Temporal Variations in the Evaluation of Television Advertisements: the Role of Key Nonverbal Cues

Siva K. Balasubramanian, University of Iowa
ABSTRACT - This study seeks to explain the variations over time in viewer evaluations (degree of liking) of television ads. A multiple regression analysis suggests that key nonverbal cues/dimensions explain a substantial proportion of these variations.
[ to cite ]:
Siva K. Balasubramanian (1990) ,"Temporal Variations in the Evaluation of Television Advertisements: the Role of Key Nonverbal Cues", in NA - Advances in Consumer Research Volume 17, eds. Marvin E. Goldberg, Gerald Gorn, and Richard W. Pollay, Provo, UT : Association for Consumer Research, Pages: 651-657.

Advances in Consumer Research Volume 17, 1990      Pages 651-657


Siva K. Balasubramanian, University of Iowa


This study seeks to explain the variations over time in viewer evaluations (degree of liking) of television ads. A multiple regression analysis suggests that key nonverbal cues/dimensions explain a substantial proportion of these variations.

The large number of studies that directly or indirectly address individuals' evaluation of television ads are dominated by post-exposure measures involving multi-item scales (e.g., attitude-toward-the-ad), which carry some hidden drawbacks. First, post:exposure measures represent global evaluations of the entire ad, and may not provide insights on how individuals react to each of the several segments that span the time-duration of the TV ad. Such insights are potentially invaluable since they could establish a scientific basis to produce TV ads that generate high evaluative impact. Each of the several segments (in a TV ad) includes numerous audio and visual variables, and the analysis of the continuous, instant-by-instant, real-time evaluations of such segments could yield important new knowledge on which variables contribute most to positive viewer impact.

Second, post-exposure evaluations of the ad could be biased by primacy-recency effects: the attitude-toward-the-ad measure obtained may be more reflective of the individual's evaluation of the ad for segments he/she is more likely to remember, rather than a "true" global evaluation of the entire ad.

The above considerations underscore the need for obtaining "online" temporal evaluative measures of the ad; in simple terms, this involves measuring the individual's reactions to the ad over time, simultaneously while he/she is being exposed to the ad. However, only a few studies have focused on over-time measures of ad-related response: Alwitt (1985) and Rothschild et al. (1988) employed overtime electroencephalographic (EEG) measures of brain activity in response to ad exposure to assess whether they reflect variations in the content of the ad; in addition, Thorson & Reeves (1986) have investigated the effects of temporal variations in viewer liking for programs/ads on memory for ads.

Although this research follows the spirit of Alwitt (1985) in that it examines the importance of variations in ad content over time, there are key differences: (a) the dependent variable used in our analysis is different: the "degree of viewer liking" toward the ad over time, and (b) we focus attention on key nonverbal cue categories that account for the variations in the dependent variable. In sum, the study attempts to provide preliminary insights on the relative importance of several nonverbal variables in influencing advertising evaluations. The choice of the explanatory variables is primarily motivated by the heightened research attention paid in recent years to the role of nonverbal variables in advertising (Hecker & Stewart 1988).

Selection of Nonverbal cue categories for analysis:

Three categories of nonverbal audio and nonverbal video cues (that characterize most TV ads) were chosen for this study: (a) music-related cues, (b) voice-related cues, and (c) camera- related cues. Prior advertising research supports the choice of these cue categories for analyses. For instance, the 18 nonverbal cue classes developed by Haley et al (1984) span all three cue categories above. Similarly, in assessing the role of ad-content on EEG activity in subjects, Alwitt (1985) has focused attention on music-variables (also see Stout & Leckenby 1988 for a discussion on the importance of music as a nonverbal element), while camera cues have been studied by both Alwitt (1985) and Rothschild et al (1988).

Selection of specific nonverbal dimensions within each cue category

Several nonverbal dimensions were selected as appropriate for analysis within each of the three nonverbal cue categories, and these are discussed next.

