Abstract
This study examines the dynamic, real-time interplay between the emotional content of political television ads and individuals’ political attitudes during ad processing based upon the Dynamic Motivational Activation (DMA) theoretical framework. Time-series cross-sectional models were developed to test the effects of three motivational inputs of emotional ads (arousing content, positivity, and negativity) and viewers’ evaluation of the featured candidates on four psychophysiological responses (heart rate, skin conductance level, corrugator electromyography, and zygomatic electromyography). As predicted by the DMA, physiological responses during ad viewing were affected by their own first- and second-order dynamic system feedback effects. These results not only support the predicted dynamic nature of the physiological system but also help disentangle message effects from the moderating and accumulating effects of the physiological system itself. Also as predicted, message motivational inputs interacted with viewers’ political attitudes to determine psychophysiological responses to the ads. Supporters of opposing political candidates showed cardiac-somatic response patterns indicative of disparate attention to the advertised information. Attentional selectivity can be a critical component in determining how information processing influences campaign message reception and effects.
Keywords
Participatory democracy is founded on the assumption of an informed public (Dahl, 1989), where understanding of cross-cutting political viewpoints is believed to be essential (Barber, 1984). To the laments of political scholars, however, political ignorance is one of the most well-known and consistent findings of contemporary political science research (Delli Carpini & Keeter, 1996). Lack of knowledge is caused, in part, by self-selective exposure to information, both mass-mediated and interpersonal, which reinforces rather than challenges an individual’s beliefs (Iyengar & Hahn, 2009). In fact, selective exposure has led to the counter-intuitive phenomenon that political polarization increases with increased availability of diverse information sources (Jacobson, 2006).
In addition to selectivity in behavioral choices, political beliefs and attitudes may bias the actual processing of political information. This selective processing may further widen political divides and contribute to polarization in the public’s knowledge and perceptions. For instance, those who support socially protective political policies show stronger physiological responses to threatening pictorial information (Oxley et al., 2008). Thus, media reports with threatening images might create a divergence in threat perceptions among liberals and conservatives. Additionally, the phenomenon called “hostile media” suggests that partisans on either side of a political issue perceive media coverage, rated neutral by nonpartisans, as hostile to their point of view (Gunther & Liebhart, 2006; Vallone, Ross, & Lepper, 1985). Perceptions of media bias could contribute to more extreme political beliefs among opposing partisans.
Extending the observation of overt selective exposure to covert selective processing, we propose that even when exposed to the exact same information, individuals, driven by their political beliefs and attitudes, process the information in a highly selective manner. This study uses real-time physiological measures and dynamic modeling to examine disparate attention to televised political campaign ads as a function of message emotional content and viewers’ attitudes toward party candidates in a presidential election. Attention is defined as the allocation process of limited mental resources, leading to selectivity in sensory information intake.
Selective attention to political ads warrants scrutiny for several reasons. First, political ads are a crucial component of our information environment during political campaigns (Cho, 2008). The most recent Wisconsin Advertising Project (2010) reports that in 2008, in the United States, more than 2.1 million political television ads were aired in the largest 100 media markets at a cost of over US$1.13 billion. Political ads offer important information to the public, such as candidates’ issue positions, and encourage further information seeking and interpersonal conversation (e.g., Benoit, Hansen, & Holbert, 2004; Pan, Shen, Paek, & Sun, 2006). Second, the majority of political ads are strategically placed in critical battleground states in an effort to inform and mobilize the apathetic, undecided, or opposing voters (Wisconsin Advertising Project, 2010). Importantly, this investment model has not carefully calculated how selective attention of voters may influence the effects of ad bombardment. Lastly, it has been suggested that saturation and dissemination of televised political ads creates a less self-selective information context than news media (Cho, 2008). Citizens are more likely to encounter dissimilar political views through news media than interpersonal communication contexts (Mutz & Martin, 2001). As control over ad exposure is presumably more difficult than control over news exposure, televised political ads seem to be one of the most important information channels for minimizing selective exposure and broadcasting political views across lines of ideological difference.
In this article, we begin with a brief review of emotional political ads and motivated attention. We argue that cognition and media processing should be theorized and analyzed as a dynamic system. A dynamic motivational model is introduced to formalize psychophysiological responses (indicative of real-time attention) as dynamic responses to the interactions between message motivational inputs and individuals’ attitudes toward the two major party candidates in the 2008 presidential election. Data from an experiment are used to test this model.
Emotional Political Ads
Emotional appeals are prevalent in political ads. Political ads are often broadly categorized as positive, negative, or comparative. Positive ads highlight the supported candidate’s merits and strengths, whereas negative ads focus on the weaknesses of the opposing candidate (Shapiro & Rieger, 1992). Comparative ads highlight both the supported candidate’s strengths and the opponent’s weaknesses (Shah et al., 2007). Although more specific distinctions exist, this emotional valence-based categorization is the most common. Research based upon this categorization has generated valuable insights on the effects of the emotional appeals of political ads (e.g., Brader, 2006; Westen, 2007).
Although this categorization is advantageous for its simplicity, it ignores important emotional dynamics across a message. For example, an ad categorized as negative is not simply negative throughout the ad. President Lyndon Johnson’s notorious negative ad during the 1964 campaign, Daisy Girl, serves as a good example. The ad opens with a young girl standing in a field, counting as she picks off the petals of a daisy. Birds chirp in the background. A narrative voice then launches into an ominous countdown, the end of which signals a nuclear explosion. Thus, as the ad unfolds across time, the emotional tone and intensity change remarkably. The first half of the ad is peaceful and pleasant. Then the countdown commences, adding suspense and emotional intensity. Finally, the nuclear explosion at the end is one of the most infamously negative scenes from a political ad. Emotional content in this ad changes from positive to negative, and from calm to frighteningly intense. Similarly, other political ads might start off negative and depressing by, for instance, presenting economic problems, only to become positive and encouraging at the end of the ad, promoting promises and solutions offered by the candidate. In addition, the tone of an ad might be negative throughout but with varying intensity. In other words, when we simply categorize an ad as positive, negative, or comparative, we have overlooked a rich amount of information on the dynamic changes of the emotional content. One negative ad can be exceedingly different from another. Furthermore, even if we keep the emotional content of ads the same or similar, we can induce notably different viewer responses by simply altering the sequential order of the emotional content (Harvey, 1990; Wang, Lang, & Busemeyer, 2011). Imagine, for example, how viewers’ responses to the Daisy Girl ad might change if the nuclear explosion were moved to the beginning of the ad.
In addition to potential over-simplification, the positive/negative/comparative ad terminology derives from a focus solely on the candidates featured in the ads or sponsoring the ads (i.e., intended emotional responses) without considering the viewers. The viewers’ predispositions should influence actual evoked responses. A negative ad attacking a particular candidate may elicit negative responses from those who support the candidate, but may cause a sense of positive agreement among those who oppose the candidate. In other words, viewers may experience complex and perhaps mixed feelings. The motivational model of emotion, reviewed below, provides a parsimonious conceptualization of this intricate experience.
