Mixed-Methods Data Analysis

1. Key Findings: The Three Student Profiles

Latent Class Analysis reveals that student attitudes toward GenAI can be grouped into three distinct behavioral clusters, summarized below. For the analysis, we used an n=178, since the rest of participants didn't have responses for all seven variables used by the model. The n values for each group are shown next to each group label, indicating that in our participant pool, the AI Resisters were the biggest group.

2. Methodology: Understanding Latent Class Analysis (LCA)

3. Latent Class Analysis Profile Matrix (Ordered by Survey Sequence)

This following table displays the average response scores (M) and descriptive labels for each student profile identified by the Latent Class Analysis (LCA). Average response scores were calculated by assigning sequential numbers to survey choices (shown in the first column), with descriptive labels mapped uniformly based on where each score falls on its scale. To give one example of the differences among the three groups notice the M values for Q12 Desire for AI Fluency and Skill Building. The AI Adopters have a very high desire (M=4.62) while AI Resisters have a very low desire (M=1.40). The table is connected to the interactive visualization in the next section. A selected table row (in highlighted color) is then displayed in the

4. Interactive Visualizations of Score Distributions

The table in the prior section shows only average values per class, not the distribution of responses per class. To see how responses are distributed per each question per class, we provide this interactive visualization. In every graph below (known as a Sina-Violin Plot), each dot represents an individual student's response score, somewhat scattered horizontally to prevent overlap. The colored outline (violin-shape) shows response volume across response levels, and the black diamond indicates the group average. Select an LCA variable on the left to see its distribution of scores across the three classes. Hover over the dots in each shape to see the number of respondents for each response choice.

5. Visualizing Responses by Profile

Once we had identified these three profiles, two members of the research team assigned the rest of survey participants to one of them, based on the open-ended responses. The size of the three classes changed as following: AI Resisters (n=105), AI Ambivalent (n=83), and AI Adopters (n=45), preserving the balance. The gallery below shows some of the survey responses this time indicating how participants in each class responded.

1 / 10

❮ ❯

Q12. Desire for AI Fluency

Desire for AI Fluency

I want to develop fluency with AI tools in order to be prepared for my future career.

6. Qualitative Analysis

As discussed in the Introduction & Methods section, we used a combination of human and GenAI skills to code the open-ended responses with qualitative codes (noting whether they are present or not in a response). Responses were assigned one or more codes, as appropriate. 32 codes were identified and applied to 819 open-ended responses. The table below shows the frequency distribution and participant engagement metrics across the qualitative coding categories. Click any column header to sort.