{"id":"b9889c64-a83e-44f3-86bb-be8c40933174","arxiv_id":"2507.12693","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Even when strategic manipulation makes the running variable jump at the cutoff, the treatment effect can be recovered by subtracting the placebo outcome's discontinuity, weighted by a local instrumental variable estimate, from the usual RDD estimate.","lead":"This paper develops a regression discontinuity method that stays valid when people manipulate the running variable around a known cutoff. It uses a placebo outcome and a placebo treatment to compute an adjustment term, so a standard falsification test becomes a correction instead of a dead end.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The causal estimand and Theorem 1 condition on D=d*, yet the assumptions permit a discontinuous D density, making P(w|D=d*) and E[·|D=d*] undefined; different one-sided conventions give different values.","rationale":"The reader's weakest_assumption focused on Assumption 6 (completeness) and the partial-linearity restriction in equation (1), but the reader's rationale separately flagged the estimand ambiguity caused by conditioning on D=d* when the density jumps. I find that this ambiguity is the most load-bearing concern: it threatens the well-definedness of the causal parameter itself, and hence of Theorem 1, before any completeness or functional-form issue is reached. The paper repeatedly uses objects like E[·|D=d*] and dP(w|D=d*) while allowing a discontinuity in the density of D; no definition, convention, or limiting argument is supplied. This is not merely a matter of notation, because the left and right limits of the conditional distribution of W (and hence of U) can differ, and the integrals in Theorem 1 will inherit that difference. The paper could repair this by defining tau0 and dP(w|D=d*) as the appropriate one-sided limit (e.g., right limit), or by assuming an atom at d*, but as written the identification result is ambiguous. The concrete test would demonstrate the ambiguity in a simple DGP. Given that the reader's conditional verdict already requires resolving this estimand ambiguity, I do not propose a change in verdict, but I elevate it to the central concern rather than the completeness or partial-linearity assumptions.","tokens_in":65917,"tokens_out":6189,"duration_ms":69101,"concrete_test":"Build a DGP satisfying Assumptions 1-6 with a discontinuous D density but no atom: e.g., f_D(d)=1 for d<0, f_D(d)=2 for d>0, U ~ Bernoulli(0.5), W = U + noise, Z binary with P(Z=1|U) = 0.2+0.6U, Y = tau*1(D>0) + U + noise, and sharp treatment A=1(D>0). Compute the right-hand side of Theorem 1 under three conventions for dP(w|D=0): the right limit (dP(w|D=0+)), the left limit (dP(w|D=0-)), and the midpoint. If the three expressions differ, or if the estimand E[Y(1,0,U)-Y(0,0,U)|D=0] cannot be assigned a value because {D=0} is null, the central identification claim is not well-defined as stated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3.1 defines the target as tau0 = E[Y(1,D,U,eta_y) - Y(0,D,U,eta_y) | D=d*]. The model explicitly allows the conditional density f(u,eta|d) to be discontinuous at d* (Section 3.1), and the running variable D is assumed to have a density that is positive but not necessarily continuous at d*. When the density jumps, {D=d*} is a null event, so the conditional distribution of (U,eta) given D=d* is not uniquely defined without specifying a version. Theorem 1 then writes the identified quantity using integrals with respect to P(w|D=d*), and Lemma 2 uses E[·|D=d*]. If the density is discontinuous, P(w|D=d*) can be taken as the right limit, left limit, or any convex combination, and these generally give different values for the RHS of Theorem 1. The paper does not define a convention, so the identification claim is not for a well-defined estimand. This is more basic than Assumption 6 or the partial-linearity condition: even if those hold, the formula in Theorem 1 may identify different objects depending on an arbitrary choice, or be undefined.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a placebo discontinuity design (PDD) for regression discontinuity settings in which the standard continuity of potential outcomes fails because agents manipulate the running variable. The structural model includes an unobserved confounder U, a placebo outcome W, and a placebo treatment Z. Under conditional independence, exclusion, a confounding bridge, and a completeness condition, Theorem 1 expresses the RDD effect tau0 as the limiting difference of bridge integrals divided by the treatment-probability discontinuity. A local linear instrumental-variable estimator is proposed; Proposition 1 shows it equals the standard RDD estimator minus a placebo-discontinuity adjustment term, and Sections 5 and 6 establish consistency and bias-corrected asymptotic normality under a partial-linearity restriction on the bridge. Appendices B and C provide relaxations of the placebo-exogeneity and treatment-assignment assumptions.","tokens_in":66164,"tokens_out":6311,"duration_ms":81946,"significance":"If the identification chain is sound, the PDD is a meaningful extension of the RDD toolkit: it converts placebo tests from diagnostics into corrections, offers point identification in a setting where Gerard et al. (2020) only obtain bounds, and the estimator decomposition is transparent and practically interpretable. The paper is also careful in other respects: the identification lemmas are detailed, the partial-linearity limitation is acknowledged in Section 5.1, and the bias-corrected inference is worked out at the usual RDD rate. The principal caveat, developed below, is that the central formula conditions on D=d* and uses P(w|D=d*) even though the model explicitly permits a discontinuous running-variable density; without a precise definition of this conditional object, the identified quantity is not well-defined.","major_comments":[{"comment":"The estimand tau0 is defined as E[Y(1,D,U,eta_y)-Y(0,D,U,eta_y)|D=d*], and Lemma 2 and Theorem 1 repeatedly use E[.