Platform Substitution Under New York City’s Food Delivery Pay Rule

Platform Substitution Under New York City’s Food Delivery Pay Rule | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Platform Substitution Under New York City’s Food Delivery Pay Rule Boon Chuan Lim This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9578165/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 4 You are reading this latest preprint version Abstract New York City’s minimum pay rule for app-based restaurant delivery workers, effective December 2023, raised industry-aggregate hourly pay from $5.05 in the last full pre-rule quarter to $21.49 by the end of 2025. Using the Department of Consumer and Worker Protection’s public quarterly tables across sixteen quarters, this paper tests a framework of regulatory substitution across compensation channels. Chow tests, supremum-Wald structural break tests, and segmented-regression interrupted time series with Newey-West standard errors and permutation-based inference all detect highly significant breaks at the rule’s implementation dates. The realised pay series shows a level shift of $11.6 per hour at Q1 2024 (Chow F = 93.8, permutation p = 0.0002). On-call hours show a level shift of − 673 thousand per week at Q2 2024 (Chow F = 16.1, permutation p = 0.0006); the sup-Wald test selects Q1 2024 (F = 33.1), indicating platforms began draining on-call capacity at enforcement onset. Tips per delivery show a level shift of −$1.71 at Q4 2023 (Chow F = 7.1) intensifying to sup-Wald F = 171.9 at Q1 2024 once user-interface redesigns at Uber Eats and DoorDash took full effect. Active workers contracted by 32 percent (Chow F = 73.3, level shift − 6.1 thousand). Each substitution arrived in a sequence keyed to which channel the regulator left open. The paper contributes a regulatory-substitution framework: when a wage rule targets one channel of platform compensation, predictable substitution along the unregulated channels follows in a sequence determined by the regulatory architecture. JEL classification: L51 (Economics of Regulation); J38 (Public Policy: Wages); J42 (Monopsony; Segmented Labor Markets); K23 (Regulated Industries and Administrative Law) regulatory substitution platform work minimum pay rules gig economy structural break tests tip suppression Figures Figure 1 Figure 2 Figure 3 Figure 4 1. Introduction When a regulator imposes a wage floor on one channel of platform compensation but leaves three other channels untouched, what happens? Standard economic intuition predicts substitution: rational platforms will adjust along the margins that remain free. The question is whether the substitution is large enough to matter for the regulation’s intended welfare transfer, and whether it follows a predictable sequence keyed to the regulatory architecture. New York City’s minimum pay rule for app-based restaurant delivery workers, in effect since December 2023 and observable through eight post-rule quarters of public industry-aggregate data, allows both questions to be answered. This paper develops and tests a framework of regulatory substitution across compensation channels. Platform compensation is a vector across multiple channels: hourly pay, tip facilitation, on-call exposure, and platform access. A wage regulation that targets one channel leaves the others as available margins for platform response. The framework predicts that single-channel wage rules will produce systematic, sequenced substitution along the unregulated channels, with the temporal ordering of substitutions determined by which margins the regulator has and has not closed at each stage of implementation. The framework yields four testable predictions in the New York setting: structural breaks in realised pay at the December 2023 enforcement (Q1 2024 in the quarterly data), in tips per delivery at the same enforcement date, in on-call hours at the April 2024 tightening, and in active worker counts following both. Each prediction is tested using Chow ( 1960 ) tests at the predicted break date, supremum-Wald structural break tests (Andrews, 1993 ) with the break date selected endogenously over a candidate window, and a segmented-regression interrupted time series specification (Bernal, Cummins, & Gasparrini, 2017 ) with Newey-West heteroskedasticity- and autocorrelation-consistent standard errors and permutation-based p-values computed over 5,000 randomizations of the intervention indicator. All four breaks are detected at conventional significance levels. Realised pay rises by $ 11.6 per hour at Q1 2024 (Chow F(2,12) = 93.8, p < 0.0001; permutation p = 0.0002). On-call hours fall by 673 thousand per week at Q2 2024 (Chow F(2,12) = 16.1, p = 0.0004; permutation p = 0.0006), with the sup-Wald test indicating that the on-call decline already began at the December 2023 enforcement. Tips per delivery fall by $ 1.71 at Q4 2023 (Chow F(2,12) = 7.1, p = 0.009), with the sup-Wald test indicating the strongest break in Q1 2024 (F = 171.9) once the interface redesigns at Uber Eats and DoorDash had been in effect for a full quarter. Active worker counts fall by 6.1 thousand at Q1 2024 (Chow F(2,12) = 73.3, p < 0.0001; permutation p = 0.0002), continuing to decline through Q4 2025. Each substitution arrives in the predicted temporal sequence. At the same time, the rule’s primary task succeeded. Realised hourly pay rose from $ 5.05 in the last full pre-rule quarter to $ 21.49 by the end of 2025, and now closely tracks the statutory floor. Total weekly industry compensation paid to workers rose from $ 19.5 million to $ 33.3 million over the same period. The rule both worked and was partially circumvented; the substantive contribution of this paper is to formalize a substitution framework that predicts the specific mechanisms of circumvention from the regulatory architecture itself, and to test that framework against the eight post-rule quarters of public data. The institutional setting is the country’s first municipal minimum pay rule for app-based restaurant delivery workers, enforced by New York City’s Department of Consumer and Worker Protection beginning December 4, 2023 at $ 17.96 per hour and subsequently raised to $ 19.56 in April 2024, $ 21.44 in April 2025, and $ 22.13 in April 2026. The rule applies to the sum of trip time and on-call time during a worker’s pay period, covers six platforms collectively responsible for essentially all app-based restaurant delivery in the city, and operates alongside parallel pay regulation for app-based for-hire vehicle drivers (briefly compared in Section 7.1 ). The contribution is twofold. First, the paper proposes the regulatory-substitution framework as a general tool for analyzing wage regulation in multi-margin compensation environments, drawing on the monopsony literature (Manning, 2003 , 2021 ) and the choice-architecture literature (Thaler & Sunstein, 2008 ; Goldin & Reck, 2020 ) to formalize platform incentives across channels. Second, the paper provides formal statistical evidence that the framework’s sequenced predictions hold in the New York case. Each successive rule change tightened one margin and revealed the next one along which platforms could substitute, producing what amounts to a sequenced regulatory experiment. Two limitations bound the analysis. First, the public DCWP tables are aggregated across all six covered platforms with no per-app breakdown, so attribution claims that require platform-specific identification rely on the regulator’s own internal analyses. Second, the worker counts are not adjusted for multi-apping; Section 6.1 presents bounds on the implied unique-individual workforce change. Section 7.1 returns to the comparison with the ride-hail regulatory regime, where trip-level data permit sharper identification. The remainder of the paper is organized as follows. Section 2 describes the institutional setting, the data, and the statistical approach. Section 3 documents that the floor binds and explains the apparent first-quarter 2024 anomaly. Section 4 analyses the on-call collapse. Section 5 analyses the tip decline. Section 6 analyses the workforce contraction and productivity rise, including a sensitivity analysis for the multi-apping confound. Section 7 discusses what these patterns imply for the design of platform-work wage rules. Section 8 concludes. 2. Setting, Data, and Statistical Approach This paper builds on three literatures. The minimum wage literature, beginning with Card and Krueger ( 1994 ) and developed by Dube, Lester, and Reich ( 2010 ) and Cengiz et al. ( 2019 ), provides the framework for studying how a wage floor binds and how affected firms respond. The platform labour literature, including Hall and Krueger ( 2018 ) and Cook et al. ( 2021 ), establishes the structural features of work mediated by app-based platforms. The monopsony literature (Manning, 2003 , 2021 ) provides the theoretical basis for expecting that wage floors in imperfectly competitive labour markets need not produce the disemployment effects predicted by competitive models. The substitution patterns documented here are consistent with predictions from this combined literature: platforms with market power over multiple compensation channels respond to single-channel regulation by adjusting along unregulated channels. A separate literature on choice architecture (Thaler & Sunstein, 2008 ; Goldin & Reck, 2020 ) is relevant to the tip-suppression mechanism in Section 5 ; the regulatory response, in the form of mandated checkout-stage tip prompts, applies the same body of theory. 2.1 The Minimum Pay Rule Local Law 115 of 2021 charged the Department of Consumer and Worker Protection with studying pay and working conditions in this sector and establishing a minimum pay rate by rule. The agency’s November 2022 study (DCWP, 2022) estimated that workers earned $ 14.18 per hour with tips and $ 7.09 without, against $ 3.06 per hour in expenses, leaving net hourly earnings of $ 11.12 with tips and $ 4.03 without. Adjusted for hours actually worked, the typical app delivery worker’s annual net earnings sat at roughly $ 12,000—below the city’s official poverty threshold for a single adult. The rule the agency adopted in June 2023 set a per-hour minimum applied to the sum of trip time and on-call time, with a phase-in over two years and an annual cost-of-living adjustment thereafter. The covered apps could comply in either of two ways. Under the Standard Method, the app’s payment to each individual worker had to meet or exceed the rate multiplied by that worker’s trip time, and the app’s aggregate payment across all workers had to meet or exceed the rate multiplied by the sum of all workers’ trip-plus-on-call hours. Under the Alternative Method, the app could pay only for trip time at a higher per-hour rate, conditional on maintaining a workforce-level utilization rate above 53 percent. Apps were free to choose. After lawsuits filed by the major apps were denied at the Appellate Division in November 2023, enforcement began on December 4, 2023 at $ 17.96 per hour. The April 2024 adjustment to $ 19.56 combined a scheduled phase-in step with a cost-of-living adjustment. The April 2025 adjustment to $ 21.44 reflected the completion of the phase-in plus a 7.41 percent inflation adjustment for the period December 2022 to December 2024. The April 2026 adjustment to $ 22.13 reflects a 3.2 percent adjustment for inflation between December 2024 and December 2025. 2.2 Data All quantitative results in this paper come from the Department of Consumer and Worker Protection’s public quarterly aggregated tables for restaurant delivery apps (DCWP, 2025), downloaded as a single XLSX file that the agency updates each quarter. The covered apps are DoorDash, FanTuan, Grubhub, HungryPanda, Relay, and Uber Eats. The four largest of these are collectively responsible for approximately 99 percent of app-based restaurant deliveries in New York City; the two smaller apps (FanTuan and HungryPanda) primarily serve East Asian restaurant niches. All measures are weekly averages within the relevant quarter. The data cover sixteen quarters from the first quarter of 2022 through the fourth quarter of 2025—four years in total, with seven full quarters before the rule took effect and eight quarters after. Table 1 reports summary statistics for the four outcome series central to the analysis: realised pay per hour, weekly on-call hours, tips per delivery, and the active worker count, separately for the seven full pre-rule quarters (Q1 2022 through Q3 2023) and the eight post-rule quarters (Q1 2024 through Q4 2025). The single transitional quarter Q4 2023, during which enforcement began on December 4, is excluded from both sub-samples. Across all four series the pre/post difference in means is large relative to the within-period standard deviation, which is the qualitative signal the formal break tests in Sections 3 through 6 quantify. Table 1 Summary statistics for the four outcome series, by sub-period. Series Pre-rule mean Pre-rule SD Post-rule mean Post-rule SD Pre/post difference Realised pay per hour ( $ ) 5.92 0.74 20.32 0.99 + 14.40 On-call hours per week (000s) 920.3 108.2 289.6 138.7 −630.7 Tips per delivery ( $ ) 3.92 0.05 1.25 0.12 −2.67 Active workers (000s) 104.7 3.5 88.0 12.7 −16.7 Notes. Pre-rule period covers Q1 2022 through Q3 2023 (n = 7 quarters). Post-rule period covers Q1 2024 through Q4 2025 (n = 8 quarters). The transitional quarter Q4 2023, during which enforcement began on December 4, is excluded from both sub-samples. Source: DCWP (2025) quarterly aggregated tables. 2.3 Two Caveats About the Aggregate Data Two limitations of the public data are important to state clearly. First, the tables are aggregated across all six covered platforms with no per-app breakdown. The Department’s subsequent tip-suppression report (DCWP, 2026) uses per-platform data internally to attribute the tip decline specifically to Uber Eats and DoorDash, but no equivalent per-app breakdown is available in the public quarterly tables. This precludes any analysis that would require comparison across platforms—for example, asking whether Relay’s restaurant-facing-only model produces different worker outcomes than Uber Eats’s consumer-facing model. Second, the worker counts are not adjusted for multi-apping—the practice of a single individual maintaining accounts on more than one platform. A worker active on both DoorDash and Uber Eats is counted twice in the total worker figure, and the hours that worker logs while concurrently signed in to both apps are counted twice in total hours. The agency’s 2022 study estimated that 56.3 percent of workers held accounts on more than one platform and that 17.7 percent of working time was logged concurrently. The Department also notes in its quarterly tables that a reduction in multi-apping is an expected consequence of the rule, since paying for on-call time gives platforms an incentive to limit it. This means the workforce contraction documented in Section 6 partially reflects falling multi-apping rates rather than entirely representing a decline in unique individuals working in the sector. Section 6.1 presents bounds under alternative assumptions. 