Emotion, Motivation, and Motivated Attention
Emotions serve an adaptive function, evolving to facilitate organisms to appropriately interact with their environments. Emotions are dispositions to act, although they do not always manifest in overt behavior (Frijda, 1987). Many discrete emotions (e.g., fear, anxiety, and happiness) can be mapped onto a dimensional space reflecting the fundamental appetitive and aversive motivational systems that have evolved to help organisms survive and thrive in their environments (for a review, see Bradley, 2000; for similar theories, see Eysenck, 1967; Fowles, 1993; Gray, 1981). The two motivational systems are activated by emotional inputs from direct experience with the physical world as well as indirectly through mediated experiences (Lang, 2006; Reeves & Nass, 1996), such as televised political ads (Bradley, Angelini, & Lee, 2007). The valence (positive, negative) of the stimulus determines which motivational system(s) is (are) activated, whereas the arousing content (i.e., intensity of the emotion) of the stimulus determines the strength of activation. Motivational activation, in turn, “initiates a cascade of sensory and motor processes, including mobilization of resources, enhanced perceptual processing, and preparation for action” (Bradley, 2009, p. 1). The appetitive system is activated by positive stimuli, which facilitates approach tendency or behaviors, including mobilizing attentional resources for sensory intake. The aversive system is activated by negative stimuli such as threats. With increased intensity of negative stimuli and greater activation of the aversive system, the attentional pattern often switches from information intake (i.e., initially to identify the threat) to information rejection (i.e., preparation for fight or flight; Lang, 2006). In many situations, the two systems are negatively correlated. However, theoretically they are independent (Cacioppo & Berntson, 1994); and depending on the context, their correlations can vary from −1 to 0 and may even be positive (e.g., Zautra, Berkhof, & Nicolson, 2002).
This motivational interpretation of emotion and the theoretical framework of motivated attention have been widely used in basic neuropsychological, psychological, and applied communication research. The two dimensions of message emotion, valence and arousing content, have been useful in parsimoniously explaining the effects of emotional messages on attentional, affective, and behavior responses (for reviews, see Ravaja, 2004; Lang, 2006). A recent study (Wang et al., 2011) using stochastic dynamic analysis found that in a user-controlled entertainment television viewing context, continuous inputs of negativity, positivity, and arousing content of programs explained the majority of variance in real-time psychophysiological responses that indicate attentional effort and autonomic arousal. Applying a motivational theoretical framework to political ads processing, Bradley et al. (2007) revealed that negative television political ads elicited reflexive preparation to move away as indicated by psychophysiological measures. A natural follow-up question is, “How do viewers’ political attitudes influence motivated selective attention?”
If our information landscape is sculpted by the general appetitive and aversive nature of the information stream, it is our individual motivational traits, emotional states, and attitudes that orchestrate how we perceive and interact with this continuously changing landscape. Individual differences in reactivity of the aversive and appetitive motivational systems moderate selective attention to various emotional stimuli, including pictures and videos (e.g., Lang, Wang, & Bradley, 2004). Additionally, differences in emotional states and attitudes can influence attention. For example, those who are anxious and depressed show an attentional bias to negative information (Mogg, Bradley, & Hallowell, 1994).
In fact, attitudes have been interpreted and studied through the motivational model of emotion. An attitude represents an evaluation of a stimulus, indicating a general orientation and direction of action toward the stimulus. Valence (i.e., positive or negative evaluation of the stimulus) and arousal (i.e., intensity or strength of the evaluation) have been recognized as important aspects of attitudes, even characterized as “currently among the best understood biological aspects of evaluation” (Cunningham, Packer, Kesek, & Bavel, 2009, p. 489). Similar to the motivational interpretation of emotion, the direction of an action or action readiness reflects attitudinal valence, whereas intensity corresponds to arousal. The peripheral nervous system (PNS) prepares the body for actions and shows various patterns of responses, even if the action is eventually inhibited (Cacioppo & Berntson, 1994; Cunningham et al., 2009). Therefore, different patterns of PNS responses, such as heart rate (HR), skin conductance level (SCL), and facial electromyography (EMG), should be affected by not only the general motivational inputs from the stimulus but also individuals’ attitudes to the stimulus. Divergent attitudes can lead to divergent allocation of attentional resources to the exact same stimuli, resulting in information intake or rejection patterns.
This study expands previous work by examining whether selective attentional processing is determined by the interactions among emotional content features of messages and individual differences in political attitudes. In a highly salient campaign, such as the 2008 presidential election, the political attitudes most likely to moderate responses to political ads are evaluations of the competing candidates. From a dynamic system viewpoint, the negativity, positivity, and arousing content of ads are considered motivational inputs because they have motivational significance that influences physiological system outputs. We analyze formal dynamic models that include or exclude the interaction effects between message motivational inputs and viewers’ evaluations of the candidates. By comparing these models, we test the following hypothesis:
More specifically, we propose that
Dynamic Motivational Processing
As argued earlier, we may gain a richer understanding of the effects of political ads by viewing them as dynamically changing inputs. A more compelling and essential reason for formalizing political ad processing as a dynamic process is that cognitive functions are dynamic and complex in nature. Dynamics are fundamental for understanding human cognition and behavior (Kelso, 1995; Ward, 2002). Political ad processing is no exception. Extant studies of message effects, however, typically employ static analysis and may therefore overlook the inherent self-generating function of human cognition. According to mounting evidence from neuroscience and cognitive psychology, along with the reinvigorating complex system approach to understanding our brains and cognition, “self-organization” is “a fundamental brain operation” (Buzsáki, 2006, p. 10). More specifically, “. . . most of the brain’s activity is generated from within, and perturbation of this default pattern by external inputs at any given time often causes only a minor departure from its robust, internally controlled program” (Buzsáki, 2006, pp. 10-11). This dynamic system approach to cognition emphasizes the nonlinear relationship between the system’s interconnected components, time dependence, and feedback loops (Buzsáki, 2006, p. 11). To accurately estimate media effects, the endogenous self-organizing and self-generating feedback effects from our cognitive systems need to be explicitly estimated and disentangled from the exogenous effects of media inputs. Time also needs to be explicitly included in both theory and analysis to prevent confounding of media effects per time unit with media effects aggregated by system feedback effects over a certain duration. Intuitively, this argument is similar to rationales for research examining effects of media contexts (e.g., Potter, LaTour, Braun-LaTour, & Reichert, 2006) and extended exposure (e.g., Thomas, 1982). At any time point, affective and cognitive responses are intimately linked to prior responses, and integrate with the prior responses to determine how an individual is affected by an exogenous stimulus at the time. To accurately estimate the effects of exogenous stimuli, such as the effects of media motivational inputs, dynamic models are needed to tease apart the influence of message content, processing system feedback, and time. Furthermore, with a solid understanding of the system elements at the level of per time unit, we then can put all of these process components together to examine how they interact and integrate with each other to generate outputs at the system level across time. A precise estimation of message effects and processing feedback effects per time unit allows greater generalizability and predictive power of the model across various manipulations and data sets, such as message stimuli with different time durations and different motivational content inputs.
