You are on page 1of 7

Published by World Advertising Research Center Farm Road, Henley-on-Thames, Oxon RG9 1EJ, UK Tel: +44 (0)1491 411000

Fax: +44 (0)1491 571188

June 2002, Issue 429

Advertising Strategy

Advertising Half-Lives for fmcg Brands and Markets


Professor Flemming Hansen and Lotte Yssing Hansen, Copenhagen Business School (CBS). Single-source data allow consumer purchases and advertising exposures to be directly linked. This makes possible very detailed and in-depth analyses of the influence of advertising on the purchase situation. Earlier studies with the data presently available have used the same retention rate for all brands studied1. However, this study concentrates on choosing the optimal retention rate, to learn whether it is possible, with good reason, to assume that there is an overall retention rate for all brands across the different product categories or whether it is necessary to estimate an optimal retention rate for each brand. As one of the explanatory variables in an advertising effect model we use an adstock calculation. To calculate an adstock value, it is necessary to determine a retention rate. In this work we estimate the optimal retention rate for 89 different fmcg (packaged goods) brands and calculate the corresponding half-life. We find that there are large variations in the advertising half-life from brand to brand and from product cat-egory to product category. We therefore question the correctness of using the same retention rate across brands and product categories. However, many factors influence the determination of the retention rate, and ideas for further research within this area are suggested. To estimate the advertising half-life we used data from Adlab, a large British single-source database covering the years 19851989.

THE DATA
Single-source means that the data coming from one source, that is a person or household, are recorded in such a way that it is possible to identify the actions of the individual sources in a unique manner. Therefore, data collected and stored as single-source data make it possible to analyse relationships at the individual level. This allows for much more detailed analysis, compared to analyses at an aggregated level, based on market data. In this work the Adlab database has been used. Adlab was a diary-based single-source panel, which was set up by Central Independent Television in co-operation with Taylor Nelson in autumn 1985 and ran through spring 1990. The panel had about 1,000 respondents reporting, among other things, TV viewing, newspaper and magazine readership, and daily purchases in 48 product categories with numerous brands in each category2. Altogether, 18 product categories with a total of 89 brands from the Adlab database were chosen for the present study. The focus was on major brands with medium or heavy advertising and with

high numbers of purchases made. In the present analysis only advertising on TV is considered. To conduct the analysis it is necessary to have a certain minimum number of observations, and therefore criteria were set up to eliminate the brands that did not meet this requirement.

METHOD
To estimate the relationship between advertising and purchasing we have used a logistic regression (logit) model. In a logit model it is possible to include numerous explanatory variables. In our initial work we learned that by including not only advertising as an explanatory variable but also promotion at time of purchase, we improved the parameter estimates significantly. Therefore the analysis we discuss here is based on a logit model with two explanatory variables, namely the advertising pressure (expressed by an adstock calculation) and promotion information at time of purchase, here called offer. An adstock calculation has the strength that it expresses the advertising pressure today as a function of, in this case, advertising up to 28 days prior to the purchase date. By using an adstock calculation with a given retention rate, the advertising that took place several days ago is given a certain lower value compared to the advertising the consumer saw the day before the purchase was made. It seems reasonable to assume that the more recently the advertising is seen, the more influence it has on the purchase situation. We are working with single-source data and the analyses we are doing are therefore on daily data which in turn means that the advertising retention rate will be estimated as a daily retention rate. That is, we initially assumed the same advertising half-life of 23 days for all brands.

ESTIMATING THE RETENTION RATE


[Readers of a nervous disposition can ignore the equations, which are shown above Ed] The analyses that have been conducted so far3 have been based on an overall daily retention rate of 0.97. The equation used to calculate the adstock value is given in Equation 1, (See Exhibit 1) where r denotes the retention rate and OTS denotes the number of opportunities to see in the given period4. In order to determine the optimal retention rate we have estimated a logistic regression model (see Equation 25) and we have defined the criteria for the optimal retention rate as being the case where the largest significance of the corresponding parameter estimate was found. In Equation 2, b0, b1 and b2 denote the respective parameter estimates for the baseline, the advertising pressure and the offer-at-purchase information respectively. t gives the probability of purchase of the given brand at time t, given the advertising pressure and the offer-at-purchase. The retention rate is an expression of how much the consumer remembers from the past. When used in our case, in Equation 1, it denotes how much the consumer remembers of advertising that has taken place in previous periods. However, the retention rate can be a difficult figure to estimate, since its interpretation is somewhat complicated. Therefore, it is often useful to use the term advertising half-life. Advertising half-life is an expression for the number of periods, in this case days, before advertising has had half its total effect. The formula for calculating this factor is given by Equation 3, in which r denotes the retention rate4. The larger the retention rate used, the more memory the consumer has of previous advertising and the longer the half-life of advertising is. Some examples of this are given in Table 1.

TABLE 1: AD RETENTION AND HALF-LIFE: THE CONNECTION


Retention rate 0.50 0.70 0.90 0.95 0.99 0.999
Source: current study

Half-life (days) 1 2 7 14 69 693

As can be seen from Table 1, the half-life of advertising increases rapidly with even small changes in the retention rate. This is very important to bear in mind when estimating and determining which retention rate to use.

RESULTS
For the 89 brands studied we have tried to estimate the optimal retention rate by studying the significance of the corresponding parameter estimate. The optimal retention rate has been defined as the most significant one. We found a very large variation among the different brands; not even within the individual different product categories can the retention rate be said to be constant. In Table 2 the results are shown for each product category. The product category is given in column 1. The retention rate range (interval) is shown in column 2, while the average product-category retention rate is given in column 3. The category half-life, calculated on the basis of the average product-category retention rate, is shown in column 4.

