ABSTRACT
The use of neuroscience methods in advertising research continues to grow, but it remains controversial. One area of neuroscience that has the potential to advance understanding of consumer decision making is neural-network analysis. The authors draw a parallel between means–end decision theory and neural-network analysis. They then apply these two theoretical perspectives to validate empirically a recognized advertising-strategy assessment (Strata) model. The results of an analysis of 240 television advertisements offer support for the neural-network-based Strata model. The article concludes with recommendations for how to improve advertising effectiveness.
MANAGEMENT SLANT
The spreading activation of a neural network underlies consumer decision making.
Means–end theory is a special case of a neural network that underlies advertisement effectiveness.
The Strata methodology can be used to develop effective advertisements that create a neural network.
The Strata methodology effectively can be used to assess finished or animatic advertisements, which results in time and cost savings.
The Strata model predicting advertisement effectiveness was validated.
INTRODUCTION
Neuroscience is an emerging field of consumer research that has garnered much interest among advertising researchers, who hope that neuroscience tools can help them better understand why customers prefer some products over others and which communications best influence customers' preferences (Fugate, 2007; Martínez-Fiestas, Del Jesus, Sánchez-Fernández, and Montoro-Rios, 2015; Nobel, 2012; Plassmann, Ramsoy, and Milosavljevic, 2012; Varan et al., 2015). This ambitious goal relies on several neuroimaging, or brain imaging, techniques. Researchers can deploy these techniques to reveal how the hidden elements of the consumer decision process are activated and to better understand how a product or an advertisement engages the pleasurable reward center in consumers' brain (Karmarkar, 2011; Nobel, 2012; Smidts et al., 2014).
In addition to increasing knowledge of brand function, neuromarketing approaches potentially are faster and cheaper than traditional advertising-research tools that ask customers directly for their thoughts, feelings, and decision-making strategies (Ariely and Berns, 2010; Varan et al., 2015). Although cost savings is one possible outcome of this evolving area of research, most experts believe that the most important potential contribution of neuroscience to advertising is its ability to guide theory generation. This process can be used to shape new models of consumer decision making and to assess traditional models of consumer responses to advertising currently in practice (Ariely and Berns, 2010; Kenning, Plassmann, and Ahlert, 2007; Plassmann et al., 2012; Smidts et al., 2014; Vakatsas and Ambler, 1999; Venkatraman et al., 2015; Yoon et al., 2012).
Neural-network research, inspired by the neural architecture of the human brain, advances the understanding of consumer decision making (Knutson et al., 2007; West, Brockett, and Golden, 1997). A neural network's ability to mimic the brain's function is the strength of its approach. From a cognitive perspective, neural-network models are consistent with a spreading-activation model of memory, which makes neural-network models well suited for representing judgment and decision making that involve the processing of information (Bhatt, 2012; Chowdhury and Samuel, 2014; Glimcher, 2009; Payne, Bettman, and Johnson, 1993; West et al., 1997). Neural-network analysis therefore seems a particularly apt approach to better understand advertising effectiveness and to assess traditional models of consumer decision making (Briesch and Rajagopal, 2010; Curry and Moutinho, 1993; Knutson et al., 2007).
The current study applies a neural-network approach to assess an established model of consumer decision making: means–end theory (Gutman, 1982). Means–end theory has been used widely in advertising research, as operationalized in the laddering methodology and the advertising-strategy assessment (i.e., Strata) methodology (Gengler and Reynolds, 1995; Reynolds and Craddock, 1988; Reynolds and Gengler, 1991; Reynolds, Gengler, and Howard, 1995; Reynolds and Gutman, 1984, 1988; Reynolds and Olson, 2001; Reynolds and Phillips, 2009; Reynolds and Rochon, 1991).
In the current study, the authors analyzed 240 television advertisements across a variety of product categories and levels of finish (i.e., animatic and finished advertisements) that served as stimuli across a diverse sample of consumers. Study participants responded to questions through the Strata methodology, a computer-driven, tailored interviewing system. The order of questioning reflected the spreading-activation model underlying neural-network theory.
