Direct Causal Effect of Gateway Course Repetition on Engineering Student Dropout: Evidence from G-Estimation of a Structural Nested Mean Model

Dropout in engineering and STEM programmes is widely associated with failure and repetition in foundational mathematics courses. However, prior research has not established whether gateway course repetition has a direct causal effect on dropout, independent of the academic history that precedes it. Students who repeat a gateway course already carry a trajectory of academic difficulty; disentangling the effect of repetition itself from the confounding influence of that prior history requires methods designed for longitudinal data with time-varying confounding. This study estimates the direct average causal effect of first-time repetition of a gateway Algebra course on three-year dropout probability, using 35 years of longitudinal administrative panel data from an Argentine public engineering faculty (FACET-UNT; entry cohorts 1982–2016; N = 22,177 students; 208,639 student-term observations). We employ G-estimation of an additive structural nested mean model (SNMM), adjusting for a six-variable time-varying confounding history capturing cumulative academic performance, enrolment patterns, and prior repetition behaviour. Uncertainty is quantified via BCa bootstrap (B = 1,000 resamples). G-estimation yields a statistically significant positive causal effect: first-time Algebra repetition increases three-year dropout probability by 5.0 percentage points (ψ̂ = 0.050; 95% BCa CI [0.021, 0.079]). This effect is substantially smaller than the causal effect of low early academic capital accumulation estimated in a parallel study using the same SNMM framework (ψ̂ = 0.253; approximately five times larger). The unadjusted dropout difference between repeaters and non-repeaters (0.5 pp) is substantially lower than the causal estimate, indicating that time-varying confounding suppresses rather than inflates the raw comparison in this setting. These findings establish that gateway course repetition has a real but modest direct causal effect on dropout. It functions as a secondary downstream mechanism operating within trajectories already shaped by earlier structural conditions, rather than as a primary causal driver. Policy implications favour a two-stage intervention model combining early trajectory support with targeted assistance at the moment of gateway repetition.