Volume 23 Number 3, 2013
Exploring Priming Effectiveness within News Stories According to Media Modality and Valence
Francesca Dillman Carpentier
~ ~ ~ ~ ~ ~ ~ ~ ~ ~
News is a complex creature. Consider online news, one of the most dominant means by which America becomes informed about its country. Few news stories employ only text when telling their story. More often than not, text is accompanied by at least one image that might or might not reinforce the meaning of the story’s headline. The complexity can grow from here, adding sound or video clips to the mix of text and image. All of these stimuli within this single story can influence how we interpret the story. More important, as news shapes our perceptions of the society we live in, these various stimuli can influence how we interpret our country.
As suggested in the quote above, numerous studies show that media content can “prime” or trigger ideas in our heads that influence our thoughts, feelings and actions without our express knowledge. Previous research also tells us that people do not process all elements of a news piece equally. Some elements stand out more than others. Some elements are remembered better than others. For example, our eyes tend to be drawn to images when presented with text, and we tend to process images more easily than text. Within modality, we tend to pay more attention to negative, as opposed to positive, information. Therefore, when positive and negative images and text are both present, which elements eclipse the other elements, outshining the other potential primes in yielding effects? In other words, in the face of multiple media stimuli available to sway us in one direction or other, which stimuli “win”? The present study evaluates how text and image elements in a news story compete to yield priming effects.
At its foundation, priming is the triggering, or activation, of a concept. This activated concept is then available to influence later thoughts and behaviors—the priming effect (see Roskos-Ewoldsen, Klinger, & Roskos-Ewoldsen, 2007 for meta-analysis). The general theoretical framework describing priming effects (e.g., Collins & Loftus, 1975; Higgins, 1996) begins with the proposal that concepts are stored in memory as nodes, which can be activated by a stimulus, making the concept ready for use by the individual. These nodes are connected to other nodes, with concepts that are closely related (e.g., children and family) having stronger connections than concepts that are weakly related (e.g., children and octopus). If a stimulus effectively triggers a node, that node activates and is now easily accessible in memory. This newly accessible concept is now available t o bias evaluations—when performing everyday evaluations, we tend to base our judgments on whatever criteria are most salient at the time.
Furthermore, if the trigger is strong enough, the activation experienced by the first node can spread to other closely connected nodes. Now, those closely connected nodes are also easily accessible in memory. Thus, the strength of the prime determines the strength, spread, and duration of the activation of nodes. Stronger primes yield stronger effects.
In media studies (see Roskos-Ewoldsen, Roskos-Ewoldsen, & Carpentier, 2008 for review), exposure to images or repeated themes within entertainment or news content have been shown to prime concepts of ethics, hostility, sexuality, or racial or gender stereotypes (e.g., Berkowitz, 1984; Domke, 2001; Valentino, Hutchings, & White, 2002). As a result, subsequent behaviors are skewed toward the activated concept (e.g., behavior was more hostile) and subsequent judgments are based more heavily on the activated concept (e.g., first impressions were influenced by racial stereotypes). Additionally, repeated exposure to certain issues in news (e.g., defense preparedness) have been shown to prime related concepts (national defense) in memory, which in turn are used as key criteria to judge political figures or political issues (see Iyengar & Kinder, 1987; Krosnick & Kinder, 1990; Pan & Kosicki, 1997).
Not surprisingly, news priming effects can carry from the specific realm of the trigger (e.g., racial or political cues to judgments about race or politics) to general evaluations of society or nation. For example, one study found that racial cues in crime news triggered concepts that influenced their views about public policy (Domke, 2001). In other words, race became a criterion participants used to evaluate public policy. The explanation for this observation lies with the ability of news to trigger multiple related concepts through spreading activation.
In addition to priming concepts, it is also possible to prime affect (Fazio, Sanbonmatsu, Powell, & Kardes, 1986; Hermans, De Houwer, & Eelen, 1994; Klauer, 1998). From the affective priming perspective, when individuals are primed with specific concepts (e.g., guns) that can spread to related concepts (e.g., violence), individuals may also be having specific affective reactions (e.g., negative affect) to the stimulus that biases subsequent evaluations of unrelated objects. In other words, the presentation of some stimulus material not only triggers that specific concept in memory, but also triggers its overall evaluation (good/bad).
A nice example that demonstrates the effectiveness and robustness of affective priming can be found in the work of Payne, Cheng, Govorun, and Stewart (2005). In their affective misattribution procedure (AMP), participants are briefly exposed to either a positively or negatively valenced prime (e.g., a smiling or angry face) and then shown a Chinese ideograph that the participants must evaluate as more or less visually appealing. The results consistently demonstrate that the valence of the prime influences the subsequent evaluations of the ideographs, such that when positive primes are presented in front of the ideograph, they are evaluated as more appealing than when negative primes are presented before the ideographs. Importantly, this effect persists even when the participants are warned that the prime can bias their evaluations.