Music category: Past researchers have focused attention on the role of music in TV ads. Two studies (Hoyer, Srivastava & Jacoby 1984; Alwitt 1985), for instance, have attempted to analyze the impact of the presence or absence of music (music as a binary independent variable) in TV commercials. However, given the richness of nonverbal dimensions associated with music, this binary approach is unlikely to produce significant insights. Indeed, Hoyer, Srivastava & Jacoby (1984) note that dummy-coding of the music variable understates the variance in the variable, thus yielding low correlations with the dependent measure; they also suggest that "more refined measurements (e.g., the amount of music) could have resulted in higher R-squares" (p. 21) in their analysis.

While Haley et al (1984) avoided the problems inherent in the dummy-variable approach to coding music, these authors acknowledge that their coding scheme for the music related nonverbal variables may have been naive, because its focus was limited to a few broad aspects: whether the music was vocal or instrumental, the amount of music included, whether the music was in the foreground or background, and the type of music employed (classical, rock, country, jazz, etc). They suggest that future research needs to employ more sophisticated codes to capture the subtle nuances of the music such as rhythms, melodies, textures, and harmonies etc.

However, our review of the relevant literature suggested that more parsimonious approaches to coding music were possible. Asmus (1983), for instance, argues that affective responses to music can be measured by activity and evaluation dimensions (see also McMullen 1982 and Trolio 1976). Apart from these two dimensions, prior research also shows a relationship between the complexity of the music stimulus and subjects' affective responses (Conley 1981; Crozier 1974; McMullen 1974). In addition, Leblanc (1980, 1982) argues that another dimension (interest toward the musical stimulus) should also be related to affective response development.

Based on these research findings, it was decided to evaluate the musical component in TV ads using the following four scales: (a) Ugly/beautiful (evaluation), (b) passive/active (activity), (c) simple/complex (complexity dimension) and (d) uninteresting/interesting (interest dimension).

Voice-related category: Based on a survey of the paralanguage literature, the following two dimensions were used to score voice-related cues: (a) unpleasant tonal quality/pleasant tonal quality, and (b) slow speech ra e/high speech rate.

Support for selecting the first scaling dimension above is derived from past research which suggests that subjects' inferences of a communication's affect depend substantially on the voice's tonal quality (Scherer 1972; also see DePaulo et al. 1980 for a review). Specifically, this line of research argues that atonal and tonal-minor stimuli evoke inferences of disgust (unpleasantness) while the tonal-major acoustic characteristic leads to judgments of pleasantness (or happiness).

Further, studies by Mehrabian and associates (see Mehrabian 1972 for a comprehensive review) and Miller et al (1976) strongly suggest that speech rate is correlated with positive inferences of message affect, thus justifying the choice-of the second scale dimension above.

Camera-related category: Alwitt (1985) and Rothschild et al (1988) have focused attention on camera-related visual-nonverbal variables e.g., zooms and cuts. Further, these variables have been studied by several communication researchers interested in the complexity of cues associated with television; it was therefore decided to include zooms and cuts as nonverbal variables of interest for our analysis.

In addition to the several nonverbal variables/dimensions discussed above, we also decided to include another key dimension highlighted in the literature on nonverbal communication: the degree of consistency between the nonverbal and verbal variables. This dimension has been shown to be crucial while inferring the affect conveyed by a message (see Mehrabian & Weiner 1967; Mehrabian & Ferris 1967).



104 student subjects participated in this study; each experimental session accommodated only one subject and lasted approximately IS minutes.

Stimulus Ads

To ensure that the data generated were free of biases due to subjects' prior brandname/company/advertisement familiarity, we considered several english language TV ads from abroad, in addition to domestic TV ads (aired outside the immediate viewing region) representing products and companies unlikely to be familiar to our subjects. A total of 13 ads were chosen from this sample for the stimulus package, based on the following criteria: (a) each ad was different from other ads in the package (so that a broad range of ad themes could be represented in the study), and (b) the ads represented product categories of interest to student subjects (to enhance the validity of the study).


A measurement device called the program analyzer was used to generate "online" evaluations of the TV ads in the stimulus package. These data represented the dependent variable, operationalized as the degree-of-liking for the ad segment the subject was being exposed to at a given instant. The analyzer used in this study consisted of a joystick linked to an IBM personal computer. The joystick served as an impromptu/surrogate Likert scale since it could be moved back and forth by the subject along a given plane, and was similar to the dial-turning equipment used by Thorson & Reeves (1986).

