1.Introduction

The diffusion of a new product takes time, typically years if not decades. Even truly innovative products that are superior to existing products will usually take years before they are widely adopted. The question is why it takes a considerable length of time before a new product saturates its market. When a new product is introduced into the market, little is typically known about its attributes by potential adopters. Furthermore, it is often the case that a large proportion of potential adopters is unaware of the introduction of the good. The dissemination of information about a new product can be slow. Thus, the efficiency of communication channels can affect the rate of new product diffusion [15, 16].

Even if the dissemination of information is rapid, product diffusion can be slowed by potential adopters’ postponment of purchase in anticipation of falling prices. It is well known, for example, that prices of many consumer electronic products, like VCRs and PCs, are falling over time. The declining price of a new product affects the rate of product adoption in two ways. It directly affects the diffusion process by enabling more consumers to afford these products while the anticipation of falling prices affects the diffusion process indirectly by influencing the timing of consumers’ purchases.

For new products that are durable, it has been Olshavsky [15] showed the statistical significance of the time coefficient in a regression of adoption rates on the time of introduction. He argued that possible explanations for this phenomenon are improvements in channels of communication, increases in marketing sophistication, and increases in consumer affluence. Qualls et al. [16] also document evidence that the overall length of a product's life cycle has become shorter for products more recently developed. As it is difficult to identify the entire product life cycle for recently introduced products, particularly the lengths of the maturity and decline stages, Qualls et al. [16] test whether the introduction and growth stages are shortening.

While the marketing literature has been productive in documenting such acceleration, it has done little to explain it. Specifically, those empirical studies in the marketing literature have not explained the interactions that occur among the various determinants of diffusion processes. The key feature of our approach is the development of an interactive empirical model of sales and price and the use of this model to identify the factors driving the acceleration of the rate of adoption of new products.

We begin by constructing an original data set of new product diffusion which includes annual data on retail sales, price, and the number of manufacturers for 61 products. The data set is unique in several respects : (1) an extensive coverage of consumer appliances introduced during the 20th century in the U.S.; (2) the use of retail sales data; and (3) an extensive coverage of changes in the number of manufacturers of each good over time. With this data set, we provide additional documentation of the empirical regularity that the speed of adoption has accelerated during the twentieth century. Furthermore, our interactive model provides a much better fit of the data than do previous non-interactive models. The main source of this acceleration is found to be a more rapidly declining price path for products introduced later inthe twentieth century. This acceleration of price declines is found to be strongly associated with two product characteristics- use of semiconductors and network externalities- and changes in the distribution of income.

2.Determinants of the Speed of Adoption

2.1.Simultaneous Determination of the Speed of Adoption and Price Change Rate

Products with faster diffusion rates seem to have faster price change rates. <Figure 1> shows that the speed of adoption is associated with the speed of price decline. As is apparent in the scatter diagram, SPD (speed of adoption) and PCH (the rate of price change) are negatively correlated.

The source of this association could be due to factors including the learning and intertemporal price discrimination. Consider the diffusion process of a new consumer durable good with learning-by-doing. Faster declines in the price of a new product make the product affordable to more consumers at a faster pace. From the viewpoint of a seller, the faster diffusion of the product makes the seller cut prices faster because the seller has to reach lower valuation consumers as the market for the new product is saturated more quickly. By selling to higher and lower valuation consumers at different points in time the producer is practicing intertemporal price discrimination. If the product is adopted more quickly, the producer accumulates production experience more quickly, and, because of learning-by-doing, this implies a faster decline in production costs. In turn, this may result in faster price declines. Thus, the effects of accelerating price change and accelerating new product adoption mutually reinforce each other. The two factors, intertemporal price discrimination and learning-by-doing, thus simultaneously determine the speed of product adoption and the rate of price decline.

Given the hypothesis that the rate of price change and the speed of adoption of a new product are simultaneously determined, we have to estimate a simultaneous equation system in order to take account of the interaction effects between the two variables. As previous empirical studies have not taken account of the interaction effects between these various factors due to their single equation specifications, the implementation of a simultaneous equation specification is one of the main contributions of this research.