2.4 Statistical Approach The data structure is a short panel: sixteen quarterly observations on industry-aggregate weekly outcomes. This structure rules out trip-level identification of the kind possible in the parallel ride-hail setting (Section 7.1 ) but is well suited to two complementary classes of inference: structural break tests, which test the null hypothesis of no break against the alternative that the data-generating process changes at one or more points in time; and interrupted time series (ITS) specifications, which estimate the magnitude of level and trend shifts at known intervention dates. Both classes are standard in regulatory-impact evaluation when the unit of analysis is a single time series rather than a panel of comparable units (Bai & Perron, 1998 ; Bernal et al., 2017 ). Structural break tests. For each outcome series, the framework predicts a break at a specific date determined by the regulatory architecture: Q1 2024 for realised pay (the first full post-rule quarter); Q4 2023 for tips per delivery (when Uber Eats and DoorDash redesigned tip flows contemporaneous with rule enforcement); Q2 2024 for on-call hours (when the on-call dilution margin was closed by the April 2024 tightening); and Q1 2024 for active workers. With the break date specified ex ante by theory, the Chow ( 1960 ) F-test compares residual sums of squares from a single OLS fit on the full sample to the sum from two sub-period fits split at the predicted break. As a robustness check the paper also reports the sup-Wald statistic of Andrews ( 1993 ), which selects the break date endogenously by maximizing the Chow F-statistic over a candidate window that brackets the predicted break. For each series the candidate window covers Q4 2023 through Q1 2025 with appropriate trimming. Interrupted time series specification. For magnitude estimation the paper uses a segmented-regression ITS specification of the form y_t = β_0 + β_1 · t + β_2 · D_t + β_3 · (t − T*) · D_t + ε_t, where t indexes quarters from 1 to 16, T* is the intervention quarter, D_t is an indicator for t ≥ T*, β_1 captures the pre-intervention trend, β_2 the level shift at the intervention, and β_3 the change in trend after the intervention (Bernal et al., 2017 ). Estimation is by OLS with Newey-West (1994) heteroskedasticity- and autocorrelation-consistent standard errors at maxlag = 2, the value selected by the Newey-West (1994) data-dependent rule for n = 16. Permutation inference. With sixteen quarterly observations, asymptotic inference is unreliable in the presence of serial correlation. The paper therefore reports permutation p-values computed by randomizing the dependent variable across the sixteen observations 5,000 times, recomputing the Chow F-statistic at the predicted break date under each permutation, and reporting the share of permutations producing a test statistic at least as extreme as the observed one (with the standard add-one finite-sample correction). The permutation procedure is robust to serial correlation and small-sample bias and provides a transparent benchmark when n = 16 is too small for asymptotic results to be relied upon. Limits of the approach. Three caveats apply. First, the tests detect breaks at the predicted dates; they cannot, on aggregate data alone, fully exclude that the breaks reflect coincident shocks unrelated to the rule. Section 5 addresses this for the tip decline by appealing to the agency’s within-industry comparison (platforms that did and did not modify their tip flows). Section 4 notes that the on-call collapse is contemporaneous with the April 2024 tightening to within one quarter and is unprecedented in the pre-rule data. Second, with sixteen observations and the largest breaks concentrated in two quarters, the tests have power against very large breaks and limited power against small ones; this is appropriate to the research question (the paper asks whether the predicted substitutions are detectable) but limits what can be said about magnitudes for which the data are simply too thin. Third, the regulator’s attribution of the tip decline to specific platforms’ user-interface choices relies on per-platform data not in the public tables; this paper presents the attribution as the regulator’s own conclusion and discusses three alternative explanations in Section 5 . Disclosure of artificial-intelligence assistance. The author used Anthropic’s Claude (model family Claude 4.7), a large language model, as an assistant during the preparation of this manuscript. Claude was used to help write the Python scripts that produced the analytical panel, the figures, and the structural break and ITS estimates from the public DCWP quarterly tables, and to produce initial drafts of expository prose that the author then reviewed, revised, and edited. All statistical results derive from the author’s own execution of the analysis scripts on the cited public dataset, and all results were independently verified by the author. All interpretations, policy claims, editorial decisions about scope and framing, and conclusions are the author’s. Claude did not contribute original research ideas, did not select the analytical approach, and is not listed as an author. The author takes full responsibility for the content of this work. 3. Did the Floor Bind? The first question is whether realised hourly pay actually rose to meet the statutory floor. Figure 1 plots realised pay per hour alongside the rule’s statutory rate across all sixteen quarters in the data. The pattern is striking. From the first quarter of 2022 through the third quarter of 2023—the seven full quarters before enforcement—pay per hour drifted in the range of $ 5.05 to $ 7.08, with no apparent trend. In the fourth quarter of 2023, the quarter during which enforcement began at mid-quarter, pay per hour rose to $ 7.57, reflecting a partial-quarter effect. In the first full post-rule quarter (Q1 2024), pay per hour jumped to $ 16.37, more than tripling from the pre-rule baseline. A reader looking at Fig. 1 might notice that the $ 16.37 realised pay in the first quarter of 2024 sits below the nominal $ 17.96 statutory floor. This is not non-compliance. The rule’s phase-in design permitted apps to maintain higher levels of uncompensated on-call time during the transition window, and the agency’s narrative report for that quarter (DCWP, 2024a) explicitly addressed the gap, noting that some apps maintained excessive levels of uncompensated on-call time during the quarter and that this was allowable and did not indicate a legal violation. Additional protections against excessive on-call time took effect on April 1, 2024. The April 2024 tightening, combined with the simultaneous rate increase to $ 19.56, produced a clean response. Pay per hour rose to $ 19.34 in the second quarter of 2024 and stayed within a tight band of $ 19.28 to $ 19.44 through the first quarter of 2025—closely tracking the $ 19.56 floor. With the April 2025 increase to $ 21.44, realised pay rose to $ 20.96 in the second quarter of 2025 and reached $ 21.49 by the fourth quarter, again closely tracking the new floor. The Chow test on the pay-per-hour series with break date set to Q1 2024 rejects the no-break null decisively [F(2,12) = 93.82, p < 0.0001]. The sup-Wald test over the Q4 2023 to Q1 2025 candidate window selects Q1 2024 as the optimal break date, with sup-Wald = 93.8, well in excess of Andrews’ ( 1993 ) one-percent critical value at the 15-percent trimming fraction. The ITS specification with intervention at Q1 2024 yields a level shift estimate of + $ 11.60 per hour (Newey-West HAC SE = 0.90; permutation p = 0.0002), with no detectable pre-intervention trend (β_1 = − 0.04, HAC SE = 0.12) and a positive post-intervention trend of + $ 0.64 per quarter (HAC SE = 0.11) tracking the April 2024 and April 2025 cost-of-living adjustments. The model R² is 0.989. The headline answer to the first question is therefore yes, with the qualification that the phase-in window operated as designed: it allowed a transition period during which the nominal rate could be partially diluted through on-call time, and once that margin was closed in April 2024, the rule has bound tightly through the most recent observation. From the third quarter of 2023 (the last full pre-rule quarter, $ 5.05 per hour) to the fourth quarter of 2025 ( $ 21.49 per hour), realised industry-aggregate hourly pay rose by 326 percent, or approximately $ 16 per hour—consistent with what the agency intended. 4. The On-Call Collapse Once the agency closed the on-call dilution margin in April 2024, the second question becomes pressing. If platforms must pay for on-call time at the same rate as trip time, what do they do? Fig. 2 shows the answer. Through the seven full pre-rule quarters and into the December 2023 enforcement window, on-call hours per week ran consistently above 750,000 and rose into the 1,000,000 to 1,072,000 range across 2023. After the December 2023 enforcement, on-call hours dropped modestly in the first quarter of 2024 (to 612,000) and then collapsed once the April 2024 tightening took effect: 305,000 in the second quarter of 2024, 240,000 in the third, 193,000 in the fourth. By the end of 2024, weekly on-call hours had fallen 82 percent from their Q3 2023 peak. The Chow test with break date Q2 2024 (the prediction implied by the April 2024 tightening) rejects no-break [F(2,12) = 16.07, p = 0.0004], and the ITS specification with intervention at Q2 2024 yields a level shift of − 672.8 thousand weekly on-call hours (HAC SE = 140.1; permutation p = 0.0006), the largest single discontinuity in any series in the dataset. The model R² is 0.903. The sup-Wald test over the Q4 2023 to Q1 2025 candidate window, which selects the break date endogenously, selects Q1 2024 (sup-Wald F = 33.11) rather than Q2 2024. This is informative rather than contradictory: it means platforms began draining on-call capacity at the December 2023 enforcement onset rather than waiting for the April 2024 tightening, and that the early action was already statistically detectable in the Q1 2024 figure of 612,000 weekly hours, even before the steeper Q2-Q4 declines. The framework’s prediction (a break tied to the April 2024 closure) is therefore confirmed in magnitude but the temporal onset is one quarter earlier than the closure date alone would suggest. Trip hours moved in the opposite direction. From the third quarter of 2023 through the fourth quarter of 2025, trip hours per week rose from 815,000 to 1,119,000—a 37 percent increase. The platforms did not reduce the volume of paid productive work; they specifically eliminated the unpaid-when-pre-rule, paid-when-post-rule on-call time that the rule’s design had targeted. Mechanically, this happened through several documented channels. The agency’s 2022 study and subsequent quarterly reports describe these explicitly: tightening of platform access controls (the industry term is “gating”) during periods of low demand, increased reliance on pre-scheduled shifts that condition future earnings on shift completion, automatic disconnection after periods of inactivity, and prioritization of high-acceptance-rate workers for available trip offers. Several of these practices have been the subject of subsequent litigation; one platform began limiting the number of workers who can be online during each hour of the day in late 2023. This pattern—trip hours flat or rising while on-call hours collapse—is the cleanest evidence in the data of what regulators sometimes call a margin shift. The rule did not change the underlying demand for delivery (which was rising independently). It did not change the technology of delivery. It did not even meaningfully change the average duration or distance of a trip. What it changed was the balance between paid trip time and what had previously been a large pool of effectively unpaid on-call time. Once that pool became expensive to maintain, the platforms drained it. The welfare implications of this margin shift are mixed. From the worker side, on-call time represents both a cost (waiting without earning) and an option value (the ability to make oneself available for trips that may or may not arrive). The pre-rule equilibrium contained a great deal of waiting that produced small per-hour earnings; the post-rule equilibrium contains less waiting but at much higher hourly pay. Workers who can secure scheduled shifts or qualify for high-acceptance-rate priority access benefit substantially. Workers who previously depended on casual, ad-hoc availability to the platforms have lost the ability to log in freely. This distributional question is real and important; the aggregate data used in this paper cannot resolve it directly. 5. The Tip Decline The second substitution margin operated in parallel with the on-call collapse. Figure 3 shows the tip-per-delivery time series across the same sixteen quarters. From the first quarter of 2022 through the third quarter of 2023, tips per delivery sat in a tight band between $ 3.85 and $ 3.98. In the fourth quarter of 2023—the same quarter enforcement of the minimum pay rule began—tips per delivery fell to $ 3.38, a 12 percent drop. By the first quarter of 2024, the figure had fallen to $ 1.44; by the fourth quarter of 2025, to $ 1.10. Total tips paid out across the industry fell from $ 10.6 million per week in the first quarter of 2023 to $ 3.9 million per week in the fourth quarter of 2025. The Chow test with break date Q4 2023 (the prediction implied by the December 2023 enforcement) rejects no-break [F(2,12) = 7.14, p = 0.009]. The ITS specification with intervention at Q4 2023 estimates a level shift of − $ 1.71 per delivery (HAC SE = 0.39; permutation p = 0.008), with no detectable pre-intervention trend (β_1 ≈ 0, HAC SE = 0.005) and a small additional negative trend of − $ 0.17 per quarter (HAC SE = 0.07) in subsequent quarters. The model R² is 0.917. The sup-Wald test over the Q3 2023 to Q1 2025 candidate window selects Q1 2024 as the optimal break date with a much larger statistic (sup-Wald F = 171.94). This pattern—a detected break at the framework’s predicted Q4 2023 date but with the strongest statistical signal at Q1 2024—is consistent with the regulator’s account: the platforms’ user-interface redesigns began in December 2023 but took a full quarter to fully reach all users and to be reflected in the consumer behaviour underlying the aggregate tipping series. The agency’s January 2026 report (DCWP, 2026) attributes this decline specifically to user-interface changes implemented by Uber Eats and DoorDash beginning in December 2023, contemporaneous with enforcement. Prior to those changes, both platforms presented the tip prompt during the checkout flow alongside fee and tax displays, with selectable percentage options. After the changes, the prompt moved to a separate post-checkout screen requiring additional navigation. The agency’s comparison platforms—Grubhub, Relay, and the smaller apps—did not modify their tip flows over the same period and exhibited stable per-delivery tip averages of approximately $ 2.17 per delivery, a figure substantially higher than the $ 0.76 average the agency reports for Uber Eats and DoorDash specifically. The agency’s estimate of cumulative worker harm from these design choices is $ 554 million over the period from December 2023 through mid-2025. This paper does not independently verify the agency’s attribution of the tip decline to specific platform interface choices. The DCWP’s analysis relies on per-platform data not available in the public quarterly tables; this paper presents the attribution as the regulator’s own conclusion. Three alternative explanations for an industry-wide tip decline are worth considering and ruling out. First, post-pandemic normalisation of consumer tipping behaviour: tips per delivery were stable in the band of $ 3.85 to $ 3.98 from Q1 2022 through Q3 2023 and showed no downward trend; the decline begins sharply in Q4 2023 contemporaneous with rule enforcement, not gradually as a normalisation story would predict, and the ITS pre-intervention trend coefficient is indistinguishable from zero. Second, inflation crowding out tip budget: order subtotals rose only modestly over the same period (from $ 27.11 in Q1 2023 to $ 29.04 in Q4 2024), and consumer fees per delivery rose substantially (from $ 4.96 to $ 5.54), but the proportional decline in tips relative to either subtotals or total order cost is much larger than what budget crowding would imply. Third, demand composition shifts: if the order mix shifted toward lower-tip categories or geographies, aggregate tips per delivery could fall without any platform behaviour change. The agency’s within-industry comparison rules this out: platforms that did not modify their tip flows showed stable tipping over the same period, while platforms that did modify them showed the documented collapse. The aggregate timing evidence cannot exclude all alternatives but is difficult to reconcile with anything other than the platform-design attribution. The arithmetic of the substitution is informative. In the first quarter of 2023, the pre-rule baseline, the typical hour of work generated roughly $ 5.89 in pay and $ 5.83 in tips, for total earnings of $ 11.72 per hour. In the fourth quarter of 2025, the typical hour generated $ 21.49 in pay and $ 2.85 in tips, for total earnings of $ 24.35 per hour. Workers gained approximately $ 15.60 per hour in pay and lost approximately $ 2.98 per hour in tips. The gain dominates the loss by a factor of more than five. This means that even after the documented tip suppression, workers are substantially better off in the post-rule equilibrium than they were in the pre-rule one. The substitution did not undo the rule’s welfare gain; it captured a significant share of what would otherwise have been a larger gain. If the pre-rule tip-per-delivery rate of approximately $ 3.85 had been preserved through the post-rule period, fourth-quarter 2025 tips per hour at the observed productivity of 2.54 deliveries per hour would have been approximately $ 9.78—rather than the observed $ 2.85. The redistribution from worker tips to platform pricing discretion is on the order of $ 7 per hour worked, or roughly 30 percent of the total compensation gain that the rule was intended to deliver. A January 2026 amendment to the Delivery Worker Laws now requires both restaurant and grocery delivery apps to present a checkout-stage tip prompt with a selectable 10 percent option and a custom-amount field, taking effect on January 26, 2026. The agency estimates that, if the platforms apply the same tip-flow design they use in markets without these restrictions, worker tip earnings will rise by approximately $ 390 million per year. Litigation by Uber Eats and DoorDash to enjoin the new requirement was denied by federal judges in late January 2026; enforcement is in its initial weeks at the time of this writing. The first quarterly tables covering periods after January 26, 2026 are expected in mid-to-late 2026 and will provide an unusually clean test of the platform-design attribution. If tips per delivery at Uber Eats and DoorDash recover toward the levels observed at the comparison platforms (approximately $ 2.17), the regulator’s attribution to user-interface design choices is supported. If tips at the two platforms remain depressed despite the restored prompts, the alternative explanations—that consumer behaviour or market conditions independently changed—become more credible. 