Dynamic Motivational Activation model (DMA, Wang et al., 2011; Wang & Busemeyer, 2007; Wang & Tchernev, 2012) takes a dynamic system approach to studying media processing and choices. Built upon motivational processing theories as reviewed earlier, DMA emphasizes the central role of motivational activation in information processing and choice behavior, and aims to explain how the effects evolve dynamically across time. DMA aims to specify and quantify physiological responses (indicating real-time cognitive and affective responses) during message processing as a dynamic function of the exogenous media or message motivational inputs as well as the endogenous feedback effects of the physiological (cognitive and affective) systems. In a recent study by Wang et al. (2011), a second-order stochastic difference equation model with delayed input effects was used to formalize several hypotheses of DMA using HR, SCL, corrugator EMG, and zygomatic EMG measures. As predicted by DMA, large variance in psychophysiological responses across time during the viewing of video clips was explained by the dynamically changing motivational inputs of the video. In addition, Wang et al (2011) estimated the time it takes the motivational inputs to reach and produce an effect on the physiological systems. Most importantly, the physiological responses showed significant first- and second-order feedback effects (i.e., lag 1 and lag 2 autoregressive terms of the dependent variable in the model), providing direct support for dynamic system message processing models. These feedback effects integrate the system’s past to the present. Through the feedback terms, responses to prior motivational inputs affect responses to the current inputs, which, in turn, affect subsequent responses to the real-time inputs. More specifically, the first-order feedback effect produces an inertial effect, while the second-order feedback effect produces an oscillation effect. These are consistent with the homeostasis characteristic of physiological systems (Stern, Ray, & Quigley, 2001) and the oscillation patterns of brain activities (Buzsáki, 2006). System feedback effects are critical in the evolution of message input effects. They determine the speed, amplitude, and duration of the integrated outputs of the dynamic system (Boker & Wenger, 2007; Harvey, 1990), while the integrated outputs are the observable behaviors in which media effects researchers are generally interested.
Following the formalization of DMA in Wang et al. (2011), this study further tests whether the following holds true.
Wang et al. have tested this hypothesis and found supporting evidence in the context of entertainment television viewing. The viewing duration in the 2011 experiment was relatively long (30 min), and viewers had continuous control over which television program to watch. The current study provides a different media environment, with shorter messages that aim to persuade rather than entertain viewers. Hence, it is necessary to test in this new context whether the previous formalization of the physiological responses as a second-order dynamic system applies to political ads processing.
Method
Pretest and Stimuli
The study was conducted at a large Midwestern university in a battleground state in the United States. For the pretest, 24 30-s televised political ads were selected from the 2008 general presidential election campaign. The ads were selected using a 2 (valence: positive, negative) × 3 (arousing content levels: arousing, moderate, calm) × 2 (candidates: McCain, Obama) factorial design, with two ads at each level. Featured candidate of an ad refers to which candidate was presented in the ad, regardless of the sponsor of the ad. For example, an ad criticizing McCain’s environmental policies is discussed as a “McCain ad.”
In total, 120 undergraduate students participated in the pretest study, with a comparable number of individuals favoring McCain, Obama, or neutral toward the two candidates. Participants came to the lab, arriving in groups of 2 to 6, and completed the study using individual desktop computers. Each participant rated the political ads using the continuous response measurement (CRM, Biocca, David, & West, 1994) implemented by MediaLab software (Jarvis, 2008). While viewing each ad on the computer, the participant simultaneously pushed left and right arrow keys on the keyboard to move a slider. The slider appeared right below the ad on the screen and its movement indicated the participant’s judgment along one of three motivational scales. The scales were (a) Arousing Content (not at all arousing to extremely arousing), (b) Positivity (not at all positive to extremely positive), and (c) Negativity (not at all negative to extremely negative). All three scales were anchored by 1 and 100 when presented on the screen. MediaLab transferred the ratings onto a scale of 0 to 2 (rounded to the hundredth decimal place) and recorded the data at 20 Hz. Participants were randomly assigned to the combinations of ads and rating scales, and the presentation order of ads was randomized for each participant. All participants made ratings of arousing content, positivity, and negativity, and no participant viewed the same ad more than once. Across participants, all 24 ads were rated on all three scales by a comparable number of people. Based upon means and medians of the CRM ratings from the pretest study, final stimuli used in the psychophysiological experiment were selected. One ad was selected at each of the 12 manipulation levels to ensure a large range of emotional content. ANOVAs showed that the manipulation of the general valence and arousing content was successful (ps < .05), which was not different between ads for the two candidates.
Participants and Procedures
The psychophysiological experiment was conducted during the last two weeks of October 2008—a few days before the 56th quadrennial U.S. presidential election. Six presentation orders were constructed using the Latin square design to counterbalance the 12 ads. Participants were randomly assigned to one of the six orders. Complete physiological data were obtained from 15 students from the same university as participants from the pretest experiment. Participants from the psychophysiological experiment shared similar demographic features as those in the pretest. They were 20 to 32 years old (M = 22.07, SD = 2.87), around half (53.33%) were males, and most were White (86.67%). None of them had participated in the pretest.
Experiments were conducted individually. After arriving at the lab and providing informed consent, the participant was prepared for the recording of physiological measures. The participant viewed two practice ads to get familiar with the experimental environment before watching the 12 stimulus ads. The participant’s physiological responses were recorded during ad viewing. After watching all the ads, electrodes for physiological measures were removed from the participant. Gender, race, and individual difference traits of motivational reactivity can affect motivational processing (Bradley, 2000; Lang, Bradley, Sparks, & Lee, 2007); and hence, these variables were measured and controlled for in data analysis. Individual differences in motivational reactivity were assessed by Approach System Activation (ASA) and Defensive System Activation (DSA) measures (Lang, Kurita, Rubenking, & Potter, 2011). Candidate Evaluation was obtained using modified questions from the American National Election Studies (ANES). Participants were asked to rate each of the major party candidates (Obama and McCain) “according to how negative or positive you feel about him” on a scale of 1 (very negative) to 9 (very positive). The order of Obama and McCain evaluation was randomized.
Physiological Dependent Variables
All physiological variables were collected using Coulbourn Instruments and a Scientific Solutions Labmaster A/D board controlled by acquisition software VPM 12.6 (Cook, 2007).
HR is controlled by the sympathetic nervous system (SNS, dominant during mobilization) and the parasympathetic nervous system (PNS, dominant during rest). Cardiac deceleration results from more dominant PNS over SNS activation, which indicates perceptual information intake and orienting. Cardiac acceleration results from more dominant SNS over PNS activity, which indicates sensory rejection, mentation or internal focus, and behavioral response or tendency (Campbell, Wood, & McBride, 1997; Lacey, 1967; Graham & Clifton, 1966). This cardiovascular pattern of attention has been supported in media research (Lang, 2006). In the current experiment, HR data were collected using two 7-mm Ag/AgCl electrodes placed on the forearms. The interval between heart beats was recorded and converted to beats per minute (BPM).
SCL measures SNS activation and is associated with motivational activation intensity (Bradley, 2000). Higher SCL indicates increased sympathetic arousal and greater motivational activation. Heightened skin conductance is a typical component of an orienting response, but this response habituates quickly. Therefore, SCL is interpreted as relating primarily to action preparation in response to the stimulus rather than perceptual information processing of the stimulus after initial orienting response (Bradley, 2009). SCL data were acquired through two 7-mm Ag/AgCl electrodes placed on the nondominant palmar surface, sampled at 20 Hz.