TABLE 2: RANGE OF RETENTION RATES, OPTIMAL AVERAGE RETENTION AND AD HALF-LIFE


Product category Deodorants Shampoo Biscuits and wafers Dairy spreads Savoury spreads Gravy makers Dog food Instant coffee Toothpaste Washing powder Tub margarine Tea bags Cat food Breakfast cereals Daily retention rate 0.010-0.999 0.015-0.999 0.055-0.999 0.360-0.999 0.605-0.970 0.825-0.999 0.695-0.999 0.885-0.999 0.875-0.999 0.825-0.999 0.920-0.999 0.970-0.993 0.930-0.999 0.960-0.999 Average daily Average product product category half-life category (days) retention rate 0.657 1.7 0.772 2.7 0.777 0.786 0.788 0.938 0.944 0.962 0.962 0.966 0.979 0.981 0.982 0.986 2.8 2.9 2.9 10.8 12.0 17.9 17.9 20.0 32.7 36.1 38.2 49.2

Toilet rolls Chocolate bars Instant soup*

0.980-0.999 0.905-0.999 0.999

0.986 0.986 0.999

49.2 98.7 692.8

* few brands; Source: current study

An overall average daily retention rate of all brands is 0.91, with a corresponding half-life of 7.4 days. It is interesting to see how large the differences are that are found within the individual product categories. Between the different product categories the variation in the retention rate is 0.6570.999, while for individual products the rate goes as low as 0.01. The fact that there are these large variations, both within and across the product categories, indicates that this is a highly uncertain and approximate measure if you try to use the same overall retention rate. Therefore a more detailed and precise analysis would most likely be obtained by estimating an optimal retention rate per brand, rather than assuming an overall average retention rate. When the individual product cat-egories are studied it is found, in most cases, that there are brands that have high retention rates and brands with low retention rates, indicating that it is a sort of zero-sum game in each category. That is, there are product category winners and product category losers, in the sense that some brands have long advertising half-lives and some brands have short advertising half-lives.

SUMMARY AND IMPLICATIONS FOR FURTHER RESEARCH


In this analysis we have estimated the optimal retention rate for 89 brands. We found that there was a very large variation between the different product categories. We also found this variation within the individual product categories, which indicates that there is no overall retention rate but this is rather a brand-specific variable and should therefore be estimated in each different case. However, one of the questions that remains is whether or not it makes sense to have (or try to use in modelling or other applications) a retention rate less than 0.5, where the advertising half-life is 1 day; or a retention rate greater than 0.99, where the advertising half-life is 69 days. It is interesting to compare our results with some of the few published advertising half-life results from other researchers6. The distribution, as published, of Goerlichs findings is shown in Exhibit 2. In comparison, our findings are illustrated in Exhibit 3. We have divided Goerlichs findings and our findings into the same intervals. In both sets of data the u shaped form appears, even though the data in Exhibit 3 are more skewed toward the right than the data from Goerlich. It would be interesting to study the extent to which this is an effect of the ten years difference in time between the collection of the data, or the differences in modelling, or if it is in the differences in the way the data are collected. Goerlichs data are stated to be based upon aggregated econometric analysis of sales data, while ours are based on single-source data. Another factor that can influence the estimation of the retention rates is the consumers degree of loyalty towards brands. The loyal consumer would probably have bought the particular brand whether or not advertising for the brand was seen prior to the purchase. We are currently working on expanding the model, with an explanatory variable that describes the individual consumers loyalty towards the given brand studied. In future work we are interested in studying the retention rate in greater detail that is, what

happens to the optimal retention rate if more variables are included in the logit model. What if we decide to use a different model how will that influence the determination of the optimal retention rate? What other criteria can be stated to determine when the optimal retention rate is found?

REFERENCES
1. LY Hansen and F Hansen: Advertising and promotion effectiveness learnings from a five-year study. 2. 3. 4. 5. 6.
Research paper No. 10, Forum for Advertising Research, Department of Marketing, Copenhagen Business School, June 2001. See in C McDonald: From frequency to continuity is it a new dawn? JOAR, Vol. 37, No. 4, 1997. F Hansen, LY Hansen and L Grnholdt: Modelling purchases as a function of advertising and promotion. IJOA, Vol. 21, No. 1, 2002. S Broadbent: One way TV advertisements work. Journal of the Market Research Society, Vol. 21, No. 3, 1979. J Neter, W Wasserman and MH Kutner: Applied Linear Regression Models. Irwin USA, 1989. B Goerlich: A consumers guide to marketing mix models. Admap, December 2001.

World Advertising Research Center

http://www.warc.com

NOTES & EXHIBITS

Flemming Hansen
Flemming Hansen is professor at the Copenhagen Business School, and head of Forum for Advertising Research. He is also an advisor to the Danish Circulation Board, to the Ministry of Communication and to a committee set up by the Danish Prime Minister.

Lotte Yssing Hansen


Lotte Yssing Hansen is a research assistant with Forum for Advertising Research at the department of marketing, CBS, and is currently working on projects regarding modelling effects of advertising using single-source data

EXHIBIT 1: EQUATIONS

Equation 1
Adstock = (1-r) ((r OTSt-1) + (r2 OTSt-2) + ... + (r28 OTSt-28))

Equation 2

Equation 3

EXHIBIT 2: DISTRIBUTION OF HALF-LIVES (GOERLICH)

Source: regrouped from Goerlich6

EXHIBIT 3: DISTRIBUTION OF HALF-LIVES (FMCG BRANDS)

Source: current study

You might also like