To test the neural-network basis of the means–end grounded Strata model, the researchers empirically assessed the strength of the linkages between the concepts (or elements) of a means–end chain and advertising effectiveness, operationalized as purchase intention. The researchers posited that a significant positive relationship would be supportive of a spreading-activation model of consumer decision making associated with a neural network. The results of this analysis support the neural-network basis of the Strata model and therefore means–end theory. This suggests that a hierarchical means–end chain representation of a consumer-preference network corresponds to a neural-network brain structure that underlies consumer decision making.
BACKGROUND
Neural Networks
A neural network is a connectionist model of brain behavior often used to understand human cognition (West et al., 1997). The interconnections, or linkages, between neurons are referred to as “synapses.” Neural-network models resemble the brain's decision-making process, whereby input neurons receive stimuli, which they then feed into a pattern-matching process that yields a decision (Bhatt, 2012; Curry and Moutinho, 1993).
The fundamental principle of a neural network is that when a neuron is activated, or fired, it then can cross a synapse gap to activate another connected neuron. In neuropsychology, neural networks are computational models intended to represent biological neural networks in the brain. Researchers use these models to solve certain kinds of problems (Briesch and Rajagopal, 2010; Chowdury and Samuel, 2014). The basic logistical calculus of a neural network is that a neuron receives inputs, processes those inputs, and generates an output (McCulloch and Pitts, 1943). In general, neural networks do not follow a linear path; researchers believe that information is processed collectively throughout the entire network of neurons, also called nodes.
With respect to decision making, the synaptic-plasticity postulate (Hebb, 1949) suggests that one's neural network is strengthened over time, becoming more efficient and efficacious as a result of repeated activation (i.e., stimuli exposure) and personal experience. Researchers thus can use neural-network analysis to understand consumer decision making with respect to a problem that a product or service solves. The approach, in addition, subsequently offers the potential to assess advertising effectiveness. More simply put, nodes are concepts or ideas that, when connected, have meaning to the consumer. This network of relevant meaning to the consumer is a means–end chain that underlies the consumer's preference and thus purchase intention.
In the next section, the authors introduce means–end theory, a well-known consumer decision-making model (Gutman, 1982), as well as its two associated research methodologies—laddering (Reynolds and Gutman, 1988; Reynolds and Phillips, 2009) and Strata (Reynolds and Craddock, 1988; Reynolds and Gengler, 1991; Reynolds and Gutman, 1984; Reynolds and Rochon, 1991). They then use this theory to draw parallels to a neural-network model of decision making.
Means–End Theory
Laddering Methodology. Means–end theory is a commonly utilized and frequently referenced framework of consumer decision making (Gutman, 1982). Previous work extended means–end theory to advertising management (Reynolds and Gutman, 1984). Researchers later developed its associated research methodology, laddering, which allowed for the collection and analysis of the hierarchical means–end chain data across four levels of abstraction (Reynolds and Gutman, 1988; Reynolds and Phillips, 2009).
The laddering methodology begins with a trained interviewer asking a series of questions of a consumer, with the goal of abstracting to the higher order meanings that drive the consumer's decision making. Each level of abstraction (i.e., node) is a concept; these concepts are linked together by meanings relevant to the consumer. Laddering therefore uncovers the nodes (i.e., concepts) and dominant connections that are relevant and meaningful and that drive the consumer's choice.
The first questions of a laddering interview usually do one or more of the following:
elicit a distinction between two stimuli or choice options (e.g., “Is your most preferred coffee brand Starbucks or Illy?”);
ask about a stated preference (e.g., “Why do you prefer Starbucks to Illy?”);
ask about actual consumption (e.g., “Why do you drink more Starbucks than Illy?”).
Then the interviewer probes the consumer's answers with some version of the question “Why is that important to you?” The interviewer uses each answer as the basis for the subsequent probe, moving the consumer up the ladder of abstraction from (1) attribute distinction to (2) functional consequences to (3) psychosocial consequences to (4) personal values.
