New 35-item scale measures how confident medical students really are with biostatistics

Medical students are expected to read the literature critically, interpret p-values, and judge whether a clinical trial actually supports the treatment it claims to test. Yet until now, educators have lacked a purpose-built instrument for measuring how competent students feel in biostatistics, the discipline that underpins evidence-based medicine. A research team from Bandırma Onyedi Eylül University and Ege University in Türkiye has now filled that gap, reporting the development and psychometric evaluation of a self-perceived Biostatistics Literacy Scale, or BLS, in BMC Medical Education.

The study addressed a deceptively simple question: how do you measure a competency that lives partly in the head? Unlike a statistics exam, which tests what a student can compute under time pressure, a self-perceived literacy scale probes self-efficacy, the confidence a learner has in their own ability to understand, apply, and critically evaluate statistical ideas. Self-efficacy matters because it shapes behavior. A student who feels incapable of interpreting a confidence interval may avoid research opportunities, disengage from journal clubs, and ultimately practice less evidence-based medicine, regardless of what they actually learned in a lecture hall.

The researchers used a rigorous three-phase methodological design covering context development of the scale, validity, and reliability. They began by generating an initial pool of 40 items intended to capture the full landscape of biostatistics literacy among medical students. This item pool was then subjected to two independent statistical trials in two different Turkish medical schools, a strategy that guards against the common pitfall of building a questionnaire that only works on the specific students who helped create it.

The first trial, an exploratory factor analysis, involved 270 students from one medical school. Exploratory factor analysis is a data-driven technique that asks, in effect, whether the pattern of answers across many questionnaire items reveals hidden underlying dimensions. The team used principal axis factoring with oblimin rotation, an approach suited to measuring constructs that are correlated with one another rather than cleanly separate. The analysis delivered a strikingly strong verdict on the data quality: the Kaiser–Meyer–Olkin measure of sampling adequacy reached 0.963, far above the conventional 0.60 threshold, and Bartlett’s test of sphericity was significant at p < 0.001, confirming that the items shared enough common variance to justify factoring at all.

That first analysis also did some pruning. Two items were removed following the exploratory phase because they failed to align cleanly with the emerging structure. What survived was a coherent three-factor architecture: Knowledge, covering students’ grasp of core statistical concepts; AI-supported Skills, a distinctly modern dimension capturing confidence in using artificial intelligence tools for statistical tasks; and Attitude, reflecting how students feel about statistics and their willingness to engage with it. Together, the three factors explained 32.16 percent of the variance in responses, a respectable figure for a multidimensional self-report instrument in the social and educational sciences.

The second trial was where the scale faced its sternest examination. A completely independent sample of 200 students from a second medical school was recruited for confirmatory factor analysis, a technique that tests whether a predefined structure, in this case the three-factor model, fits new data rather than being discovered from it. The team employed diagonally weighted least squares estimation, a modern approach well suited to ordinal questionnaire data. During this phase, three additional Knowledge items were removed in post hoc refinement, leaving a final 35-item instrument.

The fit statistics from the confirmatory analysis read like a checklist of psychometric good health. The ratio of chi-square to degrees of freedom came in at 2.065, comfortably below the 3.0 ceiling commonly accepted. The root mean square error of approximation was 0.073, within the acceptable range, and the standardized root mean square residual was 0.051, close to the ideal of zero. Most impressively, the comparative fit index reached 0.975 and the non-normed fit index 0.974, both well above the 0.95 benchmark that signals a model reproducing the observed data with high fidelity. The team also examined measurement invariance by gender, testing whether the scale functions equivalently for male and female students, and found metric and scalar invariance held, meaning scores can be meaningfully compared across gender groups.

Reliability, the consistency of measurement, was equally emphatic. Cronbach’s alpha for the total scale reached 0.976, an exceptionally high value indicating that the 35 items hang together as a coherent measure of a single overarching construct. The authors also report strong item discrimination, which means the items distinguish effectively between students at different levels of literacy rather than clustering everyone in the middle. In practical terms, the scale can detect both the struggling first-year student and the statistically confident one with precision.

The authors are careful, and rightly so, about the boundaries of their evidence. In the confirmatory sample, 82.5 percent of participants were first- or second-year students, and no fifth- or sixth-year students were represented at all. This means the current psychometric evidence primarily supports the use of the BLS among preclinical medical students in settings similar to those studied. Before the instrument can be rolled out for routine use, the researchers call for validation among senior and clinical-stage students, evaluation against appropriate external criteria such as actual statistical performance, and the development and independent validation of a streamlined short form that could be deployed more quickly in busy curricula.

Even with those caveats, the significance of the work is considerable. Biostatistics curricula in medical schools are frequently criticized as disconnected from clinical practice, and educators have had few tools to evaluate whether their teaching changes how students actually feel about and engage with statistics. A validated self-perception scale gives curriculum designers a feedback loop: administer the BLS before and after a biostatistics course, and the change in scores offers a quantifiable signal of whether the course built genuine confidence alongside technical skill. The inclusion of an AI-supported Skills factor is particularly timely, as large language models and automated analysis tools are rapidly reshaping how clinicians encounter statistics, and medical education has been scrambling to keep pace. By capturing that dimension explicitly, the Turkish team has built an instrument that reflects the reality of contemporary medical training rather than a decade-old syllabus. For a field where the stakes include how future doctors weigh evidence at the bedside, a reliable yardstick for statistical self-efficacy is more than a psychometric exercise; it is infrastructure for the next generation of evidence-based medicine.

Subject of Research: Development and psychometric evaluation of a self-perceived biostatistics literacy scale for medical students

Article Title: A self-perceived biostatistics literacy scale for medical students: development and psychometric evaluation

Article References: Konar, N. M., Suner, A., Özyürek, D., & Karaca, O. (2026). A self-perceived biostatistics literacy scale for medical students: development and psychometric evaluation. BMC Medical Education. https://doi.org/10.1186/s12909-026-10442-0

Image Credits: AI Generated

DOI: 10.1186/s12909-026-10442-0

Keywords: biostatistics, medical education, psychometrics, scale development, factor analysis, self-efficacy, evidence-based medicine, confirmatory factor analysis, reliability, measurement invariance, medical students, questionnaire validation