Table 3.

Adjusted Pearson residuals of large language model decisions.

Models and affiliation level			Q1a (adjusted Pearson residuals)		Q2b (adjusted Pearson residuals)		Q3c (adjusted Pearson residuals)		Q4d (adjusted Pearson residuals)		Q1 and Q2 (adjusted Pearson residuals)		Q3 and Q4 (adjusted Pearson residuals)
Llama 3.3-70B
	None	−1.063		0.8343		0.4576		−1.0017		−0.2288		−0.5440
	High tier	11.2757		−0.8985		−0.5720		0.5008		0.3773		−0.0712
	Low tier	−0.2126		0.0642		0.1144		0.5008		−0.1484		0.6152
Mistral-7B
	None	−0.0649		−1.0999		1.4668		0		−1.1648		1.4668
	High tier	0.1298		0.8105		−1.1734		0		0.9403		−1.1734
	Low tier	−0.0649		0.2895		−0.2934		0		0.2246		−0.2934
Gemma 2-9B
	None	0		−0.2619		−1.4974		2.8717 e		−0.2619		1.3743
	High tier	0		0.5237		0.5240		−1.6231		0.5237		−1.0991
	Low tier	0		−0.2619		0.9732		−1.2486		−0.2619		−0.2753
DeepSeek r1-distill Qwen-14B
	None	0.5377		−0.9462		0.9841		−0.3647		−0.4086		0.6195
	High tier	−1.0753		0.8280		−0.7526		0.3647		−0.2474		−0.3879
	Low tier	0.5377		0.1183		−0.2316		0		0.6559		−0.2316
Qwen 2.5-7B
	None	2.1374		−1.7008		1.2394		−1.0017		0.4366		0.2377
	High tier	−0.7125		1.7008		−1.7352		2.0033		0.9884		0.2682
	Low tier	−1.4249		0		0.4958		−1.0017		−1.4249		−0.5059

^aQ1: quartile 1.

^bQ2: quartile 2.

^cQ3: quartile 3.

^dQ4: quartile 4.

^eItalicization indicates statistically significant residuals.