6. Workforce Contraction and Productivity Rise The third substitution margin is the size and composition of the platform-engaged workforce itself. Figure 4 presents three panels showing workers, productivity, and total industry output over the sixteen quarters in the data. The active workforce, defined as the number of workers logging at least one on-call or trip period in a given week, peaked at 111,000 in the fourth quarter of 2023 and fell to 75,000 by the fourth quarter of 2025—a 32 percent contraction. Workers performing at least one trip in a week fell from 81,000 to 70,000 over the same period, a 14 percent contraction. The gap between these two figures (the workers who logged on-call time without taking a trip) compressed from 30,000 to 5,000, consistent with the on-call elimination documented in Section 4 . The Chow test on active worker counts with break date Q1 2024 rejects no-break [F(2,12) = 73.33, p < 0.0001]. The sup-Wald procedure over the Q4 2023 to Q1 2025 candidate window selects Q2 2024 as the optimal break (sup-Wald F = 77.38), marginally preferred over Q1 2024, consistent with the framework’s prediction that workforce contraction would lag the December 2023 enforcement by one to two quarters as the gating policies took effect. The ITS specification with intervention at Q1 2024 estimates a level shift of − 6.12 thousand active workers (HAC SE = 1.61; permutation p = 0.0002) plus a continuing negative trend of − 6.60 thousand per quarter (HAC SE = 0.53) through 2025. The model R² is 0.967. The combination of a level shift and a negative post-intervention trend is a useful signature of the framework: workforce contraction is not a one-time adjustment but a continuing process as platforms progressively tighten access controls. Productivity rose simultaneously. Deliveries per worker-hour—the agency’s primary productivity measure—rose from 1.41 in the third quarter of 2023 to 2.54 in the fourth quarter of 2025, an 80 percent increase. Average hours per worker rose from 17.3 to 18.1 over the same period. Both figures represent more intensive use of those workers who remained active on the platforms. Total industry output—weekly deliveries—grew from 2.65 million in the first quarter of 2023 to 3.47 million in the fourth quarter of 2025, a 31 percent increase. The pre-rule predictions made by the platforms during the rulemaking process and during litigation—that delivery volumes would collapse, that the industry would withdraw service from the city—did not materialize. Consumer demand grew steadily over the four-year window, undisturbed by either the rule or by the per-delivery consumer fee increases (from $ 4.74 in the first quarter of 2023 to $ 7.58 in the fourth quarter of 2025) that the rule produced as platforms passed costs through. What this combination means is that the same volume of delivery work is being done by fewer workers, each working slightly longer hours and substantially more productively. Whether this is welfare-improving depends on a counterfactual that the data cannot resolve directly: how many of the workers represented by the contraction in active accounts would have preferred to remain at the higher hourly pay if access had not been restricted, versus how many were casual or marginal participants who chose to exit when the platform-side incentives shifted against intermittent low-volume engagement. The aggregate data are silent on this. What can be said is that aggregate worker compensation—total earnings paid out across all workers, in dollars per week—rose from $ 19.5 million in the third quarter of 2023 to $ 33.3 million in the fourth quarter of 2025, a 71 percent increase. Whatever the distributional consequences of the workforce contraction, the total flow of dollars from platforms to workers grew substantially in absolute terms. 6.1 Sensitivity to Multi-Apping The 32 percent account-level contraction reported above conflates two distinct phenomena: an actual reduction in the number of unique individuals working in the sector, and a reduction in the practice of single individuals maintaining multiple platform accounts. The agency’s 2022 study (DCWP, 2022) provides the only available anchor for separating these. In the fourth quarter of 2021, 219,787 worker accounts at the four largest apps were held by approximately 122,104 unique individuals, implying a multi-apping multiplier of roughly 1.80. The same study estimated that 56.3 percent of workers held accounts on more than one platform. The DCWP itself notes in its quarterly tables that a reduction in multi-apping is an expected consequence of the rule, since the requirement that platforms pay for on-call time reduces the value to workers of keeping multiple apps logged in concurrently. This paper’s data cannot directly observe the post-rule multi-apping rate, but the bounds on the implied change in unique individuals can be computed under alternative assumptions. Under the assumption that the multi-apping rate was unchanged from its 2022 baseline (multiplier 1.80), the implied unique-individual workforce fell from approximately 61,700 in the fourth quarter of 2023 to approximately 41,700 in the fourth quarter of 2025—a 32 percent decline that exactly matches the account-level contraction. Under the opposite assumption that multi-apping was fully eliminated by the end of 2025 (multiplier 1.0), the implied unique-individual workforce would have risen from approximately 61,700 to 75,000, a 22 percent increase. A plausible mid-range assumption—that multi-apping fell from 56 percent to 30 percent of workers, with the multiplier moving correspondingly from 1.80 to approximately 1.40—implies a unique-individual contraction of roughly 13 percent. The bounds therefore range from a 22 percent increase to a 32 percent decrease in unique individuals working in the sector, with intermediate values most plausible. This is a wide range, and it bears emphasising that the headline 32 percent contraction reported in the narrative above is an upper bound on the unique-individual decline. A reader who interprets the contraction as evidence that one in three workers was displaced by the rule would be overstating what the data can support; the most that can be said with confidence is that platforms tightened access controls and that the share of those access changes representing displacement of unique individuals versus reduction of multi-apping cannot be separated from public data alone. Resolving this question would require either per-platform data with cross-platform identifiers or a follow-up DCWP survey of post-rule workers. 7. Discussion: A Framework of Regulatory Substitution Across Compensation Channels The patterns documented in Sections 3 through 6 share a common structure. Platform compensation in app-based work is not a single quantity but a vector across multiple channels: hourly pay, tip facilitation, on-call exposure, and platform access, each of which the platform controls and any of which can be adjusted to change a worker’s realised earnings. A wage regulation that addresses one channel of this vector—the per-hour rate in the case studied here—leaves the others as available margins for platform response. The empirical contribution of this paper is to test that platforms used all three of those margins in the New York case and that the temporal sequence of substitutions matches the framework’s prediction. The conceptual contribution is to propose this as a general framework: regulatory substitution across compensation channels predicts that single-channel wage rules in multi-channel compensation environments will produce systematic, predictable, and sequenced substitution responses keyed to which channels the regulator has and has not addressed at each stage of the rule’s implementation. Each substitution arrived in a specific temporal sequence keyed to the regulatory architecture. The break tests support this directly: the sup-Wald procedures select Q1 2024 as the optimal break for both the on-call and the tip series, but the framework’s a priori prediction—Q2 2024 for on-call (the April 2024 closure of the dilution margin) and Q4 2023 for tips (the December 2023 enforcement)—is also rejected at conventional significance levels by the Chow tests. What this means substantively is that platforms began reacting to the rule at the moment of enforcement onset (December 2023), not at the moment a particular margin was formally closed; the formal closures sharpened breaks that were already underway. The on-call-time drainage and the tip-flow redesign were both deployed as Q1 2024 strategies, with the April 2024 tightening intensifying the on-call response that was already in motion. Each substitution is, viewed individually, a rational platform response to the relevant regulatory constraint and the costs the constraint imposes. Each is also a response that reduces the welfare gain the rule was intended to produce, in different ways and for different sub-populations of workers. Taken together, they suggest that a regulatory architecture for platform work that targets only the per-hour wage rate is incomplete. The platforms control multiple compensation channels: hourly pay, tip facilitation, on-call exposure, and access to the platform itself. A wage rule that addresses only the first of these gives the platforms a clear strategic playbook for reducing its bite by adjusting the others. The implication for regulatory design is not that wage rules are ineffective—the evidence here is that the New York rule clearly worked at its primary task of raising hourly pay. The implication is rather that single-margin wage rules in multi-margin compensation environments will tend to produce predictable substitution responses, and that a complete regulatory architecture should anticipate and address the parallel margins from the start rather than chase each substitution serially after the fact. This sequenced-response pattern also suggests a research design opportunity. The April 2026 cost-of-living adjustment to $ 22.13 per hour will be the first scheduled rate change to occur after the operational substitution patterns have stabilised: on-call hours have been flat at approximately 250,000 per week for five consecutive quarters, and the tip requirement that took effect January 26, 2026 will, by Q2 2026, have produced its initial observable response in the data. Whether the April 2026 adjustment produces a clean pass-through into realised pay—or whether platforms identify yet another substitution margin—is an open empirical question. The next public quarterly tables, expected in mid-to-late 2026, will provide the first answer. 7.1 Comparison to the Ride-Hail Setting A natural comparison is with the Taxi and Limousine Commission’s parallel minimum earnings standard for app-based for-hire vehicle drivers, which has been in place since February 2019 and was substantially restructured in August 2025. The high-level contrast is informative on its own. The ride-hail rule operates as a per-trip floor, and recent evidence on the August 2025 restructuring shows that platforms responded by moving away from per-trip floor-anchoring entirely—the pre-rule spike at the floor reorganised into a more dispersed distribution. The food delivery rule analysed here operates as a per-hour floor, and platforms have not moved away from binding at it; realised hourly pay tracks the floor closely. What platforms have done instead is substitute along three other compensation channels, none of which the per-hour rule directly regulates. The two regimes thus exhibit different forms of platform response to wage regulation, both consistent with rational platform behaviour but operating at different levels of the compensation system. The two regimes share broad architectural features—both are minimums applied to a combination of engaged and on-call time—but they differ in two important respects. First, the ride-hail rule applies to a market with two dominant platforms (Uber and Lyft) and a relatively homogeneous service product, while the food delivery rule covers six platforms with substantially differentiated business models. Second, the ride-hail Commission publishes trip-level administrative data covering every dispatched ride, while the Department of Consumer and Worker Protection publishes only industry-aggregate quarterly tables. The contribution of the present paper is conceptually distinct from the ride-hail evidence. Per-trip distributional restructuring documents how concentrated platform pay decisions are around the regulatory floor, and how that concentration changes when the floor moves. The present paper documents substitution behaviour across compensation channels: how platforms reallocate worker compensation between hourly pay, on-call exposure, tip facilitation, and platform access in response to a per-hour rule. The two phenomena are adjacent but operate at different levels of platform decision-making and require different data to detect. What unifies the two settings is the broader observation that platform compensation systems are multi-dimensional, that wage regulations targeting any single dimension will tend to produce predictable adjustments in the others, and that a regulator equipped only with the targeted dimension’s data cannot fully observe the system response. The ride-hail Commission’s decision to publish trip-level data has been instrumental in enabling outside researchers to study per-trip distributional questions; the DCWP’s choice to publish only industry-aggregate tables substantially constrains what outside analysis can establish about platform-specific or worker-specific outcomes in the food delivery setting. The contrast in available evidence across the two adjacent regulatory regimes is itself informative about the value of data-publication policy as a complement to substantive rulemaking. 