Minimum effective frequency for interactive television ads
- First Online:
- Cite this article as:
- Bellman, S., Schweda, A. & Varan, D. J Direct Data Digit Mark Pract (2010) 11: 281. doi:10.1057/dddmp.2010.6
- 81 Downloads
A key task for advertising media planners is setting a frequency goal for a campaign. This study used a controlled experiment to identify the minimum effective frequency (MEF) for TV ads offering interactive response, as opposed to direct response by phone call. Participants (N=273) were exposed to ads one, three or five times. A control cell saw normal TV ads, while an interactive TV (iTV) cell saw the same ads with interactive response banners superimposed. We found the usual ‘build-up’ effects for repeated exposure on ad memory in the control cell, but there was little added value in repeat exposure for generating interactive response (ie MEF=1). Interaction rates were higher for familiar brands, but brand familiarity did not alter the effect of repetition. These results suggest that iTV ads should be placed with a reach strategy, rather than a frequency strategy.
Keywordsinteractive television advertising media planning frequency response wear-out
Lab studies have done much to clarify the conflicting findings from the field, which have been used to support repetitionist and minimalist arguments. Most importantly, in contrast to field data (eg people meter data),7 in a lab we can be sure that an ad has been seen. In the field, ads are typically viewed with low attention, and with long gaps between exposures, which together can delay the positive effects of wear-in and the negative effects of wear-out until after 10 or 15 exposures.15 A recent lab study has found that four repeat exposures in a single session, equivalent to many more repeats outside the lab, has a positive ‘build-up’ effect on intention to click.16 In addition, in the lab we can isolate the effects of repetition by controlling or manipulating the effects of product category, brand familiarity and within-category competition effects.17 For example, Campbell and Keller found that the wear-in-wear-out (inverse-U) effect of repetition applies only to unfamiliar brands.18 In our study, we varied brand familiarity to test whether it influenced the effect of repetition on the response rate for iTV advertising.
Recommendations for direct response TV ads
The main purpose of this study was to investigate whether the general rule of thumb for direct response TV ads, ‘one is enough’,20 also applies to interactive response TV advertising. We expected, therefore, that most interactive response would occur within the first two full-attention exposures. However, we did not rule out build-up effects for interactive response TV ads. For interactive response, the threshold of involvement required may be very low, since viewers can respond without leaving their chairs. Not only does this suggest that interactive response rates are higher than phone call response rates, it also suggests that interactive response might be influenced by very subtle persuasion effects, such as familiarity derived from repeated exposure.
The next section develops our hypotheses, after which we describe the lab study we used to test these hypotheses. After reporting the results, we discuss their implications for practitioners, especially media buyers, and for future research.
In this section, we develop a series of hypotheses about the effects of repetition on responses to iTV ads. We relied mainly on the literature that has investigated the effects of repetition on TV advertising. We note, though, that to the extent that the ‘call to action’ banners in iTV ads resemble web banner ads, and are therefore subject to ‘banner blindness’,23 they would have a very low probability of being seen, and more repetition would be required.
Effects of repetition on the interaction rate
Interactive response to iTV ads will display diminishing returns: the highest response rate will be associated with the first exposure, and will decline with repetition.
Moderators of the effects of repetition
Further research into the moderating effects of central versus peripheral processing on repeated messages has identified managerially controllable situations that reliably increase or decrease motivation or ability to process. For example, Batra and Ray found that motivation and ability are higher for leading brands of shopping products (photographic film, deodorants and facial moisturizers) compared to follower brands of convenience products (instant coffee, instant chocolate drink mixes and frozen pizzas).33 Consequently, brand attitudes and purchase intentions for high-motivation/ability products can exhibit an inverse-U pattern when messages are repeated, whereas for low-motivation/ability products, high repetition has an increasingly positive effect on both measures. Recent research has investigated the moderating effects of familiarity, which can affect ability to process even within a high-motivation (high-involvement) product category (eg banks, women's clothing, or health-care plans18).
