An Attitudinal Measure of Ad Effects

ABSTRACT - ASTAS is an attitudinal measure of STAS-like effects of different campaigns, brands and media. It is developed in the Nordic countries in the absence of 'real single source data’ and with a media environment dominated by print advertising. It turns out, that the measure has promising qualities, also for use in other than Nordic environments.

Citation:

Flemming Hansen (2001) ,"An Attitudinal Measure of Ad Effects", in AP - Asia Pacific Advances in Consumer Research Volume 4, eds. Paula M. Tidwell and Thomas E. Muller, Provo, UT : Association for Consumer Research, Pages: 223-229.

Asia Pacific Advances in Consumer Research Volume 4, 2001      Pages 223-229

AN ATTITUDINAL MEASURE OF AD EFFECTS

Flemming Hansen, Copenhagen Business School, Denmark

ABSTRACT -

ASTAS is an attitudinal measure of STAS-like effects of different campaigns, brands and media. It is developed in the Nordic countries in the absence of 'real single source data’ and with a media environment dominated by print advertising. It turns out, that the measure has promising qualities, also for use in other than Nordic environments.

Short-term advertising strengths (STAS) measures were introduced by John Philip Jones in Jones (1995). Since then the concept and its measurement have been one of the hottest, debated issues in the area of Marketing and Advertising Research (Ehrenberg and Scriven 1997, McDonald 1997, Broadbent et al. 1997, Hansen 1988, Spitler 1998, and Roberts 1998). This Debate has been related to media planning (recency versus effective frequency), the role of long-term effects, interaction between advertising and promotion, etc.

THE STAS CONCEPT

Basically, STAS is computed on data like these illustrated in Table 1. Here, information is available for a product regarding its purchase on a particular day, and for the same respondents, the exposure to (television) advertising within the week precding the purchase.

In its original version, STAS is computed as the ratio between purchases among advertising exposed consumers divided by the ratio of purchases among non-exposed consumers. This figure is multiplied by 100 giving a score where figures larger than 100 suggest a positive effect of varying magnitude.

Until recently, data have practically always come from electronic single source data collection systems. As it will be discussed in the paper, the critical figure is the number of respondents who have both purchased and been exposed. Unless a reasonable number of such instances can be identified, say 50 or more, the STAS computation is associated with a large statistical uncertainty.

In addition to John Philip Jones’ original US data (Jones, 1995), similar findings have been presented for the UK, Germany and France. A typical distribution of STAS scores is depicted in Table 2, where Jones’ US data are shown. On the whole, the UK figures as well as the French and German figures tend to be slightly lower than the American figures, however, the overall scope of the distribution of findings is confirmed.

In the second generation, following the introduction of the concept and the debate related to it development is going in two directions. On the one hand, some argue for more extensive use of modelling on the available single source data (Broadbent et al. 1997, McDonald 1995, Roberts 1999). Others have been concerned with the feasibility of establishing single source data using other means of data collection than electronic single source panels.The present paper deals with this development.

Here, two alternative approaches will be presented. First, the use of traditional diary-based panels is reviewed. Subsequently, the use of data collected in ad-hoc interviews with large representative samples is reported.

In both cases, two questions arise.

1. How can data be collected, particularly how can a sufficient number of 'critical responses’ be obtained?

2. What limitations apply to the data collected in these ways in terms of validity and comparability with other findings?

CONSUMER PANEL BASED STAS

The traditional panel approach is not a new one. Actually, the first STAS-like computations were carried out by Colin McDonald in 1969 based upon such data. Later, similar approaches have been attempted in Norway, and on a smaller scale in Denmark. The purpose of this has partly been to make STAS data available in markets where electronic single source data are not likely to be introduced within a foreseeable future, and partly to allow for the inclusion of effects of print media,- media which in both of these countries have a considerably larger share than television advertising. On the whole, these experiments have suffered from a very low number of observations in the 'critical cell’, that is, few observations of respondents who have both 'seen’ and 'bought’. This has been a result of the relatively small ad-hoc panels used (500B1000 respondents during 2B3 months). Also, computational problems have been considerable owing to the traditional complications associated with incomplete continuous panel reporting. Still however, the best findings (Helgesen, 1998) from Norway suggest that the STAS measures are similar to those based on electronic single source data, yet with a tendency for the scores to be slightly higher.