The laddering methodology's levels of abstraction are isomorphic to the neural-network approach of obtaining the relevant connections both between and across the decision-based nodes. The result of a laddering interview is a complete means–end chain consisting of four concepts (i.e., nodes) and three adjacent, direct connections (i.e., linkages; 1–2, 2–3, 3–4) as well as indirect connections (1–3, 1–4, 3–4). The result of a laddering study is a directed graph depicting the network of direct and indirect connections across nodes for a given sample of consumers. This graphical network is a consumer decision map because it illustrates the consumer's key decision nodes and their dominant associative connections. In addition, the pathways from bottom to top can be thought of as decision segments that advertisers can quantify to develop communication strategy (Phillips, Reynolds, and Reynolds, 2010; Reynolds, 2006).
Note that means–end theory is a hierarchical top-down approach to understanding consumer decision making. The laddering methodology, in contrast, is a hierarchical bottom-up approach whose goal is to identify the end state that defines the motivating dynamic of the decision structure. That is, the lower levels develop their importance by satisfying the respective, adjacent higher levels, whereas the laddering methodology is initiated by a distinction that is typically at the lowest, attribute level. The goal of the laddering methodology thus is to uncover a network of meanings, which also defines the association network of connections, or linkages. The consumer decision map may be viewed as a special case of neural-network model: one that focuses on only the aggregate, relevant decision-making elements, with the assumption that the levels of abstraction reflect the underlying decision-making process.
The classic means–end example is a 1984 study used to develop political strategy for the Reagan versus Mondale 1984 presidential campaign (Bahner and Fiedler, 1985; Norton, 1987). (See Figure 1 for the resulting consumer decision map from this study.) Note that in this political example, the nodes and connections were obtained through laddering interviews, and node ownership (i.e., Reagan or Mondale) was obtained from traditional polling that was conducted at the same time. This graphical representation thus is a snapshot taken at a given point in the campaign.
Because polling was used, questions about higher level values were not asked directly of the respondents; these concepts were considered too abstract. Recall that according to means–end theory, these higher ordered values are considered the defining strategic goals that give the lower level nodes their importance. (For a detailed review of means–end theory and the laddering methodology, see Phillips and Reynolds, 2009; Reynolds and Olson, 2001; Reynolds and Phillips, 2009).
In development of a communication strategy, the consumer decision map represents a strategic game board. The goal is to own as many decision nodes as you can while heeding three strategy-development rules:
Control the higher levels.
Develop strategies that connect as much as possible from bottom to top.
Neutralize, or block, your opponent's largest equities by lever-aging your own equities.
Blocking is a key factor in political-strategy development because the ability to affect preference is determined by the ability to connect the entire means–end chain. If one node cannot be activated, then the network of meaning is short-circuited, and the personal value that underlies preference is mitigated.
In the Reagan–Mondale consumer decision map example (depicted in Figure 1), two strategies immediately appear. The first is to own the “secure self/children's future” node (and above) by linking this node to the “build individual opportunity” and “economic recovery” nodes (the first two rules above), thereby blocking the social-issue disequities (i.e., negative concept nodes; the third rule). The second, alternative strategy is to own the “preserve world peace” node (the first two rules) by connecting it to “strengthen defense preparedness” (the third rule), thereby blocking the disequities on the right, foreign-policy side of the consumer decision map.
As the creator of this diagram (Norton, 1987) noted, advertising strategies were developed in this way through this decision-based framework. The strategies were specified with the means–end chain conceptualization of advertising strategy (MECCAS) framework (Reynolds and Craddock, 1988; Reynolds and Gutman, 1984) and were executed successfully (Wirthlin, 2004, pp. 142–145).
The Strata Methodology. The Strata methodology is a computer-driven, tailored interviewing system used to pretest advertising strategy and to assess and learn how advertising functions (Reynolds and Gengler, 1991; Reynolds and Rochon, 1991). Previous work (Reynolds and Rochon, 1991) drew a meaningful distinction between traditional copy-testing methods and Strata, operationalized as a neural network between the means–end levels of the MECCAS framework (Reynolds and Craddock, 1988; Reynolds and Gutman, 1984). Included in the MECCAS model are the advertising executional elements (i.e., concept nodes) and how these elements connect at adjacent levels of abstraction by the advertisement (Gengler and Reynolds, 1995).