8. Conclusion Three years of public industry-aggregate data on New York City’s app-based food delivery sector show that the December 2023 minimum pay rule worked at its primary task. Realised hourly pay rose from $ 5.05 in the last full pre-rule quarter to $ 21.49 by the end of 2025, and now closely tracks the statutory floor. Total weekly industry compensation paid to workers rose from $ 19.5 million to $ 33.3 million over the same period. Structural break tests detect highly significant breaks at the rule’s implementation dates in all four outcome series the regulatory-substitution framework predicts: realised pay (Chow F = 93.8 at Q1 2024), on-call hours (Chow F = 16.1 at Q2 2024; sup-Wald F = 33.1 at Q1 2024), tips per delivery (Chow F = 7.1 at Q4 2023; sup-Wald F = 171.9 at Q1 2024), and active worker counts (Chow F = 73.3 at Q1 2024). All level-shift estimates are statistically significant under both Newey-West HAC standard errors and 5,000-permutation inference. The data also show that platforms responded along three substitution margins: on-call hours collapsed by 82 percent (estimated level shift of − 673 thousand weekly hours), tips per delivery fell 71 percent (estimated level shift of − $ 1.71 per delivery) following user-interface redesigns at two of the four major platforms, and the active workforce contracted by 32 percent (estimated level shift of − 6.1 thousand workers, with a continuing post-intervention trend of − 6.6 thousand per quarter) while productivity rose 80 percent and total deliveries grew 31 percent. Each substitution arrived in a specific sequence keyed to which margin the regulator had left open at each stage of the rule’s implementation. None reverses the rule’s aggregate welfare gain, but each captures a substantial share of what the gain would otherwise have been. The substantive lesson is that a regulatory architecture for platform work targeting only the per-hour wage rate is incomplete in multi-margin compensation environments. The platforms control hourly pay, tip facilitation, on-call exposure, and platform access, and a rule that addresses only the first of these gives the platforms a clear and predictable strategic response. New York City’s subsequent regulatory amendments, including the January 2026 tip-protection rule and ongoing enforcement against access restrictions, represent a serial closing of these substitution margins. Whether the April 2026 cost-of-living adjustment produces clean pass-through to realised pay, or whether platforms identify a further substitution margin, will be observable in the public data within approximately six months. Declarations Funding. No funds, grants, or other support were received for the preparation of this manuscript. Competing Interests. The author declares no competing interests. The author is not affiliated with any of the platforms studied, has not received funding from them, and has no employment, contractual, or advisory relationships with the regulator (DCWP) or with any worker organisation involved in the rulemaking discussed in this paper. Data Availability. All data used in this paper are publicly available from the Department of Consumer and Worker Protection at https://www.nyc.gov/site/dca/workers/Delivery-Worker-Public-Hearing-Minimum-Pay-Rate.page. The XLSX file used for the analysis was downloaded in April 2026, representing data through Q4 2025. Analysis scripts including the Chow, sup-Wald, ITS, and permutation procedures, the cleaned analytical panel, and figure-generation code will be made available in a permanent public repository upon acceptance. No restrictions apply to either the data or the replication materials. Author Contributions. Sole authored. The author conceived the study, performed all analyses, and wrote the manuscript. Use of Artificial Intelligence. Disclosed in Section 2.4 (Statistical Approach) per Springer policy, which requires disclosure of LLM use in the Methods section or a suitable alternative. References Andrews, D. W. K. (1993). Tests for parameter instability and structural change with unknown change point. Econometrica, 61(4), 821–856. https://doi.org/10.2307/2951764 Bai, J., & Perron, P. (1998). Estimating and testing linear models with multiple structural changes. Econometrica, 66(1), 47–78. https://doi.org/10.2307/2998540 Bernal, J. L., Cummins, S., & Gasparrini, A. (2017). Interrupted time series regression for the evaluation of public health interventions: A tutorial. International Journal of Epidemiology, 46(1), 348–355. https://doi.org/10.1093/ije/dyw098 Card, D., & Krueger, A. B. (1994). Minimum wages and employment: A case study of the fast-food industry in New Jersey and Pennsylvania. American Economic Review, 84(4), 772–793. Cengiz, D., Dube, A., Lindner, A., & Zipperer, B. (2019). The effect of minimum wages on low-wage jobs. Quarterly Journal of Economics, 134(3), 1405–1454. https://doi.org/10.1093/qje/qjz014 Chow, G. C. (1960). Tests of equality between sets of coefficients in two linear regressions. Econometrica, 28(3), 591–605. https://doi.org/10.2307/1910133 Cook, C., Diamond, R., Hall, J. V., List, J. A., & Oyer, P. (2021). The gender earnings gap in the gig economy: Evidence from over a million rideshare drivers. Review of Economic Studies, 88(5), 2210–2238. https://doi.org/10.1093/restud/rdaa081 Department of Consumer and Worker Protection, City of New York. (2022). A minimum pay rate for app-based restaurant delivery workers in NYC. November 2022. https://www.nyc.gov/site/dca/workers/Delivery-Worker-Public-Hearing-Minimum-Pay-Rate.page Department of Consumer and Worker Protection, City of New York. (2024a). Restaurant delivery app data: January–March 2024. July 2024. Department of Consumer and Worker Protection, City of New York. (2024b). Restaurant delivery app data: April–June 2024. Department of Consumer and Worker Protection, City of New York. (2024c). Restaurant delivery app data: July–September 2024. Department of Consumer and Worker Protection, City of New York. (2024d). Restaurant delivery app data: October–December 2024. Department of Consumer and Worker Protection, City of New York. (2025). Restaurant delivery app data: Quarterly aggregated tables. https://www.nyc.gov/site/dca/workers/Delivery-Worker-Public-Hearing-Minimum-Pay-Rate.page Department of Consumer and Worker Protection, City of New York. (2026). Uber Eats and DoorDash engineered a $554 million pay cut: NYC is fighting back. January 2026. Dube, A., Lester, T. W., & Reich, M. (2010). Minimum wage effects across state borders: Estimates using contiguous counties. Review of Economics and Statistics, 92(4), 945–964. https://doi.org/10.1162/REST_a_00039 Goldin, J., & Reck, D. (2020). Optimal defaults with normative ambiguity. Review of Economics and Statistics, 104(1), 17–33. https://doi.org/10.1162/rest_a_00946 Hall, J. V., & Krueger, A. B. (2018). An analysis of the labor market for Uber’s driver-partners in the United States. ILR Review, 71(3), 705–732. https://doi.org/10.1177/0019793917717222 Manning, A. (2003). Monopsony in motion: Imperfect competition in labor markets. Princeton University Press. Manning, A. (2021). Monopsony in labor markets: A review. ILR Review, 74(1), 3–26. https://doi.org/10.1177/0019793920922499 Newey, W. K., & West, K. D. (1994). Automatic lag selection in covariance matrix estimation. Review of Economic Studies, 61(4), 631–653. https://doi.org/10.2307/2297912 Parrott, J. A., & Reich, M. (2018). An earnings standard for New York City’s app-based drivers: Economic analysis and policy assessment. Center for New York City Affairs, New School. July 2018. Thaler, R. H., & Sunstein, C. R. (2008). Nudge: Improving decisions about health, wealth, and happiness. Yale University Press. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 04 May, 2026 Editor assigned by journal 03 May, 2026 Submission checks completed at journal 03 May, 2026 First submitted to journal 30 Apr, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9578165","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":635736395,"identity":"77c2cd36-85ec-472c-a10d-e48ea635682a","order_by":0,"name":"Boon Chuan Lim","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6klEQVRIiWNgGAWjYBACAxDB2AAk2BsYDpCohecAkhZ8mhFaJBKQhPFpMWdvPvbg547D8uaSzx8eLvhll8cvffgA84cK3Fose46lG/aeOWy4c3aOweGZfcnFkn1pCQwHzuBx2I0cM2nGtsOMG27nMBzm7WFO3HCGx4DhYBthLfYbbh5/ANRSD9TC/4EoLYkbbjAYHOb5cRhkCwNeLUC/pEn2tqUnbzgD9Atvw/FiyR42gwNn8PgFFGISP9usbTccP/74M8+f6jx+HuaHDyrwhBgqYGxjSADRB4jVAAR/IFpGwSgYBaNgFCADAFcoXEhc4/nwAAAAAElFTkSuQmCC","orcid":"","institution":"","correspondingAuthor":true,"prefix":"","firstName":"Boon","middleName":"Chuan","lastName":"Lim","suffix":""}],"badges":[],"createdAt":"2026-04-30 14:08:31","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9578165/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9578165/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":109099823,"identity":"0f1c323e-396b-4310-8288-e0107280b4a9","added_by":"auto","created_at":"2026-05-12 14:18:45","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":87558,"visible":true,"origin":"","legend":"\u003cp\u003eRealised hourly pay rises to track the statutory floor. The dashed red line marks the December 4, 2023 effective date. Dotted vertical lines mark the April 2024 and April 2025 cost-of-living adjustments. Pay per hour is computed as total weekly pay divided by total weekly hours (trip plus on-call), aggregated across all six covered platforms.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-9578165/v1/e856adf37fe8c395d34111e4.png"},{"id":109099594,"identity":"a1c73f06-b44e-461d-aa2b-ea600c0a9685","added_by":"auto","created_at":"2026-05-12 14:17:40","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":91834,"visible":true,"origin":"","legend":"\u003cp\u003ePlatforms eliminated on-call time when forced to pay for it. Stacked area plot of weekly trip hours (lower band) and on-call hours (upper band) across all six covered platforms. The dashed red line marks the December 2023 enforcement effective date; the dotted line marks the April 2024 rate adjustment.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-9578165/v1/df043ab54689443335d903cc.png"},{"id":109099591,"identity":"4ff4ed18-bbe4-4259-b95f-d64422ef1aca","added_by":"auto","created_at":"2026-05-12 14:17:39","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":86920,"visible":true,"origin":"","legend":"\u003cp\u003eTips per delivery fell 71 percent following the December 2023 enforcement. The shaded band marks the December 2023 implementation window during which Uber Eats and DoorDash redesigned their tip flows. Average tip per delivery is computed as total weekly tips divided by total weekly deliveries.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-9578165/v1/90ddf7ef3b8d9e199441407c.png"},{"id":109099842,"identity":"685dd754-4199-433b-95ca-ed3fffb88b67","added_by":"auto","created_at":"2026-05-12 14:18:58","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":157097,"visible":true,"origin":"","legend":"\u003cp\u003eWorkforce contracted, productivity rose, deliveries grew. Panel A shows weekly active workers across all platforms (not adjusted for multi-apping). Panel B shows deliveries per worker-hour. Panel C shows total weekly industry deliveries. Dashed vertical lines mark the December 2023 enforcement effective date.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-9578165/v1/570e63db5a82f06cf67aa59d.png"},{"id":109100196,"identity":"8f05973c-2cc2-47ad-be98-1fc19abfc06d","added_by":"auto","created_at":"2026-05-12 14:20:37","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":625444,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9578165/v1/d5904a4f-1994-4ac6-a605-978819b737e0.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Platform Substitution Under New York City’s Food Delivery Pay Rule","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eWhen a regulator imposes a wage floor on one channel of platform compensation but leaves three other channels untouched, what happens? Standard economic intuition predicts substitution: rational platforms will adjust along the margins that remain free. The question is whether the substitution is large enough to matter for the regulation\u0026rsquo;s intended welfare transfer, and whether it follows a predictable sequence keyed to the regulatory architecture. New York City\u0026rsquo;s minimum pay rule for app-based restaurant delivery workers, in effect since December 2023 and observable through eight post-rule quarters of public industry-aggregate data, allows both questions to be answered.\u003c/p\u003e \u003cp\u003eThis paper develops and tests a framework of regulatory substitution across compensation channels. Platform compensation is a vector across multiple channels: hourly pay, tip facilitation, on-call exposure, and platform access. A wage regulation that targets one channel leaves the others as available margins for platform response. The framework predicts that single-channel wage rules will produce systematic, sequenced substitution along the unregulated channels, with the temporal ordering of substitutions determined by which margins the regulator has and has not closed at each stage of implementation. The framework yields four testable predictions in the New York setting: structural breaks in realised pay at the December 2023 enforcement (Q1 2024 in the quarterly data), in tips per delivery at the same enforcement date, in on-call hours at the April 2024 tightening, and in active worker counts following both.\u003c/p\u003e \u003cp\u003eEach prediction is tested using Chow (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e1960\u003c/span\u003e) tests at the predicted break date, supremum-Wald structural break tests (Andrews, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1993\u003c/span\u003e) with the break date selected endogenously over a candidate window, and a segmented-regression interrupted time series specification (Bernal, Cummins, \u0026amp; Gasparrini, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) with Newey-West heteroskedasticity- and autocorrelation-consistent standard errors and permutation-based p-values computed over 5,000 randomizations of the intervention indicator. All four breaks are detected at conventional significance levels. Realised pay rises by \u003cspan\u003e$\u003c/span\u003e11.6 per hour at Q1 2024 (Chow F(2,12)\u0026thinsp;=\u0026thinsp;93.8, p\u0026thinsp;\u0026lt;\u0026thinsp;0.0001; permutation p\u0026thinsp;=\u0026thinsp;0.0002). On-call hours fall by 673 thousand per week at Q2 2024 (Chow F(2,12)\u0026thinsp;=\u0026thinsp;16.1, p\u0026thinsp;=\u0026thinsp;0.0004; permutation p\u0026thinsp;=\u0026thinsp;0.0006), with the sup-Wald test indicating that the on-call decline already began at the December 2023 enforcement. Tips per delivery fall by \u003cspan\u003e$\u003c/span\u003e1.71 at Q4 2023 (Chow F(2,12)\u0026thinsp;=\u0026thinsp;7.1, p\u0026thinsp;=\u0026thinsp;0.009), with the sup-Wald test indicating the strongest break in Q1 2024 (F\u0026thinsp;=\u0026thinsp;171.9) once the interface redesigns at Uber Eats and DoorDash had been in effect for a full quarter. Active worker counts fall by 6.1 thousand at Q1 2024 (Chow F(2,12)\u0026thinsp;=\u0026thinsp;73.3, p\u0026thinsp;\u0026lt;\u0026thinsp;0.0001; permutation p\u0026thinsp;=\u0026thinsp;0.0002), continuing to decline through Q4 2025. Each substitution arrives in the predicted temporal sequence.\u003c/p\u003e \u003cp\u003eAt the same time, the rule\u0026rsquo;s primary task succeeded. Realised hourly pay rose from \u003cspan\u003e$\u003c/span\u003e5.05 in the last full pre-rule quarter to \u003cspan\u003e$\u003c/span\u003e21.49 by the end of 2025, and now closely tracks the statutory floor. Total weekly industry compensation paid to workers rose from \u003cspan\u003e$\u003c/span\u003e19.5\u0026nbsp;million to \u003cspan\u003e$\u003c/span\u003e33.3\u0026nbsp;million over the same period. The rule both worked and was partially circumvented; the substantive contribution of this paper is to formalize a substitution framework that predicts the specific mechanisms of circumvention from the regulatory architecture itself, and to test that framework against the eight post-rule quarters of public data.\u003c/p\u003e \u003cp\u003eThe institutional setting is the country\u0026rsquo;s first municipal minimum pay rule for app-based restaurant delivery workers, enforced by New York City\u0026rsquo;s Department of Consumer and Worker Protection beginning December 4, 2023 at \u003cspan\u003e$\u003c/span\u003e17.96 per hour and subsequently raised to \u003cspan\u003e$\u003c/span\u003e19.56 in April 2024, \u003cspan\u003e$\u003c/span\u003e21.44 in April 2025, and \u003cspan\u003e$\u003c/span\u003e22.13 in April 2026. The rule applies to the sum of trip time and on-call time during a worker\u0026rsquo;s pay period, covers six platforms collectively responsible for essentially all app-based restaurant delivery in the city, and operates alongside parallel pay regulation for app-based for-hire vehicle drivers (briefly compared in Section \u003cspan refid=\"Sec13\" class=\"InternalRef\"\u003e7.1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe contribution is twofold. First, the paper proposes the regulatory-substitution framework as a general tool for analyzing wage regulation in multi-margin compensation environments, drawing on the monopsony literature (Manning, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2003\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) and the choice-architecture literature (Thaler \u0026amp; Sunstein, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Goldin \u0026amp; Reck, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) to formalize platform incentives across channels. Second, the paper provides formal statistical evidence that the framework\u0026rsquo;s sequenced predictions hold in the New York case. Each successive rule change tightened one margin and revealed the next one along which platforms could substitute, producing what amounts to a sequenced regulatory experiment.