2.2.Exogenous Factors

Apart from the simultaneous price-diffusion relationship (to be precise, the price change rate-speed of adoption relationship), we also consider several exogenous factors that contribute to the acceleration of product adoption and price change rate. These exogenous factors can directly affect the product adoption rate or the rate of price decline.

2.2.1.Trends

There are various exogenous factors that affect the speed of adoption. First of all, there have been improvements in various information channels, which enable sellers to respond quickly to changes in markets. The improvements in information channels also contribute to the faster diffusion of information on new products. For example, computerization of various office and factory activities have accelerated business activities. Second, innovations in financial institutions such as the introduction of electronic funds transfer systems and credit cards may have loosened consumers’ liquidity constraints and accelerated transactions. Third, changes in the distribution of income in the economy can affect pricing behavior. For example, if the middle class has a larger share of total income, one would anticipate larger price changes. If national income is concentrated in a small proportion of the total population, continuous price declines of a new product would probably have minimal effects on the diffusion of the product. Fourth, the increasing competitiveness of imported products might have contributed to faster price declines of new products. Furthermore, the fact that new consumer electronics products are mostly produced by foreign manufacturers, but distributed by domestic sellers, may have intensified competition. This, in turn, may have accelerated price changes.

2.2.2.Product-Specific Factors

Other than the factors with trends, there are several factors intrinsic to each product, which probably determine the rate of price change. We characterize those factors as follows. First of all, there are tremendous variations in price level relative to income level at the time of introduction of a new product. A higher priced good in absolute value tends to show faster price change rate over time probably because there is greater scope for learning-by-doing and price discrimination over time. Second, network externalities arise in the diffusion of some type of products. As the network grows, complementary products can often be supplied at a lower price or with greater diversity if there are increasing returns to scale. This kind of benefit may have contributed to the faster speed of adoption for many consumer electronics products such as VCRs and PCs. Third, there are technological differences, in that electronic components are an increasingly important part of modern goods. Because technological advances in these components have occurred so rapidly it is possible for them to have a significant impact on the production costs of new goods, hence on their price and subsequent diffusion. For example, in the semiconductor industry, production costs have declined so rapidly that it has been difficult to maintain a stable pricing policy. This may be a factor in explaining why consumer electronics products have progressed through the product life cycle so quickly.

2.2.3.Market Structure

The rate of price change can be affected by changes in the market structure. It is believed that increases in the number of firms can accelerate the rate of price change. In general, prices are likely to change more quickly when the number of firms change faster.¹⁾

3.The Empirical Model

3.1.Data

Annual data on retail sales, price, and number of manufacturers for sixty-one consumer durable products marketed in the U.S. were collected by the author. Retail sales for major appliances, housewares, and home electronic products as reported in various summary and special issues of Merchandising Week, Merchandising, and Dealerscope Merchandising were assembled. Price data for each product are adjusted using the 1967 Consumer Price Index. Annual data on the number of producers were obtained from Thomas’ Register of American Manufacturers. Data on other important exogenous variables determining the diffusion processes of new products were compiled from various issues of Annual Statistical Abstracts of the United States.

3.2.Measurement

We fit the historical data on the diffusion of the various new durable products to the logistic function. The logistic function characterizes the typical S-shaped diffusion curve. This approach was taken by Mansfield [13] and Olshavsky [15] in measuring the speed of adoption of a new product.

Data availability varies quite significantly between different products. Therefore, we focus on the introduction and growth stages in terms of unit shipments of each product while controlling for the number of households over time. The end of the growth stage is the year when the number of household-controlled unit shipments reaches its maximum. In short, we measure the “steepness” of the S-shaped diffusion curve as the rate of adoption. As we never know what the ceiling or equilibrium value would be for each different product, we substitute the cumulative sales at the peak of unit shipment for the ceiling. Thus, we analyze the time period from the introduction of a new product to the year when unit shipment reaches its global peak, while controlling for changes in number of households. The estimated speed of adoption²⁾ is in <Table 1>, where YEAR represents “year of introduction”, i.e. the first year that a product was introduced.