There are two underlying assumptions of this means–end decision framework. First, the two lower levels (i.e., attributes and functional consequences) are product based, and the two upper levels (i.e., psychosocial consequences and personal values) represent the self-defining, personal motivations driving choice. Second, the network connecting the meanings between the product-defining and the self-defining levels of abstraction is the key to affecting brand choice underlying advertising effectiveness.
One study found that the ability of nodes and connections to predict purchase intention differed by the brand consumed most often (Reynolds et al., 1995). When consumers viewed the competitor's advertisement, connection strength was more predictive of purchase intention. The explanation posited was that considering a new product involves learning and the creation of new connections. Because a major objective of advertising is to get new customers, making connections is essential. If an advertisement only works for current customers, sales eventually will decline. For consumers viewing their brand's advertisement, the nodes predicted purchase intention, which suggests an established decision network. The findings suggest that when users see their brand's advertisement, it activates a node that, in turn, activates an existing network.
Strata is a special case of neural networks obtained by direct questioning and not by psychophysical measurement. The objective is to measure the strength of the connecting synapses between the adjacent levels of abstraction, which are thought to underlie decision making. Strata uses a unique questioning procedure that directly obtains the strength of the connections between the active concept nodes that are triggered by a specific advertisement. Strata therefore relies on three key measures:
node communication strength at each of the four levels of abstraction;
connection strength between the adjacent level nodes;
affect measures, including
outcome measures of advertising effectiveness operationalized as purchase intention, and
measures of advertisement affect (i.e., advertisement liking and entertainment).
With respect to procedure, the steps of the Strata methodology avoid some of the limitations of traditional modes of advertising research, such as complex question formations and participants being focused overly on a single advertisement. This refinement is accomplished through a computer-administered format that requires two advertisements (e.g., A and B) to be assessed together. Tasking participants to keep both advertisements in memory while responding to the node communication-strength questions avoids the problem of participants being focused too much on a single advertisement (Reynolds and Rochon, 1991).
The Strata procedure is outlined below. Note that participants view each advertisement four times—twice individually, and twice together. (See Figure 2 for a model of the Strata analysis output.)
View Advertisement A
Qualitative questions (e.g., main point)
View Advertisement B
Qualitative questions (e.g., main point)
View both advertisements (A and B): Instructions require the respondent first to answer the question “For which advertisement or advertisements is the concept ‘clearly’ communicated?” For each concept statement that is communicated clearly, the respondent then is asked whether it is “clearly” or “perfectly” communicated. This is how the simple question format is operationalized for each statement node.
Affect statements (both brand [purchase intent] and advertisement [entertaining])
Message-element statements (product attributes)
Consumer-benefit statements (functional consequences)
View both Advertisements A and B
Executional-framework statements
Leverage-point statements (psychosocial consequences)
Driving-force statements (personal values): Instructions present a graphic Venn diagram scale to explain the concept of connection, with three examples to define the scale options.
View Advertisement A
Advertisement A connection-strength questions (message element to consumer benefit) just for the “communicated nodes.”
Advertisement A connection-strength questions (consumer benefit to leverage point) just for the “communicated nodes”
Advertisement A connection-strength questions (leverage point to driving force) just for the “communicated nodes”
View Advertisement B
Advertisement B connection-strength questions (message element to consumer benefit) just for the “communicated nodes”
Advertisement B connection-strength questions (consumer benefit to leverage point) just for the “communicated nodes”
Advertisement B connection-strength questions (leverage point to driving force) just for the “communicated nodes.”
Strata Example. Strata analysis output for a given sample presents a data summary for both statements (i.e., nodes and connections, or linkages), along with affect (both product and advertisement) summary measures. Consider as an illustration the advertisement “Kid,” also from the 1984 Reagan–Bush presidential reelection campaign. This advertisement was based on the strategic option developed to negate the negative of “no war” (Reynolds, Westberg, and Olson, 1997; See Figure 1).
