COVID-19 Mortality and the Structural Characteristics of Long-Term Care Facilities: Evidence from Sweden

Abstract As in many countries around the globe, older citizens in long-term care facilities (LTCFs) in Sweden were hit hard by the Coronavirus pandemic, but mortality varied greatly between different facilities. Current knowledge about the causes of this variation is limited. This article closes this gap by focusing on the link between the structural characteristics of LTCFs—ownership, size, and staffing—and the risk of dying from COVID-19 in Sweden during 2020. Having utilized both individual- and facility-level data, our results suggest that lower staff turnover, having a nurse employed at the facility, and smaller facility size are associated with an decreased risk of dying from COVID-19. Ownership type is not directly associated with COVID-19-related mortality, but public facilities have lower staff turnover and fewer personnel with additional employment than privately run facilities, while privately run LTCFs more often have a nurse employed at the facility.


Introduction
COVID-19 took a high toll on older people living in long-term care facilities (LTCFs) in countries as diverse as the United States, Spain, and Sweden.About 5% of the residents of Swedish LTCFs died from the disease during the pandemic's first year prior to the arrival of a vaccine.Yet, as elsewhere in the world, Sweden exhibited considerable variation in death rates across nursing homes, even within the same municipalities (Coronakommissionen, 2021).It suggests that, in addition to the capacity of national and subnational governments (Capano & Lippi, 2021;Toshkov et al., 2021) and pandemic policies toward care homes (Daly et al., 2021), factors related to specific facilities played a crucial role. 1  Both scholars and practitioners have therefore asked why individuals in some LTCFs had a higher risk of dying from COVID-19 than in others.This question has brought back to the table a fundamental issue about the differential performance of public, for-profit, and nonprofit providers (Bozeman, 1987;Brown et al., 2018;Hart et al., 1997), especially in the provision of social services (Amirkhanyan, 2008;Broms et al., 2024Broms et al., , 2020;;Gingrich, 2011).This is a central issue as it concerns an increasingly large number of individuals who need to make informed choices regarding care for their children or parents, as well as making other important life choices.Yet this is also the case for policymakers and public managers who decide whether to outsource or not and, if yes, to what extent and to whom.
Advocates of market provision of publicly funded social services usually refer to the classic argument that private ownership and competition drive prices down and improve service quality (Shleifer, 1998).Nonprofits may have an advantage over both public and for-profit providers as, although they generate profit, they are legally obliged to fully reinvest it into production and organizational capabilities (Hansmann, 1980, p. 884).Critics point to numerous factors associated with the market provision of public services, such as incomplete markets and contracts, and also the low measurability of the quality of complex services, that may negatively affect service quality (Broms et al., 2020;Brown et al., 2018;Hart et al., 1997;Romzek & Johnston, 2002).
The literature has remained focused on the differences between "public" and "private" providers, but an emerging strand focuses on more subtle distinctions, for example, the differences between different forms of private providers (Broms et al., 2024).As a contribution to this emerging literature, we develop a novel theoretical argument linking the ownership type with the extent of provider profit-maximization incentives and, ultimately, service quality.Combining the insights of dimensional publicness theory (Bozeman, 1987) and contractual incompleteness (Hart et al., 1997), we argue that the intensity of profit-maximization incentives is high in private companies with a greater exposure to market forces, but lower in companies with lower exposure to market forces.Therefore, private limited companies are expected to outperform public traded ones on service quality.At the same time, due to the nondistribution constraint on residual earnings, nonprofits are least likely to let cost-cutting considerations override quality concerns, and should perform best among all nonpublic providers.
We apply this argument to evaluate a very important quality outcome-COVID-19 mortality rates in Swedish nursing homes during the first wave of the pandemic, before the mass vaccination of residents.What sets our article apart from similar studies on the link between provider ownership and COVID-19 outcomes is that (1) we explicitly treat ownership as a causal factor of the "first order," affecting COVID-19 outcomes through several channels, including facility size and staffing characteristics; and (2) we employ a unique data set, combining individual-level data on all residents of Swedish LTCFs and the structural characteristics of the nursing homes they resided in throughout the first wave of the pandemic.
