\sup$ or $\inf$ and $\varepsilon$-optimal choices replace an asserted maximizer or minimizer. For example, with value $V=\sup_{a\in\mathcal A}W(a)<\infty$, an $\varepsilon$-optimal set is $\{a\in\mathcal A:W(a)\geq V-\varepsilon\}$ for $\varepsilon>0$. Benefits, costs, risk and health or environmental outcomes can be aggregated only using declared units and conversion weights. Interpersonal utility weights, the treatment of fiscal transfers and foreign profits, discounting, and the behavioral welfare criterion are explicit inputs. A comparative-static derivative additionally requires existence and appropriate differentiability of the selected counterfactual.

Sources. Local labels [S1], [S2], and so forth restart in every question. Locators identify the relevant abstract, model, section or result at the level actually checked; a bibliographic or abstract-level verification is not represented as inspection of an entire proof. Working papers are distinguished from later publications and changing versions. Source summaries are paraphrased. All URL checks and editorial formulations refer to the audit date stated above.

Statement licence: CC-BY-4.0 · Source file

Contributions 1

Econ-533-A1 · Exploration · LLM-generated · gpt 6 astra ultra ·

GPT 6 Astra Ultra attempt on Econ-533

Remote-work output, mentoring dynamics, and unsupported scale-up

Problem ID: M54-529. Status: Unresolved. The following results concern declared restricted models.

A short-run contrast does not determine the long-run sign

The catalogue distinguishes finite-cohort effects under specified assignment policies from economy-wide adjustments in sorting, residence, entry and prices. This attempt constructs a dynamic mentoring channel, derives its discounted output effect, and identifies which network coefficients a limited experiment can recover. It makes no universal claim that onsite, hybrid or remote work is superior.

Start with one fixed team. Let $r\in\{0,1\}$ denote a sustained onsite or remote protocol, including equipment and management. Its human-capital stock evolves as \[ k_{t+1}=\rho k_t+\eta(1-r),\quad 0\le\rho<1,\quad\eta\ge0, \qquad Q_t=k_t+ar, \] with common initial stock $k_0$ and $a\ge0$. The term $a$ represents a contemporaneous remote-work output benefit, while $\eta$ represents learning obtained under the stipulated onsite protocol. This is a restricted technology, not an assumption that all mentoring requires presence. Output is measured on the same quality-adjusted scale in both regimes.

Proposition (Exact horizon response): For sustained remote versus sustained onsite work, \[ \Delta k_t=-\eta\frac{1-\rho^t}{1-\rho},\qquad \Delta Q_t=a-\eta\frac{1-\rho^t}{1-\rho}. \tag{1} \] For a discount $\beta\in(0,1]$ and finite $H$, defining $S_H(x)=\sum_{t=0}^Hx^t$, \[ \Delta V_H=aS_H(\beta) -\frac{\eta}{1-\rho}\{S_H(\beta)-S_H(\beta\rho)\}. \tag{2} \]
Proof: Subtract the two stock recursions to obtain $\Delta k_{t+1}=\rho\Delta k_t-\eta$ with $\Delta k_0=0$. Iterating yields the geometric sum in (1). Add the contemporaneous output term and sum with weights $\beta^t$ to prove (2). Writing finite sums avoids division by zero when $\beta=1$. ∎

At date zero output rises by $a$, but the limiting annual effect is $a-\eta/(1-\rho)$. For $a=1,\eta=0.3,\rho=0.8$, the initial gain is one and the limiting change is $-0.5$. At $t=5$, the change is $1-1.5(1-0.8^5)\approx-0.00848$. These are constructed parameters, not measured remote-work effects. A trial ending before the stock adjusts cannot identify (2) without a validated transition model or further follow-up.

A hybrid protocol can be represented by a fraction $r\in[0,1]$ only if the learning and contemporaneous effects are linear in that fraction. Coordinated office days may create nonlinear complementarities, so treating three remote days as exactly three fifths of a fully remote intervention adds a response restriction. The protocol must specify whether coworkers are present on the same days.

