Editor's note: Michael Wolfe is an independent marketing consultant. The work represented in this article was done while he was director of product management and analysis at Arbitron/SAMI.

While marketing researchers have long sought ways of quantifying the sales effects of advertising and other marketing stimuli, only recently have such ventures made their way out of the abstract discourses of academia to the more pragmatic world of corporate marketing. The purpose of this article is to review how the arrival of single-source marketing information over the past decade has led to some practical ways of modeling and understanding the impact of advertising and promotion on product sales.

To start with, I will illustrate, from a real-world example, some ways that marketers might want to view such information in order to understand how and why their brands perform differently in different geographic markets. I will also discuss the implications this might have on marketing strategies and how they're carried out.

Next, I will briefly review some of the more recent and in-depth research of other researchers to illustrate varying approaches to the issue of modeling the sales impact of marketing-mix variables using single-source databases. Finally, I will conclude by discussing the future of marketing-mix modeling efforts and the impact that these efforts are likely to have on more traditional marketing and advertising research.


Exhibit 1A

Similar patterns and levels of media advertising, trade and consumer promotion do not always yield the same results. The high CDI market, Seattle, was more responsive to all components of the marketing mix than was the low CDI market, Chicago. Single- source data now permits a much more accurate means of tracking and measuring the impact of marketing activities at the local level, including the development of models for measuring the effective response of different markets to advertising and promotion. Common Sense Oat Bran Waffle information from Arbitron's SAMI Scanner data and BAR Media Measurements.


Exhibit 1B


While there is probably no universally agreed upon definition of single-source marketing information (SSMI), there are some unique attributes of these databases that have enabled marketing researchers to develop more robust and promising models relating the impact of marketing-mix variables-such as advertising, price, trade and consumer promotion-on product sales and market share:

1. By definition, SSMI reports product sales and all relevant "marketing stimuli" related to advertising, price, trade and consumer promotion.

2. All measures derive from a single and uniform origin and are also reported on the same time and geographic dimensions.

3. Many of the measures derive from "electronic" means (e.g., scanning), enabling them to be more detailed, and often more precise than measuring systems requiring human intervention and accounting.

Exhibits 1A and 1B graphically illustrate how SSMI could be viewed across different geographic markets. Here, we see 39 weeks of sales (share), trade promotion, consumer promotion and advertising data (GRPs) plotted simultaneously for a brand of frozen waffles. To illustrate levels of advertising and trade promotion, the GRP and scanner-derived data on trade promotion sales to consumers were "normalized" or transformed into indexed values and plotted on the same axis. The weekly data on market-share was then plotted on a separate axis, while data on consumer promotion (coupons) were simply overlaid as discrete events. All of these were integrated on a single chart to illustrate the patterns and interrelationships of "causal" marketing stimuli on product share.

For reference, the data plotted show events occurring very soon after the brand was introduced into each market. For the most part, media weights and the pulsing manner of their delivery were found to be very similar. To the extent that these markets responded differently, however, we can begin to see a very interesting case study unfolding.

The main point of this illustration is that, despite the nearly homogeneous patterns of marketing stimuli, the sales response from each market differed considerably. In the highly developed market, Seattle, we find share gaining momentum, especially after week 21, and reaching higher sustained levels than the lesser developed market (i.e.. category per-household sales are lower) of Chicago.

Overall, a brand manager would be very interested in tracking his/her brand's performance in this fashion. Seeing that share is not building in Chicago, alterations in the frequency, mix and/or weight of advertising and promotion can be ordered up. Specifically, such information is likely to lead a marketer to shift the emphasis from developing and executing strategies at the national level to a more localized focus.


Exhibit 2


To the market researcher with a bent for modeling, more in-depth analysis can separately reveal a brand's sensitivity to promotion and advertising on a market-by-market basis. To illustrate, on Exhibit 2 a statistical technique was used to eliminate the effects of promotion from the brand's share. That process yielded a measure called "base share." The plot of base share and advertising GRPs in Exhibit 2 illustrates several points:

1. Short-term advertising effects on sales can be observed, as illustrated by the sine wave-like pattern in "base share" similar to the pulsating pattern of the advertising.

2. Sales effects of advertising are usually not immediate; although the actual length of the "lag" effect on sales might not be uniform, as seen here.

3. The "local market" is the relevant dimension for evaluating advertising and other marketing-mix variables.1

SSMI and marketing-mix models: some recent developments

Noted efforts and advances have recently been made by a number of researchers who have used single-source data to come up with their own varieties of marketing-mix models. Like other areas of advertising research, approaches to the area of marketing-mix models differ. To some, there are differences in the type of SSMI used-some use consumer panel data, while others employ store level scanner information. Still, others differ according to the level of aggregation where the data is analyzed. Here, some look at disaggregate store level information, while others look at market level data. While this is not intended to be an exhaustive or technical treatment of all recent efforts to use SSMI to develop marketing-mix models, I will briefly discuss some recent and interesting advances and applications.

