Effects of Over-Time Measures of Viewer Liking and Activity During Programs and Commercials on Memory For Commercials

ABSTRACT - Memory for commercial information was examined as a function of liking and change in liking measured over time during television viewing. Four categories of measures were taken during commercials and surrounding programs: mean liking; variance in liking, trend in liking across a segment; and change in liking between programs and commercials. Correlations were examined for simultaneous relations (memory and liking measures at the same point in time), and for proactive and retroactive relations. Significant variance in recall and recognition was explained by the over-time measures, independent of program or commercial content.


Esther Thorson and Byron Reeves (1986) ,"Effects of Over-Time Measures of Viewer Liking and Activity During Programs and Commercials on Memory For Commercials", in NA - Advances in Consumer Research Volume 13, eds. Richard J. Lutz, Provo, UT : Association for Consumer Research, Pages: 549-553.

Advances in Consumer Research Volume 13, 1986      Pages 549-553


Esther Thorson, University of Wisconsin-Madison

Byron Reeves, University of Wisconsin-Madison


Memory for commercial information was examined as a function of liking and change in liking measured over time during television viewing. Four categories of measures were taken during commercials and surrounding programs: mean liking; variance in liking, trend in liking across a segment; and change in liking between programs and commercials. Correlations were examined for simultaneous relations (memory and liking measures at the same point in time), and for proactive and retroactive relations. Significant variance in recall and recognition was explained by the over-time measures, independent of program or commercial content.


There appears to be an inverse relationship between the cognitive involvement potential of material presented immediately prior to and subsequent to the critical message and retention of message content. (Bryant and Comisky 1978, p. 73)

Based on an initial sample of 18,000 viewers.... Burke found a direct correlation between the viewer's expressed opinion of the program and this viewer's ability to recall part of the test commercials. (Priemer 1983, p. 3)

Research about program-context effects on commercials is summarized by the two results described above. When program involvement was operationalized by viewer self-report, placement in highly involving programs inhibited memory for commercials (Kennedy 1971; Bryant and Comisky 1978; Soldow and Principe 1981). When liking was similarly measured, commercials in well liked programs showed stronger memory (Priemer 1983; Krugman 1983).

The present study was designed to reconcile these findings using an "on line" measure of program involvement and liking that could be matched with memory for embedded commercials. In addition to an overall assessment of the level of liking in surrounding programs, effects on memory were hypothesized to depend on three measures of activity during viewing: (1) the variability of surrounding judgments; (2) trends over time; and (3) changes in level, variance and trend that occurred between programs and commercials. Finally, effects were not listed to proximal influences of program judgments on memory. Three different processes were hypothesized: (1) simultaneous effects (influences of message involvement on memory for commercial information at the same point in time); (2) proactive effects (influences of prior programs on subsequent memory); and (3) retroactive effects (influences of later programming on memory for earlier commercial information).

Time-dependent analyses of television viewing are not new in marketing and communication research, although few citations can be found. The most typical application in mass communication has been the study of children's responses to television and movies. In an early study by Dysinger and Ruckmick (1933/1970), children's emotional responses were indeed over time by galvanic skin response during movie watching. More recently, Krull and Husson (1979) and Anderson and Lorch (1983) have correlated visual orientation (l.e., time when eye gaze was directed at the screen vs. competing stimuli) with memory and comprehension of television programs. Although most research on program-contest effects in advertising has not used time-dependent analyses, the Bryant and Comisky study cited above varied involvingness of segments in a program, and noted their differential effects.

The scarcity of these studies stands in stark contrast to theoretical discussions about the importance of time-dependent processes in communication (Kline 1977; Krull and Paulson 1977). The extraction of meaning from television messages is best conceptualized as a series of interdependent evaluations. The successful uses of psychological concepts such as priming, serial position effects, and the processing of serial patterns (Jones 1974), attest to the importance of contextual information in encoding. Consequently, a primary assumption in this research is that mental activity occurring before, during and after the processing of a target message can dramatically influence the information retained, the length it is remembered, and the associations with past experience.

Lack of research about over-time processes probably results less from theoretical neglect than from a shortage of appropriate methods for assessing on-going reactions. The assumption that program and commercial messages are interdependent and that evaluations are mutually influential over time, suggests that exclusive use of post-viewing measures is inappropriate. Nevertheless, recording reliable evaluations of a quickly changing stimulus is difficult. In this study, subjects were asked to turn a dial as they watched television. They turned the dial up from 50 to 100 to indicate degree of liking and down from 50 to 0 to indicate dislike. A potentiometer setting was registered for each subject for every two seconds of viewing, and the over-time patterns of these mean settings comprised the measure of liking.

