The Effects of Edits on Arousal, Attention, and Memory for Television Messages: When an Edit Is an Edit Can an Edit Be Too Much?

Memory was measured using a visual recognition task. Brief video scenes were shown to the subjects who pushed the buttons on a joystick to indicate whether they had seen the scene before. Each scene was 6 frames long. Subjects viewed 120 scenes separated by 2 seconds of black. Sixty of the scenes came from messages the subjects had seen before, while the other half, foils, came from messages they had not seen before. Both the correctness and the speed of their responses were measured. The visual recognition task took 4.2 minutes.

Recognition hypotheses were tested using a signal detection analysis of the recognition data. Signal detection theory is discussed at length and applied to communication research in Shapiro, 1994. Signal detection analysis is based on the theory that finding a memory in your memory is like detecting a weak signal in the environment. Two major components affect a person's decision as to whether or not a signal (or memory) has been detected. One dimension of that decision is how sensitive one's senses are to changes in the environment. This is called sensitivity or d prime. The second dimension of that decision relates to how willing a person is to guess, that is how liberal or how conservative their decision-making strategy is. This is called the criterion bias.

To perform a signal detection analysis four values are computed for each subject: (1) the percentage of hits-that is the percentage of items subjects say they have seen before which they have in fact seen before; (2) the percentage of misses-that is, the percentage of items they said they had not seen before which they had seen before; (3) the percentage of correct rejections-that is the percentage of items they said they had not seen before that, they had not seen before, and; (4) their percentage of false alarms-that is, items they said they had seen before that they had not seen before. These values are combined to compute a subject's sensitivity (called d prime) and the subject's criterion bias. The greater a subject's sensitivity - the more accurate their memory is, both in terms of hits and correct rejections. Criterion bias is determined by the number of false alarms and misses and is interpreted as how confident a subject needs to feel about having seen an item before he or she is willing to say the item was seen before. Using signal detection analysis allows one to attribute an increase in percent correct to either improved sensitivity or a shift in criterion bias.

Attention was assessed by measuring viewers' heart rate during viewing (Lang, Newhagen, & Reeves, 1996; Martin & Venables, 1983). Research demonstrates that high attention to an external stimulus (like a television message) results in significant slowing of the heart rate (Lacey & Lacey, 1974; Lacey, Kagan, Lacey, & Moss, 1963; Lang, 1990; Lang, Newhagen, & Reeves, 1997). If viewers pay more attention to messages as the number of edits increases, then they should have slower mean heart rates (indicative of greater attention) during faster paced messages than they do during slower paced messages. Heart rate is a relatively slow responding physiological measure. Because, on average, the heart beats about one time per second, and because it may take two to three beats for a response to begin, heart rate responses are generally analyzed over time and referenced to the beginning of the stimulus. In this case, not only does the heart rate response unfold over time, but so does the independent variable manipulation. The rate of edits, that is the number of edits occurring over time, is not apparent until some time has passed. For this reason, heart rate is collected as milliseconds between beats and averaged over various lengths of time. In this study, heart rate was averaged over two 30-second periods. By the second 30-second period, the effect of rate of edits should be demonstrable relative to the initial 30 seconds. The design used for the analysis is a mixed 4 (Order of Presentation) X 4 (Edits) X 5 (Message) x 2 (Time) ANOVA. Thus, the hypothesis is for an Edits X Time interaction.

Arousal was measured in two ways: 1) viewers used the SAM (Self-Assessment Mannequin; Lang, Greenwald, Bradley, & Hamm, 1993) scale to report how aroused they felt following each message, and; 2) The frequency of skin conductance responses (SCRs) was measured during viewing (Hopkins & Fletcher, 1994).

Apparatus

The experiment was controlled by a Zenith 386 computer with a Labmaster A/D D/A board. SC was measured by placing two Beckman standard Ag--AgCl electrodes on the subject's non-dominant hand after washing the skin with distilled water to control hydration. The signal was passed to a Coulbourn SC module. SC level was sampled and recorded 10 times per second throughout message viewing. Spontaneous skin conductance responses (SCRs) greater than. 10 microsiemens were scored to obtain the frequency of SCRs per message.