In order to test whether a product’s speed of adoption is positively related to its year of introduction, we run a regression of the speed of adoption on the year of introduction. The estimated linear equation³⁾ is

Φ ⌢ i = − 14765 + 7.767 YEAR , R ¯ 2 = 0.583 − 8.97 9.22

(Figures in parentheses are t-ratios.)

The above result strongly indicates that the speed of adoption has accelerated over time and is consistent with previous empirical work.

3.3.Proxies for Exogenous Variables

A number of exogenous factors are likely to be important determinants of the new product diffusion process : 1) Improvements in communications technology may have accelerated the speed of product diffusion. 2) Innovations in financial institutions may have loosened consumers’ liquidity constraints and may have accelerated transactions. 3) Changes in income distribution, such as an increase in the share of total income by the middle class, may have accelerated price changes. 4) An increase in the competitiveness of imported products may have intensified price competition. In this section, we discuss possible proxies for these exogenous factors affecting price changes and diffusion processes.

One explanation for the acceleration of the rate of product adoption is improvements in marketing techniques and communication channels. The diffusion of information about a new product through communication channels as well as pricing, advertising, and promotion strategies on the side of sellers can critically affect the degree of the diffusion of a new product. There have been remarkable improvements in marketing techniques; the advent of electronic transactions and the availability of extensive and detailed data bases enable firms to detect and respond to competitors' pricing and promotion actions. Since the main avenue for these advances is advertising on television, we use TV saturation rates as a proxy for the level of information channels.

A second explanatory factor which may be related to the time trend in the diffusion of a new product is the loosening of liquidity constraints resulting from such financial innovations as the credit card. The income velocity of money is used as a proxy for this factor. The velocity of money is derived from the quantity of money equation : V = PY/M where V is the income velocity of money, P is price level, and Y is real income. We observe that V has increased over time. This suggests that people now hold less money relative to their incomes over time, i.e. less money is used to support a greater volume of transactions.

During the period of interest there have also been significant changes in the distribution of income. The share of income earned by people in the middle portion of the distribution has increased dramatically. Changes in income distribution may have affected price paths which in turn have influenced the speed of adoption. In this research, the thickness of the middle class is more meaningful in the growth of a new product than measures of income inequality. However, we use the percentage of aggregate income received by the top 5% of the population because it is available for all the years covered by our data set.

The phenomenon that we seek to explain may also be attributable to a technological factor noted by Gort and Klepper [8] : “The interval required for successful imitation has systematically declined over time.” Increases in the competitiveness of imported products reflect this technological factor.

3.4.Empirical Model

As mentioned in section II-a, the two factors, intertemporal price discrimination and learning-by-doing, result in the simultaneous determination of speed of adoption and price change rate. Furthermore, the information level and velocity of money directly affect the rate of adoption, while income distribution and the competitiveness of imported products affect the rate of price change. The main empirical model consists of the following simultaneous equation system, in which both of the equations are identified.⁴⁾Table .2

In equation (1), the speed of adoption is regressed on the year of commercial introduction, INF, v, and the price change variable. The price change rate captures strategic interactions among different firms as well as between firms and consumers. The variation in the rate of price change, across different products, can be attributed to the following factors: (1) differences in the distributions of consumers’ valuations; (2) differences in the potential for learning-by-doing in production; (3) differences in the strategic interactions among firms.

In equation (2), our objective is to capture the following interactive relationships : between learning by doing and the speed of diffusion; between competition among firms and diffusion; and between intertemporal price discrimination and diffusion through price dynamics. Price change rate is determined by the speed of adoption which is closely related to how fast cumulative sales increases. The NCH variable reflects the effect of changes in competition. A scatter diagram shows that there is a negative relationship between NCH and PCH. Changes in the number of firms are considered to affect the rate of price change, not vice versa. For instance, an increase in the number of producers increases the intensity of competition among firms which then accelerates the decline of price. Therefore, a change in the number of firms affects the speed of adoption indirectly through changes in price.

The variable SPD in the PCH equation captures the combined effects of the learning curve and intertemporal price discrimination, which come from (a) the learning curve literature [3, 6, 11] and (b) the durable good literature [17, 4]. The estimated coefficient of NCH reflects the effect of changes in competition on the rate of price change, while the estimated coefficient of N₀ reflects the impact of the initial market condition on pricing strategy. The YEAR variable controls for any time trend.