Before commencing the actual data collection, the subject was familiarized with the analyzer equipment with the help of a computer program that generated a visual display of a linear scale on the computer screen; this linear scale contained two anchors ("dislike" and "like" respectively). A feature of this program was that the magnitude and direction of any movement of the joystick would be correspondingly reflected in the movement of the cursor along the linear scale displayed on the computer screen. The subject was asked to practice using the joystick and its linear scale analog for 5 minutes, in order to gain "hands-on" familiarity with the sensitivity of the device. Further, since the valence of the Likert scale anchors were reinforced through pictures of a sad or a smiling face placed at either ends of the joystick plane, the subject was also familiarized with the location of the "dislike" and "like" scale anchors.

Following this familiarization procedure, the subject was exposed to 5 sample ads in an attempt to simulate the conditions characteristic of the actual study; the actual stimulus package was then shown to the subject.

The commencement of data collection via the analyzer was synchronized with the start of exposure to the stimulus package. For this study, the movement of the joystick along the vertical plane generated scale values in the range of 1 to 45 for "dislike", and between 46 and 90 for the "like" evaluations. The computer sampled the joystick position 5 times every second during the data-gathering session. Any movement of the joystick (i.e., change in the scale values along the "dislike"/"like" scale) in these small time intervals was recorded by the computer along with a record of cumulative time elapsed since the beginning of the data-gathering session.

Thus, if one could separately code the exact (cumulative) time of occurrence of any nonverbal cue of interest in the stimulus package, the data-accumulation procedure described above provides a highly accurate basis for studying temporal variations in ad evaluations directly corresponding to these occurrences of the nonverbal cue. Details of such coding exercises by expert judges are described in the next section.

After the data-gathering session, the subject responded to two self-report items. The first item measured the degree to which the subject thought the joystick movements truly reflected his/her actual evaluations of the ad. This was measured on a 5 point scale (with 1=very unlikely and 5=very likely), and the mean and standard deviation across subjects on this measure (4.223 and .625 respectively) indicated a high level of confidence among subjects that the analyzer data accurately reflected their evaluations.

The second item measured the extent to which the subject was involved with the experimental procedure pertaining to the analyzer. This was also a 5 point scale (with 1 very uninvolved and 5=very involved), and the mean and standard deviation across subjects on this measure (4.485 and .778 respectively) reflected high respondent involvement with the experimental procedure.

The subjects' responses in debriefing sessions revealed that none of them had seen any of the foreign TV ads; however, most of them had seen 4 out of the 6 domestic ads included in the stimulus package. Despite the precaution of only including domestic ads (in the stimulus package) that were not previously aired in the immediate viewing region, several subjects indicated that they were exposed to these 4 ads when they had visited contiguous viewing regions. Since subjects' evaluations of these 4 ads could be contaminated by prior brand/company/ad familiarity biases, they were excluded from subsequent analyses.

Coding Of commercials

Of the nine nonverbal dimensions discussed above (4 music-related, 2 voice-related, 2 camera-related and the nonverbal/verbal consistency dimension), we note that the camera-related dimensions (viz., zooms and cuts) are binary (present or absent) variables; the remaining variables had to be evaluated on a continuum (the research interest here was not merely on the impact of the presence of these variables, but on the magnitude of their presence). To use these dimensions as independent variables in a regression analysis, the stimulus package was coded for these variables by a panel of several expert judges (at least 2 judges in this panel had specialist expertise vis-a-vis each of the three cue categories). All the judges were blind to the purpose of this study during the coding process, which is described next.

Coding Binary Independent Variables

The judges used a "binary-coding" version of the program analyzer to code the two binary variables (viz., zooms and cuts) for all the ads in the stimulus package. This version generated a binary coded time-series based on a button-pressing procedure; judges were instructed to depress the button for each instant of time when the cue under investigation was present in the stimulus package being viewed by them. Whenever the button was depressed, the computer recorded a "1" five times every second; otherwise, the computer recorded a "0" (a similar procedure was used by Alwitt 1985).