In this advertisement, President Reagan is the voiceover saying, “A president's most important job is to secure peace, not just now, but for the lifetimes of our children.” He then says, “But it takes a strong America to build a peace that lasts,” and then, “And now we can work for a lasting peace for our children and their children to come.” He continues, “Peace is the highest aspiration of the American people. Today America is prepared for peace.”
The advertisement visuals are of children three to six years old of various racial backgrounds, engaged in activities including the following: getting a haircut, playing with a dog, in the rain, standing on a porch waving an American flag that is gently flapping in the breeze. In the final scene President Reagan is speaking to an audience, saying, “We will negotiate for it. Sacrifice for it. We will not surrender for it now or ever.”
The Strata summary has three basic components: the node strength, the connection strengths between nodes, and the affect (or advertisement effectiveness) summary. Note that in the node-scoring system, “somewhat” is interpreted as half of the “clearly” communicated score. The connection-summary score is computed from probabilistic scaling of the multiplicative combinations of the two nodes (0, 1, and 2) and connection assessment (0, 1, and 2).
The Strata assessment of “Kid” based on the node-activation and neural-connection scores suggests a tightly linked decision network (a summary score of 19) from the message elements (candidate descriptors) to the driving forces (personal values; See Figure 3). This neural-network paradigm provides a straightforward basis to assess the overall strategic effectiveness of the advertisement.
Affect scores are computed in the same way as the node scores. Although “Kid” was not seen as particularly entertaining, it was effective in terms of influencing behavioral affect, or voting intention (i.e., “more likely to vote for this candidate”). This finding suggests that the neural representation corresponds to a central (as opposed to peripheral) model of persuasion (Petty, Cacioppo, and Shumann, 1983).
RESEARCH QUESTION
The current study asks whether means–end theory, as operationalized by the Strata methodology, is a special case of a neural network that reflects consumer decision making in response to an advertisement. This research posits that it is and that, given the Strata operationalization, the strength of the linkages among the concepts of a means–end chain will be correlated positively and significantly with purchase intention. This research also hypothesizes that a significant positive correlation is supportive of a spreading-activation model of consumer decision making associated with a neural network. The methodological and analytical procedures are presented in the next sections.
RQ1: Is means–end theory, as operationalized by Strata, a special case of a neural network that reflects consumer decision making in response to an advertisement?
METHODOLOGY
Research Design and Participants
Participants were widely disparate in terms of product affiliation, usage loyalty, and demographics, across both age and gender. The average number of participants for each of the 240 advertisements was 46. Although some studies were balanced across multiple advertisements, typically participants saw the same two advertisements in this repeated-measures design. The total number of participants across all 240 advertisements was approximately 5,520.
Sample and Unit of Analysis
The sample was composed of 240 advertisements from eight countries (63 percent from the United States) across a broad range of product categories. The advertisements were a combination of finished advertisements (55 percent) and animatic advertisements (45 percent). The unit of analysis for this research was the advertisement.
Measures
The Strata methodology yielded data for the four measures described below (See Figure 3). The data for the node communication-strength and linkage-strength measures were from the largest overall “nodes-in-common” network—message element → consumer benefit → leverage point → driving force (i.e., highest total of a common means–end chain)—created by the given advertisement. Because each of the 240 advertisements was evaluated by an average of 46 participants, the value for each of the four measures for a given advertisement was the mean of the scores from these participants.
Strata Node Connection Strength. Node communication strength was measured in two steps. First, participants were asked whether the node concept was “clearly communicated” or “not at all communicated.” If the node concept was “clearly communicated,” the participant was asked whether it was “perfectly” communicated or “clearly” communicated. Node communication strength ranged from 0 to 100, with “perfectly” communicated scored as 100, “clearly” communicated scored as 62, and “not at all” communicated scored as 0. (The “somewhat clearly” value of 30 listed in Figure 2 was only an interpretive definition; the only measured score values were 0, 62, and 100.)
Strata Bridge Linkage Strength. Linkage strength was the independent variable. The researchers computed it by asking for only those nodes that were “clearly” communicated between adjacent levels of the degree of associative meaning caused by the advertisement. A Venn diagram defined the graphic rating scale.