|D=d*] and dP(w|D=d*). The model explicitly allows f(u,eta|d) to be discontinuous at d*, and D is assumed to have a density that is positive but not necessarily continuous. Under a continuous density, {D=d*} is already a null event and the conditional distribution requires a specified version; with a discontinuity, the one-sided limiting conditional distributions generally differ, so the right-hand side of Theorem 1 can take different values depending on whether P(w|D=d*) is read as the right limit, the left limit, or a mixture. The paper does not state a convention. This is not a presentation issue: without defining the estimand through one-sided limits or imposing continuity of the conditioning distribution, the claimed identification is for an object that is not well-defined. I recommend defining tau0 and all bridge integrals in Theorem 1 using explicit one-sided conditional distributions, for example the right-continuous version, and adjusting Lemma 2 and the proof of Theorem 1 accordingly.","section":"Sections 3.1-3.3, esp. Lemma 2 and Theorem 1"},{"comment":"Proposition 2 characterizes the probability limit of the estimator as tau_y_rdd - (tau_w_rdd)^T gamma_-, where gamma_- is a best local linear approximation. Corollary 1 then asserts consistency for tau0 under equation (1), exact partial linearity of the bridge. The paper does flag this limitation, but the abstract and introduction state consistency without the qualifier, and Section 4.1 introduces the partial-linear class as an approximation rather than a substantive restriction. I recommend that the paper state prominently in the abstract and introduction that the estimator is consistent for tau0 only when the confounding bridge is exactly partially linear in W; otherwise it targets a different, best-linear-approximation object.","section":"Proposition 2 and Corollary 1"}],"minor_comments":[{"comment":"The notes to Figure 2 say 'Figure 1 illustrates the relaxed assumption'; this should refer to Figure 2.","section":"Figure 2 caption"},{"comment":"The alternative estimand formula for \\tilde tau_pdd contains a bare W in the adjustment terms; please clarify whether W denotes the random vector, a sample mean, or an expectation, and define it before use.","section":"Remark 2"},{"comment":"In the sharp-design case of Theorem B.1 the formula keeps the limits as eps down to 0, while the preceding display in the fuzzy case drops them; please reconcile the two displays and state the sharp-design formula consistently.","section":"Theorem B.1"},{"comment":"The bound M in Assumption 7d is stated as existing for all d in D(epsilon); please state explicitly that M and zeta are uniform in d, which is the form used in the Lyapunov verification in the proof of Theorem 2.","section":"Assumption 7d"}],"recommendation":"major_revision","confidential_remarks":"The null-event conditioning issue is central and not fixable by a footnote: it requires redefining the estimand and the bridge integrals with an explicit one-sided convention. The authors are clearly aware of standard RDD null-event issues, as shown by Remark 1 and the Bayes-rule footnote, which makes the omission conspicuous. If they can provide a one-sided definition and show that the identification argument goes through, I would support publication; I do not see grounds for rejection, because the rest of the moment algebra is coherent and the problem is a well-posedness gap rather than an internal contradiction."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know: the main identification theorem has a hole that is more basic than the paper's own caveats. The estimator decomposition is genuinely new and elegant, but the causal parameter is not well defined as written.\n\nThe new thing is the decomposition in Proposition 1: the local IV estimator is numerically equal to the standard RDD estimate of the outcome discontinuity minus a placebo-outcome discontinuity weighted by a local IV coefficient. That is a clean, useful way to make placebo tests constructive. The identification chain through the confounding bridge is carefully argued, the proofs are detailed, and the literature review is honest and positions the work well against Gerard et al.'s bounds and Eckles et al.'s approach.