Therefore, exposure to stimulus materials can prime not only concepts in memory, but also positive or negative affect, which in turn can influence subsequent evaluations of a target in the direction of the valence of the prime even if that target is unrelated to the prime itself. In the communication setting, this effect has been studied from a variety of research perspectives, ranging from advertisements (e.g., Kamins, Marks, & Skinner, 1991; Yi, 1990) to news (e.g., Baumgartner & Werner, 2012). Regardless of the domain, the effects are consistent with the AMP procedure in that whatever affect is triggered by the stimulus materials is used when making subsequent evaluations.
Oddly, there is scant research testing the media’s ability to trigger different concepts. Most often, the study will focus on one cue (e.g., racial cue) and the cue’s effect on a particular judgment or set of related judgments. One notable exception takes into account the fact that news media contain multiple triggers (Cho, Gil de Zuniga, Shah, & McLeod, 2006). This exception was concerned with cues (immigrant status and extremism) that can converge to create an additive effect on social judgments (social intolerance). Cues consisted of words (e.g., immigrants, citizens) embedded in the text meant to trigger the assumedly related concepts. The present study is more interested, however, in how the different visual modalities of image and text might compete, resulting in the strengthening or diminishing of a general priming effect.
There is also scant research testing the media’s ability to trigger multiple affective states. As described above, in the typical affective priming procedure, a prime with a specific valence (good or bad) is used and then it is determined the extent to which that prime was used in subsequent evaluations. Unclear from previous research is what would happen if a positive and negative primed were presented together—that is, if the primes were presented in competition.
Primes in Competition
We have learned from communication research that images can convey emotions rather quickly (e.g., Newhagen, 1998), and that images can enhance or alter our interpretation of news articles by serving as emotional exemplars that highlight certain aspects of those articles, for example a crying victim (see Aust & Zillmann, 1997; Zillmann, Gibson, & Sargent, 1999). More important, adding images to text appears to attract our attention and help us remember more about the message (e.g., Levie & Lentz, 1982; see also Zillmann, Knobloch, & Yu, 2001). Perhaps this attraction manifests because images are processed more easily than text (Linde, 1983; Madigan, 1983; Teasdale et al., 1999; see also Graber, 1996). Our eyes go to the image first, and our minds take the image in as a ready-made mental picture of the event.
Because the image is “let in” faster than text, it is likely that images would be the more intense prime than would image captions or headlines. In fact, there is evidence (e.g., Domke, Perlmutter, & Spratt, 2002) that images have a unique ability to work with an individual’s existing outlooks in triggering relevant concepts that later influence evaluative outcomes, such as affect toward the government, views of the political environment, and general optimism about the future. Granted, there is a good amount of evidence that suggests the best combination for message impact is one where images and text are redundant in their meaning (e.g., Houston, Childers, & Heckler, 1987; see Zhou, 2004, for audio-visual redundancy). However, some findings (e.g., Harrison, Taylor, & Marske, 2006) suggest images alone can wield an equally powerful influence as redundant image-text combinations. Thus, because images tend to be more attention-grabbing and more easily recalled, in other words more salient than text, concepts conveyed in an image should be triggered with greater strength than concepts conveyed in a comparable text headline. Similarly, the activation of specific affect (positive or negative) should be stronger when the affective prime is an image compared to text.
Still, there might be conditions under which the text headline will outshine the image.
For example, negative information “captures our attention far more than information about positive developments” (Sheafer, 2007, p. 23). A wide array of studies have noted that negative information plays an appreciable role in influencing a variety of social judgments, biasing the judgments toward a negative valence (e.g., Kanouse & Hanson, 1971; Lau, 1985; Sheafer, 2007). These observations are often attributed to our well-known negativity bias (e.g., Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001; Ito, Larsen, Smith, & Cacioppo, 1998; Kahneman & Tversky, 1984; Rozin & Royzman, 2001). To summarize the negativity bias, we humans have a natural tendency to devote more cognitive resources t o negative than to positive information, largely due to our innate desire to survive and to guard against threats to our well-being. This bias has been demonstrated to influence numerous outcomes, from self-reported attention to memory tests to physiological indicators of attention and arousal. Namely, studies testing for valence effects tend to find that our evaluations (e.g., impression formation) are more heavily influenced by negative information than by positive information, and that we notice the valence of a stimulus more quickly when that valence is negative rather than positive (see Dijksterhuis, 2010, for review). Based on this literature, it would be expected that primes that trigger negative affect, regardless of their delivery (i.e., image versus text), will outperform primes that trigger positive affect.
Holding modality constant and looking at images only, Ito et al. (1998) examined how instinctive the negativity bias is. Participants were asked to quickly categorize positive, neutral, and negative images according to the images’ intended valence. Positive and negative images had been pretested to ensure that the images were equally extreme, probable, and arousing. Using brain function as the key indicator of initial evaluations, Ito et al. noted that “the negativity bias operates automatically at the evaluative-categorization stage” (p. 897), meaning that participants were inherently more sensitive to the negative, than to the positive, information.