The results of our analyses suggest that ownership type does not directly impact the risk of death from COVID-19, but the latter is systematically associated with the staffing and size attributes of LTCFs, which in turn affect mortality rates.Specifically, lower staff turnover, having a nurse employed at the facility, and smaller facility size are all associated with a decreased risk of death from COVID-19.

Literature review
There are three strands of literature of relevance to our research question regarding what explains variance in COVID-19 mortality across Swedish nursing homes.First, there is an ongoing academic discussion about the effect of ownership type on different publicly funded services.Second, there is a literature on the link between ownership and service quality in LTCFs and, third, on the effects of ownership on COVID-19 outcomes in LTCFs.We briefly review the findings from all three literatures.
Over the past several decades, the organization of the provision of government services in many countries around the world, and particularly in mature welfare states, has increasingly taken on the form of a quasi-market, where publicly financed services are delivered by both public and nonpublic providers.This, in turn, led to the emergence of a large empirical literature examining the effect of private ownership-either at the aggregate level as a share of market or as a direct comparison of providers with different ownership status-on a range of processes and outcomes in the provision of social services.It should come as no surprise that conflicting theoretical predictions regarding the effect of private ownership on service quality, as discussed in the introduction, are accompanied by equally inconclusive empirical findings.There are studies that find positive effects of private ownership (Bergman et al., 2016;Holum, 2018); others report negative effect (Bach-Mortensen et al., 2022, 2023) and a third strand reveals no effect or mixed findings (Blix & Jordahl, 2021;Dahlstr€ om et al., 2018;Kim, 2022).While this literature primarily exploits the public-private distinction, thus far it has fallen short of accounting for different types of private ownership (but see Broms et al., 2024).
The association between the ownership type and service quality in the context of long-term care is a well-established research area.This literature has accumulated considerable evidence, including systematic reviews and meta-analyses, showing that care quality in for-profit facilities is inferior to that in public or nonprofit LTCFs (Amirkhanyan et al., 2008;Barron & West, 2017;Comondore et al., 2009;Hillmer et al., 2005;Ronald et al., 2016;Walker et al., 2022).However, many studies have also reported mixed results, with public nursing homes performing better on some dimensions of quality and nonpublic facilities performing better on others (Hjelmar et al., 2018;Patwardhan et al., 2022;Stolt et al., 2011;Winblad et al., 2017).Furthermore, although one of the most consistent findings from this literature is that of a strong positive impact of staffing indicators on care process and outcome measures (for a review, see Spilsbury et al., 2011;Broms et al., 2021;Harrington et al., 2016), studies evaluating the link between ownership and staffing characteristics are too few for conclusive interpretation (Banaszak-Holl et al., 2018;Harrington et al., 2012).
The literature examining determinants of COVID-19 outcomes in nursing homes has focused heavily on so-called structural factors, such as the ownership type of nursing-home operators, staffing factors, and facility size.The association between operator ownership and COVID-19 outcomes is the subject of a growing literature, the findings of which have, thus far, been equivocal (for a review, see Bach-Mortensen et al. [2021]).Contrary to the findings from the pre-pandemic literature and a common media narrative, only a handful of studies have found strong empirical support for the conjecture that private ownership leads to deleterious COVID-19 outcomes.Within this literature, most studies treat ownership type as an ordinary predictor, on a par with, for example, staff numbers or qualifications.Nevertheless, some suggest, albeit implicitly, that ownership is a causal factor of the "first order," affecting COVID-19 outcomes through different channels, including facility size or staffing characteristics (He et al., 2020;Rolland et al., 2020;Shallcross et al., 2021;Weech-Maldonado et al., 2021).
