How Strongly Correlated Are the Subtests in the 2026 Indonesian University Entrance Exam (UTBK-SNBT)? Crowdsourced Estimates and Implications for National Ranking | Seberapa Kuat Subtes UTBK-SNBT 2026 Saling Berkorelasi? Estimasi dari Data Crowdsourced dan Implikasinya bagi Estimasi Peringkat Nasional
Year: 2026
Type: preprint
Abstract: The inter-subtest correlation matrix of the 2026 Indonesian university entrance examination (UTBK-SNBT) is not publicly available, and the subtest-level descriptive statistics in circulation derive from an unofficial, unverified internal table attributed to the national admissions board (SNPMB). This study aims to estimate the correlation structure across the eight subtests of the 2026 UTBK-SNBT using crowdsourced score data from platform X (N = 1,886), whilst identifying the characteristics of methodological bias inherent in this approach and the implications for national ranking estimation. The observed raw off-diagonal mean correlation was 0.300, which is moderate-to-low for a set of interrelated cognitive ability tests. Comparison with these unofficial national parameters indicates a selection bias towards higher scores, consistent with restriction of range. A Thorndike Case III adjustment via fixed-point iteration within an open-source R computational pipeline yielded a corrected population correlation of R-bar ≈ 0.425 (S_Z ≈ 67.72; 95% bootstrap interval for R-bar [0.407,0.442]). This estimate is sensitive to the unofficial national standard deviations, as changing all of them by ±10% shifts R-bar to between 0.300 and 0.519. The national score distribution also departs from normality (right-skewed with a truncated left tail), and simulation shows that the normality assumption mainly alters percentiles in the distribution tails. A confirmatory factor analysis showed that a two-factor model (Literacy and Numeracy) fit better than a single general-factor model (CFI 0.967 versus 0.932; Δχ² = 91.55, p < 0.001; φ = 0.813). Furthermore, the results indicate that platform-audience bias cannot be resolved solely through more inclusive data-collection framing.Note: An Indonesian-language abstract (Abstrak) is included in the preprint file.
Source: EdArXiv (OSF Preprints)
Author: Galuh Saputra
Language: English
Field: Decision Sciences
Domain: Social Sciences
SDG: Quality education
Open Access status: green
OpenAlex ID: W7225380330