Direct effects and policy effects under coworker exposure

For a separate static benchmark let workers form a known network, with row-normalized weights $g_{ij}\ge0$, $g_{ii}=0$, $\sum_jg_{ij}=1$. Suppose \[ Q_i(\mathbf r)=\alpha_i+a r_i+b\sum_jg_{ij}r_j. \tag{3} \] Under independent Bernoulli assignment with common probability $p$, the conditional contrast between own remote and onsite assignment is exactly $a$: coworkers' expected exposure equals $p$ in both groups. Yet changing everyone from onsite to remote changes every worker's outcome by $a+b$.

This follows directly by taking conditional expectations in (3) and then evaluating its endpoints. Thus a perfectly identified direct effect can fail to identify a full-policy effect. Different values of $b$ can be hidden if a study only reports the own-assignment contrast. With full outcome and network data, independent assignment can identify $b$ from realized exposure variation under the linear model, but its precision depends on network structure. For a worker with $n-1$ equally weighted peers, exposure variance is $p(1-p)/(n-1)$, shrinking as the team grows.

A two-stage saturation experiment makes the distinction explicit. Assign independent teams to saturation probabilities $p_L$ and $p_H$, and independently assign members conditional on saturation. The difference in average team outcomes under (3) is $(a+b)(p_H-p_L)$. Combining it with an own-assignment contrast identifies $b$ when $p_H\ne p_L$, under the fixed coefficient and no-between-team-interference assumptions. It does not identify nonlinear responses at untested saturations without additional support or smoothness.

A policy-weighting identity and its limits

Let a finite assignment vector $R$ be generated under known design law $g$, and let the target policy distribution be $\pi$. For a bounded fixed-cohort outcome $V(R)$, if $\pi(r)>0$ implies $g(r)>0$, then \[ {\mathbb{E}}_g\left[\frac{\pi(R)}{g(R)}V(R)\right]={\mathbb{E}}_\pi V(R). \tag{4} \] This is verified by summing over every assignment vector. It permits unrestricted within-team interference because the full vector indexes potential outcomes. The weighting variance may nevertheless be enormous. If $g$ independently assigns $n$ workers remote with probability one half and $\pi$ assigns everyone remote, the required endpoint probability is $2^{-n}$; one rare observation receives weight $2^n$.

If the endpoint is never assigned, (4) is unavailable. Outcomes at that unsupported vector can be changed while leaving every observed outcome fixed, providing a direct nonidentification proof. Team randomization to the desired complete policies avoids that particular support problem, at the cost of requiring independent team replications and credible separation between teams.

Retention, equivalence and economic scale-up

All earnings and promotion outcomes should be defined for the original eligible workforce, including people who leave. A rise in promotion among retained workers can result from selective departures even if no individual's promotion opportunity improves. For example, with one high-promotion and one low-promotion worker, losing the latter raises the retained mean mechanically. Following destination jobs and nonemployment preserves the cohort estimand; redefining the population as current employees does not.

A confidence interval containing zero is not an equivalence result. If the scientific margin is $\epsilon>0$, a confidence interval wholly inside $[-\epsilon,\epsilon]$ supports that prespecified equivalence claim at its coverage level; an interval such as $[-2\epsilon,2\epsilon]$ does not. For a cumulative target, discounted tail or missing-follow-up assumptions are also required. Productivity, retention and worker amenities use distinct units and should not be silently added.

Finally, economy-wide adoption changes the feasible outside options, housing and office prices, worker sorting, and firm entry. Even an experiment identifying (4) within current teams holds many of those equilibria fixed. A model connecting the experiment to a broad adoption policy must state how those prices and populations move. The dynamic formulas and exposure identities here identify specific missing assumptions; they do not establish the sign of that general-equilibrium effect.

Completion Estimate

45%

This is a post hoc, heuristic estimate for the full catalogue target. The attempt derives dynamic mentoring-output reversals, direct versus coworker-exposure policy effects, and full-assignment weighting and support limits. Full remote and hybrid outcomes still need long-term baseline-worker follow-up, validated nonlinear team protocols, family and innovation measures, and economy-wide sorting, entry, and housing responses.

References

  • [hybrid] N. Bloom, R. Han and J. Liang (2024), Hybrid working from home improves retention without damaging performance, Nature 630, 920--925. https://www.nature.com/articles/s41586-024-07500-2. The checked primary article supplies a specific hybrid-work experiment. Its setting-specific findings are not used as numerical primitives or universal long-run effects in the models above.

Go to contribution →