Table 2.
Model quartile decisions across input affiliations, chi-square test for independence P values, effect sizes, and 95% CIs.
| Models and affiliation level | Q1a, 50 (25%) | Q2b, 50 (25%) | Q3c, 50 (25%) | Q4d, 50 (25%) | P value | Cramer V (95% CI) | |||||||||
| Llama 3.3-70B | .80 | 0.05083 (0.047-0.127) | |||||||||||||
|
|
None | 37 | 148 | 15 | 0 |
|
|
||||||||
|
|
High tier | 48 | 139 | 12 | 1 |
|
|
||||||||
|
|
Low tier | 41 | 144 | 14 | 1 |
|
|
||||||||
| Mistral-7B | .63 | 0.04621 (0.027-0.120) | |||||||||||||
|
|
None | 54 | 101 | 45 | 0 |
|
|
||||||||
|
|
High tier | 55 | 112 | 33 | 0 |
|
|
||||||||
|
|
Low tier | 54 | 109 | 37 | 0 |
|
|
||||||||
| Gemma 2-9B | .08 e | 0.08408 (0.045-0.153) | |||||||||||||
|
|
None | 0 | 24 | 157 | 19 |
|
|
||||||||
|
|
High tier | 0 | 27 | 166 | 7 |
|
|
||||||||
|
|
Low tier | 0 | 24 | 168 | 8 |
|
|
||||||||
| DeepSeek r1-distill Qwen-14B | .87 | 0.04516 (0.041-0.123) | |||||||||||||
|
|
None | 3 | 73 | 113 | 11 |
|
|
||||||||
|
|
High tier | 1 | 83 | 103 | 13 |
|
|
||||||||
|
|
Low tier | 3 | 79 | 106 | 12 |
|
|
||||||||
| Qwen 2.5-7B | .05 | 0.10159 (0.071-0.161) | |||||||||||||
|
|
None | 6 | 51 | 143 | 0 |
|
|
||||||||
|
|
High tier | 2 | 69 | 127 | 2 |
|
|
||||||||
|
|
Low tier | 1 | 60 | 139 | 0 |
|
|
||||||||
aQ1: quartile 1.
bQ2: quartile 2.
cQ3: quartile 3.
dQ4: quartile 4.
eItalicization indicates relatively significant P values.