Familiar versus unfamiliar brands
For familiar brands, repetition will have no negative effect (ie a positive or flat effect) on interactive response. In contrast, for unfamiliar brands, repetition will have an inverse-U effect on interactive response — positive at first, then negative.
Speed of wear-out
Inverse-U effects of repetition on interactive response will be seen only for ads with a moderate speed of wear-out, rated in pre-tests. Fast wear-out will accelerate a decline in interactive response, whereas slow wear-out will delay this decline, or even increase response with repetition.
Sample and design
Two hundred and seventy three members of an audience panel (representative of the general public) participated individually in a one-and-a-half hour study, and each received a $20 (AUD) department store gift voucher. Participants were randomly assigned to a 2 (interactivity: normal (control) ads versus interactive ads) × 3 (ad repetition: 1, 3, or 5 exposures) × 2 (brand familiarity: familiar or unfamiliar) × 3 (wear-out: slow, moderate or fast) factorial design. Interactivity was a between-subjects factor, while all the others were within-subjects factors. We deleted data from 66 participants who had missing data or whose answers to probes in the post-test survey indicated that their response rates had been affected by experimental demand effects (ie they thought they had to respond to every ad) or an inability to understand the instructions. Half of the final sample (95 of 207 (46 per cent)) were women, and ages ranged from 18 to 86 years (M=45.23, SD=15.38). Tests of the randomness of assignment to the two between-subjects conditions, using several demographic variables (eg number of hours watching TV daily), detected no significant differences.
Pilot test results
F(4, 155)=1.73, p=0.146
F(4, 155)=2.32, p=0.059
F(4, 155)=3.54, p=0.009
The interactive versions of the ads used the impulse response format, which is the closest iTV ad format to traditional direct response ads, and has been used on the British Sky Broadcasting platform in the UK and the Wink platform in the US.43 We could have used other interactive formats, such as the dedicated advertiser location (DAL, or microsite) ads and telescopic ads tested by Bellman, Schweda and Varan.44 These ads offer a longer interactive experience, but in that study had interaction rates equivalent to the rate for impulse response ads, although the number of interactive opportunities offered was very limited (eg just one telescopic ad in a half-hour programme). We chose to use the impulse format in this study because the first few interactions with longer-duration iTV formats might exhaust the appetite for interaction, which would have obscured the effects of repetition we were hoping to observe. In this study, to equalize the offers across product categories, all the ‘call to action’ banners superimposed over the normal TV ads offered entry into a competition. The prizes for these competitions were tailored to the product category (eg to win the camcorder featured in the ad). Entry in the competition was acknowledged by a confirmation banner. Since viewers could enter each competition only once, subsequent interactions with repeated ads generated a banner telling the participant they had already entered that competition. In the results below, we count only the first response in our measure of response rate.
Procedure and measures
Approximately 24 hours after completing the study, participants who had consented to being interviewed by telephone the next day were asked to recall as much as they could about any test ads they claimed to remember seeing (164 (79 per cent of 207) consented). If the interviewer (a trained research assistant) judged that the participant had ‘proven’ recall of the ad, day-after recall was coded as 1, and 0 otherwise.50 Demographics, such as gender, age and education level, were already known for these panel members.
Even though we carried out an experiment, our response data are right-censored, that is, we do not know for sure whether participants who had not responded by the fifth exposure would have gone on to respond to a sixth exposure, or even later exposures. An appropriate method of analysis for these kinds of data is survival analysis, which typically deals with predicting the probability of ‘failure’ at a certain time. In our case, we reverse the usual interpretation of such models to predict the probability of ‘success’, that is, interactive response to a TV ad. We used the Cox regression model, as it has the advantage of being nonparametric. Because nonparametric approaches impose less structure on the data, compared to parametric approaches, they can yield a more accurate representation of the hazard rate.51, 52 (We note that we obtain similar results using a parametric inverse Gaussian survival model, but its integrated hazard function suggests that it distorts the hazard rate.) The Cox model assumes that covariates, which raise or lower the hazard of failure (response), have a fixed effect in each period of time, but this can be tested by incorporating time-varying covariates, that is, covariates whose effects change over time. If one of these time × covariate interactions is significant, then the assumption of equal effects over time has to be relaxed for that particular covariate (eg familiarity or wear-out).
