On this page

Synthesis: García-Alonso and colleagues (2024) surveyed 216 social science students in Spain on the technology acceptance model and analyzed the answer patterns as a coincidence network rather than as a regression. Perceived usefulness and credibility sit at the center of the network, and prior experience — including familiarity with video games and programming — predicts more favorable attitudes. The counter-intuitive result is the one worth carrying into training design: students who had received specific training from faculty reported lower confidence in the tool.

Key Findings

  • Usefulness and credibility organize the attitudes. The coincidence network places perceived utility and credibility centrally, which is the TAM prediction arriving in a Spanish social science faculty.
  • Prior experience predicts favour. Students with earlier exposure to ChatGPT, familiarity with video games and some programming knowledge held more positive attitudes.
  • Training is associated with less confidence, not more. Students who received specific training showed lower confidence in the tool, the opposite of what a skills-deficit account predicts.
  • Academic year and gender structure the network. Both variables appear as shaping attitudes and reported behavior rather than as controls.
  • The sample is 216 students and highly feminised. 72.3% female, from one faculty, by convenience sampling.

What the study did

The design is a convenience survey analyzed with reticular analysis of coincidences (RAC) using the CARING tool, which builds a network from co-occurrence of responses rather than estimating coefficients in a model. That choice suits exploratory description: it shows which variables cluster and which sit at the center, at the cost of the causal claims a structural model would attempt.

The training finding deserves care in interpretation. It is cross-sectional, so it can mean at least three different things: that training made students more critical and therefore less confident, that students who sought training were already the more skeptical ones, or that the training on offer was poor. The paper notes the first reading — that training raised critical awareness — which is also the most useful one for a faculty developer, because it reframes "confidence" as a possible symptom of uncritical acceptance.

Why the network view earns its place

Most TAM studies in this literature report a path model and a set of betas. The RAC approach instead asks which elements travel together, and it surfaces the demographic and experience variables as part of the structure rather than as covariates. For a faculty deciding where to intervene, that is a different kind of information: not "usefulness predicts intention" but "usefulness, credibility and prior tech experience sit together, and the trained students sit elsewhere."

The social science context also matters. These students are being trained to interpret evidence and to be skeptical of claims, and a survey that finds training lowering confidence is consistent with that training doing its job. The pedagogical implication is not to protect students from doubt but to give them the calibration skills to act while doubting.

What this means for practice

  • Instructors: do not measure the success of AI training by enthusiasm. Lower confidence after training can be the intended outcome if the training taught students what the tool gets wrong.
  • Faculty developers: target credibility explicitly. It sits at the network's center alongside usefulness, and it is the belief that outlasts a workshop.
  • Instructors: assume prior experience is unevenly distributed — gaming and programming familiarity both mattered here — and teach the tool rather than assuming it.
  • Programmes: build calibration into the curriculum: when to trust an output, what to verify, what to disclose.

Limitations

Convenience sample of 216 students in one Spanish social science faculty, 72.3% female, cross-sectional, with attitudes measured by self-report. RAC describes co-occurrence, not causation, so the training finding cannot distinguish selection from effect. The instrument relies on the technology acceptance model's constructs, which were designed for workplace systems rather than for a general-purpose text generator. Data were collected in 2024, early in ChatGPT's adoption curve.

Citation

García-Alonso, E. M., León-Mejía, A. C., Sánchez-Cabrero, R., & Guzmán-Ordaz, R. (2024). Training and Technology Acceptance of ChatGPT in University Students of Social Sciences: A Netcoincidental Analysis. Behavioral Sciences, 14(612).

Embed this page

Copy the code below to embed this page in a learning management system or other website. The embedded version shows only the page content — it hides the site header, navigation, and footer.