Empirical research on the determinants of COVID-19 outcomes in longterm care homes also suffers a number of data and methodological limitations.First, it has been predominantly based on data from the United States, with only a few contributions from different empirical settings (Brown et al., 2021;Rolland et al., 2020;Shallcross et al., 2021;Stall et al., 2020).Furthermore, the literature features only a handful of nationwide studies (Abrams et al., 2020;Gorges & Konetzka, 2020;Shallcross et al., 2021;Sugg et al., 2021) that offer reasonable grounds for generalization of the findings, even within the context of a single country.Second, most studies have utilized facility-level data only, missing potentially important individual-level variation.This is a limitation because resident health status, gender, and age are important predictors of COVID-19 outcomes (Couderc et al., 2021;Lee et al., 2021;Najar et al., 2023).Third, there is a strong, albeit not always explicit, suggestion in the literature that some structural characteristics, particularly relating to staffing qualifications and employment conditions, are the mediating factors that run between ownership type and COVID-19 outcomes.However, none of the existing studies has tested this proposition empirically.Fourth, the findings from this literature remain highly inconclusive.For example, two studies that analyzed data from the same sample and time period came to different conclusions: one found an association between for-profit ownership and higher rates of infections and deaths (Brown et al., 2021), while the other found no such association (Stall et al., 2020).
To summarize, despite a large accumulated evidence base and persistent media attention (DN, 2021;Savage, 2020), the association between type of ownership, other structural characteristics of LTCFs, and service quality, including COVID-19 outcomes, remains poorly understood.Extant literature on the effect of ownership on different outcomes in the provision of social services at large or within the context of residential elder care remains focused on the broad public-private distinction and provides no clear-cut evidence in support of the quality advantage of either ownership type.Finally, while the literature considers ownership central to the quality of care and to COVID-19 outcomes, it suffers serious data and methodological limitations.We aim to move the knowledge frontier forward by (1) nuancing the concept of ownership beyond a binary public-private distinction; (2) providing a novel theoretical framework on the link between different types of ownership and service quality; (3) explicitly theorizing the link between ownership as a causal factor of the first order, size and staffing attributes as causal factors of the second order, and how they affect COVID-19 outcomes; and (4) examining the resulting testable propositions with a novel high-quality data set (which combines the individual-and facility-level attributes) from a thus far less researched empirical setting.

Theoretical framework
Our theoretical framework for explaining the role of the structural characteristics of LTCFs in COVID-19 outcomes rests on two pillars.First, we argue that different ownership types exhibit different levels of intensity of incentives that cut costs and compromise the quality of services, which in turn affects COVID-19 outcomes.Second, we argue that ownership is a "first order" causal factor affecting COVID-19 outcomes, and staffing quality and size are "second order" causal factors.
We conceptualize COVID-19 outcomes as an indicator for a broader quality of services performed by a particular provider.Our point of departure is the well-established argument that contractual incompleteness-an omnipresent attribute of markets-enables "significant opportunities for cost reduction that do not violate the contracts, but … can substantially reduce quality" (Hart et al., 1997, p. 1,128, 1,152).Such opportunities are even more significant when it comes to the provision of complex social services because the quality of such services cannot be unambiguously described in a contract (Brown et al., 2016).Therefore, government procurement of elder care on the market inevitably leaves the government with an incomplete contract and the contractor with ample opportunities for quality-shading.
We further develop the cost-quality trade-off framework (Hart et al., 1997) by suggesting that the incentives for quality-shading are not equal for all private providers.Building on the idea that different private actors have different exposure to market forces (Bozeman, 1987), we posit that such exposure determines the intensity of the quality-shading incentives of the main types of private organizations operating on the elder-care markets.Given that three types of private organization operating on the eldercare market in Sweden in 2020-publicly traded, private limited, and nonprofit-we argue that publicly traded companies (i.e., companies whose ownership is organized via shares of equity traded on a stock exchange) have the highest intensity of quality-shading incentives, followed by private limited companies, and then by nonprofits.Nonprofits have the lowest intensity of quality-shading incentives because they are neither formed nor organized in order to generate profit.Instead, nonprofits are dedicated to pursuing mission-oriented goals (Besley & Ghatak, 2005), and this organizational form limits the use of profit to actions that further organizational purposes (Hansmann, 1980).We provide a longer discussion about the different private ownership types in Online Appendix A.