Zygomatic and Corrugator EMG have been used as an indication of emotional responses (Larsen, Norris, & Cacioppo, 2003). However, these measures are also associated with attentional effort (e.g., Cohen, Davidson, Senulis, Saron, & Weisman, 1992). Zygomatic EMG measures activities in the zygomaticus major muscle group which is located under the cheek. This muscle group is responsible for tightening of the cheek, which may indicate communication and speech tendency (e.g., Lang, Greenwald, Bradley, & Hamm, 1993; McGuigan & Rodier, 1968). The corrugator supercili muscles are located above the eyes and near the base of the eyebrows. These muscles control the lowering and raising of the eyebrows, and can index perceptual attentional effort (Cohen et al., 1992). The facial EMG data were acquired by placing a pair of 4-mm Ag/AgCl electrodes on the facial muscle sites on the left side of the face. The data were sampled at 500 Hz.
Time Series Data of Independent and Dependent Variables
For each participant, time series were created for each physiological variable at the rate of one observation per second. Each series is composed of 360 data points recorded during viewing of the 12 ads. First, a detrending procedure was carried out on the data to remove linear trends using the general linear model procedure in SAS (PROC GLM). This removes the influence of time on the data series which is not stimulus-specific and does not constitute the focus of the study (e.g., habituation and fatigue). The detrended data were standardized for each variable across all participants to facilitate interpretation and comparison of model parameters.
For independent variables, three time series were created using the medians of the CRM ratings of positivity, negativity, and arousing content acquired in the pretest, again at the rate of one observation per second. Dummy coding was used to distinguish the presidential candidate featured in each ad (OM: McCain = 0, and Obama = 1). Differences in the evaluation of the two candidates were computed for each participant (Eval = evaluation of McCain – evaluation of Obama; range = −8 to 8; M = −.60, SD = 4.73). Gender, race, and individual motivational traits (ASA, DSA, and their interaction) of viewers were controlled in all analyses.
Analysis
Time-Series Cross-Sectional (TSCS) Analysis
TSCS analysis was conducted on the time series data of all participants using PROC TSCSREG in SAS software. Advantages of TSCS include increased sample size, and simultaneous estimates of the cross-time effects of message variables and cross-sectional individual differences. In the following analyses, individual viewers were considered cross-sectional units and 360 observations during ad viewing constituted the time series for each section. The estimation method was the Fuller–Battese method (Fuller & Battese, 1974), which includes the individual- and time-related random effects to the error disturbances. The error part of the model also included an autoregressive lag 1 effect, which corrects the autocorrelation of the errors.
Model Fitting and Comparison
To test the hypotheses on dynamic interactions between messages and viewers, three groups of models were compared: (a) the message model, (b) the message and message–audience interaction model, and (c) the message–audience interaction model. All models include the first- and second-order system feedback terms (i.e., the lag 1 and lag 2 autoregressive terms of the dependent variable), but vary on inclusion of general message and message–audience interaction terms. The message models consist only of general message variables: Arousing Content (A), Positivity (P), Negativity (N), their two-way interactions (A × P, A × N, P × N), and their quadratic effects (A2, P2, N2). The message and message–audience interaction models consist of all the variables in the message models and also include interaction effects between message variables and individuals’ evaluation difference of the two candidates (Eval × A, Eval × N, Eval × P) as well as three-way interactions involving the dummy codes indicative of the featured candidate in each ad (Eval × A × OM, Eval × N × OM, Eval × P × OM). Finally, the interaction models are simpler than the second group of models, including the two-way interaction terms between message variables and candidate evaluation and the three-way interactions between message variables, candidate evaluation, and the dummy code indicating the candidate featured in the ad, but excluding general message effects.
Each of the four physiological responses was separately tested for the three groups of competing models. Following Wang et al. (2011), to identify the best delay lags for message inputs to reach the physiological system, 11 message input lagged models were estimated using lags 0 to 10 (indicating no delay to 10-s delay from the A, P, and N motivational message inputs to the elicitation of physiological responses). The input lag model with the largest regression R 2 was selected for each competing model of each physiological variable. This step reduced the 11 (message input delay) × 3 (competing models) models to three for each physiological variable, as summarized in Tables 1 through 4. This message input delay is not the focus of the current design and study. However, quantifying this delay in the models helps accurately estimate the other model parameters of interest. To test Hypothesis 1 and select the best model among the three message and audience models for each physiological response, Bayesian Information Criterion (BIC) was employed because the three models for each physiological variable are not nested and have different numbers of parameters. BIC considers both goodness of fit and complexity of models. Models with smaller BIC are preferred (Busemeyer & Deiderich, 2010; Schwarz, 1978).
Model Evaluation and Estimated Parameters for Competing HR Models.
Note: ASA = Approach System Activation; DSA = Defensive System Activation; BIC = Bayesian Information Criterion.
p < .05. **p < .10.
Model Evaluation and Estimated Parameters for Competing SCL Models.
Note: SCL = skin conductance level; ASA = Approach System Activation; DSA = Defensive System Activation; BIC = Bayesian Information Criterion.
p < .05. **p < .10.
Model Evaluation and Estimated Parameters for Competing Corrugator EMG Models.
Note: EMG = electromyography; ASA = Approach System Activation; DSA = Defensive System Activation; BIC = Bayesian Information Criterion.
p < .05. **p < .10.
Model Evaluation and Estimated Parameters for Competing Zygomatic EMG Models.
Note: EMG = electromyography; ASA = Approach System Activation; DSA = Defensive System Activation; BIC = Bayesian Information Criterion.
p < .05. **p < .10.
Results
The Interaction Model
Model parameters and model fit statistics for the four physiological responses are summarized in Tables 1 through 4. As shown, the interaction model consistently achieved the smallest BIC compared to its competing models for all four physiological responses. Thus, the interaction models are the preferred models, and Hypothesis 1 is supported. As indicated by regression R 2 , the interaction model accounts for 38.27% of variance in the HR time series across all participants, 55.67% in SCL, 33.74% in corrugator EMG, and 61.09% in zygomatic EMG. The regression R 2 is smaller than those found by Wang et al. (2011). This is expected as the current model attempts to account for variance across time as well as across all individuals (cf. models in the Wang et al. study only account for time series variance).
Message Motivational Inputs Interact With Candidate Evaluation
Message motivational inputs interact with viewers’ evaluation of the candidates to influence physiological responses, and viewers holding opposing attitudes show divergent responses (Hypothesis 2 supported). The effects are estimated by model parameters Eval × A, Eval × P, Eval × N, and their three-way interactions with the dummy code OM (indicating the featured candidate) (see Tables 1-4). The sign and size of the parameters indicate the direction and size of the effects. For illustration, these effects were simulated in a MATLAB program using the parameters of “the Interaction Model” in Tables 1 through 4. Interaction effects involving arousing content are shown in Figure 1. Figures 2 and 3 illustrate interactions involving negativity and positivity, respectively. It is worth emphasizing that these effects, estimated by the model parameters, are per time unit (i.e., per second, in this study) and are disentangled from the system’s dynamic feedback effects (discussed in more detail later).
Psychophysiological responses to Obama (on the left) and McCain (on the right) ads as a function of arousing content of the ads and viewers’ candidate evaluation.
Psychophysiological responses to Obama (on the left) and McCain (on the right) ads as a function of negativity of the ads and viewers’ candidate evaluation.
Psychophysiological responses to Obama (on the left) and McCain (on the right) ads as a function of positivity of the ads and viewers’ candidate evaluation.