Two kinds of measures were examined in the dial patterns. The most obvious measure was the mean level of the settings. This level provided an indication of how much subjects liked what they were viewing during significant time segments. Mean level of liking was defined for commercials, commercial pots, and four different parts of programs (mean level for the entire program segment, and mean level for the last two minutes, one minute, and half minute before the commercials).

A second measure was how much dial-turning subjects showed during significant time segments. Except for the possibility that extremely riveting material might "freeze" or slow dial-turning, more turning seemed likely to index mental involvement in viewed materials. Change in liking was indexed by eight measures. The first was the standard deviation of mean liking, a measure of the over-time variation around the mean for each program and commercial. Subjects who exhibited large changes in how much they liked what they were viewing would have large standard deviations.

The second change measure was the difference between means for contiguous units. The time periods used were whole programs (PR), last two minutes (TM), last minute (1M), last half minute (HM), first (C1), second (C2), and third commercial (C3), and the commercial pots that combined three commercials (P). Rather than including all possible pairs use combinations, only the following pairs were examined: PRCX, PRPX, TYCX, TMPX, lMCX, lMCX, UMCx, and HHP=. All eight pairs were examined for each of the three program-commercial pairs in the experiment.

The third change measure was the standard deviation of each of the change scores. These were again defined for each of the eight time segment pairs mentioned above and for each of the three program-pod pairs.



The fourth change measure was the absolute value of the change scores. The previous change scores included both positive and negative values; however it may be important to consider the magnitude of the changes, regardless of valence. The absolute value measures provided this index.

The fifth change measure was the mean of the differenced series for liking, or, more intuitively, their slopes. The differenced series were calculated by subtracting contiguous time points. Positive values indicate increasing scores. Negative values indicate decreasing scores. The sixth measure was the standard deviation of the slopes or the amount of change in the slopes over different time periods. The seventh measure was the difference in slopes from one time segment to the next. These were defined for the same eight program-commercial or program-pod combinations as the change scores, and for each of the three program-pod pairs. The final measure was the standard deviation of the difference in slopes, which indicated how much the trend changed during each segment. High values for this measure indicate relatively more change in the upward or downward trend of evaluations during and between segments.

Over-Time Measures and Memory for Commercials

Over-time processes have not been similarly treated in past studies and consequently, expectations about effects of the over-time measures on memory were exploratory. All hypothesized effects were conceptualized in terms of a time relation with target commercials. Liking and change in liking could be correlated with memory for subsequent commercials (proactive influence), with preceding commercials (retroactive influence), or they could be correlated with memory for commercials at the same point in time (simultaneous influence). Figure 5 summarizes the design. There are three program segments and three commercial pods. The first, second, and third pods vary in numbers of proactive, retroactive, and simultaneous influences that can potentially occur. Hypothesized effects were also conceptualized in terms of valence (positive or negative correlations between the time-dependent variables and memory).

Predictions about direction and valence of memory effects were based on assumptions about how liking and change in liking operate at two different stages of information processing: attention and memory. Program liking should enhance attention to the television screen and hence increase attention to subsequent commercials. This would in turn lead to enhanced memory for the commercials. Program liking was hypothesized to operate on input processes, which should produce only proactive effects.


H1: Mean liking should have positive proactive effects on memory for commercials.

The effect of level of liking during commercials (simultaneous effect) was harder to predict. To the extent that liking is determined by commercial execution, greater liking should be associated with stronger memory (Thorson and Friestad, in press). however, to the extent that liking commercials is determined by program context, level of liking during the commercials may be unrelated to memory. Ambiguity here precluded a specific hypothesis about simultaneous effects of level of liking.

Hypotheses about the effects of change in liking were also difficult because more dial turning (indicated by the variance and change measures) was assumed to affect both attention and memory. Viewers were presumably more attentive to television information when they showed greater dial activity (there were no instances where the programming became so compelling that the dials were "frozen"). At the same time, however, more dial turning may have indicated greater arousal, which could increase memory capacity. Dial changing during commercials, then, should be positively associated with memory:

H2: Indicators of change in liking simultaneous with commercials will be correlated positively with memory for the commercials.