\u003c/p\u003e \u003cp\u003eTwo limitations bound the analysis. First, the public DCWP tables are aggregated across all six covered platforms with no per-app breakdown, so attribution claims that require platform-specific identification rely on the regulator\u0026rsquo;s own internal analyses. Second, the worker counts are not adjusted for multi-apping; Section \u003cspan refid=\"Sec11\" class=\"InternalRef\"\u003e6.1\u003c/span\u003e presents bounds on the implied unique-individual workforce change. Section \u003cspan refid=\"Sec13\" class=\"InternalRef\"\u003e7.1\u003c/span\u003e returns to the comparison with the ride-hail regulatory regime, where trip-level data permit sharper identification.\u003c/p\u003e \u003cp\u003eThe remainder of the paper is organized as follows. Section \u003cspan refid=\"Sec2\" class=\"InternalRef\"\u003e2\u003c/span\u003e describes the institutional setting, the data, and the statistical approach. Section \u003cspan refid=\"Sec7\" class=\"InternalRef\"\u003e3\u003c/span\u003e documents that the floor binds and explains the apparent first-quarter 2024 anomaly. Section \u003cspan refid=\"Sec8\" class=\"InternalRef\"\u003e4\u003c/span\u003e analyses the on-call collapse. Section \u003cspan refid=\"Sec9\" class=\"InternalRef\"\u003e5\u003c/span\u003e analyses the tip decline. Section \u003cspan refid=\"Sec10\" class=\"InternalRef\"\u003e6\u003c/span\u003e analyses the workforce contraction and productivity rise, including a sensitivity analysis for the multi-apping confound. Section \u003cspan refid=\"Sec12\" class=\"InternalRef\"\u003e7\u003c/span\u003e discusses what these patterns imply for the design of platform-work wage rules. Section \u003cspan refid=\"Sec14\" class=\"InternalRef\"\u003e8\u003c/span\u003e concludes.\u003c/p\u003e"},{"header":"2. Setting, Data, and Statistical Approach","content":"\u003cp\u003eThis paper builds on three literatures. The minimum wage literature, beginning with Card and Krueger (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e1994\u003c/span\u003e) and developed by Dube, Lester, and Reich (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2010\u003c/span\u003e) and Cengiz et al. (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), provides the framework for studying how a wage floor binds and how affected firms respond. The platform labour literature, including Hall and Krueger (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) and Cook et al. (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), establishes the structural features of work mediated by app-based platforms. The monopsony literature (Manning, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2003\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) provides the theoretical basis for expecting that wage floors in imperfectly competitive labour markets need not produce the disemployment effects predicted by competitive models. The substitution patterns documented here are consistent with predictions from this combined literature: platforms with market power over multiple compensation channels respond to single-channel regulation by adjusting along unregulated channels. A separate literature on choice architecture (Thaler \u0026amp; Sunstein, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Goldin \u0026amp; Reck, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) is relevant to the tip-suppression mechanism in Section \u003cspan refid=\"Sec9\" class=\"InternalRef\"\u003e5\u003c/span\u003e; the regulatory response, in the form of mandated checkout-stage tip prompts, applies the same body of theory.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 The Minimum Pay Rule\u003c/h2\u003e \u003cp\u003eLocal Law 115 of 2021 charged the Department of Consumer and Worker Protection with studying pay and working conditions in this sector and establishing a minimum pay rate by rule. The agency\u0026rsquo;s November \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2022\u003c/span\u003e study (DCWP, 2022) estimated that workers earned \u003cspan\u003e$\u003c/span\u003e14.18 per hour with tips and \u003cspan\u003e$\u003c/span\u003e7.09 without, against \u003cspan\u003e$\u003c/span\u003e3.06 per hour in expenses, leaving net hourly earnings of \u003cspan\u003e$\u003c/span\u003e11.12 with tips and \u003cspan\u003e$\u003c/span\u003e4.03 without. Adjusted for hours actually worked, the typical app delivery worker\u0026rsquo;s annual net earnings sat at roughly \u003cspan\u003e$\u003c/span\u003e12,000\u0026mdash;below the city\u0026rsquo;s official poverty threshold for a single adult.\u003c/p\u003e \u003cp\u003eThe rule the agency adopted in June 2023 set a per-hour minimum applied to the sum of trip time and on-call time, with a phase-in over two years and an annual cost-of-living adjustment thereafter. The covered apps could comply in either of two ways. Under the Standard Method, the app\u0026rsquo;s payment to each individual worker had to meet or exceed the rate multiplied by that worker\u0026rsquo;s trip time, and the app\u0026rsquo;s aggregate payment across all workers had to meet or exceed the rate multiplied by the sum of all workers\u0026rsquo; trip-plus-on-call hours. Under the Alternative Method, the app could pay only for trip time at a higher per-hour rate, conditional on maintaining a workforce-level utilization rate above 53 percent. Apps were free to choose.\u003c/p\u003e \u003cp\u003eAfter lawsuits filed by the major apps were denied at the Appellate Division in November 2023, enforcement began on December 4, 2023 at \u003cspan\u003e$\u003c/span\u003e17.96 per hour. The April 2024 adjustment to \u003cspan\u003e$\u003c/span\u003e19.56 combined a scheduled phase-in step with a cost-of-living adjustment. The April 2025 adjustment to \u003cspan\u003e$\u003c/span\u003e21.44 reflected the completion of the phase-in plus a 7.41 percent inflation adjustment for the period December 2022 to December 2024. The April 2026 adjustment to \u003cspan\u003e$\u003c/span\u003e22.13 reflects a 3.2 percent adjustment for inflation between December 2024 and December 2025.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Data\u003c/h2\u003e \u003cp\u003eAll quantitative results in this paper come from the Department of Consumer and Worker Protection\u0026rsquo;s public quarterly aggregated tables for restaurant delivery apps (DCWP, 2025), downloaded as a single XLSX file that the agency updates each quarter. The covered apps are DoorDash, FanTuan, Grubhub, HungryPanda, Relay, and Uber Eats. The four largest of these are collectively responsible for approximately 99 percent of app-based restaurant deliveries in New York City; the two smaller apps (FanTuan and HungryPanda) primarily serve East Asian restaurant niches. All measures are weekly averages within the relevant quarter. The data cover sixteen quarters from the first quarter of 2022 through the fourth quarter of 2025\u0026mdash;four years in total, with seven full quarters before the rule took effect and eight quarters after.\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e reports summary statistics for the four outcome series central to the analysis: realised pay per hour, weekly on-call hours, tips per delivery, and the active worker count, separately for the seven full pre-rule quarters (Q1 2022 through Q3 2023) and the eight post-rule quarters (Q1 2024 through Q4 2025). The single transitional quarter Q4 2023, during which enforcement began on December 4, is excluded from both sub-samples. Across all four series the pre/post difference in means is large relative to the within-period standard deviation, which is the qualitative signal the formal break tests in Sections \u003cspan refid=\"Sec7\" class=\"InternalRef\"\u003e3\u003c/span\u003e through \u003cspan refid=\"Sec10\" class=\"InternalRef\"\u003e6\u003c/span\u003e quantify.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSummary statistics for the four outcome series, by sub-period.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSeries\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePre-rule mean\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePre-rule SD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePost-rule mean\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePost-rule SD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePre/post difference\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRealised pay per hour (\u003cspan\u003e$\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e20.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e+\u0026thinsp;14.40\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOn-call hours per week (000s)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e920.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e108.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e289.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e138.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026minus;630.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTips per delivery (\u003cspan\u003e$\u003c/span\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026minus;2.67\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eActive workers (000s)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e104.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e88.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e12.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u0026minus;16.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003eNotes. Pre-rule period covers Q1 2022 through Q3 2023 (n\u0026thinsp;=\u0026thinsp;7 quarters). Post-rule period covers Q1 2024 through Q4 2025 (n\u0026thinsp;=\u0026thinsp;8 quarters). The transitional quarter Q4 2023, during which enforcement began on December 4, is excluded from both sub-samples. Source: DCWP (2025) quarterly aggregated tables.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Two Caveats About the Aggregate Data\u003c/h2\u003e \u003cp\u003eTwo limitations of the public data are important to state clearly. First, the tables are aggregated across all six covered platforms with no per-app breakdown. The Department\u0026rsquo;s subsequent tip-suppression report (DCWP, 2026) uses per-platform data internally to attribute the tip decline specifically to Uber Eats and DoorDash, but no equivalent per-app breakdown is available in the public quarterly tables. This precludes any analysis that would require comparison across platforms\u0026mdash;for example, asking whether Relay\u0026rsquo;s restaurant-facing-only model produces different worker outcomes than Uber Eats\u0026rsquo;s consumer-facing model.\u003c/p\u003e \u003cp\u003eSecond, the worker counts are not adjusted for multi-apping\u0026mdash;the practice of a single individual maintaining accounts on more than one platform. A worker active on both DoorDash and Uber Eats is counted twice in the total worker figure, and the hours that worker logs while concurrently signed in to both apps are counted twice in total hours. The agency\u0026rsquo;s 2022 study estimated that 56.3 percent of workers held accounts on more than one platform and that 17.7 percent of working time was logged concurrently. The Department also notes in its quarterly tables that a reduction in multi-apping is an expected consequence of the rule, since paying for on-call time gives platforms an incentive to limit it. This means the workforce contraction documented in Section \u003cspan refid=\"Sec10\" class=\"InternalRef\"\u003e6\u003c/span\u003e partially reflects falling multi-apping rates rather than entirely representing a decline in unique individuals working in the sector. Section \u003cspan refid=\"Sec11\" class=\"InternalRef\"\u003e6.1\u003c/span\u003e presents bounds under alternative assumptions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Statistical Approach\u003c/h2\u003e \u003cp\u003eThe data structure is a short panel: sixteen quarterly observations on industry-aggregate weekly outcomes. This structure rules out trip-level identification of the kind possible in the parallel ride-hail setting (Section \u003cspan refid=\"Sec13\" class=\"InternalRef\"\u003e7.1\u003c/span\u003e) but is well suited to two complementary classes of inference: structural break tests, which test the null hypothesis of no break against the alternative that the data-generating process changes at one or more points in time; and interrupted time series (ITS) specifications, which estimate the magnitude of level and trend shifts at known intervention dates. Both classes are standard in regulatory-impact evaluation when the unit of analysis is a single time series rather than a panel of comparable units (Bai \u0026amp; Perron, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e1998\u003c/span\u003e; Bernal et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2017\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eStructural break tests. For each outcome series, the framework predicts a break at a specific date determined by the regulatory architecture: Q1 2024 for realised pay (the first full post-rule quarter); Q4 2023 for tips per delivery (when Uber Eats and DoorDash redesigned tip flows contemporaneous with rule enforcement); Q2 2024 for on-call hours (when the on-call dilution margin was closed by the April 2024 tightening); and Q1 2024 for active workers. With the break date specified ex ante by theory, the Chow (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e1960\u003c/span\u003e) F-test compares residual sums of squares from a single OLS fit on the full sample to the sum from two sub-period fits split at the predicted break. As a robustness check the paper also reports the sup-Wald statistic of Andrews (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1993\u003c/span\u003e), which selects the break date endogenously by maximizing the Chow F-statistic over a candidate window that brackets the predicted break. For each series the candidate window covers Q4 2023 through Q1 2025 with appropriate trimming.\u003c/p\u003e \u003cp\u003eInterrupted time series specification. For magnitude estimation the paper uses a segmented-regression ITS specification of the form y_t\u0026thinsp;=\u0026thinsp;β_0\u0026thinsp;+\u0026thinsp;β_1 \u0026middot; t\u0026thinsp;+\u0026thinsp;β_2 \u0026middot; D_t\u0026thinsp;+\u0026thinsp;β_3 \u0026middot; (t\u0026thinsp;\u0026minus;\u0026thinsp;T*) \u0026middot; D_t\u0026thinsp;+\u0026thinsp;ε_t, where t indexes quarters from 1 to 16, T* is the intervention quarter, D_t is an indicator for t\u0026thinsp;\u0026ge;\u0026thinsp;T*, β_1 captures the pre-intervention trend, β_2 the level shift at the intervention, and β_3 the change in trend after the intervention (Bernal et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Estimation is by OLS with Newey-West (1994) heteroskedasticity- and autocorrelation-consistent standard errors at maxlag\u0026thinsp;=\u0026thinsp;2, the value selected by the Newey-West (1994) data-dependent rule for n\u0026thinsp;=\u0026thinsp;16.\u003c/p\u003e \u003cp\u003ePermutation inference. With sixteen quarterly observations, asymptotic inference is unreliable in the presence of serial correlation. The paper therefore reports permutation p-values computed by randomizing the dependent variable across the sixteen observations 5,000 times, recomputing the Chow F-statistic at the predicted break date under each permutation, and reporting the share of permutations producing a test statistic at least as extreme as the observed one (with the standard add-one finite-sample correction). The permutation procedure is robust to serial correlation and small-sample bias and provides a transparent benchmark when n\u0026thinsp;=\u0026thinsp;16 is too small for asymptotic results to be relied upon.