Based upon the range of actual independent variable values in our experiment, the message motivational inputs in the simulation were 0 to 1.5 (on the scale of 0 to 2), and candidate evaluations were represented using the Eval scores of −6, −3, 0, 3, and 6 (on the scale of −8 to 8). Negative scores indicate favoring Obama over McCain, positive scores indicate favoring McCain over Obama, and 0 indicates a relatively neutral position in the sense that there is not clear preference for one candidate or the other. The greater the absolute value of the Eval score, the larger the difference in evaluation of the two candidates. In each graph panel, separate lines depict evaluation difference scores. From top to bottom, the four rows illustrate HR, SCL, corrugator EMG, and zygomatic EMG, respectively. The panels on the left are responses to Obama ads (i.e., ads that feature Obama, regardless of who sponsored or paid for the production and airing of the ad) and those on the right are responses to McCain ads (i.e., ads featuring McCain, regardless of sponsor). The plotted psychophysiological changes are in standardized scores.
Arousing content interacts with candidate evaluation
HR, SCL, and corrugator EMG were significantly affected by Eval × A and Eval × A × OM interactions, but zygomatic EMG was not (see Tables 1-4 for parameters and Figure 1 for illustration of the effects). Across all panels in Figure 1, we can see that when the Eval score is 0 (indicating a neutral position to the two candidates), increasing arousing content does not change physiological responses to the message. However, when evaluation is negative (indicating preference for Obama) or positive (indicating preference for McCain), increasingly arousing content does impact responses. The effects are in exactly opposite directions for supporters of the two candidates, and the larger the evaluation difference, the greater the effects. When ads become more arousing, proponents respond to their favored candidate’s ads with a decrease in HR and SCL, and an increase in corrugator EMG. While viewing the opposing candidate’s ads, the reverse pattern occurs: HR and SCL increase, and corrugator EMG decreases.
The predicted response changes are plotted in standardized scores. Comparing the panels in Figure 1, we can infer the effect sizes in terms of the portion of one standard deviation of the physiological data. The largest effect of arousing content is on HR during exposure to McCain ads. When arousing content increases from 0 to 1.5, the HR difference between strong Obama versus strong McCain proponents increases by more than 50% of one standard deviation of the HR data. HR during Obama ads differs only by around 20% of its standard deviation for strong opposing partisans. Corrugator EMG shows similar effect sizes during both candidates’ ads (around 35% of its standard deviation) as does SCL (around 20%).
Negativity interacts with candidate evaluation
All four physiological responses are affected by Eval × N and Eval × N × OM interactions, although a few were only marginally significant at p < .1 (see Tables 1-4 and Figure 2). Similar to arousing content, changes in the negativity of emotional message input has no effect on viewers with neutral evaluations, but does influence those with a preference for one candidate over the other. Negativity drives physiological reactions in opposite directions for those favoring different candidates, and again, the greater the evaluation differences, the larger the effect. There are two exceptions to this general pattern: SCL and zygomatic EMG during Obama ads show no effect of negativity regardless of candidate evaluations.
Describing the effects of message negativity in more detail, when McCain ads become more negative, McCain supporters show increased HR, SCL, and zygomatic EMG, but decreased corrugator EMG. Obama supporters demonstrate the opposite pattern of physiological responses to McCain ads. When Obama ads become more negative, Obama supporters show the same response patterns that McCain supporters show while viewing McCain ads—except that SCL and zygomatic EMG are not significantly affected. McCain supporters’ responses to Obama ads are similar to Obama supporters’ responses to McCain ads—except that SCL and zygomatic EMG are unaffected. Of the effects, corrugator EMG during Obama ads shows the largest effect size. Differences in corrugator EMG between strong Obama compared to strong McCain supporters increased by around 40% of its standard deviation when negativity increases from 0 to 1.5. The other changes are around 20% of the physiological measures’ standard deviations.
Positivity interacts with candidate evaluation
Interestingly, positivity shows very similar effects as negativity (see Tables 1-4 and Figure 3). However, the Eval × P and Eval × P × OM effects on zygomatic EMG are not significant. Another difference is that compared to negativity, positivity shows a larger divergent effect on HR (35% vs. 20% of standard deviation).
Physiological Feedback Effects
All four physiological systems have significant first- and second-order feedback effects (Hypothesis 3 supported). This means that the changes and asymptote of the physiological system’s reactions to motivational inputs depends not only on the nature of the effects of message inputs (e.g., direction and size of the effects as reported in the previous section) but also on the system’s own feedback functions. The feedback effects determine how quickly the motivational inputs affect the individual viewer’s physiological systems and how the message–individual effects evolve, accumulate, and decay across time. This is examined in detail below.
Dynamic Interplay Between Motivational Inputs and Candidate Evaluation Across Time
We have examined the interaction effects between the motivational inputs and candidate evaluation and the dynamic system feedback effects. These interaction and feedback effects are not only disentangled from one another but also teased apart from time and estimated per second. Next, we put all of these components together to examine how the system feedback effects integrate the message–audience interaction effects to generate reactions across time. Often in social and behavioral sciences, these integrated effects are the final observable behavior of the dynamic system. Importantly, they can have different—sometimes even opposite—patterns compared to the effects estimated per time unit (Busemeyer & Deiderich, 2010; Roe, Busemeyer, & Townsend, 2001).
The estimated parameters for each physiological system are entered into the proposed models. Following the common analytic strategy in time series analysis and the simulation method by Wang et al. (2011), eight combinations of the three motivational inputs (A, P, N) are selected to systematically demonstrate their effects on the physiological systems. They are (a) all three inputs are off (baseline); (b) only A is on; (c) only P is on; (d) only N is on; (e) A and P are on, but N is off; (f) A and N are on, but P is off; (g) P and N are on, but A is off; and (h) all three inputs are on. To facilitate interpretation, the magnitudes of all three motivational inputs are kept at 1.2, which can be considered as moderate on the 0-to-2 scales. These motivational inputs are controlled as a step input (i.e., turned on from zero to a fixed magnitude for a specified time duration), which is commonly used to examine the accumulation and evolution of dynamic effects (Luenberger, 1979). The step input duration is set to be 45-s each, a little longer than the actual 30-s ads in our experiment but within the range of most ads (15-60 s), so we can have a clear observation of the evolution trajectories of the effects during a realistic political ad exposure timeframe. After each step input, a 15-s zero setting (i.e., no input) is used, which allows the system to return to its natural baseline. Additionally, this zero input setting enables observation of the decay of the previous input effect and avoids confounding the subsequent input effect. Figure 4 provides a visual illustration of the eight input conditions: (a) baseline or 0, (b) A, (c) P, (d) N, (e) AP, (f) AN, (g) PN, and (h) APN. By examining how a physiological system reacts to the eight input conditions, we can systematically examine the motivational effects as estimated by model parameters (see Figures 5-8). This facilitates understanding the dynamic physiological systems with greater rigor and clarity than relying on the observed data alone. In the latter situation, entangled exogenous and endogenous influences and can be quite perplexing. A dynamic system is complex and with emergent features (Buzsáki, 2006), and formal modeling and simulation becomes essential in understanding its behavior (Busemeyer & Deiderich, 2010; Kelso, 1995; Ward, 2002; Wang et al., 2011).