Coding Continuous Independent Variables

With respect to the six remaining continuous (i.e.,interval-scaled) independent variables derived from the nonverbal cue categories, separate groups of expert judges coded-each nonverbal dimension of interest one at a time using the same procedure as the subjects in the study; the only difference in this case was that the judges focused on evaluating the nonverbal dimension of interest. In contrast, the subjects in the study were required to evaluate the ad stimulus on the "dislike"/"like" Likert scale.

Note that these six variables pertained either to the voice-related or the music-related category; further, neither category was present continuously for the entire duration of the stimulus package Since the continuous nature of the coded data generated by the analyzer did not recognize the time segments where voice or music were not present in the ad, it was necessary to avoid the inclusion of such spurious data. This could be accomplished by multiplying for each time period, the continuous time-series data generated by judges (corresponding to each continuous independent variable) with appropriate dummy time-series data on the presence (coded "1")/absence (coded "0") of voice or music in the stimulus package. Therefore, voice and music dummies were also obtained through the dummy coding procedure discussed.

Similarly, the degree of consistency between verbal and nonverbal variables in the ad was evaluated by two members of the judges panel.


We used multiple regression analysis to assess the impact of the nonverbal independent variables (continuous as well as binary) on the dependent variable (interval-scaled "dislike/like" measure). We checked for two potential problems at this point: auto-correlation among the residuals, and the data-pooling issue.

Auto-correlation among the residuals It must be noted that auto-correlation among the residuals might confound regression analysis of time-series data. Stated differently, if the evaluation of one segment of the TV ad affected how the segments following it were evaluated, the unique impact of a given set of nonverbal variables at any given time cannot be assessed accurately. The program analyzer generated data at extremely short intervals for all the variables of interest; and to minimize the autocorrelation problem, we only used data drawn at 2-second intervals from this database. The choice of this data interval is consistent with prior research analyses of "online data" (see Alwitt 1985; Thorson & Reeves 1986), and yielded 15 observations for each of the 9 ads studied, for each of the 104 subjects.



To assess the presence of autocorrelation with this time-series data base, regression analyses were conducted on data from 10 randomly selected subjects. Durbin-Watson tests indicated that autoregression was not a problem, thus providing support for using program analyzer time-series data drawn at two-second intervals.

Pooling Data for Multiple Regression Analysis

The dependent measure ("dislike"/"like" scale) has both cross-sectional (across respondents) and time-series (across time within each ad) characteristics; to perform multiple regression analysis using this variable, a decision on pooling or aggregating the database was necessary.

The pooling approach used here involved: (1) standardizing the dependent measures within each subject to minimize individual differences in such responses, and (2) averaging this standardized data across subjects for each time period. Such data pooling will maximize relationships between the nonverbal dimensions of interest and the dependent measure, since it avoids the high variance that may characterize individual subject data.

Reliability for Coding Binary Independent Variables

Phi-coefficients assess the relationship between two discontinuous dichotomous variables (Ghiselli, Campbell & Zedeck 1981), and are appropriate for estimating the inter-judge reliability of coding a binary variable. The phi-coefficients for the four binary-coded variables (at two-second intervals) given below are acceptably high:

Variable coded      Phi-coefficient

Zooms                    .89

Cuts                       .98

Music dummies     .90

(yes/no binary coding)

Voice dummies     .95

(yes/no binary coding)

Reliability for Coding Continuous Independent Variables

As already noted above, the six nonverbal dimensions (4 were music-related, 2 were voice-related) and the nonverbal/verbal consistency dimension were coded on a continuous scale. Test-retest reliabilities computed for these time-series data (coded by judges) are given in Table 1.

Results and Discussion

Table 2 summarizes the results of the multiple regression analysis with the averaged "dislike"/"like" data as the dependent measure; the independent variables were the various nonverbal dimensions described above along with the verbal-nonverbal consistency measure.

On the overall level, the predictors were able to account for a significant part of the variance in the dependent measure (R2 = .547). Table 2 presents the regression results, which are discussed next.

Voice-related category Although past research indicates that variables such as speech rate and tonal quality should be correlated significantly with the affective responses to communication, we note that the beta weights for these variables are not significant in the regression equation. A possible explanation for this may lie in the lack of variation in the data. The expert judges felt that there was little variability in these two variables over the ads in the stimulus package.