The computation of the summary index was based on a probabilistic function derived from the multiplicative model of Node(i) × Node(ii) × Linkage(i–ii), which was computed from a 0, 1, or 2 score for each node and their linkage. This calculation means the maximal linkage score possible for each connection was 8. The researchers used the likelihood of possible scores to convert the multiplicative score to a connection score. Connection scores ranged from 0 to 9, with 1 interpreted as a weak connection and 9 interpreted as a perfect connection.
Product Affect. Product affect, operationalized as purchase intention, was the dependent variable. The variable included an attitudinal item measuring how much the participant liked the brand or product after seeing the advertisement (e.g., “This advertisement gives me a more favorable view of the company or product”) and a purchase-intention item (e.g., “This advertisement makes me more likely to consider purchasing the product made by the company”). The researchers presented these items for each advertisement, and the order of presentation was randomized.
These product-affect measures were scored in a manner similar to the node communication-strength ratings above, and the scores ranged from 0 to 100. First, participants were asked whether they more likely would purchase the product. Response alternatives were “clearly” true or “not at all true.” If respondents answered, “Clearly true,” they were asked whether their purchase intention was “clearly” true or “perfectly” true.
Advertisement Affect. Advertisement-affect measures included two items that assessed how entertaining the advertisement was and how well the advertisement held the participant's attention. Advertisement affect was scored in a manner similar to the node communication-strength ratings above. The scores ranged from 0 to 100.
RESULTS
The research question was tested with correlational analysis. Correlations were calculated between the independent variable (Strata bridge linkage strength) and the dependent variable (purchase intention). Recall that the research question posited that if means–end theory, as operationalized by the Strata methodology, is a special case of a neural-network spreading-activation model of consumer decision making, Strata bridge linkage strength would be correlated positively and significantly with advertisement effectiveness, operationalized by the purchase-intention measure.
The authors calculated the Pearson zero-order correlation coefficients (rs) between the independent variable (Strata bridge connection strength) and the dependent variable (purchase intention; See Table 1). Note that the three linkage scores (A. Product Bridge, B. Personal Relevance Bridge, and C. Value Bridge) represent the largest nodes-in-common means–end chain, because this was the largest single network that could affect purchase intention. Recall that bridge linkage strength was measured only when both of the adjacent nodes were “clearly communicated.”
The sum of the three linkage strength scores (A + B + C) represents a neural-network-based strategy. The square of the correlations (r2) represents the proportion of variance in the dependent variable (purchase intention) that is explained by the independent variable (linkage strength). The authors used the Fisher r-to-z transformation to test for statistically significant differences between correlations.
All of the sample-dependent correlations (See Table 1) were significantly different from zero, including the correlation between the overall model (A + B + C) and purchase intention. In addition, the difference between the lowest level (A. Product Bridge) and the highest level (C. Value Bridge) was not statistically different. This suggests that these linkages were all of approximately equal importance (rProduct Bridge = .70 versus rValue Bridge = .61, z = 1.65, p > .05, two-tailed). The proposition that Strata methodology reflects a neural-network spreading-activation model of consumer decision making therefore is supported.
Note that the overall model (A + B + C) accounted for 62 percent of the variance in predicting purchase intention. The magnitude of this effect is meaningful considering that this model is only for the single most-connected decision network and is not a full consumer-decision map (as depicted in Figure 2). This result adds further support to earlier findings that the higher levels of means–end chains are correlated more with preference (Perkins and Reynolds, 1995; Reynolds, Gutman, and Fiedler, 1984).
Post Hoc Analysis 1: Advertisement Affect
This post hoc analysis explored the relationship of advertisement affect (operationalized as entertaining advertisement) to the aggregate network summary (A+B+C) vis-à-vis the product-affect dependent variable (operationalized as purchase intention). Many traditional copy-testing procedures use measures of how likable or how entertaining the advertisement is as a measure of advertising effectiveness. The squared correlation summary data permit a direct comparison with the purchase-intention measure. The amount of variance accounted for with respect to purchase intention was almost three times higher for the connection model than with respect to the entertaining-advertisement outcome measure (62 percent versus 21 percent; rPurchase Intention = .78 versus rEntertaining = .45, z = 5.803, p < .0001, two-tailed). This finding supports the intuitively appealing conclusion that entertainment or liking of a given advertisement is significantly less effective in predicting advertisement effectiveness than is a network-activating model.