We argue that ownership affects COVID-19 outcomes via two major causal channels: first, the attributes of the workforce and, second, the size of the facility.Long before the coronavirus pandemic, staffing was identified as a plausible causal mechanism that links ownership status with the quality of care.For example, several studies pointed to a systematic difference in the number of nursing hours or part-time personnel between forprofit, nonprofit, and public facilities (Castle & Engberg, 2006;Harrington et al., 2012;Hjelmar et al., 2018;Hsu et al., 2016;McGregor et al., 2010).The argument that the factor of staffing primarily drives variation in COVID-19 outcomes between care homes has gained ground within academia and policy circles since the beginning of the coronavirus pandemic, as illustrated by McGregor and Harrington (2020), who said "ownership matters when it comes to staffing, and staffing matters when it comes to managing outbreaks of COVID-19 in LTC facilities." Furthermore, it is plausible to assume that a relationship between ownership and COVID-19 outcomes runs though the "facility design" channel.Driven by profit-maximization incentives (Broms et al., 2024), for-profit organizations may have a bias in favor of larger units and higher occupancy rates.On the other hand, large facilities may not be an organic aim for nonprofits.The "nondistribution constraint (Hansmann, 1980) suggests that nonprofits may not be particularly interested in generating extra profit through the economy of scale afforded by large facilities.Finally, public facilities may be organized in large units because of budgetary constraints and/or centralization pressures.
In accordance with the intensity-of-incentives framework, we expect LTCFs operated by nonprofit and public organizations to have higher staffing quality and smaller facilities compared to the two types of for-profit private organizations (publicly traded and private limited providers).While staffing and size are affected by the type of ownership, they also affect COVID-19 outcomes.Regarding the factor of staffing, its effect may be disaggregated into several specific mechanisms.First, overall staffing quality is consequential for the way care homes have responded to the pandemic (Figueroa et al., 2020;White et al., 2020).Second, some scholars underscore the importance of staffing levels-usually proxied through the number of hours per resident/day-for managing the COVID-19 response (Gopal et al., 2021;Gorges & Konetzka, 2020;Harrington et al., 2020).High staffing levels may increase the likelihood of nursing homes getting a case of SARS-CoV-2, but higher staffing levels are typically associated with lower infection and death rates.For example, Gorges and Konetzka (2020, p. 2,466) have argued that "Implementing measures … , such as regular testing and cohorting of both residents and staff, is difficult without sufficient staffing levels."Third, there is an argument about the optimal mix of different types of personnel required in order to provide the key to the successful management of infection pandemics.The importance of registered nurses is rationalized by their broad responsibilities as care coordinators, and more specifically as managers of infection control and other prevention measures (Davidson & Szanton, 2020).However, the role of assistive personnel, who provide direct daily care to frail and chronically ill people in LTCFs, was also found to be important for mortality rates during the coronavirus pandemic (Gorges & Konetzka, 2020).Finally, the high presence of nonpermanent personnel (those on hourly contracts or those provided by specialized staffing agencies) and high personnel turnover, which has been found to be associated with inferior quality of care in nursing homes (Antwi & Bowblis, 2018;Castle & Engberg, 2005;Loomer et al., 2022;Shen et al., 2023), may also be relevant for the care of frail elderly people diagnosed with COVID-19.
When it comes to facility size, regardless of the reasons for having or not having larger LTCFs, size matters in relation to death from COVID-19 because larger facilities have more employees and more admissions, generating more movement of people between the facility and the surrounding community and, hence, higher infection rates and a higher risk of dying from COVID-19 for an already frail and vulnerable nursing-home population.On the other hand, it is easier to effectively enact such infection control measures as social distancing and the cohorting of residents and staff in larger facilities, (Abrams et al., 2020), thereby reducing the risk of death.
Based on these considerations, we posit that ownership is the "first order" structural factor affecting the risk of dying from COVID-19 via structural factors of "second order."Specifically, the effect of ownership runs through two channels: personnel characteristics and facility size.Better personnel quality and smaller size are likely to reduce the risk of dying from COVID-19.At the same time, ownership may affect death rates from COVID-19 via causal channels other than staffing and size, which we refer to as "direct effect."Figure 1 visualizes this theoretical framework.