Music-related Category Under the music-related dimensions, three of the four dimensions have significant regression estimates. The beta coefficients for the simple/complex and passive/active dimensions were negative, thus suggesting that as the music in the TV ad increases in complexity and/or activity dimensions, the affective evaluations tend to become more negative. This is consistent with the view of Krugman (1965) who suggests that TV is essentially a passive medium that disseminates messages in a "low-commitment" environment; therefore, the more simple or the more passive the music-related cues, the higher the affective evaluations.

On the other hand, the evaluation dimension characterized by the ugly/beautiful scale suggests that more positive evaluations of music in TV ads have a direct impact on the affective responses generated (note that the beta weight for this dimension is both positive and higher than the other independent measures). This finding is important, and-is consistent with prior research. For instance, there is some evidence that the music content in popular jingles and ads elicit positive evaluations; in other words, the more positive the evaluation of music in a commercial, the higher the likelihood of the ad also eliciting a positive evaluation. Gorn's (1982) research supports a similar inference with respect to the product advertised; in this study, a pen displayed with "liked" music elicited higher brand preference ratings relative to another pen that was paired with "disliked" music.

However, the uninteresting/interesting dimension was not significant, suggesting that affective responses to TV ads are not related to how the music-related cues perform on this dimension.

Camera-related Category Table 2 also indicates that "zooms" (a camera- related visual nonverbal cue) have a significant beta coefficient (.197); further, the positive sign of this coefficient suggests that zooms in a TV ad directly generate positive affective responses to the ad. On the other hand, the beta coefficient associated with "cuts" was not significant. A possible explanation for this could be that "zooms" typically promote greater focus on the visual stimulus, thereby increasing its informational content; however, "cuts" are synonymous with "breaks" in the visual stimulus which may disrupt continuity in the message and thereby distract the viewer. Zooms may be expected to have a positive impact on the affective responses to the ad since they enhance focus on the visual stimuli, while the distracting potential of "cuts" may not generate such impact at all.

Nonverbal/Verbal consistency dimension: While prior literature would predict a significant and positive beta coefficient for this dimension (more consistency should lead to more positive affective responses), the analysis yielded a non-significant coefficient for this variable. This result stemmed from the characteristics of the stimulus package, in that variability across the verbal/nonverbal consistency dimension was minimal (the coefficient of variation for this variable was 0.06), and the impact of this dimension was not unequivocally assessed


Although the independent variables accounted for a good proportion of the variance in the affective responses generated by ads, it must be realized that the pooling technique employed could have inflated the R2 estimate. The R2 obtained for such pooled data (involving regression on means) are typically much greater than the corresponding R2 values based on raw data. Future research could overcome this problem by estimating models (on raw cross-sectional time-series data) that satisfy rigorous statistical pooling tests (e.g., Bass & Wittink 1975).

Despite the above limitations, the study still provides useful, albeit preliminary, insights. The findings pertaining to the music-related-nonverbal dimensions are interesting. They indicate that higher the level of perceived beauty, simplicity or passivity (in that order) associated with the music component in TV ads, the higher the likelihood of generating positive evaluations of the ad. Similarly, the analyses suggest that zooms in TV ads contribute to more positive evaluations, while cuts do not register any impact on the dependent measure. These findings were interpreted as conforming to prior findings in the advertising literature.


Alwitt, L. F. (1985), "EEG Activity Reflects the Content of Commercials", in L. F. Alwitt and A. A. Mitchell (eds)Psychological Processes and Advertising Effects: Theory, Research and Applications, Hillsdale, NJ: Lawrence Erlbaum & Associates .

Anderson, D. R. (1985), "Online Cognitive Processing of Television", in L. F. Alwitt and A. A. Mitchell (eds) Psychological Processes and Advertising Effects: Theory, Research and Applications, Hillsdale, NJ: Lawrence Erlbaum Associates.

Asmus, E. P. (1983), "Experimental Analysis of Four Models of Musical Preference Development", Paper presented at the International Conference of Psychology and the Arts, University of Wales Institute of Science and Technology, Cardiff, United Kingdom, September 1983.