HR was measured by placing two Beckman mini Ag--AgCl electrodes on subjects' forearms. A ground electrode was placed on subjects' non-dominant forearm. HR was recorded using a Coulbourn bio-amplifier with filters. Heart beats were recorded as milliseconds between beats and converted to HR per second. Change scores were computed and then averaged over two 30-second periods.

Procedure

Participants were tested individually. An experimenter greeted the participant and explained that his or her heart rate and skin conductance data would be recorded using small sensors attached to forearms and hands. Each participant signed a consent form before the experiment. Participants were seated in a comfortable chair in a small room about five feet from the television monitor.

The experimenter instructed subjects to sit quietly during viewing and to pay close attention to the messages as his or her memory would be tested later. The experimenter then started the stimulus tape, and the subject viewed the messages. Before the recognition test the experimenter instructed the subject on how to use the joystick. The recognition tape was played and the subject indicated whether they had seen the scenes or not.

This experiment was conducted in conjunction with two others reported elsewhere. As a result, subjects performed six tasks during this experiment. First subjects either viewed television messages (this experiment) or read headlines on a computer screen; then they listened to a 6 minute radio message, and finally they performed the task they hadn't done first (i.e. viewed television or read headlines). Following these three stimulus presentations, subjects completed recognition tests for all three experiments in the order they saw the stimuli. The whole experiment lasted about 11/2 hours.

Analysis

All of the analyses involved the same basic 4 (Order of Presentation) X 4 (Edits) X 5 (Message) ANOVA. For the heart rate analysis, as discussed previously, an additional Time factor with two levels was added. Order of Presentation was a factor in all the analyses but there were no significant main effects of Order and no significant interactions of Order with the results reported here.

Results

Hypothesis 1

Hypothesis 1 predicted that, as the number of edits in a message increased, viewers would pay more attention, and as a result, have slower heart rates during the messages. Figure 1 shows the significant Time main effect (F(1,29)=42.99, p [is less than] .000, epsilon squared = .36(5))) and the predicted Heart Rate by Time interaction (F(3,87)=2.78, p [is less than] .046, epsilon squared = .05). Heart rate was significantly slower in the second 30 seconds of each message than it was in the first 30 seconds. That decrease is greater for messages with more edits compared to messages with fewer edits. Thus, while subjects' attention increased in the second half of the message for all messages, that increase was greater for fast and very fast paced messages compared with slow and medium messages.

[Figure 1 ILLUSTRATION OMITTED]

Hypothesis 2

This hypothesis predicted that as the frequency of edits increased, memory would increase. To test this hypothesis a signal detection analysis of the recognition data was performed. Significant main effects in the predicted direction were found for both sensitivity (d prime) (F(3,84) = 13.64, p [is less than] .000, epsilon squared = .30) and criterion bias (F(3,84) = 51.73, p [is less than] .000, epsilon squared = .62) and are shown in Figure 2. The means are given in Table 1. Thus, subjects were both more sensitive and more willing to guess during fast and very fast messages than during slow and medium messages.

[Figure 2 ILLUSTRATION OMITTED]

Table 1 Mean Criterion Bias and Sensitivity Scores by Number of Edits Rate of Edits Sensitivity Criterion Bias Slow 1.12a -.31a Medium 1.22a -.83b Fast 1.71b -.23a Very Fast 1.86b -.14c

Hypothesis 3

This hypothesis predicted that viewers' arousal will increase as the frequency level of edits increased. Arousal was measured both in terms of sympathetic nervous system activity, indexed by skin conductance, and through the use of self-report measures. As predicted, both self-report and physiological arousal increased as a function of number of edits. The main effect for Edits on the self-report data was significant (F(3,39)=27.14, p [is less than] .000, epsilon squared = .60) and is shown in Figure 3. The slow (M=3.87) and medium (M=3.60) paced messages were significantly less arousing than the fast (M=6.15) and the very fast paced (M= 7.44) messages.

[Figure 3 ILLUSTRATION OMITTED]