The Swedish case
As with most European countries, the coronavirus pandemic hit Sweden in the beginning of 2020.In mid-March 2020, the Public Health Agency of Sweden assessed the risk of community transmission as "very high" and declared the need for additional measures to stop the spread of the infection.This was effectively a declaration of the beginning of the first wave of the pandemic, which resulted in high rates of infection and death.The second wave of the pandemic began in the middle of October, again resulting in high numbers of infected and deceased.By the end of 2020, more than 10,000 individuals had died from COVID-19.The death rate was, however, not nearly as high during the course of 2021, with the death toll reaching over 14,000 by the end of the year.In other words, 2020 accounts for about 70% of deaths from COVID-19 during the first two years, which provides the rationale for the temporal focus of this article (Dahlstr€ om & Lindvall, 2021;Socialstyrelsen, 2022).
In terms of the numbers of those infected with SARS-CoV-2 and those deceased from COVID-19, Sweden is situated in the middle of the distribution among all countries but with significantly higher infection and death rates than its neighbors, Denmark, Finland, and Norway (Yarmol-Matusiak et al., 2021).Sweden adopted a strategy against COVID-19 that was different from most other countries.There were less coercive policies implemented in Sweden and the Swedish authorities did little to restrict the freedom of movement or the freedom of assembly.Many public institutions and private enterprises, which in other countries were closed, remained open for large parts of 2020.Instead, Swedish authorities relied on voluntary recommendations that were meant to achieve the same goals that other countries tried to reach with more coercive measures (Dahlstr€ om & Lindvall, 2021).Protection of vulnerable individuals within the elder-care system was, however, an important part of Sweden's response to the pandemic from the very beginning (Folkh€ alsomyndigheten, 2020).Still, by the end of April 2020, 90% of those who had died from COVID-19 were over 70 years old, and about half of those individuals lived in residential care homes.Indeed, a commission of inquiry that evaluated the government's response found the measures taken to protect vulnerable individuals to be inadequate (Coronakommissionen, 2020).
In Sweden, elder care is funded by tax contributions, but the provision is executed by both public and private providers.Elder care is comprised of assistance at home and care in LTCFs.Only the latter is the focus of this article.LTCFs are open to those elderly municipal denizens deemed by the municipal authorities as needing more care than can be provided through assistance in their own homes.Once such a decision is made, an individual gets assigned to one of the LTCFs (or, in municipalities with a so-called choice system, selects an LTCF) and receives services and care.The cost of care in LTCFs is almost exclusively covered from the municipal budget, regardless of the ownership status of the LTCF operator.Individuals admitted to nursing homes pay only a small fraction of the actual costs of their care (Ågotnes et al., 2020;Blix & Jordahl, 2021).
In line with the generally extensive level of autonomy enjoyed by Swedish municipalities, the Swedish Local Government Act of 1991 (Kommunallagen) and the Swedish Social Services Act of 2001 (Socialtj€ anstlagen) allow municipalities to provide residential elder care inhouse-i.e., staffing and managing care homes by themselves-or to buy it on the market, including the option of outsourcing the operation of public facilities.Private organizations are present in about a third of the municipalities, and their share of the local market varies from zero, often in sparsely populated and rural municipalities, to a majority, often in the municipalities of the country's metropolitan areas (Broms et al., 2020;Jordahl & € Ohrvall, 2013).On average, the proportion of LTCF residents living in publicly funded but privately run facilities is about one in five (Socialstyrelsen, 2020).
In sum, while Sweden is a typical case in terms of the pandemic dynamic and casualties (during the first two waves), it represents a type of welfare state in which the effects of the pandemic on LTCFs is yet to be studied.In terms of the organization of elder-care provision, it is the typical quasimarket observed in many other industrialized countries, but with a notable subnational variation in the provision of elder care.And while it is true that Sweden implemented a less coercive, more voluntary, strategy against COVID-19 than most other countries during 2020, protecting vulnerable individuals in LTCFs was part of the Swedish strategy from the beginning of the pandemic.