We used participants from the control sample (n=62), who viewed normal TV ads without interactive enhancements, to test for the effectiveness of our manipulations of repetition, brand familiarity and speed of wear-out. The control sample rated the familiar brands as significantly more familiar (MF=6.11 versus MU=1.39, t(60)=30.12, p<0.001, η2=0.938 (small effect=0.010, medium=0.059, large=0.138)). Familiar ads were also liked more (Aad: MF=5.39 versus MU=4.51, t(59)=5.53, p<0.001, η2=0.341).
In the following analyses, and in the survival analysis used to test the hypotheses, we treated each test ad as a separate observation for each individual, which meant that we had a total of six observations per person (N=207 × 6=1,242, 145 × 6=870 interactive, 62 × 6=372 controls). We ran a regression model with one predictor, a constant, for each of the continuous dependent variables, and the smallest Durbin-Watson statistic we obtained was 1.80 (for Ab; Aad=1.81), which indicated that there was no problem with this assumption of independence between observations.
Life table of responses by time (opportunity to respond)
% of total responses
Cox regression result
Familiarity × wear-out
Familiarity × fast wear-out
Familiarity × slow wear-out
Familiarity × time
Wear-out × time
One explanation for why we found no moderating effects of familiarity or wear-out (on the general effect of repetition on interactive response) could be that we used products that were all associated with high motivation (involvement) and ability to process. However, three of our products (nuts, salad dressing and soda) were clearly regularly purchased convenience goods (mean purchase/usage frequency=12.77 times per month), the types of products that Batra and Ray showed can be advertised repeatedly with no negative effects, because ads for these products are processed peripherally.33 In contrast, the other two products (car insurance and video camcorders) were clearly infrequently purchased shopping goods (M=0.03 times per month; a significantly lower frequency: t(60)=10.75, p<0.001, η2=0.66 (results from the control sample)). However, Campbell and Keller have shown that repeated ads for familiar brands, even of shopping products like these, can have no negative effects, because ads for familiar brands are also processed peripherally.
Another potential explanation for our results is that by superimposing interactive overlay banners over television ads, we increased their interest so that all the ads in our interactive condition were processed centrally. If our iTV ads were centrally processed, the effects of repetition would have been compressed, so that it would not be surprising to see wear-in and wear-out after just one exposure. If we had taken measures of product category involvement,53 or message involvement,54 we could have compared our interactive condition ads to the normal TV ads seen by our control condition participants, and tested whether all iTV ads are processed with greater motivation, even the ones people don't interact with. However, in previous studies, central processing has been associated with higher levels of recall.55 Our results suggest that, if anything, adding interactive overlays to TV ads reduces free recall (from 46.2 per cent for our control ads down to 37.8 per cent for iTV ads that weren’t interacted with; χ2(1)=5.37, p=0.021). Although this comparison is affected by self-selection bias (the iTV ads that people did not interact with were those they were least interested in), it is inconsistent with the hypothesis that all iTV ads are processed centrally.
We also investigated the potential moderating effects of brand familiarity, and speed of wear-out, on this main effect of repetition on interactive response. Familiar brands had a higher response rate, but familiarity did not alter the effect of repetition. In particular, ads for familiar brands were not associated with positive effects of repetition, and ads for unfamiliar brands were not associated with negative effects. This result was inconsistent with Campbell and Keller's previous research using non-interactive TV ads.18 Again, however, the key difference between our study and theirs is probably our use of a dichotomous dependent variable, interactive response, which we could observe only when the influences on this behaviour crossed a certain threshold. Campbell and Keller measured attitudes using continuous scales, rather than dichotomous ones, and therefore they could readily track fluctuations in these variables with repetition. We also observed fluctuations with repetition in attitudes to interactive ads, and the brands they advertised, in our manipulation check data. However, our data also suggest that the threshold for making an interactive response is readily achieved in the first exposure by ads for familiar brands. In contrast, ads for unfamiliar brands rarely reach this threshold, even when learning is compressed by the high-involvement situation of viewing ads in a lab. We tested and rejected an alternative explanation for our results: that superimposing interactive banners over TV ads increases the cognitive resources devoted to processing, for example by eliciting attention reflexes (orienting responses),56 which would have ensured that familiar brands were processed centrally, just like unfamiliar brands. Nevertheless, another explanation for our results is that our manipulation of familiarity was not as strong as that by Campbell and Keller. Unlike them, we used real ads, and therefore real brands, for our unfamiliar-brand ads, which may have increased their credibility. We also confounded familiarity with the potentially countervailing effects of ad execution. We return to this point below in our suggestions for future research.