Bass, Frank M. & Dick R. Wittink (1975), "Pooling Issues and Methods in Regression Analysis with Examples in Marketing Research," Journal of Marketing Research, 12 (November), 414-25.

Conley, J. K. "A psychophysical investigation of judgments of complexity in music, Psychomusicology, 1, 59-71.

Crozier, J. B. (1974), "Verbal and Exploratory Responses to Sound Sequences varying in Uncertainty Level," D. E. Berlyne (ed.) Studies in the New Experimental Aesthetics. New York: John Wiley & Sons.

DePaulo, B. M., M. Zuckerman, and R. Rosenthal (1980), "Detecting Deception: Modality Effects," L. Wheeler (ed.) Review of Personality and Social Psychology, Vol 1, Beverly Hills, CA: Sage Publications.

Ghiselli, E. E., J. P. Campbell & S. Zedeck (1981): Measurement Theory for the Behavioral Sciences, San Francisco: W. H. Freeman & Co.

Gorn, G. J (1982), "The Effects of Music in Advertising on Choice Behavior: A Classical Conditioning Approach," Journal of Marketing, 46 (Winter), 94-101, 1982.

Haley, R. I., J. Richardson & B. M. Baldwin (1984), "The Effects of Nonverbal Communication in Television Advertising", Journal of Advertising Research, 11- 18.

Hecker, Sidney & David W. Stewart (1988), Nonverbal Communication in Advertising, MA: Lexington Books.

Hoyer, W. D., R. K. Srivastava, & J. Jacoby (1984), "Sources of Miscomprehension in Television Advertising", Journal of Advertising, 13, 2, 1726.

Krugman, H. E. (1965),"The Impact of Television Advertising: Learning without Involvement," Public Opinion Quarterly, 29 (Fall), 349-356.

LeBlanc, A (1980), "Outline of a proposed model of sources of variation in musical taste", Bulletin of the Council for Research in Music Education, 1980, 61, 29-34.

LeBlanc, A. (1982), "An interactive theory of music preference", Journal of Music Therapy, 19, 2845.

McMullen, P. T. (1974), "Influence of Number of different Pitches and Melodic Redundancy on Preference Responses," Journal of Research in Music Education, 22, 198-204.

McMullen, P. T. (1982), "Connotative responses to Musical Stimuli: A theoretical explanation," Bulletin of the Council for Research in Music Education, 71, 45- 57

Mehrabian, A. & S. R. Ferris, (1967), "Inference of attitudes from nonverbal communication in two channels", Journal of Consulting Psychology, 31,248-52.

Mehrabian, A. & M. Weiner, (1967), "Decoding of Inconsistent Communications," Journal of Personality and Social Psychology, 6, 109-14.

Mehrabian, A. (1972), Nonverbal Communication, Chicago: Aldine.

Miller, N., G. Maruyuma, R. J. Beaber & K. Valone (1976), "Speed of Speech and Persuasion", Journal of Personality and Social Psychology, 34(4), 615-624.

Neibecker, B (1985), "The Validity of the Program Analyzer to Measure Emotion in Popular Music", Unpublished Manuscript, Institute for Consumer and Behavioral Research, University of Saarland, Fed. Republic of Germany.

Rothschild, M. L., Y. J. Hyun., B. Reeves., E. Thorson., & R. Goldstein (1988), "Hemispherically Lateralized EEG as a Response to Television Commercials," Journal of Consumer Research, 15, (September), 185-198.

Scherer, K. R. (1972), "Acoustic Concomitants of Emotional Dimensions: Judging Affect from Synthesized Tone Sequences", Paper presented at Eastern Psychological Association meeting, Boston, April 1972.

Stout, P. A. & J. D. Leckenby (1988), "Let the Music Play: Music as a Nonverbal Element in Television Commercials," S. Hecker and D. W. Stewart (eds.) Nonverbal Communication in Advertising, MA: Lexington Books.

Thorson, Esther., Byron Reeves (1986), "Effects of Over-Time Measures of Viewer Liking and Activity During Programs and Commercials on Memory for Commercials," Richard Lutz (ed.) Advances in Consumer Research, Volume 13, 549-553.

Trolio, M. F. (1976), "Theories of Affective Response to Music," Contributions to Music Education, 48, 1-20.