Proactive and retroactive relations between change in liking and memory operate differently. Proactive effects are thought to result from interference between information in long-term memory and new material entering from a short-term buffer (Klatzky, 1980). Retroactive effects are thought to result from interference of new items with previous items in the limited slots of the short-term buffer or in the interference caused by the transfer of old items to long-term memory (Saltz, 1971). If dial-turning enhances encoding of commercial information occurring simultaneously, then when turning occurs before a commercial, it should lead to increased competition with the incoming commercial. Competing information could be either from previous material currently in short-term buffer or material in the process of being consolidated into long-term store. When dial-turning occurs after a commercial, it should also lead to more competition with prior information. Prior material is either occupying the short-term buffer or is itself being consolidated into long-term store. New information could preclude either of these processes and lessen the probability of creating a successful long-term trace of a previous commercial.


H3: Proactive and retroactive indicators of change in liking will be correlated negatively with memory for commercials.

The predictions are constrained to the presentation of commercial pods in various program segments. Here, three methods of three commercials each were embedded in comedic, violent, and sexual program segments presented to viewers in counterbalanced orders (Thorson, Reeves, Schleuder, Lang, and Rothschild 1985). This allowed determination of degree, valence, and direction of relations between the time-dependent variables and memory.



Forty-six middle-class adults (number of males - 19), from Madison, WI, were tested. The subjects ranged in age from 20 to 50 and were members or friends of a community group. Payment for their participation was donated to the group.

Commercials and Program Sequences

The stimulus sequences consisted of 27.5 minutes of material: two minutes of TV black, 21 minutes of program segments (approximately 7 minutes each of comedy (Happy Days), violence (A-Team), and sexual content (segments from the movies Peter Proud and Shampoo), and 4.5 minutes of 30-second commercials. The order of presentation was: TV black (one minute), Program 1, Pot 1 (3 commercials), Program 2, Pod 2, Program 3, Pod 3, and TV black (one minute). Three orders of the programs and commercials were used to counterbalance for order effects.


Groups of one to 10 subjects were instructed to use the potentiometer and were given the opportunity to practice. Before viewing the stimulus videotape, subjects watched two five-minute segments of feature stories and sports to acclimate to the viewing room. Each of the practice segments was followed by three thirty-second commercials.

After viewing the practice tape and the stimulus tape, subjects were asked to complete a questionnaire about the three program segments. A second questionnaire asked for free recall of the nine commercials in the stimulus tape. Subjects were asked to list for each commercial the appropriate product category, brand name, claims made, and executional information about the scenes, actors, and stories. Subjects were asked to include any information recalled even if not all information was available. To encourage thorough recall, a bonus of 104 was added to the participation fee for each commercial recalled.

A third questionnaire asked for recognition of the nine commercials. There were three pages for each commercial, each page presenting five selections, one of which was the target. The first page listed product categories, the second. brand names, and the third, claims.

In a final activity, subjects responded to an adjective checklist test for each advertised product (the results of which are not relevant to the present analysis). After completion of all the questionnaires, subjects were thanked and excused.


The recall and recognition questionnaires were scored for accuracy. This resulted in eight memory measures for each commercial pot: total recall plus recognition score, recall of product category, brand name, claim, and execution, and recognition of product category, brand name, and claim.

The potentiometer data were standardized for each subject. Time series for each of the time-dependent variables were then constructed. For mean liking, standard deviation of mean liking, trend, and standard deviation of trend, there was a score generated for each commercial, each pod, and four program means for each of the three programs: whole program, last two minutes, last minute, and last half minute. For changes in mean liking, absolute value of the changes, and standard deviations of the two variables, there were eight scores generated for each pod by combining the four program intervals with pods or the first commercial in each pod.


Potentiometer Patterns Over Time

Figures 2 and 3 show the means of the potentiometer values across 825 2-second time intervals in the first order. These data are presented to illustrate visually the nature of the over-time changes in liking. In each series there is a clear distinction between the commercials and the program segments. These aggregate changes indicate that program-commercial break points represent significant shifts in the series and that there is substantial individual variation in the type of change present. For the males, there was always a large drop at the beginning of each commercial pot, with a slow rise across the ensuing 90 seconds. The pattern was more complicated for the females. The first pot produced a marked trop, the second pot, a marked increase, and there was a small drop in the third pod. There were also differences in the mean level of the standardized readings, with females lower for sex and violence and higher for comedy. The figures also show more up and down movement for females than males.

