In order to assess the feasibility of obtaining and determining the structure of valence and arousal ratings in a large, population-based, observational, epidemiologic study, the analyses reported here explored the relative importance of four variables of interest (age, sex, depression, and anxiety) for valence and arousal ratings of a small number of pre-selected unpleasant, neutral, and pleasant words. The variables were selected based on previous studies which indicated associations of those variables with affective quality perception or experience, established examples of which include an age-related positivity effect , increased emotion recognition capabilities of women , and dysfunctional affective experience and perception during states of depression or anxiety . Valence and arousal ratings were employed, because these two dimensions represent essential qualities of affective experience and affective quality perception [4, 10]. Written words were used as stimuli, because they provide easy and unobtrusive access to the human affect interface . Moreover, words enable the control for certain psycholinguistic measures known to influence cognitive processing .
Compared to previous valence and arousal investigations, the present study is characterized by certain features, which have the potential to extend existing findings. First and foremost, while previous studies had most often taken the form of laboratory experiments, the present study utilized the setting of a large, population-based, observational, epidemiologic study, the BiDirect Study . Second, while previous studies often included rather small samples, the present study analyzed ratings of more than 700 participants. Third, while previous studies usually included young (often predominantly female) university students, the BiDirect Study had recruited a sex-balanced sample of middle-aged to elderly adults randomly invited from the general population. Finally, while previous studies had usually focused on only one or two variables of interest, the present study explored the roles of four relevant explanatory variables.
From our point of view, the rating patterns observed in the present study (Table 1, Additional file 1: Figures S1 and S2) indicated that it was feasible to obtain valence and arousal rating data of acceptable quality in the context of the BiDirect Study. Taking into account existing methodological differences (e.g. the number of words to be rated, the rating context, or the underlying population), this view is further supported by the comparison of the present rating results with ratings from previous studies [46, 54] for overlapping nouns (Additional file 1: Table S1). The ratings obtained in the present study consistently fall in between those from Kissler et al.  and Kanske et al.  and exhibit similar variances, despite large heterogeneity of the rating population. However, the predictors used in the present study systematically explained only very little variance in the data, with significant effects occurring only for arousal, but not for valence.
There is an ongoing debate regarding the relationship between valence and arousal, and it is currently unresolved whether valence is better represented by a bipolar or by two unipolar dimensions [4, 7, 8, 12]. However, it can be stated that the average valence and arousal ratings collected here (Table 1; Additional file 1: Figures S1, S2) exhibited the expected distributions with regard to the pre-defined word categories. Moreover, the mean ratings appear to be rather stable (Table 1). Thus, altogether the present results do not argue against a further pursuit of emotional (affective) epidemiology . This may be worthwhile given that emotions and affect matter for health and health perception : frequent or chronic experience of negative emotions (e.g. sadness, fear) and affect may influence the development of somatic conditions such as e.g. infectious disease  or lung dysfunction  and may exacerbate chronic somatic diseases such as diabetes , arthritis , cancer , and cardiovascular disease . Frequent experience of positive emotions (e.g. enjoyment) and affect, in contrast, likely exert protective effects on health , e.g. by reducing the risk of developing cardiovascular disease .
The present study found that the proportions of variance explained by the specific linear model applied (which included the predictors age, sex, depression, and anxiety, plus the adjusted variables education and examiner) were in a range between 1.3 and 5.4% and thus very low. In the valence dimension, the model even failed to reach significance for each of the three words categories (due to this finding, we refrain from interpreting any effects regarding the valence dimension). This indicates that the valence ratings were largely unaffected by the selected predictors, and that there may be other, potentially more relevant variables that were not included in the present study. Besides the inherent affective properties of the words themselves, likely candidates include situational, personality-, language-, or culture-related factors.
The modeling results regarding the arousal ratings for the three word categories showed small effects, which were slightly larger for unpleasant (5.4% explained variance) and neutral (4.6%) compared to pleasant (3.5%) words. Interestingly, the explanatory variables of interest as a group tended to explain the most variance in case of arousal by neutral words (Table 4, Fig. 2). This could be taken as a hint that the arousal dimension of affective quality perception, especially if accessed by rather neutral (or possibly “ambiguous”) stimuli, may be comparably well-suited to detect subtle inter-individual affective tendencies or differences.
Figure 2 (right column) displays that the relative importance patterns (i.e. the degrees of variance explained by the four explanatory variables of interest relative to each other) qualitatively differed between the three word categories in the arousal dimension. Thus, the very same variables tended to explain more or less variance depending on how a group of people had pre-categorized the utilized word stimuli (unpleasant, neutral, or pleasant). This impression is corroborated by the statistical effects regarding the individual predictors.
In the present study, (trend) effects of sex on arousal ratings were found for all word categories (Table 2). In case of unpleasant words, female sex was associated with increased arousal. This effect was also detected in all sensitivity analyses (Additional file 1: Tables S2, S3, and S4). In case of neutral and pleasant words, female sex was associated with decreased arousal. Noteworthy, the latter effects were not found in the third sensitivity analysis (Additional file 1: Table S4), where additional (possibly non-compliant) participants had been excluded.
Sex has previously been found to influence affective perception  and experience [20, 81,82,83,84]. Previous findings indicated that women are possibly more emotionally reactive and receptive. Moreover, sex-related differences were more consistently observed with respect to unpleasant emotions [85, 86]. This is in line with the present result that the sex effect on arousal ratings seemed to be most consistent in case of the unpleasant words. Furthermore, sex-specificity during emotional processing was reflected in a tendency of women to “prefer” unpleasant stimuli, which was in contrast to a preference for pleasant stimuli in men . This is in line with the present finding that women rated the unpleasant words to be more arousing, but the pleasant (and neutral) words to be less arousing compared to men.
