Does Infrastructure Investment Pay Off Differently by Governance Type? A Small-Sample Heterogeneous-Effects Analysis of NAAC-to-NIRF Trajectories

Abstract

A recurring policy question in Indian higher education is whether the same quality-assurance investment yields different competitive payoffs depending on an institution's governance type: whether it is centrally funded, State funded, or a self-financing Private/Deemed university. This paper examines one specific version of that question empirically: does a university's NAAC-assessed Infrastructure and Learning Resources score (Criterion 4) predict a subsequent gain in its NIRF overall ranking score, and does that relationship differ across governance types? Using 48 Indian universities for which both a NAAC criterion-level snapshot and at least two subsequent NIRF ranking cycles (2016-2025) were available, we split institutions into higher- and lower-infrastructure groups by a median split on Criterion 4 and estimated the average and type-moderated association with annualised NIRF score change, first with an unadjusted comparison of means, then with a covariate-adjusted linear model, and finally with an explicit treatment-by-governance-type interaction term. We found no statistically detectable overall association between infrastructure standing and subsequent NIRF gains (adjusted effect = 0.37 points/year, SE = 0.49, p = .45), and no statistically detectable heterogeneity by governance type (largest interaction term p = .66). The only robust predictor of annualised change was the institution's own baseline NIRF score, which was negatively associated with subsequent gains (b = -0.06, p = .047), consistent with a regression-to-the-mean or ceiling effect rather than any effect of infrastructure investment itself. Because the governance-type subgroups are small (as few as 6 Central universities with usable longitudinal data), we treat these null results as inconclusive rather than as evidence of no effect, and we set out the sample size a confirmatory replication would need. All analyses were conducted in Python..