[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85958-en":3,"doc-seo-85958-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},85958,13056703019404,"Miles","https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0",8,"Research & Report","Representation Learning for Semiparametric Causal Mediation Analysis under No Essential Heterogeneity","A two-stage estimator is proposed for structural mediation parameters by combining deep representation learning with G-estimation under the “No Essential Heterogeneity”(NEH) assumption. The UNIT method uses TARNet to learn shared covariate representations across treatment arms and to approximate the heterogeneity-dependent component of the weight function in the G-estimating equation (Zheng and Zhou, 2015), enabling identification despite unmeasured mediator–outcome confounding. Simulations with non-Gaussian covariates and nonlinear mediator effects show improved precision of mediation coefficients (median SE ratios 1.45–1.51 at n≥2000) without added bias or coverage loss.","Representation Learning for Semiparametric Causal Mediation  \nAnalysis  \nunder No Essential Heterogeneity  \nRoberto Faleh∗  \nMethods Center University of Tübingen, Germany  \nSofia Morelli  \nMethods Center University of Tübingen, Germany  \nHolger Brandt  \nMethods Center University of Tübingen, Germany  \narXiv :2607 . 10540v1 [ stat .ML] 12 Jul 2026  \nAbstract  \nWe propose a two-stage estimator for structural mediation parameters that combines deep representation learning with G-estimation under the “No Essential Heterogeneity”(NEH) assumption.  \nWe term the method UNIT, for Unmeasured-confounding-robust NEH-based Identification with TARNet. In the first stage, TARNet estimates the heterogeneous effect of a randomized treatment on a mediator by learning a shared covariate representation across treatment arms. The resulting conditional average treatment effect (CATE) estimate provides a plug-in approximation to the heterogeneity-dependent component of the weight function entering the G-estimating equation of Zheng and Zhou (2015), which identifies the structural parameters even in the presence of unmeasured mediator–outcome confounding. We show that more accurate first-stage representation learning can yield a more informative plug-in weight and thereby improve the precision of the structural parameter estimator. In simulations with non-Gaussian covariates and nonlinear mediator effects, TARNet weights reduce the Stage-2 standard error of the mediation coefficient (median SE ratios 1.45–1.51 at n ≥ 2000) compared to the classical approach, at no cost to bias or coverage.  \nKeywords: causal mediation, G-estimation, no essential heterogeneity, representation learning, structural mean models, TARNet.  \n1 Introduction  \nThe primary goal of many randomized experiments is to estimate the effect of a treatment on an outcome (Hernánand Robins, 2020) . In psychology, economics, and biology, researchers are often interested not only in estimating overall treatment effects, but also in studying the underlying causal mechanisms and understanding how an intervention exerts its effects (VanderWeele, 2015) . Such an investigation is especially relevant when the objective is, for instance, to evaluate a scientific theory or to guide and inform the design of future interventions. A frequently studied setting (Rijnhart et al., 2021; Holland, 1988) involves a mediator M, an intermediate variable that is influenced by the treatment and, in turn, affects the outcome, potentially together with a set of baseline covariates. Randomization of the treatment T removes confounding between the treatment and the outcome. However, for mediator variables, the situation becomes more complex. Randomization of the treatment does not, in general, imply randomization of the mediator, since the mediator may be influenced by processes that occur after the initial treatment assignment (Bullock et al., 2010) . The consequence is that the causal effect of the mediator on the outcome can be confounded. In this context, several technical approaches are available and, in general, this does not represent a major methodological concern (Wodtke and Zhou, 2026) . A more serious problem arises when some of these confounding variables are not measured (Ten Have and Joffe, 2012; Imai et al., 2010) . This is a plausible scenario when, for example, the trial was originally designed to estimate only the overall treatment effect and was not designed to support a causal mediation analysis. In other circumstances, the confounders can be potentially known but unethical, prohibitively expensive, or difficult to measure (Brandt, 2020) . As a result, researchers could have extensive knowledge about the causal mech  \nanism and a rich set of covariates but still lack confounders needed for the identification of the indirect effects. For ∗ Corresponding author: roberto-faleh[at][uni-tuebingen.de](uni-tuebingen.de).  \nPreprint.  \nexample, an online retail platform may be interested in examining","cbCaisbtenJtaDjZ","https://ap.wps.com/l/cbCaisbtenJtaDjZ","pdf",541346,3,1,30,"English","en",105,"# Introduction\n## Problem setting and motivation\n## Challenges from unmeasured mediator–outcome confounding\n## Alternative causal frameworks (SNMs and G-estimation)\n# Method overview\n## UNIT two-stage estimator with TARNet\n## Role of NEH and identification via weight function\n# Simulation results\n## Precision improvements and robustness checks","[{\"question\":\"What is the main contribution of the UNIT method?\",\"answer\":\"UNIT is a two-stage estimator for structural mediation parameters that integrates deep representation learning with G-estimation under the No Essential Heterogeneity (NEH) assumption.\"},{\"question\":\"How does TARNet help with mediation analysis in the presence of unmeasured confounding?\",\"answer\":\"TARNet learns a shared covariate representation across treatment arms to estimate a CATE, which is used as a plug-in approximation for the heterogeneity-dependent weight component in the G-estimating equation.\"},{\"question\":\"What do simulation results show about precision, bias, and coverage?\",\"answer\":\"When covariates are non-Gaussian and mediator effects are nonlinear, TARNet-based weights reduce the Stage-2 standard error median ratios (1.45–1.51 at n≥2000) compared with the classical approach, without increasing bias or harming coverage.\"}]",1784207375,76,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"representation-learning-for-semiparametric-causal-mediation-analysis-under-no-essential-heterogeneity","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":20},"https://docshare.wps.com/document/research-report/",{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/representation-learning-for-semiparametric-causal-mediation-analysis-under-no-essential-heterogeneity/85958/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-26","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What is the main contribution of the UNIT method?","Question",{"text":75,"@type":76},"UNIT is a two-stage estimator for structural mediation parameters that integrates deep representation learning with G-estimation under the No Essential Heterogeneity (NEH) assumption.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does TARNet help with mediation analysis in the presence of unmeasured confounding?",{"text":80,"@type":76},"TARNet learns a shared covariate representation across treatment arms to estimate a CATE, which is used as a plug-in approximation for the heterogeneity-dependent weight component in the G-estimating equation.",{"name":82,"@type":73,"acceptedAnswer":83},"What do simulation results show about precision, bias, and coverage?",{"text":84,"@type":76},"When covariates are non-Gaussian and mediator effects are nonlinear, TARNet-based weights reduce the Stage-2 standard error median ratios (1.45–1.51 at n≥2000) compared with the classical approach, without increasing bias or harming coverage.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,122,127,130,134],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":22,"slug":121},"research-report",{"id":123,"doc_module":4,"doc_module_name":46,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":46,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":46,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":46,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]