2.4. Data analysisIn order to accou

2.4. Data analysis
In order to account for the nested structure in the data, analyses are based on hierarchical linear modeling (HLM) (Raudenbush et al., 2004). Subjective stress, sCort, sAA, and items on music listening (music episode (yes/no), perceived valence, perceived arousal, reasons for music listening) were considered level-1 variables. In analyses concerning the relationship between music episode and sCort, time of day has been added as predictor (UC = −0.45, t(363) = −16.57, p ≤ 0.001). However, in analyses concern- ing valence/arousal and reasons for music we refrained from adding time of day as predictor for two reasons: First, the number of level-1 observations is already relatively small, as only music episodes could enter analyses. Therefore we aimed at models which were parsimonious in order to avoid type-B error inflation. Second, as music listening was most often reported in the late afternoon and evening (both in the data set in which all observations were included (UC = 0.03, t(2119)=2.914, p=0.004) as well as in the data set in which only the observations were included from partici- pants who provided saliva samples (UC = 0.05, t(361) = 2.151, p = 0.032)), we assume that the role of diurnal variations in sCort was kept to a minimum in these analyses. At the individual level (level-2), the intercept (ˇ0) was modeled as a function of gender (