Previous theories provide us with expectations of the signs of the coefficients of each variable. Putting together various literatures, we summarize expected signs of variables in <Table 3>.

If the factors elaborated so far explain the acceleration of the speed of adoption of a new product, the YEAR variable should be insignificant. One anticipates that the inclusion of a proxy for the level of communicative efficiency would reduce the effect of the year of introduction significantly, as YEAR and INF move together. Improvements in communicative efficiency, as reflected in TV saturation rates, could accelerate the speed of adoption. Thus, INF and SPD may also have a positive relationship. Increases in the speed of transactions and the relaxation of liquidity constraints may also accelerate the speed of adoption. Thus, V and SPD could be expected to have a positive relationship. The effect of PCH on SPD is expected to be negative as faster price declines enable more consumers to purchase a new product by relaxing the capital constraint.

While the variable NCH in the PCH equation should be negative, the variable N₀ in PCH equation is expected to be insignificant because the inclusion of the effects of changes in market structures would be considered to be more important for change (as opposed to level) of price.

We believe that changes in income distribution may influence firms' price discrimination strategies. A decrease in the share of aggregate real income received by the top 5% of the population may cause prices to decline more rapidly as firms seek to sell to the middle class. Therefore, the percent of income received by top 5% of the population is expected to be positively related to the rate of price change. The larger the share of total income by the middle class, the larger price changes would be. Thus, IDS and PCH are anticipated to have a positive relationship. Increases in the competitiveness of imported products or increases in the ease of imitation may also accelerate price declines. Thus, it is likely that IMP and PCH have a negative relationship.

The estimated coefficient of SPD in PCH equation should be negative because the larger is learning in production and the more extensive are intertemporal price discrimination practices, the larger one expects the rate of price decrease to be.

3.5.Weaknesses and Limitations

Recognition of some weaknesses in this analysis is in order. First of all, many of the exogenous variables are highly correlated. This is partly because proxies are used for certain economic variables and many of these proxies are affected by the time trend. One may also question whether the proxies are direct and relevant measures of the economic factors that they are used to represent.

Second, the number of firms is partly endogenous. This is because in the long run firms’ entry decisions depend on how profitable it would be to operate in a particular industry. Therefore, potential entrants take account of the expected profit levels in future periods, as well as the profits in the current period. However, we regard price and output as more important endogenous variables than the number of firms, because the number of firms does not change within a short period of time.

Third, we are dealing with only one of several alternative measures of the rate of adoption. One could argue that the saturation rate, rather than cumulative output, should be used in the estimation of the rate of adoption, especially for consumer durable goods. The saturation rate tells us the proportion of consumers (or households) that have adopted a certain new product, while cumulative output tells how many units have been sold. However, because of data availability, most of the previous studies have also used output data.

4.Empirical Results

Using the three-stage least squares estimation method (3SLS),⁵⁾ we have obtained the following regression results for equations (1) and (2). Figures in parentheses are t-ratios.

The results show that the factors determining PCH indirectly change SPD. Besides the indirect factors, direct changes in trend variables also accelerate SPD. The results also document the strong and consistent significance of the income distribution. The price of a new product declines more quickly when the middle class possess a larger share of income. The velocity of money was also strongly and consistently significant, however it should be noted that the velocity of money is itself believed to be influenced by the level of communicative efficiency. Improvements in communication channels affect the velocity of money, which in turn influences the speed of adoption of a new product. Thus, INF is also found to be an important factor in the determination of the speed of adoption of a new product, but once V is included, the coefficient on INF decreases dramatically. As was noted in section III, many proxy variables are affected by the time trend. The IMP variable turns out to be insignificant. This may be because D (the dummy variable for electronic and/or video/audio products) has already explained the competitiveness of imported products over the years. We note that P₀ and D are negatively related to PCH. This implies that electronic products and products with higher initial prices relative to income level tend to have faster decreasing prices. The estimated coefficient of YEAR is insignificant; P₀, D, IDS, V , and INF explain both price declines and speed of diffusion very well.