Data and method
To investigate how death from COVID-19 among LTCF residents may relate to the characteristics of their facilities, we leverage multilevel data, combining individual-level information on the residents of Swedish LTCFs with facility-level data.This data collection, which was conducted as part of the SWECOV-project, included access to several administrative registers.The sample frame for the individual-level data derives from the National Board of Health and Welfare's (in Swedish Socialstyrelsen, henceforth NBHW) register of social services for the elderly and people with disabilities (SoL), and covers all individuals approved to reside at an LTCF as of February 2020 (N ¼ 83,083).Our study period spans from March to December 2020.Importantly, since the first LTCF resident was administered a vaccine dose in late December 2020, this period forms a discrete, pre-vaccination, stage of the pandemic.
Complementing register data includes information on whether a resident contracted SARS-CoV-2 and died from COVID-19, as well as socioeconomic and medical information, such as age, biological sex, country of birth, education level, and comorbidities (see Online Appendix B for more detail).
Our facility-level data on ownership type derives from Statistics Sweden's Central Business and Workplace Register.The variable ownership captures the legal status of the company operating a nursing home.It takes on four values: public, publicly traded, private limited, and nonprofit.Publicly operated LTCFs housed the overwhelming majority (79%) of the individuals in the sample.
As for facility staff, we consider a number of potentially important personnel-related factors, mainly using register-based labor statistics (RAMS) from Statistics Sweden. 2 In the theory section we explain that, based on the literature's accumulated knowledge, we would like to capture four aspects of staff composition at the facility level: staffing quality (White et al., 2020), staffing levels (Gorges & Konetzka, 2020), the influence of different types of personnel (Davidson & Szanton, 2020), and the permanency of staff (Antwi & Bowblis, 2018).The five personnel characteristics that we include are, therefore, whether a nurse and/or a manager was employed at a given LTCF in December 2019, indicative of on-site availability of health care expertise and higher-level decision-making competence 3 ; the number of care personnel, indicative of staffing levels; the share of care personnel with at least one additional source of income, a proxy for part-time employment; and share of care staff with income from the same LTCF during all months of 2019, a proxy for a (low) turnover.By capturing three different personnel categories (managers, nurses, and care personnel), we also get an indication of how these different categories are related to the outcome in question.
Turning to the facility size variable, it is calculated as the number of residents in a given LTCF.
Online Appendix B.3 provides a full description of variables.
We matched individuals to facilities by linking information on individuals' place of residence in Statistics Sweden's Population Register with LTCF visiting addresses in the Business and Workplace Register.When large clusters of unmatched individuals with LTCF decision were registered as residing at addresses nearly corresponding to a LTCF visiting address, we undertook complementary hand-coding to increase the match rate (a detailed description of the matching procedure and its results is provided in Online Appendix B.1).This resulted in a data set covering 51,576 residents (62% of our population of residents) in 2,085 nursing homes.The availability of this data set-hitherto unparalleled in scope-allowed us to estimate the risk of dying from COVID-19 for residents of Swedish LTCFs, taking important individual and facility characteristics into account.

Empirical analysis
To estimate staffing quality and size as a function of operator ownership, we ran a series of OLS (using size and all staffing indicators except nurse/ manager employed at facility as outcome variables) and logistic (nurse/ manager employed at facility as outcome variable) regressions on the facility-level data.To ascertain that regional differences do not confound the focal relationship, this facility-level analyses included county fixed effects with standard errors clustered at the same level.The model estimating (logged) number of care staff also included the (logged) number of residents, thus producing an estimate of staff density.
To predict the risk of COVID-19 death as a function of ownership and other facility-level factors, we conducted a set of survival analyses, including a number of individual-level covariates identified by the extant literature as risk factors for dying from COVID-19: age (years), sex, comorbidities (cardiovascular, lung or kidney disease; hypertension; diabetes; and dementia), region of birth and education (at most compulsory or at least secondary).This broad battery of controls allows us to control for variation across both biological and socioeconomic factors that could potentially affect the assignment to certain types of facilities and mortality risks of COVID-19.Online Appendix B.4 describes all the control variables in more detail.