Post Hoc Analysis 2: Advertisement Finish
The second post hoc analysis explored the relative differences between the independent samples of finished and animatic advertisements. (All of the correlations are shown in Table 2.) All correlations were statistically different from zero. Among the overall (A + B + C) correlations, there were no significant differences in terms of the variance accounted for (58 percent versus 68 percent; rFinished = .76 versus rAnimatic = .83, z = −1.46, p > .05, two-tailed).
There was also no significant difference between the product bridge and the personal-relevance bridge. There was, however, a significant difference among value-bridge correlations, with the animatic advertisements explaining more variance in purchase intention than the finished advertisements (52 percent versus 29 percent; rFinished = .54 versus rAnimatic = .72, z = −2.31, p < .05, two-tailed). This difference suggests that the value-bridge linkage was more important for animatic advertisements.
A possible interpretation is that animatic advertisements are more involving and require more cognitive processing than finished productions, which are more passive. Perhaps this contrast is similar to radio as more of an active medium, because it allows the listeners to construct their own mental imagery. In a finished television advertisement, in contrast, the characters and action do not leave as much to the imagination.
An important implication of this finding is that advertising can be assessed in rough stages before production expenses are incurred. An important finding in the results is that the connection bridges (A + B + C) relative to purchase intention explained more than seven times the variance explained by the entertaining-advertisement measure (58 percent versus 8 percent; rFinished—Purchase Intention = .76 versus rFinished—Entertaining = .29, z = 7.25, p < .00001, two-tailed).
For the animatic advertisements, there were no statistical differences among the three linkage-level correlations. There also was not a statistical difference between the measures of overall neural strategy (A + B + C) as contrasted with the finished advertisements. This result also supports the prior finding (Reynolds and Gengler, 1991) that researchers can assess animatic advertisements to estimate how well an advertisement will deliver against an a priori strategy. A strategic assessment also can highlight where an advertisement can be improved, with respect to increasing the magnitude of its connections before production expenses are incurred. For animatic advertisements, purchase intention again was found to explain significantly more variance than the entertaining-advertisement measure, but by a slightly smaller magnitude (68 percent versus 49 percent; rAnimatic—Purchase Intention = .83 versus rAnimatic—Entertaining = .70, z = 4.26, p < .0001, two-tailed).
DISCUSSION
The goal of this research was to explore the validity of means–end theory, as operationalized by the Strata research methodology, for assessing television advertising. The Strata framework integrates three areas of interest to advertising researchers: means–end decision theory, neuromarketing, and assessment of advertising effectiveness. The Strata methodology has participants directly assess the strength of the connections or associations triggered by a given advertisement execution. This system of connections corresponds to the creation of a neural network that is postulated to underlie decision making.
Analysis of 240 television advertisements across a range of countries and product categories yielded data measuring the strength of cognitive linkages (neuroscience) among levels of abstraction (means–end theory) caused by advertising stimuli. The aggregated data positively and significantly correlated with advertising effectiveness, operationalized as the level of purchase intention for the advertised product.
The first post hoc analysis contrasted advertisement affect, operationalized as an entertaining advertisement, with the neural-based strategy model predicting advertising effectiveness, operationalized as purchase intention. The results of this direct comparison suggest that the neural model is substantially more predictive of advertising effectiveness than is a traditional, entertainment-based copy-testing assessment approach. An additional post hoc finding is that the strategic assessment of animatic advertisements was consistent with those of finished advertisements. This suggests that early research on advertisement ideas can lead to significant increases in advertising effectiveness and production-cost savings. The overall conclusion of this analysis using the Strata research model is that means–end theory is a special case of neural networks, which directly suggests the validity of the strategy-development process utilizing this underlying theoretical perspective.