\u003c/p\u003e \u003cp\u003eLimits of the approach. Three caveats apply. First, the tests detect breaks at the predicted dates; they cannot, on aggregate data alone, fully exclude that the breaks reflect coincident shocks unrelated to the rule. Section \u003cspan refid=\"Sec9\" class=\"InternalRef\"\u003e5\u003c/span\u003e addresses this for the tip decline by appealing to the agency\u0026rsquo;s within-industry comparison (platforms that did and did not modify their tip flows). Section \u003cspan refid=\"Sec8\" class=\"InternalRef\"\u003e4\u003c/span\u003e notes that the on-call collapse is contemporaneous with the April 2024 tightening to within one quarter and is unprecedented in the pre-rule data. Second, with sixteen observations and the largest breaks concentrated in two quarters, the tests have power against very large breaks and limited power against small ones; this is appropriate to the research question (the paper asks whether the predicted substitutions are detectable) but limits what can be said about magnitudes for which the data are simply too thin. Third, the regulator\u0026rsquo;s attribution of the tip decline to specific platforms\u0026rsquo; user-interface choices relies on per-platform data not in the public tables; this paper presents the attribution as the regulator\u0026rsquo;s own conclusion and discusses three alternative explanations in Section \u003cspan refid=\"Sec9\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eDisclosure\u003c/strong\u003e \u003cp\u003eof artificial-intelligence assistance. The author used Anthropic\u0026rsquo;s Claude (model family Claude 4.7), a large language model, as an assistant during the preparation of this manuscript. Claude was used to help write the Python scripts that produced the analytical panel, the figures, and the structural break and ITS estimates from the public DCWP quarterly tables, and to produce initial drafts of expository prose that the author then reviewed, revised, and edited. All statistical results derive from the author\u0026rsquo;s own execution of the analysis scripts on the cited public dataset, and all results were independently verified by the author. All interpretations, policy claims, editorial decisions about scope and framing, and conclusions are the author\u0026rsquo;s. Claude did not contribute original research ideas, did not select the analytical approach, and is not listed as an author. The author takes full responsibility for the content of this work.\u003c/p\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"3. Did the Floor Bind?","content":"\u003cp\u003eThe first question is whether realised hourly pay actually rose to meet the statutory floor. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e plots realised pay per hour alongside the rule\u0026rsquo;s statutory rate across all sixteen quarters in the data.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe pattern is striking. From the first quarter of 2022 through the third quarter of 2023\u0026mdash;the seven full quarters before enforcement\u0026mdash;pay per hour drifted in the range of \u003cspan\u003e$\u003c/span\u003e5.05 to \u003cspan\u003e$\u003c/span\u003e7.08, with no apparent trend. In the fourth quarter of 2023, the quarter during which enforcement began at mid-quarter, pay per hour rose to \u003cspan\u003e$\u003c/span\u003e7.57, reflecting a partial-quarter effect. In the first full post-rule quarter (Q1 2024), pay per hour jumped to \u003cspan\u003e$\u003c/span\u003e16.37, more than tripling from the pre-rule baseline.\u003c/p\u003e \u003cp\u003eA reader looking at Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e might notice that the \u003cspan\u003e$\u003c/span\u003e16.37 realised pay in the first quarter of 2024 sits below the nominal \u003cspan\u003e$\u003c/span\u003e17.96 statutory floor. This is not non-compliance. The rule\u0026rsquo;s phase-in design permitted apps to maintain higher levels of uncompensated on-call time during the transition window, and the agency\u0026rsquo;s narrative report for that quarter (DCWP, 2024a) explicitly addressed the gap, noting that some apps maintained excessive levels of uncompensated on-call time during the quarter and that this was allowable and did not indicate a legal violation. Additional protections against excessive on-call time took effect on April 1, 2024.\u003c/p\u003e \u003cp\u003eThe April 2024 tightening, combined with the simultaneous rate increase to \u003cspan\u003e$\u003c/span\u003e19.56, produced a clean response. Pay per hour rose to \u003cspan\u003e$\u003c/span\u003e19.34 in the second quarter of 2024 and stayed within a tight band of \u003cspan\u003e$\u003c/span\u003e19.28 to \u003cspan\u003e$\u003c/span\u003e19.44 through the first quarter of 2025\u0026mdash;closely tracking the \u003cspan\u003e$\u003c/span\u003e19.56 floor. With the April 2025 increase to \u003cspan\u003e$\u003c/span\u003e21.44, realised pay rose to \u003cspan\u003e$\u003c/span\u003e20.96 in the second quarter of 2025 and reached \u003cspan\u003e$\u003c/span\u003e21.49 by the fourth quarter, again closely tracking the new floor.\u003c/p\u003e \u003cp\u003eThe Chow test on the pay-per-hour series with break date set to Q1 2024 rejects the no-break null decisively [F(2,12)\u0026thinsp;=\u0026thinsp;93.82, p\u0026thinsp;\u0026lt;\u0026thinsp;0.0001]. The sup-Wald test over the Q4 2023 to Q1 2025 candidate window selects Q1 2024 as the optimal break date, with sup-Wald\u0026thinsp;=\u0026thinsp;93.8, well in excess of Andrews\u0026rsquo; (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1993\u003c/span\u003e) one-percent critical value at the 15-percent trimming fraction. The ITS specification with intervention at Q1 2024 yields a level shift estimate of +\u003cspan\u003e$\u003c/span\u003e11.60 per hour (Newey-West HAC SE\u0026thinsp;=\u0026thinsp;0.90; permutation p\u0026thinsp;=\u0026thinsp;0.0002), with no detectable pre-intervention trend (β_1\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.04, HAC SE\u0026thinsp;=\u0026thinsp;0.12) and a positive post-intervention trend of +\u003cspan\u003e$\u003c/span\u003e0.64 per quarter (HAC SE\u0026thinsp;=\u0026thinsp;0.11) tracking the April 2024 and April 2025 cost-of-living adjustments. The model R\u0026sup2; is 0.989.\u003c/p\u003e \u003cp\u003eThe headline answer to the first question is therefore yes, with the qualification that the phase-in window operated as designed: it allowed a transition period during which the nominal rate could be partially diluted through on-call time, and once that margin was closed in April 2024, the rule has bound tightly through the most recent observation. From the third quarter of 2023 (the last full pre-rule quarter, \u003cspan\u003e$\u003c/span\u003e5.05 per hour) to the fourth quarter of 2025 (\u003cspan\u003e$\u003c/span\u003e21.49 per hour), realised industry-aggregate hourly pay rose by 326 percent, or approximately \u003cspan\u003e$\u003c/span\u003e16 per hour\u0026mdash;consistent with what the agency intended.\u003c/p\u003e"},{"header":"4. The On-Call Collapse","content":"\u003cp\u003eOnce the agency closed the on-call dilution margin in April 2024, the second question becomes pressing. If platforms must pay for on-call time at the same rate as trip time, what do they do? Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e shows the answer.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThrough the seven full pre-rule quarters and into the December 2023 enforcement window, on-call hours per week ran consistently above 750,000 and rose into the 1,000,000 to 1,072,000 range across 2023. After the December 2023 enforcement, on-call hours dropped modestly in the first quarter of 2024 (to 612,000) and then collapsed once the April 2024 tightening took effect: 305,000 in the second quarter of 2024, 240,000 in the third, 193,000 in the fourth. By the end of 2024, weekly on-call hours had fallen 82 percent from their Q3 2023 peak.\u003c/p\u003e \u003cp\u003eThe Chow test with break date Q2 2024 (the prediction implied by the April 2024 tightening) rejects no-break [F(2,12)\u0026thinsp;=\u0026thinsp;16.07, p\u0026thinsp;=\u0026thinsp;0.0004], and the ITS specification with intervention at Q2 2024 yields a level shift of \u0026minus;\u0026thinsp;672.8 thousand weekly on-call hours (HAC SE\u0026thinsp;=\u0026thinsp;140.1; permutation p\u0026thinsp;=\u0026thinsp;0.0006), the largest single discontinuity in any series in the dataset. The model R\u0026sup2; is 0.903. The sup-Wald test over the Q4 2023 to Q1 2025 candidate window, which selects the break date endogenously, selects Q1 2024 (sup-Wald F\u0026thinsp;=\u0026thinsp;33.11) rather than Q2 2024. This is informative rather than contradictory: it means platforms began draining on-call capacity at the December 2023 enforcement onset rather than waiting for the April 2024 tightening, and that the early action was already statistically detectable in the Q1 2024 figure of 612,000 weekly hours, even before the steeper Q2-Q4 declines. The framework\u0026rsquo;s prediction (a break tied to the April 2024 closure) is therefore confirmed in magnitude but the temporal onset is one quarter earlier than the closure date alone would suggest.\u003c/p\u003e \u003cp\u003eTrip hours moved in the opposite direction. From the third quarter of 2023 through the fourth quarter of 2025, trip hours per week rose from 815,000 to 1,119,000\u0026mdash;a 37 percent increase. The platforms did not reduce the volume of paid productive work; they specifically eliminated the unpaid-when-pre-rule, paid-when-post-rule on-call time that the rule\u0026rsquo;s design had targeted.\u003c/p\u003e \u003cp\u003eMechanically, this happened through several documented channels. The agency\u0026rsquo;s 2022 study and subsequent quarterly reports describe these explicitly: tightening of platform access controls (the industry term is \u0026ldquo;gating\u0026rdquo;) during periods of low demand, increased reliance on pre-scheduled shifts that condition future earnings on shift completion, automatic disconnection after periods of inactivity, and prioritization of high-acceptance-rate workers for available trip offers. Several of these practices have been the subject of subsequent litigation; one platform began limiting the number of workers who can be online during each hour of the day in late 2023.\u003c/p\u003e \u003cp\u003eThis pattern\u0026mdash;trip hours flat or rising while on-call hours collapse\u0026mdash;is the cleanest evidence in the data of what regulators sometimes call a margin shift. The rule did not change the underlying demand for delivery (which was rising independently). It did not change the technology of delivery. It did not even meaningfully change the average duration or distance of a trip. What it changed was the balance between paid trip time and what had previously been a large pool of effectively unpaid on-call time. Once that pool became expensive to maintain, the platforms drained it.\u003c/p\u003e \u003cp\u003eThe welfare implications of this margin shift are mixed. From the worker side, on-call time represents both a cost (waiting without earning) and an option value (the ability to make oneself available for trips that may or may not arrive). The pre-rule equilibrium contained a great deal of waiting that produced small per-hour earnings; the post-rule equilibrium contains less waiting but at much higher hourly pay. Workers who can secure scheduled shifts or qualify for high-acceptance-rate priority access benefit substantially. Workers who previously depended on casual, ad-hoc availability to the platforms have lost the ability to log in freely. This distributional question is real and important; the aggregate data used in this paper cannot resolve it directly.\u003c/p\u003e"},{"header":"5. The Tip Decline","content":"\u003cp\u003eThe second substitution margin operated in parallel with the on-call collapse. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e shows the tip-per-delivery time series across the same sixteen quarters.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFrom the first quarter of 2022 through the third quarter of 2023, tips per delivery sat in a tight band between \u003cspan\u003e$\u003c/span\u003e3.85 and \u003cspan\u003e$\u003c/span\u003e3.98. In the fourth quarter of 2023\u0026mdash;the same quarter enforcement of the minimum pay rule began\u0026mdash;tips per delivery fell to \u003cspan\u003e$\u003c/span\u003e3.38, a 12 percent drop. By the first quarter of 2024, the figure had fallen to \u003cspan\u003e$\u003c/span\u003e1.44; by the fourth quarter of 2025, to \u003cspan\u003e$\u003c/span\u003e1.10. Total tips paid out across the industry fell from \u003cspan\u003e$\u003c/span\u003e10.6\u0026nbsp;million per week in the first quarter of 2023 to \u003cspan\u003e$\u003c/span\u003e3.9\u0026nbsp;million per week in the fourth quarter of 2025.\u003c/p\u003e \u003cp\u003eThe Chow test with break date Q4 2023 (the prediction implied by the December 2023 enforcement) rejects no-break [F(2,12)\u0026thinsp;=\u0026thinsp;7.14, p\u0026thinsp;=\u0026thinsp;0.009]. The ITS specification with intervention at Q4 2023 estimates a level shift of \u0026minus;\u003cspan\u003e$\u003c/span\u003e1.71 per delivery (HAC SE\u0026thinsp;=\u0026thinsp;0.39; permutation p\u0026thinsp;=\u0026thinsp;0.008), with no detectable pre-intervention trend (β_1\u0026thinsp;\u0026asymp;\u0026thinsp;0, HAC SE\u0026thinsp;=\u0026thinsp;0.005) and a small additional negative trend of \u0026minus;\u003cspan\u003e$\u003c/span\u003e0.17 per quarter (HAC SE\u0026thinsp;=\u0026thinsp;0.07) in subsequent quarters. The model R\u0026sup2; is 0.917. The sup-Wald test over the Q3 2023 to Q1 2025 candidate window selects Q1 2024 as the optimal break date with a much larger statistic (sup-Wald F\u0026thinsp;=\u0026thinsp;171.94). This pattern\u0026mdash;a detected break at the framework\u0026rsquo;s predicted Q4 2023 date but with the strongest statistical signal at Q1 2024\u0026mdash;is consistent with the regulator\u0026rsquo;s account: the platforms\u0026rsquo; user-interface redesigns began in December 2023 but took a full quarter to fully reach all users and to be reflected in the consumer behaviour underlying the aggregate tipping series.\u003c/p\u003e \u003cp\u003eThe agency\u0026rsquo;s January 2026 report (DCWP, 2026) attributes this decline specifically to user-interface changes implemented by Uber Eats and DoorDash beginning in December 2023, contemporaneous with enforcement. Prior to those changes, both platforms presented the tip prompt during the checkout flow alongside fee and tax displays, with selectable percentage options. After the changes, the prompt moved to a separate post-checkout screen requiring additional navigation. The agency\u0026rsquo;s comparison platforms\u0026mdash;Grubhub, Relay, and the smaller apps\u0026mdash;did not modify their tip flows over the same period and exhibited stable per-delivery tip averages of approximately \u003cspan\u003e$\u003c/span\u003e2.17 per delivery, a figure substantially higher than the \u003cspan\u003e$\u003c/span\u003e0.76 average the agency reports for Uber Eats and DoorDash specifically. The agency\u0026rsquo;s estimate of cumulative worker harm from these design choices is \u003cspan\u003e$\u003c/span\u003e554\u0026nbsp;million over the period from December 2023 through mid-2025.\u003c/p\u003e \u003cp\u003eThis paper does not independently verify the agency\u0026rsquo;s attribution of the tip decline to specific platform interface choices. The DCWP\u0026rsquo;s analysis relies on per-platform data not available in the public quarterly tables; this paper presents the attribution as the regulator\u0026rsquo;s own conclusion. Three alternative explanations for an industry-wide tip decline are worth considering and ruling out. First, post-pandemic normalisation of consumer tipping behaviour: tips per delivery were stable in the band of \u003cspan\u003e$\u003c/span\u003e3.85 to \u003cspan\u003e$\u003c/span\u003e3.98 from Q1 2022 through Q3 2023 and showed no downward trend; the decline begins sharply in Q4 2023 contemporaneous with rule enforcement, not gradually as a normalisation story would predict, and the ITS pre-intervention trend coefficient is indistinguishable from zero. Second, inflation crowding out tip budget: order subtotals rose only modestly over the same period (from \u003cspan\u003e$\u003c/span\u003e27.11 in Q1 2023 to \u003cspan\u003e$\u003c/span\u003e29.04 in Q4 2024), and consumer fees per delivery rose substantially (from \u003cspan\u003e$\u003c/span\u003e4.96 to \u003cspan\u003e$\u003c/span\u003e5.54), but the proportional decline in tips relative to either subtotals or total order cost is much larger than what budget crowding would imply. Third, demand composition shifts: if the order mix shifted toward lower-tip categories or geographies, aggregate tips per delivery could fall without any platform behaviour change. The agency\u0026rsquo;s within-industry comparison rules this out: platforms that did not modify their tip flows showed stable tipping over the same period, while platforms that did modify them showed the documented collapse. The aggregate timing evidence cannot exclude all alternatives but is difficult to reconcile with anything other than the platform-design attribution.