Moreover, since community transmission of SARS-CoV-2 varied substantially within Sweden during 2020, we also stratified baseline hazards by county (N ¼ 21).Indeed, counties also display considerable variation in the propensity of privately run LTCFs.Figure 2 shows considerable intercounty variation for both community transmission and the share of privately operated LTCFs-and, notably, that Sweden's largest county, Stockholm, displayed the highest incidence of both-illustrating the necessity of accounting for community transmission in the analysis.
As such, our main analysis estimates how LTCF characteristics affect the risk of dying from COVID-19 during 2020, conditional upon both individual and contextual factors.Although our empirical strategy is therefore inherently based on selection of observables, and our results should be seen as correlational rather than strictly causal, combining individual-level covariates with higher-level stratification serves as a demanding model that severely limits the room for biased estimates in the present case.Substantively, a primary source of such bias would arise from endogenous selection, whereby certain types of residents freely and actively sort into different categories of LTCFs.In reality, the possibility of this happening is very limited in the present context, for a number of reasons, which furthermore makes the Swedish case an appropriate environment to study the differences between public and private public-goods provision in general.
As mentioned above, since LTCF fees are capped at a low level, living cost is here a trivial factor, meaning a low level of separation along socioeconomic lines, at least within a given municipality 4 or county.Further, in all but a small number of municipalities that employ choice systems, placement at a given facility is subject to a municipal case officer's discretion.In combination with frequent queues for LTCF placement, this leaves very little room for the applicant to pick and choose between individual facilities.On the demand side, the fact that applicants are old and often afflicted by cognitive disabilities like dementia further limits the likelihood of endogenous selection to specific LTCFs.Reassuringly and in line with these arguments, we find that, accounting for county, variation in individual-level covariates across the different types of LCTFs is minor (see Figure B5 in the Online Appendix).In sum, given that our model accounts for both higher-level (county and municipal) factors pertaining to market structure, community transmission of COVID-19, and institutional variation in elder-care related factors, and individual-level factors like age and dementia, we believe our estimates are plausibly free of notable amounts of bias.

Results
We begin by estimating differences at the LTCF level in staffing quality and facility size as a function of operator ownership.Figure 3 reports the estimates for staffing quality (panels a-e) and size (panel f) by ownership type.The results show that there is little difference between provider types with regard to the number of care staff (panel 3[a]).Facilities with publicly traded ownership have the lowest number of care staff, but the estimate is statistically significant (p < 0.05) only compared to public ownership.Public facilities display a much higher share of care staff employed during the full year: 70% compared with 40-50% for the private categories (panel Note.Estimates are predicted values of operator ownership with 95% CI, with standard errors clustered at the county level.All models control for county fixed effects.Estimates in panels a-c and f are from OLS; and in panels d and e from logistic regression.The model predicting the logged number of care staff includes controls for the logged number of residents to account for facility size.A table with the full results is available in section C.1 of the Online Appendix.

3[b]
).Similarly, public facilities have a lower share of care staff with other sources of income: one-in-ten versus one-in-four (panel 3[c]).On the other hand, it is rather uncommon for public LTCFs to keep a nurse employed at the facility (pr�0.4),while private providers routinely do so (the corresponding probability ranges from 0.7 to 0.8).Moreover, publicly traded operators are about half as likely to have a manager employed at the facility proper (pr.�0.25) as any other provider (pr.�0.5).Finally,panel 3(f) shows that the difference in size between the operators is not statistically significant.
Continuing the investigation into the differences in individual-level risk of COVID-19 death as a function of operator ownership, we begin by looking at the unadjusted relationship through Kaplan-Meier curves for each of the four categories throughout the study period.Due to the old age and frailty of LTCF residents, many (N ¼ 11,547) died from non-COVID-19 causes, and were therefore censored from the study.Other reasons for censoring included address change (N ¼ 1,494) and termination of the decision for eligibility for care in LTCF (N ¼ 79).35,874 residents continued until the end of the period of study.