Limitations and Future Research
This study is not without its limitations. Chief among them are the challenges of creating a realistic experience in a laboratory setting. In such settings, participants frequently are set up as experts rather than consumers. Strata's computerized procedures required participants to consider two advertisements simultaneously to avoid focusing solely on one advertisement. Strata, however, does not expose participants to advertising clutter, as they would experience in a more realistic setting. Indeed, this benefit of the repeated-measures design, along with the design's other known benefits (e.g., time and cost savings), must be tempered by its disadvantages—namely, the possibility of carry-over effects.
The benefits of using a standardized purchase-intention question that is administered easily across cultures and product categories also comes at a cost: It means that researchers do not have a behavioral measure that could be tailored more to the specific product category or participant (Woodside, 2016). As digital data of actual behavior become more prevalent, researchers will need to rely less on self-reported measures of behavioral intentions, thereby avoiding many of the veracity problems associated with these measures.
Given this set of findings, an obvious next step is to determine which physiological activations correspond to the different judgment tasks, including both product (e.g., purchase intentions) and advertisement (e.g., entertaining) affect, and the relationship to node activation and connection judgments. This relationship should be examined for each level of abstraction underlying the decision process. This understanding well may provide the defining insight to advance the field of neuroadvertising research. Future research should move beyond directional testing of associations and should make use of asymmetric modeling methods to test for predictive validity using holdout samples.
The models underlying advertising effectiveness very well may vary by product category. Future research should be aimed at identifying relevant differences at the linkage level as well as understanding the relative importance of the entertainment value of the execution. It is possible that different product categories require different advertisement (weighting) models.
Both laddering research and the development of product-specific statements used in Strata traditionally have required costly one-on-one in-depth interviews. Advances in technology now permit Internet-based, one-on-one interviewing (e.g., Stream), which has reduced the time and cost limitations of laddering studies without compromising quality (Reynolds and Phillips, 2009). In addition, the Stream software includes an online coding function and a decision-segmentation methodology (Phillips et al., 2010; Reynolds, 2006). A likely next step will be the use of artificial intelligence software to conduct decision-based laddering interviews.
Managerial Implications
This article has validated the hypothesis that means–end theory, as operationalized in the questioning format of the Strata methodology, is the special case of neural networks that underlies decision making related to advertising's goal of generating purchase intent for a given product, service, or political candidate. These results hold implications for practice.
The most significant implication is the importance of focusing on the linkages, given that the key to effective advertising is to cause the associative connections among the four nodes. Asking the creative team to specify, as precisely as possible, what in a proposed execution will result in the linkages being made provides a process to focus meaningful discussion and ongoing learning. The creative challenge (as depicted in Figure 4) asks how to make the neural connections.
The creative team must be specific as to which executional elements will make the desired connections between adjacent key levels of nodes that define the decision-affecting network. The question-labeled boxes should focus the team on how to develop effective advertising by communicating a complete network of neural associations. This translation of the key nodes that define the consumer-decision network into a MECCAS model format subsequently will allow the team to assess, through Strata, how well a given execution delivers again the designated key elements and the strength of connections that are made.
ABOUT THE AUTHORS
Thomas J. Reynolds is professor emeritus at the University of Texas – Dallas and managing director of Strategic Research, Development and Assessment. He has provided marketing-strategy consulting to numerous Fortune 500 companies in more than 25 countries and to political campaigns, including Ronald Reagan's 1984 U.S. presidential reelection campaign. His research has been published in Journal of Consumer Research, Journal of Marketing Research, and Review of Marketing Research, among many others.
Joan M. Phillips is dean and professor of marketing, Andreas School of Business, Barry University, Miami, Florida, USA. Her research interests include improving questionnaire-based methodologies and understanding how personal values and the valence of brand attitudes impact consumer decision-making and choice. Her research has been published in Journal of Consumer Research, MIT Sloan Management Review, Review of Marketing Research, Journal of Public Policy & Marketing, Marketing Letters, Journal of Marketing Theory and Practice, and Qualitative Market Research: An International Journal.
- Received October 25, 2016.
- Received (in revised form) October 16, 2017.
- Accepted December 4, 2017.
- Copyright© 2019 ARF. All rights reserved.