\u003c/p\u003e \u003cp\u003eThe arithmetic of the substitution is informative. In the first quarter of 2023, the pre-rule baseline, the typical hour of work generated roughly \u003cspan\u003e$\u003c/span\u003e5.89 in pay and \u003cspan\u003e$\u003c/span\u003e5.83 in tips, for total earnings of \u003cspan\u003e$\u003c/span\u003e11.72 per hour. In the fourth quarter of 2025, the typical hour generated \u003cspan\u003e$\u003c/span\u003e21.49 in pay and \u003cspan\u003e$\u003c/span\u003e2.85 in tips, for total earnings of \u003cspan\u003e$\u003c/span\u003e24.35 per hour. Workers gained approximately \u003cspan\u003e$\u003c/span\u003e15.60 per hour in pay and lost approximately \u003cspan\u003e$\u003c/span\u003e2.98 per hour in tips. The gain dominates the loss by a factor of more than five.\u003c/p\u003e \u003cp\u003eThis means that even after the documented tip suppression, workers are substantially better off in the post-rule equilibrium than they were in the pre-rule one. The substitution did not undo the rule\u0026rsquo;s welfare gain; it captured a significant share of what would otherwise have been a larger gain. If the pre-rule tip-per-delivery rate of approximately \u003cspan\u003e$\u003c/span\u003e3.85 had been preserved through the post-rule period, fourth-quarter 2025 tips per hour at the observed productivity of 2.54 deliveries per hour would have been approximately \u003cspan\u003e$\u003c/span\u003e9.78\u0026mdash;rather than the observed \u003cspan\u003e$\u003c/span\u003e2.85. The redistribution from worker tips to platform pricing discretion is on the order of \u003cspan\u003e$\u003c/span\u003e7 per hour worked, or roughly 30 percent of the total compensation gain that the rule was intended to deliver.\u003c/p\u003e \u003cp\u003eA January 2026 amendment to the Delivery Worker Laws now requires both restaurant and grocery delivery apps to present a checkout-stage tip prompt with a selectable 10 percent option and a custom-amount field, taking effect on January 26, 2026. The agency estimates that, if the platforms apply the same tip-flow design they use in markets without these restrictions, worker tip earnings will rise by approximately \u003cspan\u003e$\u003c/span\u003e390\u0026nbsp;million per year. Litigation by Uber Eats and DoorDash to enjoin the new requirement was denied by federal judges in late January 2026; enforcement is in its initial weeks at the time of this writing. The first quarterly tables covering periods after January 26, 2026 are expected in mid-to-late 2026 and will provide an unusually clean test of the platform-design attribution. If tips per delivery at Uber Eats and DoorDash recover toward the levels observed at the comparison platforms (approximately \u003cspan\u003e$\u003c/span\u003e2.17), the regulator\u0026rsquo;s attribution to user-interface design choices is supported. If tips at the two platforms remain depressed despite the restored prompts, the alternative explanations\u0026mdash;that consumer behaviour or market conditions independently changed\u0026mdash;become more credible.\u003c/p\u003e"},{"header":"6. Workforce Contraction and Productivity Rise","content":"\u003cp\u003eThe third substitution margin is the size and composition of the platform-engaged workforce itself. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e presents three panels showing workers, productivity, and total industry output over the sixteen quarters in the data.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe active workforce, defined as the number of workers logging at least one on-call or trip period in a given week, peaked at 111,000 in the fourth quarter of 2023 and fell to 75,000 by the fourth quarter of 2025\u0026mdash;a 32 percent contraction. Workers performing at least one trip in a week fell from 81,000 to 70,000 over the same period, a 14 percent contraction. The gap between these two figures (the workers who logged on-call time without taking a trip) compressed from 30,000 to 5,000, consistent with the on-call elimination documented in Section \u003cspan refid=\"Sec8\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eThe Chow test on active worker counts with break date Q1 2024 rejects no-break [F(2,12)\u0026thinsp;=\u0026thinsp;73.33, p\u0026thinsp;\u0026lt;\u0026thinsp;0.0001]. The sup-Wald procedure over the Q4 2023 to Q1 2025 candidate window selects Q2 2024 as the optimal break (sup-Wald F\u0026thinsp;=\u0026thinsp;77.38), marginally preferred over Q1 2024, consistent with the framework\u0026rsquo;s prediction that workforce contraction would lag the December 2023 enforcement by one to two quarters as the gating policies took effect. The ITS specification with intervention at Q1 2024 estimates a level shift of \u0026minus;\u0026thinsp;6.12 thousand active workers (HAC SE\u0026thinsp;=\u0026thinsp;1.61; permutation p\u0026thinsp;=\u0026thinsp;0.0002) plus a continuing negative trend of \u0026minus;\u0026thinsp;6.60 thousand per quarter (HAC SE\u0026thinsp;=\u0026thinsp;0.53) through 2025. The model R\u0026sup2; is 0.967. The combination of a level shift and a negative post-intervention trend is a useful signature of the framework: workforce contraction is not a one-time adjustment but a continuing process as platforms progressively tighten access controls.\u003c/p\u003e \u003cp\u003eProductivity rose simultaneously. Deliveries per worker-hour\u0026mdash;the agency\u0026rsquo;s primary productivity measure\u0026mdash;rose from 1.41 in the third quarter of 2023 to 2.54 in the fourth quarter of 2025, an 80 percent increase. Average hours per worker rose from 17.3 to 18.1 over the same period. Both figures represent more intensive use of those workers who remained active on the platforms.\u003c/p\u003e \u003cp\u003eTotal industry output\u0026mdash;weekly deliveries\u0026mdash;grew from 2.65\u0026nbsp;million in the first quarter of 2023 to 3.47\u0026nbsp;million in the fourth quarter of 2025, a 31 percent increase. The pre-rule predictions made by the platforms during the rulemaking process and during litigation\u0026mdash;that delivery volumes would collapse, that the industry would withdraw service from the city\u0026mdash;did not materialize. Consumer demand grew steadily over the four-year window, undisturbed by either the rule or by the per-delivery consumer fee increases (from \u003cspan\u003e$\u003c/span\u003e4.74 in the first quarter of 2023 to \u003cspan\u003e$\u003c/span\u003e7.58 in the fourth quarter of 2025) that the rule produced as platforms passed costs through.\u003c/p\u003e \u003cp\u003eWhat this combination means is that the same volume of delivery work is being done by fewer workers, each working slightly longer hours and substantially more productively. Whether this is welfare-improving depends on a counterfactual that the data cannot resolve directly: how many of the workers represented by the contraction in active accounts would have preferred to remain at the higher hourly pay if access had not been restricted, versus how many were casual or marginal participants who chose to exit when the platform-side incentives shifted against intermittent low-volume engagement. The aggregate data are silent on this. What can be said is that aggregate worker compensation\u0026mdash;total earnings paid out across all workers, in dollars per week\u0026mdash;rose from \u003cspan\u003e$\u003c/span\u003e19.5\u0026nbsp;million in the third quarter of 2023 to \u003cspan\u003e$\u003c/span\u003e33.3\u0026nbsp;million in the fourth quarter of 2025, a 71 percent increase. Whatever the distributional consequences of the workforce contraction, the total flow of dollars from platforms to workers grew substantially in absolute terms.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e6.1 Sensitivity to Multi-Apping\u003c/h2\u003e \u003cp\u003eThe 32 percent account-level contraction reported above conflates two distinct phenomena: an actual reduction in the number of unique individuals working in the sector, and a reduction in the practice of single individuals maintaining multiple platform accounts. The agency\u0026rsquo;s 2022 study (DCWP, 2022) provides the only available anchor for separating these. In the fourth quarter of 2021, 219,787 worker accounts at the four largest apps were held by approximately 122,104 unique individuals, implying a multi-apping multiplier of roughly 1.80. The same study estimated that 56.3 percent of workers held accounts on more than one platform.\u003c/p\u003e \u003cp\u003eThe DCWP itself notes in its quarterly tables that a reduction in multi-apping is an expected consequence of the rule, since the requirement that platforms pay for on-call time reduces the value to workers of keeping multiple apps logged in concurrently. This paper\u0026rsquo;s data cannot directly observe the post-rule multi-apping rate, but the bounds on the implied change in unique individuals can be computed under alternative assumptions.\u003c/p\u003e \u003cp\u003eUnder the assumption that the multi-apping rate was unchanged from its 2022 baseline (multiplier 1.80), the implied unique-individual workforce fell from approximately 61,700 in the fourth quarter of 2023 to approximately 41,700 in the fourth quarter of 2025\u0026mdash;a 32 percent decline that exactly matches the account-level contraction. Under the opposite assumption that multi-apping was fully eliminated by the end of 2025 (multiplier 1.0), the implied unique-individual workforce would have risen from approximately 61,700 to 75,000, a 22 percent increase. A plausible mid-range assumption\u0026mdash;that multi-apping fell from 56 percent to 30 percent of workers, with the multiplier moving correspondingly from 1.80 to approximately 1.40\u0026mdash;implies a unique-individual contraction of roughly 13 percent.\u003c/p\u003e \u003cp\u003eThe bounds therefore range from a 22 percent increase to a 32 percent decrease in unique individuals working in the sector, with intermediate values most plausible. This is a wide range, and it bears emphasising that the headline 32 percent contraction reported in the narrative above is an upper bound on the unique-individual decline. A reader who interprets the contraction as evidence that one in three workers was displaced by the rule would be overstating what the data can support; the most that can be said with confidence is that platforms tightened access controls and that the share of those access changes representing displacement of unique individuals versus reduction of multi-apping cannot be separated from public data alone. Resolving this question would require either per-platform data with cross-platform identifiers or a follow-up DCWP survey of post-rule workers.\u003c/p\u003e \u003c/div\u003e"},{"header":"7. Discussion: A Framework of Regulatory Substitution Across Compensation Channels","content":"\u003cp\u003eThe patterns documented in Sections \u003cspan refid=\"Sec7\" class=\"InternalRef\"\u003e3\u003c/span\u003e through \u003cspan refid=\"Sec10\" class=\"InternalRef\"\u003e6\u003c/span\u003e share a common structure. Platform compensation in app-based work is not a single quantity but a vector across multiple channels: hourly pay, tip facilitation, on-call exposure, and platform access, each of which the platform controls and any of which can be adjusted to change a worker\u0026rsquo;s realised earnings. A wage regulation that addresses one channel of this vector\u0026mdash;the per-hour rate in the case studied here\u0026mdash;leaves the others as available margins for platform response. The empirical contribution of this paper is to test that platforms used all three of those margins in the New York case and that the temporal sequence of substitutions matches the framework\u0026rsquo;s prediction. The conceptual contribution is to propose this as a general framework: regulatory substitution across compensation channels predicts that single-channel wage rules in multi-channel compensation environments will produce systematic, predictable, and sequenced substitution responses keyed to which channels the regulator has and has not addressed at each stage of the rule\u0026rsquo;s implementation.\u003c/p\u003e \u003cp\u003eEach substitution arrived in a specific temporal sequence keyed to the regulatory architecture. The break tests support this directly: the sup-Wald procedures select Q1 2024 as the optimal break for both the on-call and the tip series, but the framework\u0026rsquo;s a priori prediction\u0026mdash;Q2 2024 for on-call (the April 2024 closure of the dilution margin) and Q4 2023 for tips (the December 2023 enforcement)\u0026mdash;is also rejected at conventional significance levels by the Chow tests. What this means substantively is that platforms began reacting to the rule at the moment of enforcement onset (December 2023), not at the moment a particular margin was formally closed; the formal closures sharpened breaks that were already underway. The on-call-time drainage and the tip-flow redesign were both deployed as Q1 2024 strategies, with the April 2024 tightening intensifying the on-call response that was already in motion.\u003c/p\u003e \u003cp\u003eEach substitution is, viewed individually, a rational platform response to the relevant regulatory constraint and the costs the constraint imposes. Each is also a response that reduces the welfare gain the rule was intended to produce, in different ways and for different sub-populations of workers. Taken together, they suggest that a regulatory architecture for platform work that targets only the per-hour wage rate is incomplete. The platforms control multiple compensation channels: hourly pay, tip facilitation, on-call exposure, and access to the platform itself. A wage rule that addresses only the first of these gives the platforms a clear strategic playbook for reducing its bite by adjusting the others.\u003c/p\u003e \u003cp\u003eThe implication for regulatory design is not that wage rules are ineffective\u0026mdash;the evidence here is that the New York rule clearly worked at its primary task of raising hourly pay. The implication is rather that single-margin wage rules in multi-margin compensation environments will tend to produce predictable substitution responses, and that a complete regulatory architecture should anticipate and address the parallel margins from the start rather than chase each substitution serially after the fact.\u003c/p\u003e \u003cp\u003eThis sequenced-response pattern also suggests a research design opportunity. The April 2026 cost-of-living adjustment to \u003cspan\u003e$\u003c/span\u003e22.13 per hour will be the first scheduled rate change to occur after the operational substitution patterns have stabilised: on-call hours have been flat at approximately 250,000 per week for five consecutive quarters, and the tip requirement that took effect January 26, 2026 will, by Q2 2026, have produced its initial observable response in the data. Whether the April 2026 adjustment produces a clean pass-through into realised pay\u0026mdash;or whether platforms identify yet another substitution margin\u0026mdash;is an open empirical question. The next public quarterly tables, expected in mid-to-late 2026, will provide the first answer.\u003c/p\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e7.1 Comparison to the Ride-Hail Setting\u003c/h2\u003e \u003cp\u003eA natural comparison is with the Taxi and Limousine Commission\u0026rsquo;s parallel minimum earnings standard for app-based for-hire vehicle drivers, which has been in place since February 2019 and was substantially restructured in August 2025. The high-level contrast is informative on its own. The ride-hail rule operates as a per-trip floor, and recent evidence on the August 2025 restructuring shows that platforms responded by moving away from per-trip floor-anchoring entirely\u0026mdash;the pre-rule spike at the floor reorganised into a more dispersed distribution. The food delivery rule analysed here operates as a per-hour floor, and platforms have not moved away from binding at it; realised hourly pay tracks the floor closely. What platforms have done instead is substitute along three other compensation channels, none of which the per-hour rule directly regulates. The two regimes thus exhibit different forms of platform response to wage regulation, both consistent with rational platform behaviour but operating at different levels of the compensation system.