A mixed framework for causal impact analysis under confounding and selection biases: a focus on Egra dataset
Abstract
We employed the structural causal model’s (SCM) backdoor adjustment criteria, with the do-calculus intervention, which is deduced from a well-structured direct acyclic graph (DAG) to overcome the confounding and selection biases issue in the presence of all measured sets of covariates in an individually identical distribution (i.i.d). The Potential outcome’s (PO) stable unit treatment value assumption (SUTVA) is employed to show that the treatment considered is homogenous across the i.i.d dataset. The strong ignorability assumption, which is structured to handle confounding and optimally deal with selection biases, with their inference estimation matching techniques with roots in statistics is similarly used. Thus, we leverage the strength of the SCM framework in handling confounding bias (by identifying the right sets of covariates for adjustment) and that of the PO framework in handling selection bias (by correctly matching treated and untreated units in the dataset). The mixed framework is the first to be implemented on Early Grade Reading Assessment (EGRA) intervention program datasets, and the EGRA program dataset implemented is called “Strengthening Education in North-East” Nigeria (SENSE-EGRA) an initiative of the American University of Nigeria with funding from the United States Agency for International Development (USAID). The SENSE-EGRA SCM was validated for correctness using conditional independent test (CIT) criteria. The causal inference impact estimates were implemented with 4 estimators. i.e., the ordinary least square (OLS) regression, the propensity score weighting (PSW), propensity score stratification (PSS), and the propensity score by matching (PSM). The average treatment effects (ATE) for both the OLS and the PSW estimators are inconsistent, i.e., 0.661 and 4.931 respectively, which biases are occasioned by the presence of nonlinearity in the dataset and the estimators’ inability to handle the overlap condition effectively. The PSS and the PSM produced more consistent and reliable estimates for the ATE and the average treatment effects for the treated (ATT), which are 1.592, 1.535, 2.173, and 1.977 respectively. © 2023, The Author(s), under exclusive licence to Bharati Vidyapeeth's Institute of Computer Applications and Management.
Author(s)
Ayem G.T., Ajibesin A., Iorliam A., Nsang A.S.
Year
2023
Countries
Nigeria
Language
English
Research Method
Quantitative
Article Type
Peer-Reviewed Articles
Keywords
Primary education | Foundational Learning | Literacy | Reading | Assessment | evaluation intervention | Excel Import
Source Link
Full Citation Style
Research Team
Ayem G.T.
Ajibesin A.
Iorliam A.
Nsang A.S.
© 2026 AERD Database System. All Rights Reserved. Developed by Strategic Technologies.