One approach is represented through the efforts of Edward Dittus and his consulting company, Marketing Media Assessment of Westport, Connecticut. Dittus begins his analysis by collecting integrated and weekly scanner sales, promotion, and local media advertising data at the market level. This model basically uses a simultaneous equation solution to ferret out the effect of advertising and promotion.

Dittus has used his modeling approach with a number of major packaged goods firms and directs his efforts specifically at helping them improve the productivity of their marketing and media plans. Dittus' model uses single-source data to quantify the sales effects of various media decisions. It is designed to specifically help companies determine the best strategies for answering critical marketing-mix questions.

Some of these include:

1) the relative sales contributions of advertising, promotion and pricing, and how increases or decreases would affect brand volume,

2) the timing constraints of advertising, including carryover effects and the optimal length of advertising hiatus periods and

3) the best advertising mix and daypart strategies.2

Another recent advance in marketing-mix models has been developed by Dennis Bender of the A.C. Nielsen Company.3 Bender's approach uses "scanner sales" data at the individual store level. Rather than filtering out promotion at the market level, Bender focuses on store level data and employs a simultaneous equation solution known as "vector transfer function regression." Overall, Bender's approach focuses on a fairly large array of marketing-mix independent variables, ranging from trade and consumer promotion to pricing and media weight, frequency, mix and quality variables. By using store-based disaggregate data, Bender contends that his model avoids biases sometimes encountered in accurately measuring the separate effects of promotion and advertising on sales.

In other words, when different marketing stimuli tend to move together, it is difficult to accurately separate or isolate the effects of each. In addition, "aggregation biases" also occur even at the market level which mask and/or distort advertising response. After developing the model using store level scanner data, Bender also uses household panel data as a supplement in order to focus on how household purchase history for a product and/or category effects responses to these marketing-mix variables.

Finally, another interesting approach is represented by Drs. Fred Zufryden and James Pedrick of USC. They look at household purchase data along with TV metered viewing behavior from these same households using panel data from NPD/Nielsen. Using a multinomial logit model, Zufryden and Pedrick isolate individual household responses to a host of marketing-mix stimuli. Given the direct linkage to metered TV viewing, individual responses to different advertising reach and frequency levels can be evaluated.4 Overall, Zufryden and Pedrick's approach models "the impact of advertising media plans and other marketing variables on performance measures that relate to brand choice probabilities and market-level consumer purchase dynamics."5

Single-source data and the future of marketing-mix models

When SSMI came on the scene in the early 80s, attention was first focused on developing models for evaluating some of the more obvious and short-term marketing-mix variables such as trade promotion. Now, thanks to the research of those cited here and others, efforts have expanded to looking at the broadest array of marketing-mix variables such as consumer promotion, media advertising and all of these factors combined.

While recent approaches to marketing-mix models differ, all illustrate how single-source marketing information has enabled researchers to quantify the impact of promotion in advertising in ways not deemed possible or practical ten years ago.

Because of more precision and uniformity in the measuring of marketing stimuli and sales simultaneously, effects of the different marketing "levers" can now be isolated and quantified down to the market and household level. The relative ease with which this information can be brought together, along with the greater computing power now available for using sophisticated modeling techniques, has permitted marketing-mix models to actually be used by the corporate marketing world. The result is that some marketing managers are actually using these models to aid in their planning and decision making.

The practical implications of these efforts are many. Media planners and marketing managers now have the tools for improving the efficiency of their total marketing plans and budgets, enabling them to better target their media and promotion dollars. The old cliché about "not knowing which half of the advertising budget is actually working for you" is being challenged. As manufacturers demand more accountability for how all marketing funds are spent, these models are likely to become more prominent.

Notes
1. Schroer, James C., "Ad Spending: Growing Market Share, "Harvard Business Review, Jan./ Feb. 1990, pp. 44-48

2. Dittus, Edward & Kopp, Marty, "Advertising Accountability in the 1990s: Moving from Guesswork and Gut Feelings," Advertising Research Foundation Media Research Workshop, May 8, 1990.

3. Bender, J. Dennis, "Measuring the Advertising-Sales Relationship: Meters, Measurement of Advertising Audiences, and New Analytics, D Advertising Research Foundation Behavioral Research and Single-Source Data Workshop, Jun. 26-27, 1990.

4. Pedrick, James H., & Zufryden, Fred S., "Evaluating the Impact of Advertising Media Plans: A Model of Consumer Purchase Dynamics Using Single-Source Data," Marketing Science (TIMS/ORSA), Winter, 1991 (to be published).

5. Ibid, p. 1.