The third empirical test of the model of density dependence in rates of organizational mortality concerns the population of firms publishing news- papers in the San Francisco Bay area (in collaboration with Glenn Carroll). Partial likelihood (PL) is used in much the same manner as just described for the semiconductor firms. Lifetimes are broken into yearly segments, producing 30,781 observations. The values of all covariates are updated at the beginning of each calendar year: age, density, number of foundings in the prior year, and the index of political turmoil. All but the last observation for each publisher is censored on the right. For the 1,837 publishers that ceased publishing during the period of study, the last observation ends with an event; the spell is not censored. For the 333 publishers still in existence at the end of 1975, the last observation is censored on the right as well. The specification of dependence on time since entry in the mortality rate is an approximation to the Weibull model obtained by including the log-age at the beginning of each yearly spell as a covariate in the log-linear model, as described earlier.
Figure 11.7 shows the fluctuations in the mortality of newspaper publishers over the period. The peak period for mortality is slightly later than that for foundings. Indeed, it is partly a reflection of the fluctuations m foundings because young newspaper publishers have a much higher rate of mortality than older ones, as we noted in the previous chapter.
Figure 11.7 Mortality of newspaper publishing firms by year (San Francisco Bay area)
Now we explore the consequences of adding measures of density to models for the rate of mortality of newspaper publishers.74 Table 11.6 contains estimates of the basic model in equation (11.1). These estimates tell that both aging and density had strong effects on the mortality of publishers. Since the effects of aging in this population have been considered at length elsewhere, we concentrate on the effects of density. As predicted, the effect of density on the mortality rate is non-monotomc. The first-order effect is negative, and the second-order effect is positive. Both effects differ significantly from zero at the .01 level.
According to these estimates, the mortality rate fell and then rose with increasing density over the historical range. The minimum rate, at N = 282, was only a third as large as the rate at zero density. From this point on, the mortality rate rose gradually with increasing density. By the time the historical maximum (N = 377) was reached, the rate was 13 percent higher than the minimum rate. Figure 11.8 illustrates the estimated relationship between density and the mortality rate.
As was the case with labor unions but not with semiconductor firms, we found no evidence that the number of recent failures in the population affected the failure rate of newspapers (estimates are not reported here). This result stands in sharp contrast to the findings for the founding rate in this population, which showed a strong dependence on the number of recent foundings.
Finally, as Carroll and Huo (1986) found, political turmoil increased the mortality of newspaper publishing firms significantly. The estimated effect implies that the mortality rate rose almost three-fold during years charac- terized by riots, ethnic violence, and violent labor unrest.
Figure 11.8 Effect of density on mortality rate of newspapers (estimates from Table 11.6)
Source: Hannan Michael T., Freeman John (1993), Organizational Ecology, Harvard University Press; Reprint edition.