Context: Prolonged disconnection from school and work represents major setbacks during the transition to adulthood and is a distinct feature of the developmental trajectories of many disadvantaged youths, especially those from a marginalized racial background (Hong and Chung 2022; Shanahan 2000). Differential schooling experiences are hypothesized mechanisms that widen racial disparity (e.g., Conwell 2021; Hanushek et al, 2019; Jarvis and Okonofua 2020; Jencks and Phillips 1998). Yet disadvantaged youths alienated from education are the ones who can reap the most benefit when engaged (Brand and Xie 2010). To what extent could the racial disparity in youth disconnection from school and work be reduced through reducing the racial gap in earlier schooling? Unlike the conventional descriptive Blinder-Kitagawa-Oaxaca decomposition (Blinder 1973; Kitagawa 1955; Oaxaca 1973), a "causal decomposition analysis" partitions the between-group disparity in an outcome (youth disconnectedness, denoted as ) into a component that could be reduced through conceivable interventions on malleable factors (earlier schooling experiences, denoted as ) and a component that would remain. The key is to simulate a distribution of the counterfactual outcome for each group when hypothetical interventions would have reduced or even eliminated the between-group gap in . The results are credible when the causal relationship between and within each group can be identified under plausible assumptions (DiNardo, Fortin, and Lemieux 1996; Huber 2015; Jackson 2021; Jackson and VanderWeele 2018; Li and Li 2023; Lundberg 2022; Park et al. 2023; Park, Kang, and Lee 2023; VanderWeele and Robinson 2014; Yamaguchi 2015; 2017; 2019). Objectives: This study reveals "endogenous confounding" and "incomplete common support" as major obstacles in causal decomposition analysis. We propose a rank-preserving transformation of endogenous confounders to not only address these challenges, but also to simulate an equity-oriented intervention. Analyzing the National Longitudinal Study of Youth 1997 (NLSY 97) data, we investigate the potential of reducing the racial gap in youth disconnectedness through such an intervention in earlier schooling. Estimand: Let S = 1 denote Black youths and S = 0 for White youths. Corresponding to the substantive research questions, we define two estimands: (1) Reduction in racial disparity in youth disconnectedness should the distribution of earlier schooling experiences of Black youths become comparable to that of Whites: [delta][subscript reduced] = E[Y(M)|S = 1] - E[Y(M*)|S = 1], where M* represents the counterfactual earlier schooling experiences under the hypothetical intervention. 2) Remaining racial disparity in the outcome despite the equalization efforts in schooling: [delta][subscript remain] = E[Y(M*)|S = 1] - E[Y(M)|S = 0]. Methodological Challenges: Identification of the counterfactual quantity E[Y(M*)|S = 1] requires adjustment for confounding covariates that are either independent of race (e.g., gender, denoted as X) or distribute differently across the racial groups (e.g., family SES, denoted as A). Youths growing up in lower-SES households tend to struggle more in school and later become more disconnected from school and work on average. Moreover, due to the legacy of racial discrimination, Black youths are over-represented among the lower-SES. Some researchers have proposed a lottery type intervention that unconditionally or conditionally randomizes M to equate its distribution between race (e.g., Jackson and VanderWeele 2018; Lunberg 2022); others have proposed an affirmative action (AA) type intervention that equalizes the conditional odds of a binary between race (Opacic et al. working paper; Zhou & Pan, 2023). Yet the existing literature has generally overlooked the issues related to "endogenous confounding" and "incomplete common support." Unconditionally randomizing M is unrealistic as it ignores how M is predicted by A and X; equating the conditional distribution or the conditional odds of M between race within levels of A would fail to achieve racial equity in M due to the racial difference in A. Moreover, when the range of M does not completely overlap between race within levels of A and X, the counterfactual information Y(M*) would be lacking for individuals not within the common support. Rank-Preserving Transformation as a Solution: Our proposed alternative strategy involves a rank-preserving transformation of the endogenous confounder A. Let A[subscript s] denote the standardized score or the percentile score of A representing an individual's relative standing within racial group s for s = 1, 1. Importantly, A[subscript s] is not endogenous to S. Through weighting, we simulate a hypothetical world in which the counterfactual distribution of M* for Black students would be made equal to the observed distribution of M among their White counterparts who share the same covariate values X = x and A[subscript s] = a. This resembles an equity-oriented intervention aiming to overcome the standing racial (dis)advantage in A as we make Pr(M* = m|S = 1, X = x, A[subscript s=1] = a = Pr(M* = m|S =0, X = x, A[subscript s=1]. The weighting strategy easily accommodates multidimensional M. For Blacks with M = m, X = x, and A[subscript s=1] = a, [equation omitted]. The conditional average counterfactual outcome E[Y(M* = m)|S = 1], X = x, A[subscript s] = a] for Blacks will be identified by the conditional average observed outcome of their Black counterparts E[Y(M = m)|S = 1], X = x, A[subscript s] = a]. Yet when there is incomplete common support, especially when certain values of M = m are observed among Whites but not Blacks who share the same covariate values x and a, the counterfactual E[Y(M* = m)|S = 1], X = x, A[subscript s] = a] cannot be identified for the Blacks. This is the case when an equity-oriented intervention would reduce but not eliminate the racial gap in M. Analytic Results: NLSY 97 includes a nationally representative sample of youths born between 1980 and 1984. Our outcome measure is maximum duration (# weeks) of disconnection from school and work in 2007. Measures of schooling experiences between 1998 and 2006 include 8 grade GPA, high school GPA, # days suspended from school each year, # months missed school each year, # grades repeated or skipped, and the highest level of education attained. Figure 1 shows that, if through hypothetical interventions, Black youths would have the same schooling experiences as their White counterparts, their average maximum duration of disconnection would be reduced from 7.6 weeks to 5 weeks; the racial disparity in disconnection would be reduced from three weeks to half a week, the latter being statistically indistinguishable from zero. Conclusion: Causal decomposition analysis promises to inform interventions for promoting equity. The validity and theoretical significance of results are contingent upon the plausibility of identification assumptions. Yet even in the absence of omitted confounding, the analyst must take great care in examining the common support and appropriately handling endogenous confounders.