--- title: Cumulative Cross-World Weighted Effects url: https://www.emergentmind.com/topics/cumulative-cross-world-weighted-effects type: topic --- # Cumulative Cross-World Weighted Effects Searching arXiv for the cited papers and closely related longitudinal causal inference work. Cumulative cross-world weighted effects are longitudinal causal estimands for comparing two static treatment regimes when standard longitudinal positivity fails. They are defined as weighted averages of the potential outcome contrast $Y(\overline a_T)-Y(\overline a_T^\prime)$, with weights that accumulate over time and depend simultaneously on the treatment process under both counterfactual treatment histories [2507.10774]. The construction is intended to preserve the contrast between the two regimes themselves rather than substituting an implementable stochastic policy, but this comes at the cost of targeting a non-implementable cross-world quantity [2507.10774]. The topic also intersects with recent work on clone-censor-weighting (CCW), which shows that weighting procedures for treatment initiation windows can implicitly target hybrid or impossible interventions assembled from incompatible treatment histories, thereby clarifying why cross-world weighting is both interpretable in mechanistic terms and problematic in policy terms [2404.15073]. ## 1. Formal longitudinal setup and estimand The proposed framework is longitudinal with $T$ time points and observed data \[ Z=(X_1,A_1,X_2,A_2,\dots,X_T,A_T,Y), \] where $X_t$ are time-varying covariates, $A_t\in\{0,1\}$ are binary treatments, and $Y$ is the final outcome [2507.10774]. Treatment histories are denoted $\overline A_t=(A_1,\dots,A_t)$, covariate histories are $\overline X_t=(X_1,\dots,X_t)$, and the history before treatment at time $t$ is \[ H_t=(\overline X_t,\overline A_{t-1}) \] [2507.10774]. Potential outcomes are defined in an NPSEM: \[ X_t=f_{X,t}(A_{t-1},H_{t-1},U_{X,t}),\qquad A_t=f_{A,t}(H_t,U_{A,t}),\qquad Y=f_Y(A_T,H_T,U_Y) \] [2507.10774]. For a regime $\overline a_T=(a_1,\dots,a_T)$, the counterfactual outcome is $Y(\overline a_T)$, and the counterfactual covariate history is $\overline X_t(\overline a_{t-1})$ [2507.10774]. The cumulative cross-world weighted effect is \[ \psi(\overline a_T,\overline a_T^\prime) := E\!\left( \{Y(\overline a_T)-Y(\overline a_T^\prime)\} \prod_{t=1}^T w_t\!\left[p_t\{\overline X_t(\overline a_{t-1})\}\right] w_t^\prime\!\left[p_t^\prime\{\overline X_t(\overline a_{t-1}^\prime)\}\right] \right) \] [2507.10774]. Here the natural propensity scores are \[ p_t\{\overline X_t(\overline a_{t-1})\} = P\!\left\{A_t(\overline a_{t-1})=a_t\mid \overline X_t(\overline a_{t-1})\right\}, \] \[ p_t^\prime\{\overline X_t(\overline a_{t-1}^\prime)\} = P\!\left\{A_t(\overline a_{t-1}^\prime)=a_t^\prime\mid \overline X_t(\overline a_{t-1}^\prime)\right\} \] [2507.10774]. Thus the estimand is not merely a causal contrast between two regime-specific means; it is a weighted contrast whose weighting functional itself is cross-world. A plausible implication is that the estimand should be understood as selecting, across time, the portions of the two counterfactual trajectories that remain jointly informative when positivity fails. The paper states this point more directly by emphasizing that the weights adapt “cumulatively across timepoints and simultaneously across both counterfactual treatment histories” [2507.10774]. ## 2. Cumulative weighting across counterfactual worlds The “cumulative” feature is the product structure \[ \prod_{t=1}^T w_t\!\left[p_t\{\overline X_t(\overline a_{t-1})\}\right] w_t^\prime\!\left[p_t^\prime\{\overline X_t(\overline a_{t-1}^\prime)\}\right] \] [2507.10774]. Weighting therefore responds at every time point to longitudinal nonpositivity rather than only at baseline. The paper gives several explicit weighting choices [2507.10774]: | Choice | $w_t$ | $w_t^\prime$ | |---|---|---| | No weighting | $1$ | $1$ | | Weight toward only the target regime | $p_t$ | $1$ | | Weight toward only the comparator regime | $1$ | $p_t^\prime$ | | Overlap weighting | $p_t$ | $p_t^\prime$ | | Trimming | $\mathbb{1}(p_t\ge \varepsilon)$ | $\mathbb{1}(p_t^\prime\ge \varepsilon^\prime)$ | A concrete trimming example is \[ w_t[p_t]=\mathbb{1}(p_t>\varepsilon),\qquad w_t^\prime[p_t^\prime]=\mathbb{1}(p_t^\prime>\varepsilon^\prime) \] [2507.10774]. A smooth trimming choice used in the empirical application is \[ w_t(\pi_t)=1-e^{-20\pi_t},\qquad w_t^\prime(\pi_t^\prime)=1-e^{-20\pi_t^\prime} \] [2507.10774]. The key restriction on the weights is \[ \pi_t(\overline x_t)\pi_t^\prime(\overline x_t)=0 \implies w_t\{\pi_t(\overline x_t)\}w_t^\prime\{\pi_t^\prime(\overline x_t)\}=0 \] [2507.10774]. This requirement ensures that if either relevant propensity score is zero, the combined weight is also zero. In substantive terms, paths unsupported in either counterfactual world are removed from the weighted functional. This suggests that weighting is not simply a variance-control device. It determines which regions of longitudinal state space contribute to the estimand and therefore partly defines the target itself. ## 3. Identification without standard