Recall and Recognition as a Function of Pod Position

Mean memory levels also varied across commercial pods. There was a strong serial position effect, with memory significantly higher (p<.05) for Pods 1 (X = 8.0) and 3 (X = 9.1) than for Pod 2 (X = 6.4).

Mean Level of Liking and Memory

Hypothesis 1 suggested liking would have positive proactive effects. Because change scores from the whole program, the last two minutes, one minute, and half-minute were highly correlated, only one of the four was counted if there was a significant correlation with memory. As can be seen in Table 1, 15 mean level of liking variables were correlated significantly with the memory measures, and all were positive. There were, however, also simultaneous and retroactive correlations, and while the retroactive ones were primarily positive, the simultaneous relations were both positive and negative. This result is inconsistent with the notion that liking operates only proactively on attentional processes. The simultaneous effects are particularly puzzling because liking during a commercial can apparently damage memory for it. It may be, however, that if the memory measures had been at the level of the individual commercial rather than pods, the liking would have been positively correlated with one or two of the commercials, and the negative correlations have resulted from within-pod retro- or proactive relations.





Changes in Liking and Memory

Hypothesis 2 suggested that change in liking simultaneous with commercials should be positively associated with memory for them. Again, Table 1 shows support for the hypothesis. Because change scores from the whole program, the last two minutes, one minute, and half-minute were highly correlated, one of the four was counted if there was a significant correlation with memory. While there were only nine significant correlations between simultaneous change in liking variables and memory, seven of these were positive.

Hypothesis 3 suggested that retroactive and proactive change-in-liking values would be negatively associated with memory. As can be seen in Table 1, this prediction was supported. Of 38 significant proactive correlations, 29 (76%) were negative. Of 31 retroactive correlations, 22 (71%) were negative.

Comparison of the number of significant correlations for each of the eight change measures shows absolute value of the change scores and the standard deviation of mean liking to be strongest (15 significant correlations each). Both the change in liking and standard deviation of change in liking showed 11 significant correlations. The standard deviation of the difference in slopes had 9 significant correlations. The other three measures were weaker.

Prediction of Memory from Mean Liking and Change in Liking

Table 2 shows the squared multiple correlations for each of the preference measures. To reduce the number of independent variables entered in the equations, only scores for entire programs, commercial pods, and changes between programs and pods were w ed. These variables were entered step-wise into the equations (P-to-enter criterion was p<.05). Each multiple correlation was significant with between one and ten predictors of performance. These variables accounted for between 11 and 83 percent of the variation in performance (R2-.51). >. The predictions for recall were marginally better than those for recognition (R2 for recall - .52; R2 for recognition - .48) even though there was greater variation in how well recall performance could be predicted in the three different pods. Most notably, the predictions of recall performance in Pods 1 and 3 were better than for Pod 2 for information about product, brand. and scene. As noted before, Pot 2 also had the lowest level of performance in all three viewing orders.




The most important results in the present study were the strong proactive and retroactive relations between memory and the time-dependent variables, wean level of liking, and measures of change in liking. Mean liking showed predominantly positive effects, while the change in liking measures showed mostly negative effects. Further, liking and change in liking predicted significant percentages of the variation in performance scores, independent of commercial or programming content.



Most research in advertising and mass communication assumes that evaluations of messages as they occur are independent of prior and subsequent judgments, and that memory for media information can be predicted by consideration of simultaneous judgments alone. In contrast, the present study emphasized the influence of the viewing process. The time-dependent effects argue that watching television is best conceptualized as a series of interdependent program segments and messages, each influential backward and forward in time, as well as at the same point in time. In fact, the present study showed a greater number of significant predictions that were distant compared to predications that were proximal.

It is also important to note that the success of the performance predictions was not dependent on the products, brands, arguments, or executions of the commercials. The results were unchanged for three different viewing orders in which each of three programs and nine commercials occupied different positions. A substantial amount of variation in performance was attributable merely to the position of messages relative to other content, and to the nature of an individual's reaction to surrounding messages. It is likely, therefore, that studies that do not approximate a natural viewing situation by providing a context for the evaluation of messages could possibly overemphasize content effects at the expense of effects attributable to over-time processes.