\u003c/p\u003e \u003cp\u003eThe two regimes share broad architectural features\u0026mdash;both are minimums applied to a combination of engaged and on-call time\u0026mdash;but they differ in two important respects. First, the ride-hail rule applies to a market with two dominant platforms (Uber and Lyft) and a relatively homogeneous service product, while the food delivery rule covers six platforms with substantially differentiated business models. Second, the ride-hail Commission publishes trip-level administrative data covering every dispatched ride, while the Department of Consumer and Worker Protection publishes only industry-aggregate quarterly tables.\u003c/p\u003e \u003cp\u003eThe contribution of the present paper is conceptually distinct from the ride-hail evidence. Per-trip distributional restructuring documents how concentrated platform pay decisions are around the regulatory floor, and how that concentration changes when the floor moves. The present paper documents substitution behaviour across compensation channels: how platforms reallocate worker compensation between hourly pay, on-call exposure, tip facilitation, and platform access in response to a per-hour rule. The two phenomena are adjacent but operate at different levels of platform decision-making and require different data to detect.\u003c/p\u003e \u003cp\u003eWhat unifies the two settings is the broader observation that platform compensation systems are multi-dimensional, that wage regulations targeting any single dimension will tend to produce predictable adjustments in the others, and that a regulator equipped only with the targeted dimension\u0026rsquo;s data cannot fully observe the system response. The ride-hail Commission\u0026rsquo;s decision to publish trip-level data has been instrumental in enabling outside researchers to study per-trip distributional questions; the DCWP\u0026rsquo;s choice to publish only industry-aggregate tables substantially constrains what outside analysis can establish about platform-specific or worker-specific outcomes in the food delivery setting. The contrast in available evidence across the two adjacent regulatory regimes is itself informative about the value of data-publication policy as a complement to substantive rulemaking.\u003c/p\u003e \u003c/div\u003e"},{"header":"8. Conclusion","content":"\u003cp\u003eThree years of public industry-aggregate data on New York City\u0026rsquo;s app-based food delivery sector show that the December 2023 minimum pay rule worked at its primary task. Realised hourly pay rose from \u003cspan\u003e$\u003c/span\u003e5.05 in the last full pre-rule quarter to \u003cspan\u003e$\u003c/span\u003e21.49 by the end of 2025, and now closely tracks the statutory floor. Total weekly industry compensation paid to workers rose from \u003cspan\u003e$\u003c/span\u003e19.5\u0026nbsp;million to \u003cspan\u003e$\u003c/span\u003e33.3\u0026nbsp;million over the same period. Structural break tests detect highly significant breaks at the rule\u0026rsquo;s implementation dates in all four outcome series the regulatory-substitution framework predicts: realised pay (Chow F\u0026thinsp;=\u0026thinsp;93.8 at Q1 2024), on-call hours (Chow F\u0026thinsp;=\u0026thinsp;16.1 at Q2 2024; sup-Wald F\u0026thinsp;=\u0026thinsp;33.1 at Q1 2024), tips per delivery (Chow F\u0026thinsp;=\u0026thinsp;7.1 at Q4 2023; sup-Wald F\u0026thinsp;=\u0026thinsp;171.9 at Q1 2024), and active worker counts (Chow F\u0026thinsp;=\u0026thinsp;73.3 at Q1 2024). All level-shift estimates are statistically significant under both Newey-West HAC standard errors and 5,000-permutation inference.\u003c/p\u003e \u003cp\u003eThe data also show that platforms responded along three substitution margins: on-call hours collapsed by 82 percent (estimated level shift of \u0026minus;\u0026thinsp;673 thousand weekly hours), tips per delivery fell 71 percent (estimated level shift of \u0026minus;\u003cspan\u003e$\u003c/span\u003e1.71 per delivery) following user-interface redesigns at two of the four major platforms, and the active workforce contracted by 32 percent (estimated level shift of \u0026minus;\u0026thinsp;6.1 thousand workers, with a continuing post-intervention trend of \u0026minus;\u0026thinsp;6.6 thousand per quarter) while productivity rose 80 percent and total deliveries grew 31 percent. Each substitution arrived in a specific sequence keyed to which margin the regulator had left open at each stage of the rule\u0026rsquo;s implementation. None reverses the rule\u0026rsquo;s aggregate welfare gain, but each captures a substantial share of what the gain would otherwise have been.\u003c/p\u003e \u003cp\u003eThe substantive lesson is that a regulatory architecture for platform work targeting only the per-hour wage rate is incomplete in multi-margin compensation environments. The platforms control hourly pay, tip facilitation, on-call exposure, and platform access, and a rule that addresses only the first of these gives the platforms a clear and predictable strategic response. New York City\u0026rsquo;s subsequent regulatory amendments, including the January 2026 tip-protection rule and ongoing enforcement against access restrictions, represent a serial closing of these substitution margins. Whether the April 2026 cost-of-living adjustment produces clean pass-through to realised pay, or whether platforms identify a further substitution margin, will be observable in the public data within approximately six months.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding.\u0026nbsp;\u003c/strong\u003eNo funds, grants, or other support were received for the preparation of this manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests.\u0026nbsp;\u003c/strong\u003eThe author declares no competing interests. The author is not affiliated with any of the platforms studied, has not received funding from them, and has no employment, contractual, or advisory relationships with the regulator (DCWP) or with any worker organisation involved in the rulemaking discussed in this paper.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability.\u0026nbsp;\u003c/strong\u003eAll data used in this paper are publicly available from the Department of Consumer and Worker Protection at https://www.nyc.gov/site/dca/workers/Delivery-Worker-Public-Hearing-Minimum-Pay-Rate.page. The XLSX file used for the analysis was downloaded in April 2026, representing data through Q4 2025. Analysis scripts including the Chow, sup-Wald, ITS, and permutation procedures, the cleaned analytical panel, and figure-generation code will be made available in a permanent public repository upon acceptance. No restrictions apply to either the data or the replication materials.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions.\u0026nbsp;\u003c/strong\u003eSole authored. The author conceived the study, performed all analyses, and wrote the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eUse of Artificial Intelligence.\u0026nbsp;\u003c/strong\u003eDisclosed in Section 2.4 (Statistical Approach) per Springer policy, which requires disclosure of LLM use in the Methods section or a suitable alternative.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAndrews, D. W. K. (1993). Tests for parameter instability and structural change with unknown change point. Econometrica, 61(4), 821\u0026ndash;856. https://doi.org/10.2307/2951764\u003c/li\u003e\n\u003cli\u003eBai, J., \u0026amp; Perron, P. (1998). Estimating and testing linear models with multiple structural changes. Econometrica, 66(1), 47\u0026ndash;78. https://doi.org/10.2307/2998540\u003c/li\u003e\n\u003cli\u003eBernal, J. L., Cummins, S., \u0026amp; Gasparrini, A. (2017). Interrupted time series regression for the evaluation of public health interventions: A tutorial. International Journal of Epidemiology, 46(1), 348\u0026ndash;355. https://doi.org/10.1093/ije/dyw098\u003c/li\u003e\n\u003cli\u003eCard, D., \u0026amp; Krueger, A. B. (1994). Minimum wages and employment: A case study of the fast-food industry in New Jersey and Pennsylvania. American Economic Review, 84(4), 772\u0026ndash;793.\u003c/li\u003e\n\u003cli\u003eCengiz, D., Dube, A., Lindner, A., \u0026amp; Zipperer, B. (2019). The effect of minimum wages on low-wage jobs. Quarterly Journal of Economics, 134(3), 1405\u0026ndash;1454. https://doi.org/10.1093/qje/qjz014\u003c/li\u003e\n\u003cli\u003eChow, G. C. (1960). Tests of equality between sets of coefficients in two linear regressions. Econometrica, 28(3), 591\u0026ndash;605. https://doi.org/10.2307/1910133\u003c/li\u003e\n\u003cli\u003eCook, C., Diamond, R., Hall, J. V., List, J. A., \u0026amp; Oyer, P. (2021). The gender earnings gap in the gig economy: Evidence from over a million rideshare drivers. Review of Economic Studies, 88(5), 2210\u0026ndash;2238. https://doi.org/10.1093/restud/rdaa081\u003c/li\u003e\n\u003cli\u003eDepartment of Consumer and Worker Protection, City of New York. (2022). A minimum pay rate for app-based restaurant delivery workers in NYC. November 2022. https://www.nyc.gov/site/dca/workers/Delivery-Worker-Public-Hearing-Minimum-Pay-Rate.page\u003c/li\u003e\n\u003cli\u003eDepartment of Consumer and Worker Protection, City of New York. (2024a). Restaurant delivery app data: January\u0026ndash;March 2024. July 2024.\u003c/li\u003e\n\u003cli\u003eDepartment of Consumer and Worker Protection, City of New York. (2024b). Restaurant delivery app data: April\u0026ndash;June 2024.\u003c/li\u003e\n\u003cli\u003eDepartment of Consumer and Worker Protection, City of New York. (2024c). Restaurant delivery app data: July\u0026ndash;September 2024.\u003c/li\u003e\n\u003cli\u003eDepartment of Consumer and Worker Protection, City of New York. (2024d). Restaurant delivery app data: October\u0026ndash;December 2024.\u003c/li\u003e\n\u003cli\u003eDepartment of Consumer and Worker Protection, City of New York. (2025). Restaurant delivery app data: Quarterly aggregated tables. https://www.nyc.gov/site/dca/workers/Delivery-Worker-Public-Hearing-Minimum-Pay-Rate.page\u003c/li\u003e\n\u003cli\u003eDepartment of Consumer and Worker Protection, City of New York. (2026). Uber Eats and DoorDash engineered a $554 million pay cut: NYC is fighting back. January 2026.\u003c/li\u003e\n\u003cli\u003eDube, A., Lester, T. W., \u0026amp; Reich, M. (2010). Minimum wage effects across state borders: Estimates using contiguous counties. Review of Economics and Statistics, 92(4), 945\u0026ndash;964. https://doi.org/10.1162/REST_a_00039\u003c/li\u003e\n\u003cli\u003eGoldin, J., \u0026amp; Reck, D. (2020). Optimal defaults with normative ambiguity. Review of Economics and Statistics, 104(1), 17\u0026ndash;33. https://doi.org/10.1162/rest_a_00946\u003c/li\u003e\n\u003cli\u003eHall, J. V., \u0026amp; Krueger, A. B. (2018). An analysis of the labor market for Uber\u0026rsquo;s driver-partners in the United States. ILR Review, 71(3), 705\u0026ndash;732. https://doi.org/10.1177/0019793917717222\u003c/li\u003e\n\u003cli\u003eManning, A. (2003). Monopsony in motion: Imperfect competition in labor markets. Princeton University Press.\u003c/li\u003e\n\u003cli\u003eManning, A. (2021). Monopsony in labor markets: A review. ILR Review, 74(1), 3\u0026ndash;26. https://doi.org/10.1177/0019793920922499\u003c/li\u003e\n\u003cli\u003eNewey, W. K., \u0026amp; West, K. D. (1994). Automatic lag selection in covariance matrix estimation. Review of Economic Studies, 61(4), 631\u0026ndash;653. https://doi.org/10.2307/2297912\u003c/li\u003e\n\u003cli\u003eParrott, J. A., \u0026amp; Reich, M. (2018). An earnings standard for New York City\u0026rsquo;s app-based drivers: Economic analysis and policy assessment. Center for New York City Affairs, New School. July 2018.\u003c/li\u003e\n\u003cli\u003eThaler, R. H., \u0026amp; Sunstein, C. R. (2008). Nudge: Improving decisions about health, wealth, and happiness. Yale University Press.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"journal-of-regulatory-economics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"rege","sideBox":"Learn more about [Journal of Regulatory Economics](http://link.springer.com/journal/11149)","snPcode":"11149","submissionUrl":"https://submission.nature.com/new-submission/11149/3","title":"Journal of Regulatory Economics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"regulatory substitution, platform work, minimum pay rules, gig economy, structural break tests, tip suppression","lastPublishedDoi":"10.21203/rs.3.rs-9578165/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9578165/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eNew York City’s minimum pay rule for app-based restaurant delivery workers, effective December 2023, raised industry-aggregate hourly pay from $5.05 in the last full pre-rule quarter to $21.49 by the end of 2025. Using the Department of Consumer and Worker Protection’s public quarterly tables across sixteen quarters, this paper tests a framework of regulatory substitution across compensation channels. Chow tests, supremum-Wald structural break tests, and segmented-regression interrupted time series with Newey-West standard errors and permutation-based inference all detect highly significant breaks at the rule’s implementation dates. The realised pay series shows a level shift of $11.6 per hour at Q1 2024 (Chow F = 93.8, permutation p = 0.0002). On-call hours show a level shift of − 673 thousand per week at Q2 2024 (Chow F = 16.1, permutation p = 0.0006); the sup-Wald test selects Q1 2024 (F = 33.1), indicating platforms began draining on-call capacity at enforcement onset. Tips per delivery show a level shift of −$1.71 at Q4 2023 (Chow F = 7.1) intensifying to sup-Wald F = 171.9 at Q1 2024 once user-interface redesigns at Uber Eats and DoorDash took full effect. Active workers contracted by 32 percent (Chow F = 73.3, level shift − 6.1 thousand). Each substitution arrived in a sequence keyed to which channel the regulator left open. The paper contributes a regulatory-substitution framework: when a wage rule targets one channel of platform compensation, predictable substitution along the unregulated channels follows in a sequence determined by the regulatory architecture.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eJEL classification: \u003c/strong\u003eL51 (Economics of Regulation); J38 (Public Policy: Wages); J42 (Monopsony; Segmented Labor Markets); K23 (Regulated Industries and Administrative Law)\u003c/p\u003e","manuscriptTitle":"Platform Substitution Under New York City’s Food Delivery Pay Rule","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-12 14:15:13","doi":"10.21203/rs.3.rs-9578165/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2026-05-04T19:44:22+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-05-04T02:29:00+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-05-04T02:28:00+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of Regulatory Economics","date":"2026-04-30T13:57:45+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"journal-of-regulatory-economics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"rege","sideBox":"Learn more about [Journal of Regulatory Economics](http://link.springer.com/journal/11149)","snPcode":"11149","submissionUrl":"https://submission.nature.com/new-submission/11149/3","title":"Journal of Regulatory Economics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"7cabaf0b-aa76-4100-9fb7-38099287aa04","owner":[],"postedDate":"May 12th, 2026","published":true,"recentEditorialEvents":[{"type":"reviewersInvited","content":"8","date":"2026-05-04T19:44:22+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-05-04T02:29:00+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-05-04T02:28:00+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of Regulatory Economics","date":"2026-04-30T13:57:45+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-12T14:15:14+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-12 14:15:13","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9578165","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9578165","identity":"rs-9578165","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}