TABLE 4.

Linear mixed-effects analysis for models aiming to predict liking by the four aesthetic experience variables 3D impression, emotional value, interestingness, and powerfulness.

Tested model	df	AIC	logLik	R2	p(χ2)
Base Model #0	8	1442	−713	0.589
Model #1a	13	1433	−704	0.604	0.0021
Model #1b	13	1402	−688	0.628	<0.0001
Model #2	18	1400	−682	0.637	0.0305
Model #2	Estimate	t	df	p	Cohen d
FE 3D impression	0.014	<1	388.5	0.7262, n.s.	–
FE emotional	0.075	1.50	314.5	0.1335, n.s.	–
FE interesting	0.425	4.23	6.4	0.0047	0.3818 “medium”
FE power	0.363	5.34	8.7	0.0005	0.4820 “medium”

Base Model #0 contains only these four variables as fixed factors, plus participants and artworks as random intercepts. Model #1a and Model #1b add random slopes for interestingness and powerfulness by artworks and participants, respectively. Model #2 combines Models #1 and #1b by adding random slopes by artworks and participants. Best-fitting model, while being parsimonious, is indicated by bold face. FS, fixed slopes (fixed factors); df, degrees of freedom; R², coefficient of determination, based on the likelihood-ratio test; p(χ²), probability of accepting a significant effect despite a non-existent difference regarding the more complex versus the one-step less complex model. For the best-fitting model, statistics about fixed effects are given in detail. Effect sizes (expressed as Cohen d) are qualified according to the suggestions of Cohen (1988).