Finally we consider the result of fitting the same basic models to data on founding rates of organizations that published local newspapers in the San Francisco Bay area from 1840 to 1975. This organizational form provides interesting contrasts with the two just considered. As with labor unions, records on newspaper firms go back well over a hundred years. And like labor unions, newspapers have been heavily involved in political events and movements (Carroll 1987). But like semiconductor manufacturing firms, newspaper publishing organizations are business organizations that produce a material product in competitive markets using a technology that has changed frequently.
Carroll (1985, 1987) reconstructed the life histories of all newspapers published in seven urban areas in the United States. The analysis reported here uses data from one of these areas, the San Francisco Bay area (San Francisco, Oakland, and San Jose), by far the largest of the areas studied.51 Using a variety of historical sources, as noted in Chapter 7, Carroll recorded dates of founding (initiation of publication), mortality (cessation of publication) or right-censoring for a total of 2,170 newspaper organizations. As in the other two studies, this research considers the entire history of the population.
Figure 9.11 plots the fluctuations in numbers in this population over the period 1840-1975. Density grew more or less linearly over the period up to about 1920. At this point, the number of newspaper publishers began a gradual decline, with a sharp drop in the 1940s and a temporary rebound during the 1950s. It is worth noting that growth in density of newspaper organizations does not track the growth of population and commercial activity in the region. Density peaked while these resource dimensions were just beginning explosive growth; density leveled off and then declined during a time when newspaper circulation and advertising revenues showed continued strong growth. Thus, this pattern of growth and decline in density does not appear to reflect fluctuations in availability of resources or obsolescence of social and material technologies.
Figure 9.11 Density of newspaper publishing firms by year (San Francisco Bay area)
Figure 9.12 shows yearly fluctuations in the number of foundings in this population. Notice that the peak period of foundings was the 1890s, and that the number of foundings dropped sharply during the period of high density. This comparison suggests that the relationship between density and the founding rate may have been non-monotonic in this population.
Estimates of models for founding rates are obtained by treating the process as a point process and using the intervals between events to estimate the rate with partial likelihood estimators, which is one of the methods used earlier in analyzing founding rates of unions. Here the observa-tions are the durations between the 2,170 foundings. All but the last of these spells ends with an event; the last is censored on the right at the end of 1975. Carroll and Huo (1986) previously analyzed yearly counts of foundings in this population without considering effects of density. They used models in which the count in any year was regressed on the count for the previous year and the square of the previous year’s count of foundings. They found that lagged foundings tended to have the non-monotonic pattern of effects found earlier in studies of the ecology of Argentinean and Irish newspaper populations (Delacroix and Carroll 1983). They also found that the number of foundings was affected significantly by several environmental covariates and that counts of foundings rose significantly during years of political turmoil (mainly anti-Chinese riots, general strikes, and strike-related vio- lence). So we include in our analyses a variable that distinguishes years of such political turmoil.
Figure 9.12 Foundings of newspaper publishing firms by year (San Fran- cisco Bay area)
Table 9.8 reports PL estimates of models of density dependence in founding rates in the San Francisco Bay area newspaper publisher popula- tion. The first column contains estimates of our basic model of density dependence (9.4a). The model also contains the effect of political turbu- lence. Again we find that the effect of density was non-monotonic. The estimated first-order effect of density is positive, and the second-order effect is negative. Both point estimates differ significantly from zero at the .01 level. In addition, political turbulence raised the rate significantly, as Carroll and Huo (1986) found with a different model and estimator. According to the estimate in column 1, the founding rate of newspaper publishing firms was 40 percent higher in years of marked political turbulence.
The second column contains estimates of a model that adds the effect of foundings in the prior year, as in equation (9.6a). The effects of density still have the predicted signs, and both point estimates still differ significantly from zero. Prior foundings also have the predicted effect on the founding rate: the first-order effect is positive and the second-order effect is negative. Both of these point estimates also differ significantly from zero.
In exploring alternative specifications, we found that a better fit can be obtained by using a log-quadratic effect of density.52 In the log-quadratic model the log of the founding rate is a linear function of density and its square. This model has essentially the same qualitative interpretation as the one discussed to this point. Column 3 in Table 9.8 reports estimates of this model. Again the relationship between the founding rate and density is non- monotonic: the first-order effect is positive and the second-order effect is negative. Both point estimates are significant at the .01 level. The effects of prior foundings are essentially unchanged from those in column 2.
In both columns 2 and 3 the estimated effect of political turmoil is much lower than in column 1, and the point estimates are smaller than their standard errors. We are not sure why adding the flow of recent foundings depresses the effect of political turmoil, given the fact that Carroll and Huo (1986) find a significant effect of this variable in models that include the effects of prior foundings. Perhaps the difference is due to the fact that our models also contain the effect of density, or it may reflect differences in the estimators used or in other details of the specifications.
The estimated effects of density in Table 9.8 imply that the effect of density is non-monotonic within the observed range of density.53 In the case of the best-fitting model, the estimated effects of density in the third column show that the founding rate rose with increasing density until N = 206 and then declined. The founding rate at this maximum was 41 percent larger than at N = 0. At the historical maximum of density (N = 377) the rate was only 17 percent higher than the rate at zero density. Put differently, the founding rate in a population with 377 newspaper publishers was about the same as that in one with 70 newspaper publishers.
However, the estimates in Table 9.8 imply that growth in the founding rate has been monotonic over the historical range of variation in number of recent foundings. That is, the estimates tell that the founding rate was at a maximum when B = 51, while the historical high was B = 47. So these results suggest that the founding rate increased at a decreasing rate with the number of foundings in the previous year.
Source: Hannan Michael T., Freeman John (1993), Organizational Ecology, Harvard University Press; Reprint edition.