ATTITUDINAL STAS

Faced with these problems, it has been attempted in Denmark to develop a procedure based upon personal interviews. This is done on a continuous CAPI data collection system on which a traditional print and television ad recall measurement is administered. In 199899 various experiments have been made by adding recent purchase' questions to this survey. Based upon this, experiences have been obtained with the computation of STAS-like scores.

Different data bases have been collected, testing the effects of asking about purchases in more extended periods than one day, about store visits versus brand purchases and exploring the effects of different media. To ask about ad-exposure and recall of recent purchases, gives data that must be interpreted as reflecting attitudes as well as actual behaviour. Thus, the scores have been labelled 'Attitudinal Short Term Advertising Strength' (ASTAS).

The first attempt to establish attitudinal STAS scores was done by adding a simple purchase question: 'Have you bought brand X yesterday?' to a standard ad recall survey where people for a precoded list of brands were asked whether they had seen television commercials or print advertising recently.

These scores, like the Norwegian panel based scores, tended to be higher than what has been reported from electronic based single source channels. There are several reasons for this.

The first attempt to establish attitudinal STAS scores was done by adding a simple purchase question: 'Have you bought brand X yesterday?' to a standard ad recall survey where people for a precoded list of brands were asked whether they had seen television commercials or print advertising recently. These scores, like the Norwegian panel based scores, tended to be higher than what has been reported from electronic based single source channels. There are several reasons for this.

TABLE 1

STAS COMPUTATION

TABLE 2

STAS SCORES FROM THE US, UK AND GERMANY (JONES 1998)

TABLE 3

DANISH STAS MEASURES, TV + PRINT (11 BRANDS AND 3 STORES)

TABLE 4

NUMBER OF RESPONDENTS IN CRITICAL CELL WITH 1 AND 3 DAYS' RECALL

First by asking for exposure lately you are likely to include more exposure than what occurred the last week, which is measured with single source based STAS computations.

Secondly, there is a bias relating to the respondent. Those respondents claiming to have seen advertising for a particular brand and bought the same brand are likely to be consumers who see more advertising and buy more brands than the average consumer.

Thirdly, just as people who have seen a particular advertising are more likely to buy the brand, people who have bought the brand may be more likely to see advertising for the brand, or to claim that they have done so.

All this has implications. Whereas, as discussed with regular STAS measures (Jones, 1995), a score larger than 100 is expected to reflect positive effects of advertising, "100" is not necessarily corresponding to the neutral value for the ASTAS scores. If it was, it would imply that the scores were nominally scaled. This may be the case with traditional STAS scores, but it can certainly not be expected to be the case with attitudinal STAS scores. Still, however, higher scores suggest more effective advertising, and smaller scores are interpreted as indications of advertising working less efficiently. This is particularly the case, when some of the before mentioned bias is minimised. One can expect brands in different product areas to perform differently. Closer inspection of the data shows that different levels exist for detergents, soft drinks, tabloid newspapers etc. These different levels, however, do not imply that the attitudinal STAS measure is not a useful advertising effect measure. Only they suggest that meaningful comparisons must be made between related brands, between same brands over time, between media or between whole product categories.

THE SIZE OF THE CRITICAL CELL

Obviously ASTAS scores are only meaningful if a sufficient number of observations are available. Particularly, the critical cell: Those who have purchased and seen advertising is important. In many analyses of STAS data, this cell is very small when computed for single brands. Therefore, most findings published, are given not on a brand level, but as percentiles or in other aggregated fashions.

With attitudinal STAS measures, there are several ways in which the size of critical cell can be improved. Obviously, it can be done by increasing the number of interviews (but thereby normally also costs and the length of the data collection period) more critical cell observations can be obtained.