Teasdale et al. (1999) went a step further and pitted images with text in a way that varied the affective meaning of the pairing. Images with positive captions were tested against images with negative captions. Also included were images with captions that resulted in mixed valence, or ambiguity, in terms of the combined affective message. Results showed that different areas of the brain were activated when participants were trying to understand the negative image-caption pairs in comparison to the positive image-caption and mixed image-caption pairs, suggesting pure negative messages are processed differently than mixed or positive messages. There is also something different about how we process pure negative information as opposed to simply having a message that matches in valence (message redundancy).
This body of studies, combined with the research examining affective priming, suggests that valence might be one way to manipulate priming intensity, with negative primes providing a stronger trigger than positive primes. Therefore, one could hypothesize that purely negative primes would yield stronger priming effects than either mixed primes or positive primes. It is also plausible that the negativity bias is strong enough that a mixed-valence message would overshadow a pure positive message in influencing later judgments.
However, an important issue not directly addressed in the above studies is where the negative component of a mixed message is presented. In the Teasdale et al. (1999) study, the caption provided participants with the means for interpreting the image-caption pairing in terms of a specific valence. Each image by itself was neutral in terms of emotional appeal. What happens if image and text both have a distinct affective message, yet the messages do not necessarily reinforce each other? Which image-text pairing will yield the greatest effect?
Purpose of Study
It stands to reason that delivering a prime via an object that is naturally salient to us would boost any effect the prime might have on social judgments. Salience, in this case, refers to an element that readily captures our attention and is easily recalled. Because news often employs multiple types of elements when delivering a story, it is useful to discern which of the many elements that are presented will have the greatest effect on us, in order to more fully understand the various influences news has on our outlooks of community and country.
In efforts to begin a path of study on competing elements within an online news story, we are focusing our attention in this investigation on images versus text headlines, as well as negative versus positive valence. This focus is brought because many online stories have a headline in addition to at least one image, and many stories contain both negative and positive information. In addition, the basic comparisons between image and text, and between negative and positive information, are well-established in terms of ease of recall and initial attention.
Following the tradition of studies of news media on evaluations of politics, political figures, policy, and the like, we seek to determine how the treatment of different topics portrayed in online news are contributing to the individual’s overall assessment of the country. However, rather than focusing on the topics themselves, we are interested in the way the topic is presented—the modality (image, text) and valence of the prime. Valence, in particular, should be an important variable when considering how and why news shapes our thoughts about what is happening in our world, specifically in how we feel about what is occurring. Therefore, we are investigating how the way a prime is portrayed will moderate the extent to which an individual’s feelings about their society will contribute to their overall assessment of the country.
In attempts to explore the “which one wins” question, we offer two experiments using online news pages to test primes that vary in valence (positive, negative) and modality (images versus headlines or captions). In the first experiment, we examine the effectiveness of a prime if it is presented as an image only or headline only, as well as if both an image and headline are present. Primes in this case are in the context of a single news story. In the second experiment, we evaluate the effectiveness of competing primes in a more naturalistic setting of having the primes exist in conjunction with other content that might further draw the eye. The following general hypotheses are derived from the reviewed literature:
In a 2 (image valence; positive vs. negative) x 2 (text valence; positive vs. negative) between-subjects experiment with added image-only (positive or negative) and text-only (positive or negative) conditions, participants viewed a news webpage displaying one story that featured a positive image, with a positive headline, a positive image with a positive headline, a positive image with a negative headline, a negative image with a positive headline, a negative image with a negative headline, a negative headline, or a negative image. Headlines related to the story but did not directly describe the image. For conditions with an image only, a text line indicated that the story had been continued from a previous page (thus explaining the absence of a headline). Only a segment of the full story was viewable in the conditions, the segment of which was held consistent across condition. After viewing their randomly assigned news page, participants completed an online questionnair e that included the study’s dependent measures.
One hundred thirty-three undergraduates were recruited to participate from lower-level communication courses at a large, southeastern public university. There were more women (n = 109) than men in the sample, with the majority being Caucasian (n = 110) and in their fourth year of school (n = 75). Average age was 20.8 (SD = 1.6). Participants were randomly divided among the eight conditions, with 73 in the experimental conditions (positive image/positive headline = 19, negative image/positive headline = 14, positive image/negative headline = 19, negative image/negative headline = 21) and 60 in the control conditions (positive image only = 13, negative image only = 18, positive headline only = 16, negative headline only = 13).
Twelve news webpages were created for the main study. Prior to this creation, a pretest was conducted using an additional 40 participants recruited from the same population as the main sample. In this pretest, 12 images were tested using Bradley and Lang’s (1994) Self-Assessment Manikin (SAM) to gauge participants’ affective responses to the different images. Four images significantly differed on the positive-negative dimension of the SAM, but not on the low arousal-high arousal dimension—an important stipulation, as effects of valence should be separated from effects of excitation. Two of the four images featured a child (head and shoulders of a happy child, head and shoulders of an angry child), and two were sports related (wide shot of players celebrating, wide shot of players arguing). Worthy of note, both happiness and anger are generally considered to be approach, as opposed to avoidance, emotions (see Elliot, 2008).
