We alternatively discuss results with our two dependent variables: VACadmin and VACeff, respectively.
Drivers of dissemination of vaccines (VACadmin)
Table 2 reports results with VACadmin as the dependent variable during the very early stages of the vaccine rollout. In the data as of January 12, 2021, a little over 3000 persons per 100,000 population were vaccinated on average. The related results are reported in Table 2 with four different modeling variations.
The results show that more prosperous states and states with greater rural populations had greater vaccine administrations. The resulting coefficients are statistically significant in all the models estimated. In terms of relative magnitudes, the elasticity of VACadmin with respect to RGDPpc turns out to be 0.8, while that with respect to RURAL is about a fourth of that at 0.2.Footnote 13 From a policy perspective, of course, changing the composition of the rural population is very slow, time-consuming, and often politically challenging. Furthermore, smaller states (POP) seem to be more adept at vaccinating their populations, consistent with the notion of a relatively better coordination, networking, and communication in such states.
When medical personnel are considered in Models 1.2 and 1.4, we see that the coefficients on PHYSICIANS and NURSES are statistically insignificant, whereas that on the broader measure, HEALTHworkers, is positive and significant (at the 10% level). This finding alludes to the importance of other health workers - e.g., appointment schedulers, nursing assistants, etc. These resources are especially important in the context of COVID vaccinations with inherent short shelf-life, scarcity, and refrigeration/transportation requirements. In addition, states with more nursing homes were no different from others in vaccine administration.
Turning to structural issues with respect to governmental involvement in the health system of states, states with a centralized health system were no different from others with regard to vaccinations. Surprisingly, state and local governments spending greater amounts on health and hospitals tended to have lower vaccine administrations.Footnote 14 This finding can be viewed in the context of a greater government size (involvement) leading to some decision-making lethargy or to a lack of coordination between federal and local governments - especially relevant in the context of COVID-19 vaccines (e.g., transportation, logistics, execution, etc.).
With regard to the other controls, states with Democrats as governors, those with more elderly populations, with relatively greater COVID-19 death rates, more nursing homes, and with more health workers were no different from others when it came to administering vaccines.Footnote 15
Later, in Section 4.3, we shall examine how these results change as the initial vaccination period has elapsed, along with a change in government leadership at the federal level. We turn next to examining the determinants of vaccine efficiency.
Drivers of vaccine administration efficiency (VACeff)
The results with VACeff as the dependent variable using January 12, 2021 data are reported in Table 3. The average percent of vaccines that were delivered over what was received in each state was about 37% (Table 1). Vaccine efficiency is especially relevant in the case of the COVID-19 vaccines, not only to save lives but also to conserve and prevent a scarce resource from going waste (because unused vaccines that have been on the shelf for a while lose their efficacy).
The results for vaccine efficiency provide some interesting contrasts with those with vaccine administration. More prosperous states again had some comparative advantages in administering efficiency; however, the coefficient on RGDPpc was statistically significant in two of the four models estimated. On the other hand, vaccination efficiency was no different in more rural states and the administration advantages of smaller states were largely absent when it came to efficiency - the coefficient on POP was negative and marginally significant in Model 2.4.
More significant differences in administration and efficiency emerge with respect to the impact of COVID-19-related deaths. States with a greater number of deaths had greater vaccination efficiency. It may be the case that states with more deaths also had more hospitalization, enabling better and faster alternative uses of unused vaccines or it may have to do with the politics of expedited vaccinations in states facing adverse press reports with more deaths.
Again, as in Table 2, vaccination efficiency was greater in states with more health workers, but not just with more nurses and physicians. On the other hand, both structural centralization of state health systems (CENTRALIZED) and fiscal decentralization tended to undermine efficiency.
Efficiency was, however, higher in states with more nursing homes, in the early phases of the vaccine rollout. Whether this is still the case, when the initial focus on nursing home vaccinations becomes less relevant over time, will be examined in the following section. Thus, some of the negative efficiency effects of a state’s direct involvement are countered by nursing homes and overall health employment.
As in Table 2 with regard to vaccine administration, states with elderly populations and with Democrat governors were no different from others with regard to vaccination efficiency.