Eight conditions of motivational inputs with arousing content, positivity, and negativity being on and off during different time periods (input magnitude = 1.2).
Heart rate dynamic responses to motivational inputs in Obama ads (left panels) and McCain ads (right panels).
Skin conductance level dynamic responses to motivational inputs in Obama ads (left panels) and McCain ads (right panels).
Corrugator electromyography dynamic responses to motivational inputs in Obama ads (left panels) and McCain ads (right panels).
Zygomatic electromyography dynamic responses to motivational inputs in Obama ads (left panels) and McCain ads (right panels).
The simulated effects are shown in Figures 5 through 8. In each figure, the eight input conditions are represented as letters at the bottom. The corresponding step input durations are highlighted in grey. The five panels on the left are dynamic physiological responses to ads featuring Obama, whereas the five panels on the right show responses to McCain ads. From top to bottom, candidate Eval scores range from strong support of Obama (Eval = −6) to strong support of McCain (Eval = 6). A consistent pattern emerges across the four physiological measures.
First, the figures illustrate the dynamic nature of the physiological systems. Onset and offset of a motivational input do not instantaneously bring the system to its equilibrium state. Instead, upon exposure to motivational message input, it takes time for the physiological system to reach its equilibrium state. Similarly, when external motivational message input is turned off, it takes time for the system to decay back to its baseline. During this dynamic effect growth and decay, a key role is played by system feedback. Feedback effects integrate the motivational input effect to generate the dynamic trajectories depicted in Figures 5 through 8.
Second, it is interesting to note that when integrated across time by system feedbacks (i.e., Figures 5-8), the influence of candidate evaluation on motivational inputs effects appears different from its per time unit effect alone (i.e., Figures 1-3). As shown in each of Figures 5 through 8, the five panels on the left (i.e., responses to ads featuring Obama) demonstrate similar response directions to the eight input conditions. However, from the top to the bottom panel (i.e., increasingly favorable attitudes toward McCain), the response magnitude monotonically diminishes. A similar phenomenon is observed among the five panels on the right (i.e., responses to ads featuring McCain). Here, the response magnitude attenuates from the bottom to the top panel (i.e., increasingly supportive of Obama). The two panels in the middle (i.e., neutral candidate evaluation) show similar response patterns. These results indicate that people respond intensely to ads featuring their favored candidate, but are generally less responsive to ads featuring the opponent. The powerful integration and moderation effect of system feedbacks is clearly demonstrated by comparing this integrated effect pattern to the motivational inputs and candidate evaluation interaction effects per second described earlier (Figures 1-3). First, a small system input effect can accumulate through system feedback effects, growing into a much larger effect across time. As shown in the top left panel in Figure 1, when arousing content increases from 0 to 1.2, HR of Obama supporters (thickest line) drops about .036 of one standard deviation. However, when arousing input effect is accumulated by feedback effects across time, it grows many times larger, producing a HR deceleration of nearly .25 of one standard deviation (Condition “A” in the top left panel of Figure 5). Second, the exogenous input effect, aggregated and moderated by system feedback effects, may manifest in a direction different from the per unit time estimation of the input effect. This, too, is the case in our data. In general, system feedback effects attenuate the opposing trend of motivational input effects among people with opposing candidate evaluations. System feedback effects energize responses to ads featuring the favored candidate and attenuate, instead of reverse, responses to ads featuring the opponent. This is even more evident in a supplementary simulation, in which the exact same model and model parameters used to generate Figures 5 through 8 were implemented, but system feedback effects were turned off (see the first author’s website for the supplementary simulation). When feedback effects were turned off, motivational input effects across time were opposite among people with opposing candidate evaluations. This is exactly consistent with the per time unit effects illustrated in Figures 1 through 3.
Third, arousing content, positivity, and negativity influence different physiological systems in different ways. As shown in Figure 5, the HR pattern predicted by the dynamic model coincides with previous research measuring HR in response to emotional media messages. Arousing content (Condition “A”) elicits large HR deceleration. Nonarousing (i.e., calm) positive and negative messages (Conditions “P,” “N,” and “PN”) elicit HR acceleration, and negativity elicits slower HR than positivity (Lang et al., 1993). However, if arousing content is added to these conditions (Conditions “AP,” “AN,” and “APN”), HR becomes slower. Previous analyses have seldom closely examined coactive messages (Conditions “PN” and “APN,” both of which contain positivity and negativity simultaneously). This dynamic model clearly shows HR acceleration when it is calm (Condition “PN”), but HR deceleration when arousing content is added (Condition “APN”). This replicates findings by Wang et al. (2011).
The SCL dynamic model simulation is shown in Figure 6. Similar to previous findings (e.g., Wang et al., 2011), the SCL model is most responsive to arousing content. Surprisingly, however, the presence of arousing content (Conditions “A,” “AP,” “AN,” and “APN”) causes a decrease in SCL. This is contrary to many previous findings using static methods, which generally find that arousing content leads to an increase in SCL. The findings here are also different from Wang et al.’s results, in which the only response of SCL to the eight input conditions was an SCL increase in response to arousing content (Condition “A”). Some speculation of these aberrant SCL findings seems warranted. The current model parameter estimations are based upon observations from participants viewing relatively short, highly persuasive stimuli (30-s political ads). In contrast, Wang et al.’s parameter estimation was based upon data from self-controlled 30-min entertainment television viewing. Additionally, when the political ad experiment was conducted, the political campaign had been going on for several months, with Hillary Clinton’s concession of the Democratic Party nomination to Obama in early June marking the beginning of the general election race (Franz & Ridout, 2010). It is possible that viewers had habituated to the arousing content of campaign ads, perhaps even showing emotional responses similar to annoyance or boredom (Hastings, Stead, & Webb, 2004). It is also interesting to note that SCL is more complicated during the viewing of ads featuring McCain compared to those featuring Obama. During the former, an input of positivity and/or negativity (Conditions “P,” “N,” and “PN”) caused an increase in SCL among McCain supporters and a decrease among Obama supporters.
Comparing the dynamic corrugator and zygomatic EMG side by side (Figures 7 and 8), the two measures react to the eight input conditions in a generally opposite way, which is consistent with psychophysiological theories and extant empirical evidence. Arousing content (Condition “A”) increases corrugator but decreases zygomatic EMG. Positivity (Condition “P”) decreases corrugator but increases zygomatic EMG. Negativity (Condition “N”) also decreases corrugator but barely affects zygomatic EMG. It is interesting to note that coactive messages (Conditions “PN” and “APN”) have a clear impact on the two facial EMG measures. When the message is calm (Condition “PN”), corrugator activity decreases but zygomatic activity increases, but when the message is arousing (Condition “APN”), the increasing or decreasing activity is largely reduced.
Discussion
Consistent with a wealth of research on motivated attention (Bradley, 2009; Lang, 2006) and previous tests of DMA (Wang & Busemeyer, 2007; Wang et al., 2011), this study supports the central role of motivational activation in the allocation of attentional resources, which dynamically changes as a message unfolds. Furthermore, this study extends previous DMA work by theorizing dynamic motivational influences of both message components and viewers’ attitudes. It provides evidence of disparate attention among supporters of the two opposing political candidates. This confirms previous work theorizing mediated cognition as interactions between messages and individuals (e.g., Lang, 2006; Southwell, 2005), but is the first attempt to test the interactions using dynamic models of physiological data.
