Introductory videos tailored to student interests boost data science test scores

A five-year experiment at the University of Tsukuba has found that something as simple as the first video a student watches can measurably change how well they learn data science. Researchers report that when first-year students were shown introductory videos connected to the research topics of their own colleges, their scores on an objective test of course content rose by roughly 16 percent.

The course in question, called simply Data Science, has been mandatory for all first-year students at the University of Tsukuba since 2019, regardless of whether they study literature, medicine, sports science, engineering or art. Each academic year, about 50 lecturers taught about 50 parallel classes of the same course. The course ran for ten weeks between October and December, with two 75-minute periods each week. The first weeks covered data management, the middle weeks focused on data analysis, and video lectures on big data, the internet of things, cyber-physical systems and machine learning punctuated the schedule. Students practiced with exercises built in Microsoft Excel. Classes were held face-to-face in 2019, 2022 and 2023, and on-demand in 2020 and 2021 because of the COVID-19 pandemic.

The research team, led by Yoshito Hirata, Kazuto Fukuchi, Naoya Todo, Etsuko T. Harada and Jun Sakuma, framed their work around challenges raised in the data science education literature: how to design balanced course content, how to include field-specific application knowledge, and how to teach students with weaker mathematical or computational preparation. Because data science sits at the intersection of statistics, computation and domain knowledge, the authors argue that the way those intersections are introduced may matter most for students who arrive with less quantitative training.

In a preliminary analysis of the 2019 cohort, the researchers used a multi-layered regression analysis, dividing questionnaire items into six ordered groups and fitting multivariate linear regressions with the lm function in R Studio. They found that developing an interest in data science was fundamental to deeper understanding: interest was linked to how many technical terms students believed they could explain, and the objective test score was related to the existence of a related introductory video, the number of exercises completed and the number of technical terms students felt able to explain. Subjective understanding, in turn, was governed by the presence of a related introductory video, interest in data science, satisfaction with the course and self-assessed term knowledge.

Building on those correlations, the team introduced two measures from 2020 onward. First, they prepared a growing library of introductory videos, chosen year by year under time and budget constraints, covering as many of the university’s colleges and specialties as possible. Second, they designed an exercise in which students answered a questionnaire about topics chosen to interest them, then analyzed the anonymized answers themselves, usually in the second week, with the results returned to them as exercise material from the sixth week of the same year or the second week of the following year.

The video effect showed up most clearly in what the researchers describe as natural experiments. In 2020, at colleges where related introductory videos were shown for the first time, the objective test score increased by 15.1 percent on average; for scores of 4 or higher on the six-point test, the odds ratio was 1.72 with a p-value of 0.00036. In 2021, the team expanded the video library to cover specialties previously omitted and found that in the newly covered colleges the objective score tended to rise by 16.4 percent on average, with an odds ratio of 2.5 for scores of 4 or higher and a p-value of 0.00016. By contrast, colleges where a related video had already been introduced by 2020 showed an odds ratio of 1.29, and colleges still without a related video showed an odds ratio of 0.95, which was not statistically significant.

The effect also depended on students’ prior preparation. Participants who had not studied Math II or Math B in Japanese high school, subjects covering calculus and statistical inference respectively, were more likely to be motivated by the introductory videos. The authors suggest that a well-tuned introduction may act as a bridge for students with weaker mathematical backgrounds, helping them see why data science matters in their own field before the technical content begins. Students who reported being motivated by the videos were also 1.6 times more likely to say they could explain the term primary data and 1.4 times more likely to explain stacked area chart, terms connected to real-world examples shown in the videos.

The survey-based exercise produced its own measurable effects. Students who reported being motivated by analyzing their own questionnaire answers were 1.6 times more likely to say they could explain multiple choice answer format and 1.5 times more likely to explain contingency table, both concepts they had practiced directly in the exercise. Comparing odds ratios across the study’s tables, the researchers found the impact of the survey exercise was slightly greater than that of the introductory videos, both on explainable technical terms and on subjective understanding. Because the two measures can be applied independently, the team concludes that using both together would be appropriate in practice.

Across the datasets from 2019 to 2023, objective test scores, developed interests, course satisfaction and subjective understanding saturated by academic year 2021, according to the study’s figures. The effects of the videos and exercises on subjective interest appeared stronger during the in-person years of 2022 and 2023 than during the on-demand years of 2020 and 2021, but the authors caution they could not rule out alternative explanations, such as year-by-year improvements in course materials. They note that the relationship between students’ interests and objective test scores was significant in the on-demand years but not in the in-person years, while most other outcomes showed nearly comparable effects across both formats.

The researchers are careful about the limits of their work. They describe the study as exploratory, conducted at a single university, and say the findings should be examined more rigorously through a randomized trial. They also acknowledge that they could not evaluate how much self-referencing, the memory advantage of analyzing data about oneself, contributed to the exercise’s success, because the study lacked a proper control group. Raw student-level data cannot be shared for privacy reasons, as they are owned by the University of Tsukuba. Still, the authors hope the results will encourage lecturers worldwide to craft course introductions that cover students’ diverse interests, noting that the resolution of the college, rather than the whole university, appears to be the right granularity for introductory stories that spark intrinsic motivation from the very first class.

Subject of Research: How interest-matched introductory videos and self-referential survey exercises affect data science learning among university students

Article Title: Targeting students’ interests to facilitate their learning of data science

Article References: Hirata, Y., Fukuchi, K., Todo, N., Harada, E. T., & Sakuma, J. (2026). Targeting students’ interests to facilitate their learning of data science. Discover Data, 4(1), Article 1. https://doi.org/10.1007/s44248-026-00101-6

Image Credits: AI Generated

DOI: 10.1007/s44248-026-00101-6

Keywords: data science education, University of Tsukuba, intrinsic motivation, introductory videos, student interests, objective test scores, survey exercises, self-referencing, higher education, regression analysis, Targeting, students