Disparate Effect Of Missing Mediators On Transportability of Causal Effects

Vishwali Mhasawade, Rumi Chunara
Proceedings of the Fourth Conference on Causal Learning and Reasoning, PMLR 275:757-771, 2025.

Abstract

The transport of mediation effects is an important question for upstream interventions, such as targeted development of parks to improve population health, as their effect on populations is mediated by factors like physical activity, which can vary from place to place. However, upstream treatment effect estimates could be biased when mediator variables are missing for the population where the effect is to be transported. We study this issue of the impact of missing mediators on transported effects, motivated by challenges in public health, wherein mediators are commonly missing but not at random. We propose a sensitivity analysis framework to quantify the impact of missing mediator data on transported mediation effects, identifying when the conditional transported mediation effect becomes insignificant for subgroups with missing data. Applied to longitudinal data from the Moving to Opportunity Study, a large-scale housing voucher experiment, this framework demonstrates the sensitivity of transported mediation effects to data missingness. In particular, we quantify the effect of missing mediators on transport effect estimates of voucher receipt in childhood, an upstream intervention on living location. We then assess the subsequent impact on the risk of mental health or substance use disorder mediated through parental health across sites. Our findings highlight that missing mediators can disparately impact effect estimates across population subgroups and provide a tangible understanding of how much missing data can be withstood for unbiased effect estimates in such mediated settings.

Cite this Paper

BibTeX
@InProceedings{pmlr-v275-mhasawade25a, title = {Disparate Effect Of Missing Mediators On Transportability of Causal Effects}, author = {Mhasawade, Vishwali and Chunara, Rumi}, booktitle = {Proceedings of the Fourth Conference on Causal Learning and Reasoning}, pages = {757--771}, year = {2025}, editor = {Huang, Biwei and Drton, Mathias}, volume = {275}, series = {Proceedings of Machine Learning Research}, month = {07--09 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v275/main/assets/mhasawade25a/mhasawade25a.pdf}, url = {https://proceedings.mlr.press/v275/mhasawade25a.html}, abstract = {The transport of mediation effects is an important question for upstream interventions, such as targeted development of parks to improve population health, as their effect on populations is mediated by factors like physical activity, which can vary from place to place. However, upstream treatment effect estimates could be biased when mediator variables are missing for the population where the effect is to be transported. We study this issue of the impact of missing mediators on transported effects, motivated by challenges in public health, wherein mediators are commonly missing but not at random. We propose a sensitivity analysis framework to quantify the impact of missing mediator data on transported mediation effects, identifying when the conditional transported mediation effect becomes insignificant for subgroups with missing data. Applied to longitudinal data from the Moving to Opportunity Study, a large-scale housing voucher experiment, this framework demonstrates the sensitivity of transported mediation effects to data missingness. In particular, we quantify the effect of missing mediators on transport effect estimates of voucher receipt in childhood, an upstream intervention on living location. We then assess the subsequent impact on the risk of mental health or substance use disorder mediated through parental health across sites. Our findings highlight that missing mediators can disparately impact effect estimates across population subgroups and provide a tangible understanding of how much missing data can be withstood for unbiased effect estimates in such mediated settings.} }
Endnote
%0 Conference Paper %T Disparate Effect Of Missing Mediators On Transportability of Causal Effects %A Vishwali Mhasawade %A Rumi Chunara %B Proceedings of the Fourth Conference on Causal Learning and Reasoning %C Proceedings of Machine Learning Research %D 2025 %E Biwei Huang %E Mathias Drton %F pmlr-v275-mhasawade25a %I PMLR %P 757--771 %U https://proceedings.mlr.press/v275/mhasawade25a.html %V 275 %X The transport of mediation effects is an important question for upstream interventions, such as targeted development of parks to improve population health, as their effect on populations is mediated by factors like physical activity, which can vary from place to place. However, upstream treatment effect estimates could be biased when mediator variables are missing for the population where the effect is to be transported. We study this issue of the impact of missing mediators on transported effects, motivated by challenges in public health, wherein mediators are commonly missing but not at random. We propose a sensitivity analysis framework to quantify the impact of missing mediator data on transported mediation effects, identifying when the conditional transported mediation effect becomes insignificant for subgroups with missing data. Applied to longitudinal data from the Moving to Opportunity Study, a large-scale housing voucher experiment, this framework demonstrates the sensitivity of transported mediation effects to data missingness. In particular, we quantify the effect of missing mediators on transport effect estimates of voucher receipt in childhood, an upstream intervention on living location. We then assess the subsequent impact on the risk of mental health or substance use disorder mediated through parental health across sites. Our findings highlight that missing mediators can disparately impact effect estimates across population subgroups and provide a tangible understanding of how much missing data can be withstood for unbiased effect estimates in such mediated settings.
APA
Mhasawade, V. & Chunara, R.. (2025). Disparate Effect Of Missing Mediators On Transportability of Causal Effects. Proceedings of the Fourth Conference on Causal Learning and Reasoning, in Proceedings of Machine Learning Research 275:757-771 Available from https://proceedings.mlr.press/v275/mhasawade25a.html.

Related Material