The dynamic analysis extends theoretical understanding of the message–audience interactions in two ways. First, it disentangles the message–audience interaction effects from the information processing system feedback effects and also from time. The message–audience interaction effects per second suggest that opposite candidate evaluations lead to opposite attentional responses to the same message. Second, it examines how individuals’ attitudes to candidates interact with message motivational content to affect attention across time. The processing system feedback effects accumulate the motivational input effects, creating a larger effect among supporting viewers while attenuating the effect among viewers supportive of the opposing candidate.
The Cardiac-Somatic Coupling of Attention to Emotional Messages Featuring the Favored or the Opposed Candidate
Increasing political polarization has been a growing concern for scholars and the public (Mutz & Martin, 2001). The present study suggests that in addition to selective exposure, selective attention to political information may contribute to this polarization. Despite reduced exposure selectivity in a high-intensity election campaign (Cho, 2008), attentional selectivity can be a critical component in determining how information processing influences campaign message reception and effects. As revealed by the integrated, dynamic effects figures (Figures 5-8), viewers with neutral candidate evaluations respond in the same way to both candidates’ ads. However, for those who already prefer one candidate or the other, ads featuring a favored candidate elicit intense responses, whereas ads featuring the opponent result in decreased responsiveness. Furthermore, the larger the candidate evaluation difference, the more divergent an individual’s responses are to a favored and opposed candidate’s message. This divergent pattern is even more dramatic when examining the message–individual interaction alone (Figures 1-3).
From these figures, we can see a clear pattern of attentional selectivity based upon the sensory information intake-rejection interpretation (e.g., Lacey, 1959, 1967) and the cardiac-somatic coupling theory of attention (Cohen et al., Obrist, Webb, Sutterer, & Howard, 1970). On one hand, the cardiac-somatic coupling response pattern of attention suggests that increasing external attention or sensory intake is associated with more dominant PNS activation, leading to slower HR (Lang, 2006). Simultaneously, corrugator muscles tighten and zygomatic muscles relax to facilitate perception by minimizing communication distraction and noise (Bartoshuk, 1956; Haagh & Brunia, 1984). On the other hand, sensory rejection is associated with more dominant SNS activity, leading to faster HR, decreased corrugator activity, and increased zygomatic activity. The information rejection mode helps prevent external disruption and facilitates internal focus and mental activities (e.g., imagination, decision making, counter-argument) or facilitates preparation for behavioral responses (e.g., fight or flight, communication, and signaling). This cardiac-somatic coupling of attention is consistently revealed in the figures illustrating HR and facial EMG responses. Again, response patterns to the same motivational message inputs vary largely by the viewers’ attitudes to the candidates.
Interestingly, depending on the viewers’ attitudes, the cardiac-somatic responses are elicited by different dimensions of emotional content of the ads (see Figures 1-3). When ads featuring the favored candidate become more arousing, or when ads featuring the opposed candidate become more negative or positive, an information-intake mode is activated. Viewers show decreased HR and zygomatic activity, and increased corrugator activity. When ads featuring the favored candidate become more negative or positive, or when ads featuring the opposed candidate become more arousing, viewers show increased HR and zygomatic activity, and decreased corrugator activity. This information-rejection response pattern suggests that viewers may be starting to engage in internal mental activities or action preparation. One possible explanation is that with Election Day on the immediate horizon, viewers were already quite familiar with their favored candidate. Viewers might have been mostly intaking the excitement delivered in ads with arousing content, whereas valenced information might have elicited mental activities or behavioral tendencies, such as counterarguing criticisms or elaborating on supportive messages about their favored candidate. When viewing ads featuring an opposed candidate, participants may have attempted to encode valenced information in an effort to counterargue against or criticize the opponent’s viewpoints. When the ad is arousing, execution of mental activities or behavioral tendencies (e.g., arguments or counterarguments) may occur, which interrupted the encoding process. More interestingly, however, the dynamic system feedback effects intensify responses to ads featuring the favored candidate, but attenuate responses to the ads featuring the opponent (Figures 5-8). At the integrated effect level, viewers are responsive to ads featuring their favored candidate but rather irresponsive to ads featuring the opponent. Also, it is interesting to note that response patterns of both candidates’ supporters to their favored and opposed candidate are not perfect mirror images of each other. Perhaps this indicates mutual influences of ideological differences and physiological reactivity (Oxley et al., 2008).
These findings on divergent attention during political information processing suggest a potential source for gaps in public opinion and political perceptions, such as the hostile media effect (Huge & Glynn, 2010; Gunther & Liebhart, 2006), and the increasingly intensified negativity of Democrats’ and Republicans’ evaluations of an opposing party president (Abramowitz, 2010; Jacobson, 2006). People not only selectively expose themselves to information that reinforces rather than challenges their beliefs (Iyengar & Hahn, 2009) but also attend to the political information environment in a highly selective manner. Our data suggest that individuals are more responsive to information about their favored candidate. This finding does not support the common tactic of using negative ads attacking an opponent to mobilize supporters or leaning voters. Furthermore, it seems that individuals are drawn by the excitement about their favored candidate, and quickly engage in arguments and action tendencies if the information is valenced. Further replications are needed—particularly connecting the findings to postexposure, long-term political perceptions. However, it seems likely that in addition to selective exposure, selective attention to the same political information may contribute to increased bias in political perceptions.
By accounting for the influence of political predispositions, interactions between political attitudes and message content, and the dynamic nature of message processing, some inconsistent results in the extant political ads literature might be resolved. An important and lingering question addressed by current research is whether negative or positive ads have an advantage in capturing or retaining audience attention during message exposure. Research generally relies on various memory measures as indicators of attention or learning. Yet, empirical research has revealed inconsistent findings. For instance, Shapiro and Rieger (1992) found that arguments in negative ads were better remembered than those in positive ads. However, Geer and Geer (2003) showed that recall did not differ between negative and positive political ads, and Basil, Schooler, and Reeves (1991) found that positive ads were more likely to be remembered than negative ads. Perhaps, these seemingly disparate findings could be reconciled by acknowledging the dynamic nature of both ads and information processing. The context of the information (e.g., the nature of its preceding information, the length of the ad) might significantly affect whether negative or positive information is advantageous for ad memory. Future research expanding on the current study might include memory or thought listing measures to examine whether arguments or counterarguments actually occur during the information rejection mode, and also further specify the nature of mental activities or behavioral tendencies during ad viewing. For instance, Meirick (2002) found that comparison ads prompted more counterarguments than negative ads. Postexposure thought listing and real-time psychophysiological measures might illuminate whether the ad content that prompts counterarguing by audience members occurs during the information rejection mode as indicated by the cardiac-somatic response patterns.
Effects of coactivation of both motivational systems present another interesting finding. Scant empirical data exist on how coactivation of both motivational systems affects physiological responses and cognitive processes. However, these contexts may have important implications for political ads research because the widely used comparison ads are likely coactive messages. In addition to HR findings that are consistent with those found by Wang et al. (2011), this study finds somatic facial EMG data that couples with HR data. The data can be interpreted from the cardiac-somatic coupling patterns of attention (Lacey, 1967; Obrist et al., 1970). When the coactive message is calm (Condition “PN”), HR accelerates, corrugator activity decreases, and zygomatic activity increases, indicating a sensory rejection mode. When the coactive message is arousing (Condition “APN”), the increasing or decreasing activity is attenuated, which suggests that the PNS may become more active, competing with the SNS to quiet down the body and starting to facilitate sensory intake. These speculations should be further tested using experiments specifically designed to test coactive motivational activation, such as during watching comparison ads (Shah et al., 2007).
