Table 3.

Standardized multiple linear regression results by architecture.

Model (parameters)	βEPSa	βsent	F test (df)	P valueb	Multiple R2
RoBERTa_base (125M)	0.04c	0.78c	2034 (22, 497)	<.001	0.6197
GatorTron_base (345M)	0.05c	0.79c	2236 (22, 497)	<.001	0.6417
RoBERTa_large (355M)	0.05c	0.76c	1918 (22, 497)	<.001	0.6057
GPT-2_large (774M)	–0.01	0.89c	4685 (22, 497)	<.001	0.7896

^aEPS: entities per sentence.

^bIndividual predictor P values for Beta_sent were <.001 for all models. P values for Beta_EPS were <.001 in all cases except for the GPT-2_large model where EPS was not a significant predictor (P=.184)

^cPredictor results are significant at the P<.01 level.