\n\nThe soft spot is the estimand itself. tau0 is defined as E[Y(1)-Y(0)|D=d*]. The paper explicitly allows the density of the running variable to jump at d*, and the conditional distribution of unobservables to be discontinuous. But then {D=d*} is a null event, and E[\\cdot|D=d*] is not uniquely defined without choosing a version. Theorem 1 writes the identified quantity using integrals with respect to P(w|D=d*), never specifying whether this is the left limit, right limit, or something else. The two one-sided choices generally give different values, and the formula may identify different objects depending on that arbitrary convention. This is not a cosmetic issue; it undermines the identification claim as stated. The paper flags the partial-linearity requirement for consistency and inference, but never flags this side-of-cutoff ambiguity. The reader's conditional verdict is right: the algebra is sound, but the causal parameter needs an explicit convention.\n\nMinor: Corollary 1 and Theorem 2 require exact partial linearity of the bridge in W; the paper acknowledges this in Section 4 but it is easy to miss that it is an assumption rather than derived. The absence of any simulation or empirical application is a real gap for an estimator with this many moving parts.\n\nWho should read it: applied econometricians running RDD with manipulation concerns, and theorists working on proxy variables. It deserves a serious referee. The authors need to define tau0 as a one-sided limit (the right-limit version matches their estimator) and adjust proofs accordingly. Once that is done, the paper becomes a solid contribution.","headline":"A novel RDD correction via placebo outcomes, but the causal estimand conditions on a null event without a side convention—fix the estimand and it's a solid paper.","tokens_in":66712,"tokens_out":3939,"would_cite":false,"duration_ms":48749,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A placebo outcome and placebo treatment let regression discontinuity designs recover treatment effects even when the running variable is strategically manipulated.","keywords":["regression discontinuity design","placebo outcome","negative control","causal inference","bias-corrected inference","local instrumental variable","confounding bridge","strategic manipulation"],"falsifier":"In a Monte Carlo study with a known treatment effect, a continuous unobserved confounder, and only a scalar placebo treatment, the completeness condition fails; if the PDD estimate does not concentrate on the true $\\tau_0$ while a bounds approach still covers it, the paper's point-identification claim is refuted. Separately, estimating the bridge nonparametrically and testing the partial-linearity restriction would reveal whether $\\hat{\\tau}_{\\mathrm{pdd}}$'s limit is $\\tau_0$ or a best-linear approximation.","tokens_in":65663,"feed_emoji":"🎯","tokens_out":7073,"duration_ms":72375,"temperature":0.7,"pith_summary":"Regression discontinuity designs normally require that expected potential outcomes are continuous at the cutoff. This paper claims that this requirement can be dropped if the researcher has a placebo treatment and a placebo outcome available. The paper identifies the treatment effect at the cutoff, despite strategic manipulation of the running variable, through a confounding-bridge adjustment, and proposes a local instrumental variable estimator that decomposes into a standard RDD estimate minus an adjustment term built from the placebo outcome's discontinuity. If correct, the method turns the common practice of testing for placebo discontinuities into a correction that recovers the causal effect instead of merely flagging invalidity.","feed_headline":"New estimator recovers RDD effects despite strategic sorting","feed_subtitle":"When strategic behavior breaks the usual continuity at the cutoff, placebo variables restore point identification of the treatment effect.","key_machinery":"The confounding bridge $h(d-d^*, w)$, defined by $\\mathbb{E}[Y\\mid D=d,U] = \\int h(d-d^*, w)\\,dP(w\\mid D=d,U)$, is the object that carries the argument: it converts unobserved confounding into an integral equation in observed quantities once the placebo treatment $Z$ is used as an instrument. The local instrumental variable estimator approximates the bridge as partially linear, $h(d-d^*,w) = g(d-d^*) + w^{\\top}\\gamma$, and the finite-sample equivalence $\\hat{\\tau}_{\\mathrm{pdd}} = \\hat{\\tau}^{y}_{\\mathrm{rdd}} - (\\hat{\\tau}^{w}_{\\mathrm{rdd}})^{\\top}\\hat{\\gamma}_{-}$ shows exactly how the placebo outcome's discontinuity adjusts the standard RDD estimate.","core_discovery":"The central claim is that the RDD parameter $\\tau_0 = \\mathbb{E}[Y(1,D,U,\\eta_y)-Y(0,D,U,\\eta_y)\\mid D=d^*]$ is identified even when the distribution of unobserved confounders $U$ jumps at the cutoff, provided expected potential outcomes are continuous conditional on $U$, a placebo outcome $W$ and placebo treatment $Z$ satisfy exclusion and selection conditions, and a confounding bridge $h$ exists and is unique. Theorem 1 shows $\\tau_0$ equals the limiting difference of integrals of $h_+$ and $h_-$ against the placebo-outcome distribution at the cutoff, divided by the discontinuity in treatment probability; with completeness (Assumption 6) the bridge is identified from observed data. The proposed local instrumental variable estimator is numerically equal to the standard RDD estimator for the outcome minus the product of the placebo-outcome RDD discontinuity and a weight $\\hat{\\gamma}_{-}$; the paper proves consistency and, after robust bias correction, asymptotic normality at rate $n^{-2/5}$ at