Based on the results of the pretest, news pages were created containing either a positive or negative “child” or “sports” photograph paired with a positive or negative accompanying headline. No appreciable differences were found when exploring effects of story topic on subsequent evaluations, nor were any significant interactions between story topic and other control variables (e.g., gender) found. Therefore, “child” and “sports” conditions were collapsed and no further discussion of gender will be presented in this article. Image-only and headline-only versions were also created for each valence. The viewable story that accompanied each image and caption was neutral in tone and consistent across all conditions for its category (sports or children). The only other viewable elements on the page were the header and sidebar taken from the Seattle Times website. This was done in order to re-create the look of an authent ic screen capture of a Seattle Times webpage. Figure 1 shows one example of the manipulated stimuli.
Sessions were conducted in a computer laboratory containing 20 computers outfitted with a keyboard and mouse, thus accommodating 20 participants. The computers were located in open cubicles that aided in preventing one participant from seeing another participant’s condition. The delivery of the instructions, stimuli, and measures were all performed online.
Upon entering the laboratory, each respondent was instructed to sit at any computer, the computers of which had already been assigned to a condition. After all respondents were seated and consent forms were signed, respondents were then instructed to follow the instructions on the computer screen. Throughout the session, the experimenter remained in the computer room to answer questions, assist students with any problems they encountered, and ensure that respondents completed the entire online study.
The first portion of the study served primarily to collect demographic information. The participants then proceeded to view a screenshot of a page from the Seattle Times website. The participants were instructed to view the page just as they would any other online news page and click the “next” arrow whenever they were finished viewing.
After clicking next, participants continued to measures assessing their overall feelings and opinions about the country, from elements of society to politics to other domestic issues. Participants also completed items intended to mask the intent of the study. Upon completion of the session, respondents were debriefed and given course credit for their participation.
The primary dependent variable in this study was participants’ overall assessment of the state of the country. Response choices were on a 7-point scale ranging from 1 = Very Negative to 7 = Very Positive, M = 3.97, SD = 1.02. This item represents the subsequent evaluation used to test for priming effects.
Among the criteria available to predict the overall assessment were participants’ views of the educational system in the United States (1 = Needs Repair...7 = Works Well, M = 2.99, SD = 1.65), the economic state of the United States (1 = Optimistic...7 = Pessimistic, M = 4.66, SD = 1.36), violence in society (1 = Very Little...7 = A Lot, M = 5.22, SD = 1.08), and politicians’ understanding of the public (1 = Not at All...7 = A Lot, M = 3.27, SD = 1.26). All items were then coded so that lower scores indicated negative views and higher numbers indicated positive views.
With respect to the prime, which was intended to activate positive or negative affect, participants were asked to rate how they felt about society, using eight affect-based response choices inspired by the Positive and Negative Affect Schedule, or PANAS (Watson, Clark, & Tellegen, 1988). Response choices were on the following 7-point semantic differential scales: Good-Bad, Positive-Negative, Secure-Insecure, Lots of Faith-Little Faith, Safe-Unsafe, Joyful-Sad, Hopeful-Despair, and Happy-Angry. All items were coded so that higher scores indicated more positive feelings about society. These items (α = .84) were then averaged into a composite score of general affect toward society (1 = extremely bad, 7 = extremely good, M = 4.21, SD = .93). This composite was the key variable used to assess the prime’s (positive or negative affective activation) effect on an overall assessment of the state of the country. Specifically, we sought to test the extent to which participants’ affective responses contributed to the prediction of their overall assessment of the country, depending on the valence and modality of the prime.
The zero-order correlation between affect toward society and overall country assessment across all conditions was .53 (p < .01).
In line with political priming investigations, a hierarchical regression analysis was used to test how much the prime contributed as a criterion in the subsequent overall assessment of the country, the prime being an affective cue of the image and/or headline (see Table 1). All continuous variables were mean centered for analysis (Aiken & West, 1991). No multicollinearity issues were detected in any analysis (Pearson correlations between independent variables ranged from |r| = .01 to .24).
First, in order to test for any pure negativity effects, the image-only condition was analyzed by itself, as was the text-only condition. Entered in the first step of each model were control variables and main effects. Control variables for all analyses were participants’ views of education, the economy, violence and politicians. Gender was initially considered as a control variable, but as it yielded no appreciable variation, it was dropped from the analyses. Main effects for the image-only analysis were the participants’ affect toward society and type of image (negative = 0, positive = 1). Type of image was replaced with type of text for the text-only analysis. Entered in Step 2 was the two-way interaction between image (or text) and affect toward society.