More specifically, previous findings regarding effects of sex on arousal ratings of words were mixed . Redondo et al.  and Gilet et al.  did not find any sex-related differences. Soares et al.  found that women gave higher mean arousal ratings overall. Partly in line with the present results, Grunwald et al.  found that women gave higher mean arousal ratings for pleasant and particularly for unpleasant words. Also partly in line with the present results, Söderholm et al.  found that women rated unpleasant and neutral words as more arousing compared to men.
In the present study, an effect of age on arousal ratings was found for the neutral words (Table 2). Older age was associated with increased arousal ratings. This effect was also detected in all applicable sensitivity analyses (Additional file 1: Tables S2, S4). Moreover, this effect was the statistically most robust effect among effects of individual predictors of interest.
Aging has previously been found to influence affective perception  and experience, whereat the age-related positivity effect [81, 82, 88] is one prominent example. Findings regarding an influence of age on arousal ratings of words were again mixed, which may at least partly be due to methodological differences between studies . Using a SAM scale, Keil and Freund  found that older adults gave higher arousal ratings for unpleasant words compared to young adults. However, using a seven-point Likert scale (“very relaxed” to “very tensed”), Grühn and Smith  found that older adults gave lower arousal ratings for unpleasant words and higher arousal ratings for pleasant words compared to young adults. Using a seven-point Likert scale (“very calming” to “very arousing”), Söderholm et al.  reported complex results. Using a seven-point Likert scale (“very relaxed” to “very tensed”), Gilet et al.  found that middle-aged and older adults gave higher mean arousal ratings compared to young adults; this finding is basically in line with the results of the present study. Eventually, using a six-point Likert scale (“not at all intense” to “very intense”), Grunwald et al.  found no age-related differences in arousal ratings for unpleasant or pleasant words. However, in line with the findings of the present study, older adults gave higher arousal ratings compared to young or middle-aged adults for neutral words.
In the present study, age was modeled as a continuous variable, enabling extrapolation: for instance, the effect size point estimate for age observed here (0.033; Table 2) indicates that on average, holding the other variables in the model constant, becoming 30 years older (at least in the age range covered) would go along with an increase in average arousal ratings of neutral words of one point on the nine-point scale applied here.
In the present study, an effect of anxiety on arousal ratings was found for the neutral words (Table 2): anxiety was associated with increased arousal. This effect was also found in all applicable sensitivity analyses (Additional file 1: Tables S2, S4). Notably, no effects of depression on arousal ratings were found in any of the word categories in any of the analyses (Table 2; Additional file 1: Tables S2, S3, and S4).
The literature regarding influences of anxiety on arousal ratings of words is sparse. A study by Kanske et al.  probed the influence of depression and anxiety on valence and arousal ratings of auditorily presented words. The results indicated that higher anxiety scores were associated with higher arousal ratings of unpleasant words (r = 0.43). In line with the results of the present study, which showed a positive association between anxiety and arousal by neutral words (Table 2), the correlation between anxiety scores and arousal ratings of neutral words reported in Kanske et al.  was also positive (r = 0.11; however, this correlation was not significant).
Taken together, the results of the present study indicated that across the three word categories, with respect to the predictors of interest that were explored here, it seems that sex tended to be important in principle, while the importance of age and anxiety was specific for non-toned (i.e. neutral) information. Noteworthy, anxiety tended to explain more variance than depression (Fig. 2), although both were binary variables. However, this finding should be interpreted with caution, bearing in mind the strong comorbidity of depression and anxiety , a reality which was also reflected by the association between the depression and anxiety predictor variables (Additional file 1: Figure S3). However, the number of participants who fell into respective categories (depression only: N = 62; anxiety only: N = 52; both depression and anxiety: N = 55) was very similar, and the age and sex distributions were largely comparable between categories. Thus, the finding that anxiety stood up to depression in the present context of an affective quality perception task may be informative.
Strengths, limitations, and future directions
In the present study, a large, sex-balanced sample of middle-aged to elderly participants, who were randomly invited from the general population of a mid-sized, north-western European university city, was analyzed. Overcoming small and often sex-dysbalanced participant samples of previous studies, the present study investigated valence and arousal ratings of unpleasant, neutral, and pleasant German nouns and simultaneously explored the relative importance of four relevant variables with regard to six relevant outcomes of affective quality perception along those dimensions. However, certain limitations need to be considered. First, given the observational data, unmeasured or residual confounding cannot be ruled out. Second, the very low proportions of explained variance suggest that there may be other, potentially more relevant explanatory variables in addition to those considered here. Candidates include e.g. certain personality dimensions, most likely extraversion and neuroticism [4, 11]. Third, potential data acquisition problems (deficient task comprehension, non-compliance) might have, in addition to idiographic variability, contributed to observed “deviant” ratings and non-monotonic frequency distributions. Moreover, the data were collected exclusively in a Western cultural context. Furthermore, the perceived communicative context, which was not considered here, may play an important role (e.g. [91, 92]). A further limitation to age as a predictor is that the sample did not include young adults. Finally, the words used here as stimuli are somewhat heterogeneous, as they include both emotion-terms (i.e. hate, envy) as well as emotion-laden terms (e.g. corpse, holidays). Previous research has shown processing differences between those two categories (e.g. ) that might extend to processing malleability by individual differences . With regard to future directions, it is noteworthy that follow-up studies could use more, more frequent, less heterogeneous, and less abstract word stimuli and an implicit instead of an explicit task (e.g. the so-called emotional stroop paradigm ).