Self-Assessment Manikin (SAM) Scale

• Self-Assessment Manikin (SAM) Scale is a non-verbal, pictorial tool that measures affect by quantifying valence and arousal on a 1–9 scale.
• It supports both retrospective and continuous evaluation methods, enabling dynamic tracking of emotional states in immersive settings.
• Empirical findings demonstrate robust agreement and sensitivity of SAM measures, highlighting its effectiveness in reducing recall bias.

The Self-Assessment Manikin (SAM) is a non-verbal, pictorial assessment instrument developed to quantify core affective responses along several psychological dimensions. SAM provides numeric and visual measures of affective states induced by a stimulus, enabling systematic comparison of emotional responses in experimental and applied contexts. Traditionally, SAM captures three principal affective dimensions—valence, arousal, and dominance—using easily interpretable manikin figures rated on a 1–9 scale. Recent adaptations have leveraged the scale for both retrospective (post-stimulus) and continuous (in-stimulus) emotion measurement, especially in immersive media research (Voigt-Antons et al., 2020).

1. Conceptual Framework of the SAM Scale

SAM operationalizes emotional experience across three dimensions (Bradley & Lang, 1994): valence (pleasure–displeasure), arousal (calm–excited), and dominance (control–submissiveness). Each affects a distinct axis of emotional appraisal:

• Valence: Ranges from highly negative to highly positive affective responses.
• Arousal: Ranges from low (calm, sleepy) to high (excited, alert) levels of activation.
• Dominance: Ranges from feelings of subjugation or passivity to experiences of control or empowerment.

In the implementation described in “Comparing emotional states induced by 360° videos via head-mounted display and computer screen,” only the valence and arousal axes are measured, consistent with a growing trend to focus on these two dimensions in media and affective computing research (Voigt-Antons et al., 2020).

2. Administration Modalities and Visual Constructs

The two primary administration modalities are retrospective (post-stimulus) and continuous (in-stimulus):

• Retrospective Evaluation: Participants assign discrete integer ratings immediately after each stimulus presentation. This is achieved by selecting a single figure or point on a standard SAM pictogram for each dimension.
• Continuous Evaluation: Participants adjust or select positions in real-time on a two-dimensional arousal-valence coordinate plane that is displayed throughout stimulus presentation. The horizontal axis encodes valence, the vertical axis arousal, each spanning the 1–9 SAM scale.

The following table summarizes the representation and modality attributes for each evaluation mode:

Modality	Visual Representation	Interaction Method
Retrospective	Standard SAM pictograms (valence, arousal)	Mouse (screen); VR control
Continuous	2D arousal-valence grid (axes labels, no figures)	Mouse (screen); VR control

The implementation omits the dominance dimension; continuous rating replaces multiple discrete selections with free sampling across the affective space (Voigt-Antons et al., 2020).

3. Protocol and Data Acquisition

Retrospective SAM ratings are solicited immediately after each 60 s stimulus. For screen-based presentation, users interact with the SAM scale using a mouse; for head-mounted display (HMD) presentation, interaction occurs within VR using a controller. Participants receive definitions and a practice session before trials.

Continuous SAM ratings use a persistent arousal-valence grid during stimulus presentation. Participants are permitted to record self-assessment via clicks—any number, at any time. Each click records the position corresponding to participants’ current affective state.

Data extraction for continuous ratings involves computing the time-averaged score for each participant $p$, video $v$, system $s$, and method $m$:

$$\overline{R}_{p,v,s,m} = \frac{1}{N_{p,v,s,m}} \sum_{i=1}^{N_{p,v,s,m}} r_i$$

where $N_{p,v,s,m}$ is the total number of clicks and $r_i$ are the ratings across the interval (Voigt-Antons et al., 2020).

4. Analytical Approaches and Statistical Models

Data reduction aggregates ratings to obtain mean valence and arousal per participant, video, system, and method:

$$\overline{V}_{p,v,s,m} = \frac{1}{K} \sum_{k=1}^{K} v_k, \quad \overline{A}_{p,v,s,m} = \frac{1}{K} \sum_{k=1}^{K} a_k$$

where $K=1$ for retrospective ratings and $K=N$ for continuous ratings.

The experimental design applies a 2 (presentation system: HMD vs. screen) $\times$ 2 (rating method: retrospective vs. continuous) repeated-measures ANOVA for each dependent variable (valence, arousal, presence):

$$Y_{p,v,s,m} = \mu + \alpha_s + \beta_m + (\alpha\beta)_{s,m} + \varepsilon_{p,v,s,m}$$

Effect sizes are quantified via generalized eta squared ($\eta_G^2$); F-ratios are computed as $$F_{\text{effect}} = \frac{\text{MS}_\text{effect}}{\text{MS}_\text{error}}$$ with appropriate degrees of freedom.

Agreement between retrospective and continuous ratings is quantified via intra-class correlation coefficients (ICC), specifically using the two-way mixed absolute-agreement average-measures model (Voigt-Antons et al., 2020).

5. Empirical Findings in Immersive Media Assessment

Key results on SAM-based assessment of 360° video experiences via HMD and screen are as follows:

• Valence: No significant main effect of presentation system nor rating methodology; both retrospective and continuous modes produce highly correlated valence assessments (ICC = 0.80, $p<.05$).
• Arousal: Significant main effect of rating method; retrospective arousal ratings on screen presentation are systematically lower than continuous scores (retrospective $M = 4.70$, $SE = 0.30$; continuous $M = 5.64$, $SE = 0.29$; $F(1,13)=5.58$, $p=0.034$, $\eta_G^2=0.30$). No comparable difference emerges under HMD presentation (retrospective $M = 5.30$, continuous $M = 5.41$).
• Agreement: Good agreement is observed for arousal across methods (ICC = 0.673, $p<.05$). This suggests the scale robustly captures dynamic affective states but may underestimate arousal when administered retrospectively in less immersive settings (Voigt-Antons et al., 2020).

6. Modifications and Methodological Implications

Implemented adaptations include the exclusion of the dominance dimension and the introduction of a 2D arousal-valence grid for continuous, in-media rating. Pictograms are omitted in the continuous mode, retaining only numeric/label axes. This approach allows for in situ emotional tracking without disrupting presence or immersion, especially in VR contexts.

A plausible implication is that continuous, embedded self-assessment—particularly via a 2D plane—may offer advantages for recording rapid fluctuations in affective state, reducing recall bias and increasing ecological validity. The observed high ICCs between methods further indicate construct consistency across retrospective and continuous administration, despite mean rating discrepancies in specific contexts.

7. Future Directions and Considerations

The continuous, 2-dimensional SAM variant provides a promising framework for affective measurement in virtual environments and other high-immersion media settings. The possibility of integrating such continuous self-report measures with physiological or behavioral indices may further enhance sensitivity to complex emotional dynamics. Moreover, systematic investigation of dominance (currently omitted) and the refinement of real-time sampling interfaces could yield richer multidimensional affective profiles.

In summary, the SAM scale, particularly its two-dimensional adaptation, demonstrates methodological flexibility and robust psychometric properties under experimental manipulation of media presentation and rating method, supporting its continued adoption in affective and immersive media research (Voigt-Antons et al., 2020).