With two exceptions, we found increasing efficiency across the full range of scale examined, in each of the 16 country-intervention pairs. Although the shape and strength of the association varied by country and intervention type, we found cost per key output and scale to be strongly and negatively correlated in almost all instances - with the fraction of variation in efficiency explained by scale ranging from 19% - 96%. Doubling in scale resulted in reductions in unit costs averaging 34.2% across all country-intervention pairs and ranging from 2.4% to 56%.
Over the scale that we examined, we saw no up-turn in cost per key output except for STI and PMTCT programs in India. The latter was due to only one data point (Figure 6). This suggests that for programs in similar demographic and epidemic settings one would expect to observe increasing efficiency at least up to similar levels of output. This is an encouraging finding as it suggests that the current global HIV prevention program will become less costly over a wide range of expansion. However, this finding must be tempered by the finding of an up-turn in unit cost reported in the previous study of SW programs in India , and by the up-turn reported here in PMTCT and STI programs. It must also be tempered by unpublished data from India by co-author Dandona using PANCEA methods for a more recent fiscal year that suggest an up-turn for VCT and SW programs. However, it is important to note that with simple linear functions, the regression trends were downward sloping in all cases.
There are a number of possible and non-mutually exclusive causes of the systematic economies of scale observed in the PANCEA data. One is that busier facilities are able to distribute their fixed costs over more cases, thus lowering their unit costs. A second is that busier facilities are better positioned to take advantage of potential lower prices through bulk purchases, sharing of services and other advantages of scale that lead to greater efficiency. Yet a third mechanism by which large scale may reduce unit costs is that program administrators learn how to integrate HIV prevention activities with routine services, reduce personnel "down-time," reduce supply loss and breakage, and generally deploy their resources more efficiently. This learning occurs over time and is thus an "economy of time" rather than of scale per se. However, in an era of program expansion, time and scale are highly correlated.
The first explanation, i.e., that unit costs decline as fixed costs are distributed over more output, is an arithmetic and programmatic truism, and therefore must contribute to the observed scale effects. The extent of that contribution depends on the portion that fixed costs constitute of total costs. The larger the fixed cost portion, the stronger the scale effect will be. Thus, it may be possible to increase efficiency by either increasing demand for services or finding ways to reduce costs that have been treated as fixed. Because inputs were market priced, we introduced a bias toward not being able to detect economies of scale arising from the ability to make bulk purchases. The third explanation, namely that program managers learn how to optimize deployment of the resources over time, and thus with scale, is supported by longitudinal analyses we conducted in PANCEA and other analyses we are conducting  (and unpublished data).
The purpose of the current analysis was to document the relationship between scale and unit cost and not to formally assess the relative contribution of each of these three causes. Our team is planning further analyses of the underlying determinants of efficiency using multivariate regression techniques to shed light on this question. We are currently analyzing longitudinal data on about 20% of the PANCEA sites and applying multivariate analyses to the existing data set in an effort to address these questions more definitively.
Our analysis has a number of important limitations. The data presented here are cross-sectional. Rather than documenting the change in unit costs over time within expanding programs, we tabulated the unit costs of many programs simultaneously and measured the association between scale and unit costs among them. Although these analyses are strongly suggestive of economies of scale within programs, they do not demonstrate it directly. In addition, this study does not use multivariate statistical methods to show the relative contribution of other possible predictors of unit costs, or to control for possible confounding factors. In particular, this paper does not assess the possible role of service quality as a mediator between unit cost and service volume. The effects of quality on unit costs are unclear. Higher quality services may require more resources and thus raise costs. It may also increase a programs reputation, thereby lowering outreach and promotion costs. By attracting more clients, it may also contribute to lowering unit costs via other economy of scale effects. We did not observe service delivery directly, and meaningful measures of quality vary by intervention type. However, for VCT programs, the largest program sub-set, we regressed cost per client on numerous indicators of program quality. We found no statistically significant associations between indicators of service quality and unit costs. For these reasons, it is conceivable, if unlikely, that the differences observed reflect program differences other than scale, but reflected in scale.
Because we did not observe an upturn in unit costs with larger scale for the vast majority of the programs, we are unable to address the question of what the optimal size of different types of prevention modalities might be. That is, we cannot identify optimal facility size, i.e., when it is more efficient to open a new facility in a nearby community rather than continue to expand services in an existing one. We believe that the lesson for now is, more services will usually reduce cost per service. Micro-economic theory suggests that an upturn in unit costs may be a signal to policy makers that the number of HIV prevention service providers should be increased in the relevant setting .