The same can be accomplished by extending the length of the "time window" in which purchase questions are asked. Asking for purchases within the last three days, rather than for the last day may make good sense for many fast moving consumer goods. Actually, in the second experiment, one of the purposes was to learn about the effect of doing this. Here, three brands were included, that had been included in the initial study also.

By comparing the number of purchase responses, relative to the number of interviews in the two samples, it was found that there was a small tendency to overstate purchases when asked about three days rather than what could be expected by comparing with three times the score for 'yesterday'. This difference, however, was not statistically significant. Consequently, it was decided to go along with the three day questioning interval.

Actually, one can argue for the use of different purchasing (and advertising exposure intervals) for different products. With products with very low purchasing frequency, longer purchase intervals than three days could make sense. When doing this, of course, comparability with brands measured with shorter purchasing intervals is lost, but again, the most valuable use of attitudinal STAS measures are to be found in comparisons between related brands and for same brands in different time periods.

EXTENDED SAMPLE SIZE

Since the initial results were so promising, it was decided to start data collection with 23 brands. This was done in the first quarter of 1999. For television viewing, it provided a total database of 3778 respondents.

Here, plain ASTAS measures are computed for the 23 brands with related confidence intervals. Here a specially developed test, providing confidence intervals for ASTAS (and STAS) scores, is used. In this proceeding, the testing of significance of differences is also possible between scores for different brands or between the score of a brand in different periods (Olsen and Nilsson, 1999). It appears that the intervals are generally large, and of course the larger, the smaller the critical cell is

Still, the confidence intervals are of a magnitude making it meaningful to test statistical significance of the findings. Here almost all of the chi-square values for tests of significance suggest a significant advertising effect. Only with very small critical values or with ASTAS scores close to one, this is not the case.

TABLE 5

STAS SCORES TV, N=3874

As discussed earlier, consumers who see advertising for a brand are more frequent shoppers than consumers that do not. Consequently, they are also likely to buy other products more frequently, and we would expect an artificial ASTAS score, computed for purchases of any other brand than brand x. Relative to exposure to brand x advertising to larger than 100.

Using the largest data bases available, with the 3878 interviews, recent TV-ad exposure and purchases, the usefulness of adjusted ASTAS measures is demonstrated. It is argued that it provides much more sensitive and useful advertising effect measures than traditional ad or brand recall, attitude, preference or purchase intention questions.

Actually, this is also the case. This upwards bias reflect the response bias discussed earlier. This, we can adjust for by computing an average "false" ASTAS scores for all other brands as an effect of exposure to advertising for brand x, and divide the true ASTAS with this, (Jones, 1998).

ADJUSTING FOR MEDIA EXPOSURE

With television campaigns, often planned in bursts, it is possible to disregard observations made in periods, where television viewing can not possibly have occurred. This adjustment refine the reporting of ASTAS scores for television. Since similar adjustments rarely can be made meaningfully for print media, it is however recommendable to do so only in connection with campaigns where television is the dominant media. The general effect of this correction is a slight downward adjustment of the ASTAS scores.

TABLE 6

TV VERSUS PRINT, ADJUSTED SCORES, N=2845

ASTAS FOR PRINT

It is perfectly possible to compute ASTAS scores for print advertising. As suggested before, these scores can not be adjusted for actual media exposure, since particularly magazine advertising may provide exposure over extended periods of time.

When comparing unadjusted ASTAS scores for print with similar scores for television, an interesting picture emerges. Like in the Norwegian and in the Danish panel cases, print scores higher than television. STAS scores for print have not been made available internationally. It is not possible to conclude whether this is a general, or a special Nordic phenomenon, or whether it reflects the ASTAS measurement procedures. However, findings from Starch show a similar tendency. One may speculate, that with fast moving consumer goods, as they have been studied in ASTAS surveys, the

impact of special offers in daily newspapers and of free sheets, combined with deals in stores, may account for this effect. It is noteworthy, thus, that the adjustment introduced in connection with television scores gives a larger downwards adjustment for print, than for television, suggesting that print viewers on the whole may be more active shoppers of advertised brands.