For the image-only analysis, Step 1 yielded a significant model, although no specific predictors were statistically significant, R2 = .39, F(6, 24) = 2.61, p < .05. The addition of the two-way interaction did not improve upon the model, ΔR2 = .01, ΔF(1, 23) < 1. For the text-only analysis, affect toward society (the prime; B = 1.52, SE B = .28) significantly predicted participants’ overall assessment of the country in Step 1, R2 = .66, F(6, 22) = 7.05, p < .05. Adding the two-way interaction in Step 2 did not improve upon the model, ΔR2 = .001, ΔF(1, 21) < 1. Affect toward society remained the only significant predictor, B = 1.61, SE B = .50.
Next, a more thorough test of the hypotheses was conducted using the conditions in which image was presented along with text. Entered in the first step of this hierarchical regression model were the above control variables and main effects of affect toward society, image valence (negative = 0, positive = 1) and text valence (negative = 0, positive = 1). Entered in Step 2 were all two-way interactions pertaining to affect toward society and type of prime (e.g., affect toward society X image valence, image valence X text valence). Entered in Step 3 was the three-way interaction accounting for affect toward society and type of prime—the test of whether the prime’s effect on overall assessment was moderated by both modality and valence.
Views of politicians and affect toward society best predicted participants’ overall assessment of the current state of the country in Step 1, when type of prime was not taken into account, R2 = .42, F(7,65) = 6.71, p < .001. The addition of the two-way interactions did little to improve upon the initial model, ΔR2 = .02, ΔF(3,62) < 1. However, the addition of the three-way interaction (B = 1.29, SE B = .46, p < .01) yielded a significant improvement, suggesting that priming effects are explained by both negativity and image biases, Step 3 ΔR2 = .06, ΔF(1,61) = 7.75, p < .01.
Simple slope analyses were performed to examine the significant image valence X text valence X affect toward society interaction. Specifically, the contribution of affect toward society in overall assessments was evaluated for each combination of image and text. The slopes created by the analysis were anchored by the observed mean centered minimum (= -2.97) and maximum (= 1.66) scores for affect toward society. As seen in Figure 2, the contribution of affect toward society in predicting overall evaluations was largest for the valence-redundant combinations, negative image-negative text B =.80, SE B = .27; positive image-positive text B = .80, SE B = .38. The mixed-valence combinations had the smallest slopes, negative image-positive text B = .09, SE B = .43; positive image-negative text B = .23, SE B = .23.
According to subsequent one-tailed t-tests (see Aiken & West, 1991, p. 17), in support of H1c, both of the valence-redundant slopes significantly differed from zero, negative image-negative text t(11) = 2.93, p < .05; positive image-positive text slope, t(13) = 2.13, p < .05. In contrast, H1a and H1b were not supported as neither of the mixed-valence combinations differed from zero, suggesting the type of prime either had no effect in these conditions or the image and text valences cancelled each other out in influencing overall assessments, negative image-positive text t(6) < 1; positive image-negative text slope t(11) < 1.
Additional one-tailed t-tests showed that both valence-redundant slopes significantly differed from each of the mixed-valence slopes, ts = 5.00 to 7.07, dfs = 17 to 24, p < .01. The valence-redundant slopes did not differ from one another [t (24) < 1], nor did the mixed-valence slopes differ from one another, t (17) = -1.06, ns.
According to the results of Study 1, the valence of the prime did not appear to moderate priming effects when controlling for the modality of the prime (image-only or text-only). When modalities were seen in competition, both the negative image-negative text and positive image-positive text primes acted similarly, such that the relationship between affect and overall assessments was seen most clearly under conditions of valence redundancy. Put simply, image and text of the same valence yielded a prime of greater intensity, compared to image and text of differing valence. This supports H1c, which predicted the redundancy conditions would outperform the mixed-valence conditions.
This general finding does not support the proposition that priming effects would be moderated by both the well-known negativity bias (H1a) and also-established image bias (H1b), such that negative image-negative text and negative image-positive text combinations should outperform positive image-negative text and positive image-positive text combinations. Rather, according to Study 1, the answer to the question of which prime wins is a simple “neither.”
Contextualizing these results within the affective priming literature, it would appear that when the valences of the image and text primes were redundant, the prime was able to trigger the specific affect (positive or negative) that would then bias subsequent evaluations. However, when the primes were presented alone (i.e., the image or text only conditions), the affective prime was not strong enough to bias subsequent evaluations. Similarly, when the affective nature of the primes were mixed and therefore in competition with the other, there were no priming effects. It is possible, then, that the combination of a positive and negative affective prime when presented together simply cancels the other out.