Feedback Effects of the Physiological Systems
Consistent with the findings by Wang et al. (2011), this study provides evidence of the first- and second-order feedback effects of the four PNS physiological measures. The system feedback effects determine the dynamic nature of message processing and motivational effects. First, they determine the speed of message effect activation and decay. As shown in the model simulation figures (Figures 5-8), changes in motivational inputs do not elicit instantaneous changes in physiological responses. Rather, because of feedback terms, it can take anywhere from fractions of a second to a number of seconds before a physiological response system reaches its equilibrium state. Second, the feedback effects determine the final level of growth and the cumulative effect produced by a message motivational input. As illustrated by a simulation example of a first-order system provided by Wang et al. (2011), if message inputs and their effects parameters are kept the same, dramatically different maximum and cumulative effects can be produced by slightly varying the first-order feedback parameter. Although the system dynamics are much more complicated when the system has an additional second-order feedback, which produces an oscillation, the powerful influence of these feedback effects is demonstrated in that simple example.
In the current study, it is interesting to compare the estimated system input effects (Figures 1-3) to the integrated system outputs (Figures 5-8). As discussed earlier, the system feedback effects accumulate the motivational inputs, producing larger effects, but they may also alter effect directions by attenuating the opposing patterns among people with opposite attitudes. It is a common feature of dynamic complex systems that unexpected effects at the integrated system level can emerge from interactions between the interconnected system components (Busemeyer & Deiderich, 2010; Ward, 2002; Wang et al., 2011). For example, a well-known decision-making phenomenon, preference reversal, is explained by a computational theory called decision field theory from a dynamic system approach (Roe, Busemeyer, & Townsend, 2001). The authors show that in a dynamic decision system formulated by their theory, a system input with options A and B, where greater utility is assigned to A than B, can generate an unexpected system output in which B is preferred over A because of dynamic interactions between the attention and evaluation components in the system. This is similar to our case, where motivational input effects interact with system feedback effects, generating aggregated effects at the system level that appear in a different direction from the motivational inputs effects alone.
The evidence of system feedback effects suggests three important implications for message design and message effects research. First, consideration of message variables alone may provide only a partial picture of message effects. Observed message effects usually are an integration of both message effects and the processing system’s feedback effects. Message variables controlled by media production practitioners, such as emotional content and production features, produce their effects through human processing systems. It is important to carefully study how these controllable message production features can achieve certain outcomes through the moderation of dynamic processing systems. To accomplish this, researchers need to keep in mind that message manipulations are probably not, in many cases, linearly related to outcomes. As shown in the current study and the study by Wang et al. (2011), the processing system feedback effects contribute to the nonlinear growth and decay of the message effects.
Second, consideration of time duration is critical to understanding message effects. The dynamic nature of the message effects observed in the present study indicates that effect sizes (and sometimes even directions) depend crucially on the status of the message processing system on the effect evolution trajectory when research observations are made. Cumulative message effects depend on how long the system has been activated, whether it has reached its equilibrium state, and if so, for how long. Empirical studies with the same experimental design and stimulus manipulation but different stimulus durations and/or different measurement timings may find divergent or even conflicting results based upon static analysis such as ANOVA. On the other hand, dynamic analytical tools allow time to take central stage in examining message effects. By estimating message variable effects per time unit, this procedure allows researchers to disentangle message variable effects from the confounding effects of message duration. Furthermore, dynamic models separate the effects caused by message variables from the moderating and cumulating effects of the information processing system feedback effects.
Last, the demonstrated dynamic nature of message processing emphasizes the importance of contextual effects in the field of communication. As specified by the feedback terms, preceding stimuli or contextual factors can affect media processing, further illustrating that media processing depends on prior history and experience. This is consistent with classic media effects theories, such as excitation transfer theory (Zillmann, 1971). The formal dynamic models tested here can help researchers further understand how communication messages create contexts for one another, such as how the processing of ads is affected by program context (e.g., Potter et al., 2006) and how a political conversation can be influenced by group dynamics and school/family environment (e.g., Hively & Eveland, 2009).
Limitations and Future Directions
Several limitations of this study warrant consideration. First, with limited data points, we have restricted our models to include only the key interaction terms between the linear main effect of motivational inputs and candidate evaluation. Future studies should obtain larger data samples to test more complicated interactions, such as those involving quadratic motivational terms and more sophisticated political attitude constructs, such as partisanship, political ideology, or attitudes toward specific public policy issues. Second, contrary to much previous literature, SCL found in this study showed a decreasing response to motivational inputs (see Figure 6). Wang et al. (2011) also found that SCL was rather irresponsive based upon dynamic analysis. One possibility is that uncontrolled structural features of the messages, such as music and camera changes, might have influenced skin conductance (e.g., Potter & Choi, 2006). Future studies should consider controlling these factors. In addition, skin conductance changes were more responsive to positive and negative motivational inputs in ads featuring McCain, but this change was not observed during the viewing of Obama ads. It is unclear why this pattern emerges. Future research might explore the underlying reasons producing these results. Third, participants’ evaluations of the two candidates were measured following message exposure. This procedure was done in an attempt to avoid possible contextual effects of candidate evaluation measures on the sensitive physiological measures. The impact of the stimulus ads on candidate evaluation was expected to be minimal because the stimuli included a mix of randomly ordered ads both for and against each candidate. However, the study did not include an experimental check of the effects of the ads on postexposure evaluations. Future studies might measure candidate evaluations (or other political attitude constructs) before exposure to the stimuli. This might be accomplished by including a prescreening session.
Finally, this study focuses on the cardiac-somatic coupling pattern of attention during political ads processing. Future research should examine other key components of information processing, such as comprehension, memory, and attitude change, which would provide a more comprehensive understanding of the dynamic interactions between messages and attitudes (Lee, Roskos-Ewoldsen, & Roskos-Ewoldsen, 2008). In addition, different types of political messages and communication channels, such as debates (Holbert, LaMarre, & Landreville, 2009), conversations (Hayes, 2007), and online forums (Knobloch-Westerwick & Meng, 2009), introduce different degrees of interactivity and information dynamics. Other types of political messages could be tested and compared with political ads processing using the DMA framework. More importantly, motivational content can affect selective exposure in a dynamic way (David, Song, Hayes, & Fredin, 2007; Wang et al., 2011; Wang & Tchernev, 2012), and means of information acquisition (e.g., active selectivity or passive exposure), in turn, can affect cognitive processing (Wise & Kim, 2008). Therefore, a natural next step is to integrate dynamic selective exposure and selective attention to examine their mutual influences (Slater, 2007). This would provide further insight into how citizens dynamically interact with political information streams to construct a personalized political information environment.
Footnotes
Acknowledgements
The authors thank the three anonymous reviewers for their valuable comments.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was partially supported by the National Science Foundation (Grant No. SES 0818277) to the first author.