CAPI VERSUS CATI

In the data reported so far, CAPI interviews, where lists of the brands in question are shown to the respondents, have been used. Purchase questions have been phrased before advertising exposure questions, with a considerable number of other questions in between. An opportunity for at the same cost to increase the number of responses in the critical cell, or to reduce costs, would be to use CATI interviews. This would also make it easier to use disproportionate sampling. The more interviews conducted, the more efficient disproportionate sampling will be. Since Gallup, Taylor Nielsen, Sofras in Denmark conduct a very large number of CATI interviews per year (app.60,000), the possibility exists, when using this technique, to increase the number Of respondents in the critical cell significantly with the use of disproportionate sampling, this is done by asking the advertising exposure question only to respondents who have made a purchase. Here it is important, whether the CATI interview provide findings in line with those of the CAPI procedure. This has been tested with 1007 respondents in February 1999, for the 23 brands already analysed. Data are shown in Table 7.

TABLE 7

CAPI VERSUS CATI, TV-ASTAS SCORES

Only in few cases do the CATI scores deviate significantly from the CAPI score. There is however an overall tendency for the CATI score to be slightly lower than the CAPI score. When this is adjusted for, the differences are less expressed. Thus, it is perfectly feasible to work with CATI based ASTAS scores, proving large samples and particularly large critical cells.

TRACKING WITH ASTAS

Basically, the ASTAS score is an attitudinal measure of advertising effects. It does not allow itself for computations of added sale relative to different levels of spending. It only reflects attitudinal responses in the market, but it does so in a way, which is probably more closely linked with subsequent purchasing behaviour than any other existing attitudinal measure, such as purchasing intentions, brand preferences, top brand awareness etc. Therefore, it has promising implications for tracking purposes. It can be used for-tracking in its own right. It is possible to compute ASTAS scores for periods with high versus periods with no advertising, for periods with strong competitive advertising versus periods without competing advertising etc. As long as data collecting goes on continuously, it is possible to define any period of relevance or any particular competing brand in a relevant period, and then to carry out meaningful ASTAS computations. The insight gained from doing so on a continuous basis has a great potential. With more frequently purchased brands, it may even be possible to relate changes in ASTAS to media planning, and thereby provide an alternative continuous tracking measure. Also, as long as data collection takes place continuously, it is possible to add other relevant tracking questions to the questionnaire used, including such things as message comprehension, brand linking, brand awareness, purchase intentions etc. With such a repertoire of measures built up around the ASTAS score ideal, conditions emerge for gaining better insight into the way in witch advertising functions.

REFERENCES

Broadbent, S., J.Z. Spittler and K. Lynch (1997), Building Better TV Schedules: New Light from the Single Source, Journal of Advertising Research, Vol. 37, 4 July-August.

Ehrenberg, A. and J. Scriven (1997), Added Values or Propensities to Buy, Admap, Jan., pp. 11-13.

Hansen, F. (1998), "Advertising Testing" in ESOMAR Handbook of Market and Opinion Research, Advertising research: testing communication effects.

Helgesen, T. (1998) Single Source Data from Panel Voters. Paper presented at the 1998 ASI Seminar, Hamburg, June 1998.

Jones, J. Ph. (1995), When Ads Work: New Proof that Advertising Triggers Sales. New York: Lexington Books/The Free Press.

Jones, J. Ph. (1998), How Advertising Works, Publications

McDonald, C. (1995), Where to look for the Most Trustworthy Evidence, Short-term Advertising Effects are the Key. Admap, The Advertising Association, January.

McDonald, C. (1969), Relationships between Advertising Exposure and Purchasing Behaviour, Market Research Society Conference, pp. 67-98.

Olsen, J.K. and Nilsson O.S.(1999), Measuring Consumer Response to Price Deals in the Grocery Sector. Research Paper, The Marketing Institute, Copenhagen Business School.

Roberts, A. (1997), Linking Sales to Advertising Activity, Admap, The Advertising Association, January, QEII Conference Centre, London.

Spittler, J.Z. (1998) Untangling the Confusion of TV Scheduling Theories, in Esomar Proceedings from "The world wide Media Research Seminar, " Mexico City

----------------------------------------