Upon examining the design of Study 1, however, the prime was presented in a fairly isolated manner, in that there were no other stimuli on the page to compete with the primes themselves. This limits the ecological validity of the study, as it is rare to come across a webpage that only has one main piece of visual information. It also limits the ability to find negativity or image effects, as these effects might be more apparent when individuals have more options as to where they devote their cognitive resources. In other words, if negativity and image biases help us filter information, we must have enough information to filter in order for these biases to guide our selective information processing. After all, it is well documented in the communication literature that there is a very limited set of resources available to process mediated messages (Lang, 2000). A second study was therefore conducted with a more ecologically valid and complex webpage used as the stimulus.
Study 2 was conducted to explore whether the priming effects obtained in Study 1 could be replicated when the prime was embedded within a more natural and complex website featuring other news stimuli. According to Lang (2000, 2006), people are information processors and the basic components of processing are to perceive stimuli, attend to those stimuli, and then turn them into mental representations. Although we are all information processors, we all also have a limited and fixed pool of mental resources and attention that can be used at any given moment. When more information is presented than can be processed, or when only a little bit of attention is being allocated to a given task, we generally rely on cost-saving measures, such as paying more attention to images or negative information. By creating a more complex web environment for the second study, it is believed that less attention will be given to each individual element within the page, thereby creating a scenario more likely to cause certain elements of the page (e.g., images) to work as more effective affective primes.
As with Study 1, image and text manipulations were embedded within a homepage of the Seattle Times website. For Study 2, the webpage included multiple stories, rather than just the one story pertinent to the manipulation. The procedure mirrored that of Study 1. Regarding design changes, no image-only or text-only conditions were used in Study 2, as results from Study 1 clearly demonstrated that it is the primes working together that yielded the notable effects.
In a 2 (image valence; positive vs. negative) X 2 (text valence; positive vs. negative) experimental design, respondents were randomly assigned to one of four conditions. Depending on the condition, participants viewed a website that featured either a positive image with positive caption, a positive image with negative caption, a negative image with positive caption, or a negative image with negative caption. After viewing the website, participants completed the same questionnaire items as in Study 1.
Sixty-eight undergraduate students at a large, southeastern public university participated in the experiment. The participants were recruited from lower-level communication courses and ranged in age from 19 to 24, M = 20.2, SD = .98. There were more women (n = 58) than men, with the majority being Caucasians (n = 62) in their junior (n = 31) or sophomore (n = 25) year.
The same images and stories from Study 1 were used in Study 2. The key addition for Study 2 was that there were two other stories included on the page, both of which were neutral in affective appeal and held consistent across condition. One of these buffer stories contained a simple bar graph and the other had no accompanying image. Story location on the page was systematically varied to control for potential order effects.
In attempts to recreate the look of an authentic screen capture of a Seattle Times webpage, menu bars and part of an advertisement were also included on the page. Figure 3 shows one example of the manipulated stimuli.
The same measures used in Study 1 were again used in Study 2. These measures were: current state of the country (M = 3.37, SD = 1.19), views on education (M = 2.60, SD = 1.39), the economy (M = 2.51, SD = 1.33), violence (M = 4.57, SD = 1.04), and politicians (M = 3.32, SD = 1.22), and affect toward society (M = 3.75, SD = .86). Pearson correlations between independent variables ranged from |r| = .00 to .20). Affect toward society and overall country assessment correlated across condition at r = .53, p < .01.
A hierarchical regression analysis was performed using the same configuration of control variables and effects as the Study 1 analysis that included the three-way affect toward society X image valence X text valence interaction. In Step 1 (see Table 2), economic views and affect toward society initially were the best predictors of overall assessments of the current state of the country, R2 = .41, F(7,57) = 5.62, p < .001. The addition of the two-way interactions did little to improve upon the initial model, ΔR2 = .01, ΔF(3,54) < 1. However, the addition of the three-way interaction (B = 1.60, SE B = .07, p < .05) yielded a significant improvement, suggesting that priming effects again were moderated by both negativity and image biases, ΔR2 = .05, ΔF(1,53) = 4.69, p < .05.
Simple slope analyses were again performed to examine the significant three-way interaction, slopes being anchored by the observed mean centered minimum (= -2.37) and maximum (= 2.87) scores for affect toward society. As Figure 4 shows, the slope is largest for the positive image-positive text combination (B = 1.04, SE B = .33), followed by the negative image-negative text (B = .69, SE B = .20) and negative image-positive text (B = .56, SE B = .33) combinations. The positive image-negative text combination had the smallest slope, B = -.42, SE B = .56.
According to subsequent one-tailed t-tests (see Aiken & West, 1991, p. 17), the positive image-positive text slope (the highest positive slope) was significantly different from zero [t(8) = 3.17, p < .01], as was the negative image-negative text slope, t(7) = 3.54, p < .01. As with Study 1, this supports H1c. In partial support of H1a and H1b, the negative image-positive text slope approached statistical significance, t(5) = 1.70, p = .07. The positive image-negative text slope did not differ from zero, t(5) < 1, suggesting the prime either had no effect in this condition or the image and text valences cancelled each other out in influencing overall assessments.