Robustness checks using alternative timing of vaccine rollouts
As the information on vaccine rollouts is coming out with regularity, we conducted a robustness check by measuring the two dependent variables at an alternative date - February 2, 2021. Besides checking for robustness at an alternative time period where states had greater experience in confronting the challenges faced by the vaccine rollout, the second date also accounts for a change in the government ideology with a change in the presidency and the balance of power at the federal level and in states across the Union.
The corresponding results, using variants of Tables 2 and 3, are reported in Table 4. Overall, we find that the results for the latter period are mixed, with some support for earlier findings and some remarkable differences.
First, more prosperous states continued to see vaccine administration gains, but now such states had no efficiency advantages. Quantitatively, the elasticity of VACadmin with respect to RGDPpc is almost half in the latter period compared to the former period (specifically, 0.4 in Model 1.1A, compared to 0.8 in Model 1.1). Qualitatively similar was the case for more rural states (although the coefficient on RURAL is marginally significant in Model 2.3A).
Second, less populous states showed greater administration and efficiency than larger states. While the results for vaccine administration support earlier findings, those with efficiency now are stronger than the earlier period. On the other hand, the effects of COVID-19-related deaths on efficiency are not evident in the latter period.
Third, and perhaps most striking, are the results with respect to the medical variables. All three, SlhHOSP, CENTRALIZED, and nHOMES, had no impact on vaccine efficiency, while the negative sign on SlhHOSP supported earlier results with regard to vaccine administration. Further, the negative impact of nursing homes on vaccine administrations is likely due to nearly all states broadening the priority for vaccinations and most nursing home populations already being vaccinated in the United States.Footnote 16
Finally, as shown in Tables 2 and 3, the influences of Democrat governors and elderly populations continue to be statistically insignificant.Footnote 17
Whereas the data on vaccination success will emerge over time, this study provides intermediate information when there is still time to fine-tune the vaccination process, with positive implications for avoiding unnecessary loss of life.
Impact of a legacy of corruption
The corruption risks associated with all phases of the COVID-19 vaccine rollout, from manufacture to allocation and distribution, have been well recognized by international bodies (United Nations Office on Drugs and Crime ).Footnote 18 In the present context, state corruption can potentially have an impact on vaccinations, with the effects being either positive or negative, depending upon whether enabling or retarding effects of corrupt activity prevail (see Goel et al. ). Since U.S. corruption is hard to detect, especially concurrently, we consider two measures of state-level corruption lagged over two different time periods: (a) CorruptSR: a five-year average of per-capita federal corruption convictions in a state; and (b) CorruptLR: a corresponding 10-year average (see Table 1 for details). Goel et al.  have discussed how the tension between the scale and speed of vaccine roll out might interject with corruption.
Key questions addressed in this respect are:
These results, inserting the two corruption variables alternatively into Models from Table 2-4, are reported in Table 5. Since most of the other findings are in broad agreement with earlier results, we will focus on the corruption variables.
Overall, the impact of corruption is consistent with the greasing theory - states with a greater legacy of corruption seem relatively better adept at vaccination administration and efficiency. This might have to do with weak institutions in such states, where strict vaccination mandates (e.g., who to vaccinate on priority, how to ship vaccines and at what temperatures, policing, etc.) better enable administration and efficiency of vaccines. Furthermore, the presence of corruption might increase vaccinations when clinics are open longer/special hours, or allow the jumping of queues via the payment of bribes. It could also be the case that more corrupt states are better able to inflate vaccination success data to show better performance.
We do, however, find differences in the impact of corruption, especially those of CorruptSR, across vaccines administration and vaccine efficiency - with VACeff positively and significantly impacted by a greater corruption legacy (albeit at the 10%-level in Model 3.3a). Viewed differently, in the latter vaccination period (data from February 2, 2021), short-term corruption legacy significantly and positively impacts vaccination efficiency, but not administration.
Relatively speaking, the effects of short-term corruption are stronger than longer-term corruption - effects of corruption tend to dissipate over time. Thus, the answer to the second question posed above is a no. Quantitatively, comparing Models 3.1a and 3.2a, the elasticity of VACadmin with respect to CorruptSR is somewhat smaller than that with respect to CorruptLR (0.08 versus 0.10), respectively (both based on vaccination data from January 12, 2021). It would be interesting to see how these findings change as data on concurrent corruption emerges.