positivity Identification is developed under NPSEM and strong sequential randomization: \[ U_{A,t}\ \perp\ \{\underline U_{X,t+1},\underline U_{A,t+1},U_Y\}\mid H_t, \qquad t=1,\dots,T \] [2507.10774]. The paper notes that this is stronger than ordinary sequential exchangeability because the estimand depends on cross-world quantities [2507.10774]. Under NPSEM and strong sequential randomization, if positivity has held up to earlier times, then the natural propensity scores are identified by observed-data quantities: \[ P\{A_t(\overline a_{t-1})=a_t\mid \overline X_t(\overline a_{t-1})\} = P(A_t=a_t\mid \overline X_t,\overline A_{t-1}=\overline a_{t-1}) =:\pi_t(\overline X_t), \] with an analogous expression for $\pi_t^\prime(\overline X_t)$ under $\overline a_T^\prime$ [2507.10774]. The main identification formula is \[ \psi(\overline a_T,\overline a_T^\prime) = \int E(Y\mid \overline a_T,\overline x_T) \prod_{t=1}^T w_t\{\pi_t(\overline x_t)\}w_t^\prime\{\pi_t^\prime(\overline x_t)\} \,dP(x_t\mid \overline A_{t-1}=\overline a_{t-1},\overline x_{t-1}) \] \[ - \int E(Y\mid \overline a_T^\prime,\overline x_T) \prod_{t=1}^T w_t\{\pi_t(\overline x_t)\}w_t^\prime\{\pi_t^\prime(\overline x_t)\} \,dP(x_t\mid \overline A_{t-1}=\overline a_{t-1}^\prime,\overline x_{t-1}) \] [2507.10774]. The paper explicitly states that no standard positivity assumption is required for identification because the design of the weights replaces it [2507.10774]. The framework distinguishes positivity from overlap of covariate distributions through the density ratio \[ \rho_t(\overline X_t) = \frac{dP(X_t\mid \overline A_{t-1}=\overline a_{t-1},\overline X_{t-1})} {dP(X_t\mid \overline A_{t-1}=\overline a_{t-1}^\prime,\overline X_{t-1})} \] and the condition \[ P\{\rho_t(\overline X_t)\in (0,\infty)\}>0,\qquad \text{for every }t \] which the paper terms partial common support [2507.10774]. Partial common support is weaker than full overlap, but the paper states that it ensures the weighted functional remains informative about the causal contrast [2507.10774]. ## 4. Interpretation, non-implementability, and the cross-world problem The paper presents cumulative cross-world weighted effects as mechanism-relevant but not policy-relevant in the sense of a feasible intervention [2507.10774]. The reason is structural: the estimand depends on a product of weights evaluated under two counterfactual worlds simultaneously, and one cannot implement an intervention that generates both $\overline X_t(\overline a_{t-1})$ and $\overline X_t(\overline a_{t-1}^\prime)$ for the same person [2507.10774]. The resulting tradeoff is explicit. If positivity holds, the ordinary causal contrast \[ E\{Y(\overline a_T)-Y(\overline a_T^\prime)\} \] is both interpretable and implementable. If positivity fails, implementable stochastic or modified policies can be identified, but those may conflate regime effects with downstream effects on intermediate treatments and covariates [2507.10774]. The cumulative cross-world weighted effect instead preserves the pure contrast of the two static regimes, but only as a non-implementable cross-world quantity [2507.10774]. The paper also gives a null-preservation property: \[ P\{Y(\overline a_T)=Y(\overline a_T^\prime)\}=1 \quad\Rightarrow\quad \psi(\overline a_T,\overline a_T^\prime)=0 \] [2507.10774]. This establishes that the estimand respects a strong form of causal nullity despite its non-implementable character. Closely related concerns appear in the analysis of CCW for treatment initiation windows. There, a regimen such as “start treatment prior to day 30” is shown to estimate the potential outcome under a two-stage intervention where “A) prior to day 30, everyone follows the treatment start distribution of the study population and B) everyone who has not initiated by day 30 is forced to initiate on day 30” [2404.15073]. When earlier initiators are used to represent those forced to initiate at the end of the window, the target may require assigning a day-30 starter the prior exposure history of someone who started earlier, producing a cross-world or impossible intervention [2404.15073]. This parallel is conceptually important: in both settings, weighting may preserve a mechanistic contrast while moving away from interventions that could be literally enacted. ## 5. Partial common support and collapse to zero A central substantive limitation is that identification alone does not guarantee informativeness. The paper states that if for some $t$, \[ P\{\rho_t(\overline X_t)\in(0,\infty)\}=0, \] then \[ \psi(\overline a_T,\overline a_T^\prime)=0 \] regardless of the true potential outcome contrast [2507.10774]. The argument is that when the relevant conditioning event has probability zero, the corresponding propensity score is set to zero, so the combined weight is zero on those paths; if no common support remains, the entire weighted functional vanishes [2507.10774]. Conversely, under positive-probability overlap at each timepoint, the paper states that the functional can preserve sign: if \[ P\{Y(\overline a_T)>Y(\overline a_T^\prime)\}=1, \] then \[ \psi(\overline a_T,\overline a_T^\prime)>0, \] and if \[ P\{Y(\overline a_T)