Additional one-tailed t-tests showed that nearly all slopes significantly differed from one another at p < .01 (ts = 3.58 to 8.29, dfs = 10 to 15). The negative image-negative text and negative image-positive text slopes were statistically the same, t(12) = 1.34, ns.
As with Study 1, the results of Study 2 support H1c, which suggests that image-text redundancy should yield significant priming. Indeed, once again, it is those conditions when the affective primes are redundant (both image and text are positive or negative) that produced clearly significant affective priming results.
Unlike Study 1, though, there also appeared to be an interaction between mode and valence that influences the priming effect. Study 2 results suggest that when the prime exists in an environment in which there is more competing stimuli for the individual to filter, valence redundancy is not the sole force allowing the prime to emerge. Whereas the positive image reinforced with positive text did wield appreciable strength in priming participants, both negative image conditions also yielded significant priming effects. This lends support to both H1a and H1b as the results suggest that both negativity and image biases are in operation, in which negative images appear to have overshadowed positive text. Put simply, when negative information was presented as an image, there was affective priming regardless of whether the text accompanying the image was negative or positive; when the image was positive, though, affective priming only operated when the accompanying text was also positive, as there was no significance when the positive image was paired with the negative text.
It is noteworthy that, as in Study 1, the positive image-negative text condition yielded no priming effects. It would appear that having a positive image is not powerful enough alone to override the negative text, but at the same time, the negativity of the text is not enough to override a positive image. This can be interpreted as further corroboration that the negativity bias works in concert with the image bias to influence priming effects under conditions in which the individual is faced with various points of information. More generally, simple reinforcement of a message (in this case, valence redundancy) is not a sufficient explanation if multiple stimuli are competing for the individual’s cognitive resources.
These results conform to the general predictions made from the limited-capacity approach (Lang, 2000). It could be argued that within this second study, there was more information presented to the participants, who in turn relied more heavily on mental shortcuts than when the webpage only presented the single story (as in Study 1). This increase in information complexity could explain the emergence of the negativity and image bias in the second study as fewer mental resources were allocated to the processing of each element presented on the webpage.
This assertion is supported by the research of Bornstein and D’Agonstino (1992, 1994). Their research, which examines the role of fluency and affect, has demonstrated that when affective primes are presented subliminally, the priming effects tend to be stronger than when the affective primes are presented supraliminally. These results can be interpreted as meaning that affective primes are strongest when participants’ direct attention and awareness of them is lowest—that is, the affective prime operated most effectively when it was allowed to operate entirely automatically as it was outside of the awareness of the participants. In some ways, this is what was accomplished in Study 2. By adding more stories and other images and text for participants to look at, less attention was focused on the primes themselves as their attention was divided among many things. When their att ention was divided in this manner, they were more susceptible to the automatic nature of the image and negativity biases.
In summary, the results from Study 2 demonstrate that participants were able to be affectively primed when the image and text were redundant in valence. Unlike the first study, though, when the image was negative and the text was positive, a negativity and image bias emerged such that participants were affectively (negatively) primed. The image and negativity bias both seem to be in operation, though, as their working in conjunction would explain the lack of significance for the positive image-negative text condition.
Two studies were used to explore how priming effects might be moderated by the valence (positive, negative) or modality (image, text) of the prime itself, in efforts to begin testing which prime “wins” (see Bargh, 2006) when presented in competition with another prime. Competing hypotheses were proposed with regard to negativity biases, image processing biases, and message redundancy. If the negativity bias best explained priming effectiveness, any prime with negative information (negative image-negative text, negative image-positive text, positive image-negative text) should outperform all-positive primes (positive image-positive text). However, if image biases worked in conjunction with the negativity bias, image should intensify negativity, such that negative image-negative text and negative image-positive text would outperform positive image-negative text and positive image-positive text. Still, if priming effectiveness was best ex plained by simple redundancy, valence-matched configurations would be expected to equally outperform mixed-valence configurations. Findings suggest that priming effectiveness is explained by both redundancy and the interaction between negativity and image biases, but that these explanations depend on the complexity of the information environment.
In both studies, valence redundancy produced significant priming effects, suggesting quite simply that two primes are better than one, whether the primes are presented in a fairly information-sparse or information-busy environment. The additive nature of the primes was further clarified by the lack of significant results when the images or text were presented alone. In isolation, the images or text—regardless of valence—did not produce any affective priming. It was only when the primes were put together that effects were obtained. This finding is consistent with, and perhaps a slight extension of, previous research that demonstrates effectiveness of message redundancy in influencing or educating audiences. The finding is also consistent with direct tests of priming intensity, which vary the length, duration, or frequency of the prime to evaluate strength of the effect (e.g., Dillman Carpentier, Roskos-Ewoldsen, Roskos-Ewold sen, 2008).