Evaluations of commercial messages are typically assessed using judgments attributable to entire messages or to a portion of particular interest (usually content features like language, judgments of people, etc.). These evaluations are similar to the calculations of mean levels of liking in the present study. These are not the only relevant judgments, however. The present results indicate that the structure of evaluations - regardless of the particular messages to which the evaluations apply - are also significantly related to performance. The important structural evaluations were variability in response (standard deviation measures), trend of response (mean of differenced responses), and the change measures.


Anderson, Daniel R. and Elizabeth P. Lorch (1983), "Looking at Television, Action or Reaction?" In Children's Understanding of Television, eds. Jennings Bryant and Daniel R. Anderson, new York: Academic Press.

Bryant, Jennings and Paul W. Comisky (1978), "The Effect of Positioning a Message Within Differentially Cognitively Involving Portions of a Television Segment on Recall of the Message," R""an Communication Research, 5, 63-75.

Dysinger, Wendell S. and Christian A. Ruckmick (1933/1970), The Emotional Responses of Children to the Motion Picture Situation, new York: Arno Press and The New York Times.

Jones, Mari Riess (1974), "Cognitive Representations of Serial Patterns," in Human Information Processing: Tutorials in Performance and Cognition, ed. Barry R. Kantovitz, Hillsdale, NJ: Lawrence Erlbaum.

Kennedy, John R. (1971), "How Program Environment Affects Television Commercials," Journal of Advertising Research, 11. 33-38.

Klatzky, Roberta (1975), Human Memory: Structures and Processes, San Francisco: W.H. Freeman.

Kline, F. Gerald (1977), "Time in Communication Research, in Strategies for Communication Research, eds. Paul M. Hirsch, P.V. Miller and F. Gerald Kline, Beverly Hills: Sage.

Krull, Robert and Albert S. Paulson (1977), "Time Series Analysis in Communication Research," in Strategies for Communication Research, eds. Paul M. Hirsch, P.V. Miller, and F. Gerald Kline. Beverly Hills: Sage.

Krull, Robert and William Husson (1979), "Children's Attention: The Case of TV Viewing," in Children Communicating: Media and Development of Thought, Speech, Understanding, ed. Ellen Wartella, Beverly Hills: Sage.

Priemer, August B. (1983), "The Use of Qualitative Evaluation: Problems , Pitfalls, and Potentials," paper presented at the Association of National Advertisers Television Workshop, New York, February 15.

Saltz, Eli (1971), The Cognitive Bases of Human Learning, Homewood, IL: Dorsey Press.

Soldow, Gary F. and Victor Principe (1981), "Response to Commercials as a Function of Program Context," Journal of Advertising Research, 21, 59-65.

Thorson, Esther and Marian Friestad (in press), "The Effects of Emotion on Episodic Memory for TV Commercials," in Advertising and Consumer Psychology, eds. Alice Tybout and Pat Cafferata, NY: Lexington Press.

Thorson, Esther, Byron Reeves, Joan Schleuter, Annie Lang, and Michael L. Rothschild (1985), "Effects of Program Context on the Processing of Television Commercials," in Proceedings of the American Academy of Advertising, et. Nancy Stephens, Tempe AZ: Arizona State University.

Note. The authors thank the American Broadcasting Companies, Inc. and Guy Lometti, Director of Social Research, ABC, for their support of the research. Thanks are also due to Annie Lang for her assistance with the data files.



Esther Thorson, University of Wisconsin-Madison
Byron Reeves, University of Wisconsin-Madison


NA - Advances in Consumer Research Volume 13 | 1986

Share Proceeding

Featured papers

See More


Alternative “Facts”: The Effects of Narrative Processing on the Acceptance of Factual Information

Anne Hamby, Hofstra University
David Brinberg, Virginia Tech, USA

Read More


So-Bad-It’s-Good: When Consumers Prefer Bad Options

Evan Weingarten, University of California San Diego, USA
Amit Bhattacharjee, Erasmus University Rotterdam, The Netherlands
Patti Williams, University of Pennsylvania, USA

Read More


The Preference for Simultaneity: When Different Events Happen to Different People at the Same Time

Franklin Shaddy, University of Chicago, USA
Yanping Tu, University of Florida, USA
Ayelet Fishbach, University of Chicago, USA

Read More

Engage with Us

Becoming an Association for Consumer Research member is simple. Membership in ACR is relatively inexpensive, but brings significant benefits to its members.