MSE-optimal bandwidths.","pith_inferences":["The point-identification claim rests on a completeness condition that is untestable and will often fail when $U$ is continuous and $Z$ is scalar; in such settings the method likely degrades to partial identification, so its practical scope is narrower than the title suggests.","The consistency and normal-theory results require the confounding bridge to be exactly partially linear in $W$; if the bridge is nonlinear, the estimator converges to a best-linear approximation of the bridge, and the gap to $\\tau_0$ is not recovered by the stated theory.","Because $\\hat{\\gamma}_{-}$ is a ratio of covariances, a placebo outcome only weakly related to $U$ will inflate the variance of the adjustment term, mirroring weak-instrument behavior in IV; researchers should report the first-stage strength of $Z$ for $W$.","The same placebo-corrected template could be applied to other discontinuity designs, such as kink designs or multi-cutoff settings, whenever a pre-treatment proxy for the confounding variable is available."],"forward_implications":["Researchers who currently report placebo-outcome discontinuities as evidence that an RDD is invalid can now use those same discontinuities in an adjustment term that recovers the treatment effect.","When the placebo outcome shows no jump at the cutoff, the estimator numerically reduces to the standard RDD estimate, so the method nests conventional practice.","The target parameter is the treatment effect on the treated at the cutoff; only the left limit is adjusted, so the estimator does not require estimating a bridge on both sides symmetrically.","Robust bias-corrected inference is valid at MSE-optimal bandwidths, so confidence intervals can be constructed without undersmoothing."],"supporting_citations":[{"why":"Defines the standard RDD identification framework and the local linear discontinuity estimators that the PDD estimator builds on and reduces to.","marker":"Hahn et al., 2001"},{"why":"Introduces the negative-control proxy-variable identification strategy for unmeasured confounders that the paper adapts to the RDD cutoff.","marker":"Miao et al., 2018"},{"why":"Provides the confounding-bridge formalism used in Assumption 5 and Lemma 3 to move from unobserved $U$ to observed $(Z,W)$.","marker":"Tchetgen Tchetgen et al., 2024"},{"why":"Supplies the local linear regression technique that motivates the boundary-bias-reducing estimator.","marker":"Fan and Gijbels, 1992"},{"why":"Develops the robust bias-corrected inference procedure that Theorem 2 extends to the placebo discontinuity design.","marker":"Calonico et al., 2014"},{"why":"Establishes completeness conditions for nonparametric instrumental variable identification, which underpin Assumption 6's uniqueness argument.","marker":"Newey and Powell, 2003"},{"why":"Connects RDD continuity to the conditional distribution of unobservables, the failure of which is the target of this paper.","marker":"Lee and Lemieux, 2010"},{"why":"Gives bounds for treatment effects under manipulation, which the paper complements with point-identification conditions.","marker":"Gerard et al., 2020"}],"fun_headline_variants":["Placebo bridge restores RDD treatment effects","RDD survives strategic sorting with placebo adjustment","Placebo variables rescue RDD from manipulation","Local IV estimator fixes RDD with placebo outcome"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that no nonzero function of the unobserved confounder $U$ has conditional mean zero given the running variable and placebo treatment; if this completeness condition fails, multiple bridge functions solve the observed equations and $\\tau_0$ is not point identified.","fun_headline_variants_meta":{"raw":{"variants":["Placebo bridge restores RDD treatment effects","RDD survives strategic sorting with placebo adjustment","Placebo variables rescue RDD from manipulation","Local IV estimator fixes RDD with placebo outcome"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000235,"raw_usage":{"total_tokens":1478,"prompt_tokens":900,"completion_tokens":578,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":516,"completion_tokens_details":{"reasoning_tokens":521}},"tokens_in":516,"tokens_out":578,"duration_ms":6738,"temperature":1.0,"reasoning_tokens":521,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T16:43:25.019197+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"In a Monte Carlo study with a known treatment effect, a continuous unobserved confounder, and only a scalar placebo treatment, the completeness condition fails; if the PDD estimate does not concentrate on the true $\\tau_0$ while a bounds approach still covers it, the paper's point-identification claim is refuted. Separately, estimating the bridge nonparametrically and testing the partial-linearity restriction would reveal whether $\\hat{\\tau}_{\\mathrm{pdd}}$'s limit is $\\tau_0$ or a best-linear approximation.","supporting_citations":[{"cited_title":"(2001, Section 4.1) imply that e⊤ 0(R⊤ 1 K−R1)−1R⊤ 1 K−W 𝑝 →(𝛽 𝑤 −,0)⊤=lim 𝜖↓0 E 𝑊⊤ 𝑖 |𝐷 𝑖 =𝑑∗−𝜖 54 1 ℎ𝑛 e⊤ 1(R⊤ 1 K−R1)−1R⊤ 1 K−W 𝑝 →(𝛽 𝑤 −,1)⊤","cited_arxiv_id":null,"evidence_quote":"Defines the standard RDD identification framework and the local linear discontinuity estimators that the PDD estimator builds on and reduces to."}],"review_version":1}