Econometric Measurement of the Duration of Advertising Effect on Sales: A Comment

Abstract

In a recent article, Darral G. Clarke (1976) examined relevant econometric literature on duration of cumulative effect of advertising expenditures on sales. He concluded that the duration intervals derived from annual models are too long (p. 355) and, moreover, that these intervals are longer when estimates are made from annual rather than from quarterly or bimonthly data. I do not question first conclusion, but have some serious doubts about way in which these estimates were obtained. Although Clarke discusses many studies besides mine (1971), I examine his conclusions largely on basis of that article simply on grounds of greater familiarity. However, criticism clearly relates to other studies as well. An alternative to Clarke's model based on my 1971 study is presented to estimate duration of advertising effects. The two models then are compared in an effort to identify possible bias in Clarke's model. Finally a question is raised about effect of using discrete unit (rather than continuous) time intervals on lag in impact of advertising. Omitted variables constitute a nagging problem of econometric research in general, and of empirical studies of advertising effects in particular. Advertising expenditures by competitors can have, and presumably do have, a significant impact on effectiveness of a firm's own advertising efforts. Hence, a firm's sales may depend not only on its own advertising expenditures, but on those of its competitors as well. In fact existence of multiple advertising effects seems to be rule. In general, sales appear to be a function of advertising effects of other firms in same industry and in competing industries, and even of a firm's own advertising program in previous periods. One method of capturing dynamics of such situations is to substitute a model with (at least) two distributed lagged variables for single-variable model used by Clarke.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 12 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.