Reply to the Rejoinder on Taxonomy of Health Networks and Systems: A Reassessment

Abstract

I offer a brief reply to the Bazzoli, Shortell, and Dubbs thoughtful response to my critique of their taxonomy of hospital systems. The authors addressed the issue of taxonomic bias by suggesting that because only 25 percent of system hospitals (reported by Cuellar and Gertler 2003) are not in clusters, complications attributable to spatial dispersion might be minimal. Recall, the bias stems from the inability of the statistical computations to distinguish systems that report low levels of service capacity sharing among hospitals because they are spatially dispersed from those that simply do not engage in capacity sharing. First, I would note that Cuellar and Gertler counted urban hospitals only, thereby leaving out roughly 900 nonurban system hospitals, few of which would be in clusters. Thus, combining these with the 25 percent of urban nonclustered hospitals, the total comes to be that around 50 percent of system hospitals are not in clusters. Second, whatever the number, its significance depends on how evenly the nonclustered hospitals are distributed across systems. Looking at 2001 American Hospital Association (AHA) data, 4 percent of urban hospitals in systems labeled “centralized” were not in clusters, which compares with 37–60 percent for “decentralized” and “independent” systems, respectively—some very significant differences. Thus, high inequalities in the distribution of nonclustered system hospitals together with the large numbers of nonclustered hospitals provide ample basis for concluding: (1) significant bias exists in the taxonomy and (2) the taxonomy probably classifies systems more by their patterns of spatial dispersion than by any dimension of system structure. As to the second issue—whether or not measures based on local sharing of service capacity validly capture differences in system structures (setting aside the issue of bias)—the authors cite the Alexander et al. (2003) finding of a correlation between direct measures of system centralization and the taxonomy codes. Obviously, correlation and causality are not equivalents. Indeed, the Alexander et al. finding appears to be a clear example of a spurious correlation (otherwise known as illusory correlation or the “lurking variable”), a situation in which two otherwise unrelated variables are correlated because each is related to a third unmeasured variable. There are a number of possibilities for that third variable (e.g., system ownership type), but spatial dispersion is a prime candidate. The taxonomic codes are indisputably related to spatial dispersion (see the above as well as Table 1 in my commentary). Additionally, it is reasonable to infer that system structure (as measured by Alexander et al.) also is strongly related to spatial dispersion. Large systems, which in the taxonomy tend to be coded as decentralized, are widely dispersed geographically. And, it is well recognized ceteris paribus that organizations operating within diverse, complex, and differentiated environments—of which spatial dispersion is a special case—favor organizational decentralization (Lawrence and Lorsch 1967; Mintzberg 1979; Porter 1999). Finally, to adopt the Alexander et al. finding, one must satisfactorily answer the following question: What theoretical rationale supports the claim that patterns of service capacity sharing (upon which the taxonomic codes are based) in the diverse markets in which the highly complex multihospital systems are located determine system choices about corporate structure? In their discussion of networks versus systems, the authors note that networks were seen early on as alternatives to ownership-based models. It is true that many joined networks to avoid being caught up in the rolling wave of mergers and acquisitions. However, it fast became apparent that networks offered inadequate legal, administrative, and structural arrangements for strategic objectives to be accomplished. Consequently, the networks have tended to resolve into specialized collaborative entities. Indeed, the 1999 AHA special survey of systems and networks shows that network and system memberships often overlapped, which is consistent with the two being compliments, not alternatives. Regardless, it would seem self-evident that as loosely coupled organizations are far more decentralized, nonhierarchical, and nonstrategic than systems, undifferentiated application of the taxonomic measures, methods, and definitions to both models cannot be justified.