SPD = 0.038 YEAR 1.34 + 0.749 INF 1.33 + 51.42 V 2.38 − 15.83 PCH − 3.96 R 2 ¯ = 0.758 PCH = − 0 . 394 P 0 − 2.93 − 3 . 731D − 2.65 + 0 . 016 N 0 1.52 − 0 . 024 NCH − 0.34 + 0 . 155IDD 2.65 − 0 . 165IMP − 0.30 − 0 . 0188SPD 2.85 R 2 ¯ = 0.728

The dummy variable, D (consumer electronic products), can also be regarded as the aggregation of a number of other variables with economic implications. Specifically, the products within the group of “consumer electronic products” can be broken down into two subgroups, which are not mutually exclusive. The first is a product group using semiconductors, which represents a technological distinction. Products with and without semiconductors are generally differentiated by the following factors : ease of imitation, the number of potential competitors, and differences in learning curves. The second group is composed of products with network externalities, which is a feature of the user side. The dummy variables for each group of products are denoted TECH and NE, respectively. <Table 1> shows the classification of products in the data set.

The following results show the significance of both network externalities and the technological characteristics attributable to the use of semiconductors in production. The effect of network externalities on the rate of price change is notably stronger than that of the technological factor. This is probably because most of the products with higher proportions of semiconductors in total production costs are products with network externalities.

SPD = 0.029 YEAR 0.98 + 0.769 INF 1.34 + 60.70 V 2.73 − 13.36 PCH − 3.38 R 2 ¯ = 0.755 PCH = − 0 . 394 P 0 − 2.61 − 2 . 885NE − 2.35 − 3.210 TECH − 1.77 − 0 . 017 N 0 1.56 − 0.030 NCH − 0.42 + 0 . 117IDS 1.87 − 0 . 125IMP 0.23 − 0 . 018SPD − 3.13 R 2 ¯ = 0.754

5.Conclusions

This study is the first attempt to use a formal empirical model to investigate the observation that the speed of new product adoption has accelerated over time. Studies of new product diffusion have been done in both the marketing literature and the economics literature. The former is mainly composed of empirical work, while the latter is mainly theoretical. In the marketing literature, the above phenomenon has been frequently documented, but there has been no attempt to understand why it is true. On the other hand, in the economics literature, several strands of research have tried to explain the factors affecting new product diffusion, but there has been relatively little interaction between the different areas. Moreover, empirical studies have been particularly rare in the economics literature. This paper remedies these deficiencies by bringing together the various factors determining diffusion and examining their interactions, and by stylizing the process of product diffusion as an interactive system.

The empirical results show that the main source of acceleration are faster declines in prices. Faster price declines make the product affordable to more consumers within a given period of time. Based on theories of intertemporal price discrimination and learning-by-doing, the association between the speed of adoption and the speed of price decline was explained.

Faster price declines are attributed to several product characteristics as well as changes in income distribution. Above all, the introduction of consumer electronic products in more recent years can be regarded as the most important factor in accelerating price declines. Consumer electronic products are technologically different from non-electronic goods, in that semiconductors are important components. As the price of semiconductors has dropped rapidly, the falling production costs can be rapidly incorporated to the price of consumer electronic goods. Furthermore, most of the recently introduced consumer electronic products have network externalities, and many products with network externalities require complementary products. A complementary product becomes more readily or cheaply available as more people have the main product.

There seem to be a number of variables that explain the changing pace of product diffusion processes. Largely, these variables are related to improvements in marketing techniques and communication channels. In the marketing literature, these factors have been considered as the main source of acceleration of speed of adoption. These factors also turned out to be significant in this study. However, one major difference between previous studies and this study is that the former focuses only on the factors that operate directly on the speed of adoption, while this study incorporated factors that work through price changes as well as the factors that work directly on the speed of adoption.

One of the weaknesses of this analysis is that many of the exogenous variables are highly correlated, as many of the proxies are strongly related with the time trend. Thus, it is likely that certain variables that appeared to be insignificant are in fact significant. This study also treats the number of firms as exogenous. In future work, the interaction effects among the speed of entry of new firms and the speed of price change and the speed of adoption should be further examined.