We had expected that ads that wore out fast, in pre-tests, would definitely not show any build-up effects in their response rates, whereas ads that were very slow to wear out might still encourage interactive response after repeated exposure. However, speed of wear-out, like familiarity, did not alter the main effect of repetition. Again, this suggests that ads that are interacted with achieve the threshold for this behavioural response in the first exposure, and for ads that aren’t interacted with the first time, further repeats do not raise behavioural intention to anywhere near this threshold level.
Implications for practitioners
Our results also suggest that interactive response campaigns should be very short in duration. Returns from repeat insertions will rapidly become unprofitable. To extend the life of a campaign, multiple offers (‘calls to action’) could be used, as different offers might be attractive to different segments within the same audience over time.
Since frequency makes little difference to response, iTV advertisers should instead use a reach strategy, aiming to reduce duplication of exposure as much as possible.57 With new technologies such as DVRs increasing the rates of ad avoidance, it is important for advertisers to maximize the yield from the few exposures they are able to achieve. An interactive response to a single TV exposure deepens a viewer's experience with the advertised brand, and increases the impact of that exposure, potentially offering all the benefits of repeated exposure (persuasion to act) in a single exposure.
Our findings apply to the typical situation for media planners in which repetition is booked ahead of schedule. If advertisers are in the position of being able to send out exposures on the basis of response data at the individual household level, which is definitely a possibility with iTV, they could take advantage of beta-binomial models for estimating the likelihood of the household responding, given that no response has been seen so far.58, 59, 60
Limitations and suggestions for future research
Lab exposure can be predictive of in-market success, and models of repetition effects in the field can be calibrated using data from lab experiments.63, 38 However, as in all lab studies, our findings should be transferred to the real world with caution. As we argued above, we used a lab study to compress the effects of repetition into a practical number of exposures. In the field, where people pay less attention to ads than our participants did in the lab, the number of insertions required to achieve the equivalent of one full exposure in our lab may be many times higher than one.64 Our rate of response was also much higher than it would be outside the lab. However, we offered estimates above of how many additional exposures might be needed to reproduce our results in the field, based on Rossiter and Danaher's recommendations.20
Finally, a limitation of this study is that we investigated only one type of interactive response ad: the impulse response format. Other formats exist for iTV ads, such as the dedicated advertiser location (DAL), offered by British Sky Broadcasting in the UK, which allows viewers to leave the programme and immerse themselves in a brand-building interactive experience, consisting of multiple pages of text and video, similar to a small website. For these ads, direct response may not be the focus: advertisers may be happy simply to build favourable brand awareness. Repetition may have different effects on responses to these types of ads, and repeated interactions may be meaningful responses. Future research should investigate the effects of repetition on long-experience iTV ads such as the DAL format.
The results of this lab study into the effects of repetition on response to interactive TV ads imply that the conventional thinking associated with repeat message exposure does not fit the interactive proposition. People will either interact or they won’t. Repeat exposure doesn’t help. In strategic terms, this suggests that interactive campaigns should be short, and that interactive advertisers should focus on reach rather than frequency. Although the number of insertions required to achieve the equivalent of one full exposure in the lab will vary in real life, advertisers can use the results of this article to plan their media buys for iTV ads.
This research was supported by the member companies of the Beyond: 30 consortium and by the Australasian Cooperative Research Centre (CRC) for Interaction Design.