Regarding negativity versus image biases, in Study 1, where only the story featuring the primes was presented, competing primes of mixed valence appeared to cancel each other out, suggesting that neither the negativity nor image bias were robust enough to carry the effect. In other words, having participants simultaneously primed with both positive and negative affect appears to have not caused any bias one direction or the other in any subsequent evaluations.
However, in Study 2’s information-complex environment where other stories were presented alongside the prime-relevant story, the negativity bias and image bias interacted to influence priming effects beyond redundancy. When the negative prime was presented as a text and matched with a positive image, no effects were found. However, the effectiveness of the negative image-positive text condition was not far behind the negative image-negative text condition in producing the anticipated effects. Thus, it appears that when individuals are faced with more stimuli fighting for attention, the negativity and image processing biases emerge to help sort out which information warrants our cognitive effort. In this way, we are more susceptible to affective priming whenever we are devoting less attention to the surrounding environment.
There are several limitations to this study that should be acknowledged. First, the samples were drawn entirely from an undergraduate population. Although this is fairly standard practice and there is no theoretical reason that age should influence these results, having a more diverse age range would help generalize the findings. Second, the samples had more Caucasians and females; having a more diverse subject pool would also help to generalize the results. Third, the samples were not particularly large for the type of analysis conducted; however, the overall models and interactions were significant, which suggests our sample size was sufficient. Some of the secondary analyses performed produced marginally significant results, though, and it is believed with a slightly larger sample, those results would have been significant.
Finally, it is possible that the size difference between the image and the text in the presentations could have influenced the results of this investigation. This is more of a concern with Study 2 than with Study 1, as the text in Study 1 was more comparable to its image in terms of screen area. Likewise, the positioning of the text and image might have influenced the findings, albeit the text in Study 1 was above the image and the text in Study 2 was below the image. In future research that examines competing effects of modality, taking care to produce equivalent sizes and vary the text and image positions would help eliminate any confounds that might arise from one prime being more eye-catching than another due to its placement.
Despite these limitations, the present study highlights an area of priming research related to priming intensity that is in need of additional attention. As news provides us with a sea of elements through which we must navigate to fully comprehend the story, it is important to try to understand which elements have the largest impact on us, whether we recognize that impact or not. Whereas substantial research has shown that media messages can influence subsequent decision-making, little research has looked into what happens when message elements compete for our information processing resources. As such, this study represents a ”second-generation” investigation of priming, shifting the focus from showing if a priming effect exists to asking when and how primes can interact. By answering such second-generation questions, perhaps a more complete understanding of media effects will begin to emerge and the lessons learned from this media ef fects research will become more accessible to industry professionals who might use these lessons in practice. Certainly, we will learn more about the ways in which news can shape how we think about our nation.
Bornstein, R.F., & D’Agostino, P.R. (1994). The attribution and discounting of perceptual fluency: Preliminary tests of a perceptual fluency/attributional model of the mere exposure effect. Social Cognition, 12, 103-128.
Higgins, E.T. (1996). Knowledge activation: Accessibility, applicability and salience. In E.T. Higgins & A.W. Kruglanski (Eds.), Social psychology: Handbook of basic principles (pp. 133-168). New York, NY: The Guilford Press.
Ito, T. A., Larsen, J. T., Smith, N. K., & Cacioppo, J. T. (1998). Negative information weighs more heavily on the brain: The negativity bias in evaluative categorizations. Journal of Personality and Social Psychology, 75, 887–900.
Kanouse, D.E., & Hanson, L.R., Jr. (1971). Negativity in evaluations. In E.E. Jones, D.E. Kanouse, H.K. Kelly, R.E. Nisbett, S. Valins, & B. Weiner (Eds.), Attribution: Perceiving the causes of behavior (pp. 47-62). Morristown, NJ: General Learning.
Roskos-Ewoldsen, D.R., Klinger, M., & Roskos-Ewoldsen, B. (2007). Media priming. In R.W. Preiss, B.M. Gayle, N. Burrell, M. Allen, & J. Bryant (Eds.), Mass media theories and processes: Advances through meta-analysis (pp. 53-80). Mahwah, NJ: Lawrence Erlbaum.
Roskos-Ewoldsen, D.R., Roskos-Ewoldsen, B., & Carpentier, F.D. (2008). Media priming: An updated synthesis. In J. Bryant & M.B. Oliver (Eds.), Media effects: Advances in theory and research (3rd ed., pp. 74-93). Mahwah, NJ: Lawrence Erlbaum.
Teasdale, J.D., Howard, R.J., Cox, S.G., Ha, Y., Brammer, M.J., Williams, S.C.R., & Checkley, S.A. (1999). Functional MRI study of the cognitive generation of affect. American Journal of Psychiatry, 152, 209-215.
Watson, D., Clark, L.A., & Tellegen, A. (1988). Development and validation of brief measures of positive and negative affect: The PANAS scales. Journal of Personality and Social Psychology, 54, 1063-1070.
Copyright 2013 Communication Institute for Online Scholarship, Inc.
This file may not be publicly distributed or reproduced without written permission of