Figure 4 shows that, by the end of the study period, residents in publicly operated facilities have a 95% survival probability.This is significantly higher than for residents in both publicly traded and private limited categories, for which the survival rates are around 93%. Further, residents in nonprofit-run LTCFs display a significantly lower survival probability (90%) than residents in all other categories.In order to account for confounding factors, we also ran two separate sets of Cox proportional hazard regressions, with standard errors clustered at the LTCF level, considering the highly uneven pattern of incidence of the disease by facility. 5The first modeled the risk of dying from COVID-19 as a function of operator ownership and a second in which this was modeled as a function of the LTCFs' size and staffing characteristics.
Model 1 in Table 1 reports predicted hazard ratios for the three private ownership categories, compared to the reference category, public ownership, when adjusting for the full set of individual-level covariates.The reported estimates suggest that residents of LTCFs operated by nonprofits and publicly traded companies are at a higher risk of dying from COVID-19 than residents of public nursing homes, while the estimate for private limited companies is not statistically significant.However, in model 2, which stratifies baseline hazards by county, the differences in predicted hazard ratios by ownership type are statistically not significant.This finding suggests that the type of ownership is not systematically associated with the risk of dying from COVID-19.Instead it suggests that the observed differences in risk between different operator types are an artifact of private facilities being more common in regions more severely hit by the pandemic (see Figure 2).
Table 1 also reports estimated hazard ratios for the size and staffing variables, adjusted for individual-level covariates, without and with county strata (models 3 and 4, respectively).The results for size suggest that residents in larger facilities experienced a statistically significant (p < 0.000) higher risk of dying from COVID-19, with hazard ratios between 1.3 and 1.4, depending on whether county strata are included.This translates into a 1% increase in the number of residents, predicting a 0.31% increase in the expected risk of COVID-19 death.
The results regarding staffing quality are mixed.The estimate for care staff, full year, measuring the turnover among care staff, is statistically significant (p < 0.05) in both the unstratified and stratified models.The size of the coefficient in the latter (model 4; HR ¼ 0.56) means that a one percentage-point increase in full-year staff corresponds to a 0.44% lower predicted risk of COVID-19 death.Further, having a nurse employed at the LTCF is associated with a 17% lower risk of death (HR ¼ 0.83, p < 0.05 in model 4), when accounting for county-level strata.
While the hazard ratio for the variable care staff with other income is statistically significant and large (HR ¼ 3.47, p < 0.000) in model 3, it falls below the accepted level of statistical significance when community spread is accounted for in model 4. The coefficients for the remaining staffing variables-number of care staff employed and having a manager employed at the facility-are not statistically significant.
In sum, the proposition that facility size matters for the risk of dying from COVID-19 finds strong support in the data, but when it comes to staffing quality, only a stable cadre of care staff and, organizationally, the physical presence of nurses are associated with reduced risks of COVID-19.

Robustness
We scrutinized the sensitivity of the reported results in several ways.Section C of the Online Appendix provides a full description of the robustness tests and their results.
One reasonable concern is that the limited match rate of 62% could result in estimates that are unrepresentative of the target population of all LTCF residents.In an effort to rule out potential bias arising from an inability to match every eligible recipient of residential care with a nursing home, we replicated the main analysis on a number of alternative samples.First, cognizant that the limited matching of residents to facilities may be a result of unobserved factors, and thereby is a source of bias that risks skewing our estimates in unknown directions, we first reran the original analysis on a subsample with a much higher match rate.Using a separate municipal register of LTCF residents from the country's largest municipality, Stockholm, we obtained a match rate of 94%.Rerunning the analysis using the original data but restricted to LTCF residents in Stockholm, and comparing this with the results of the better-matched municipal information, garnered highly similar results.Second, we focused on subsamples of the original data in a targeted fashion by excluding several groups of individuals who were particularly difficult to match to a nursing home: residents with a recent LTCF decision and individuals with partners.To this end, we also excluded LTCFs with a small number of matched individuals overall.Results for these better matched subsamples were highly similar to the original results.Taken together, these varying approaches to adjusting the original sampling strategy assuage concerns that the limited match rate of 62% would incur biased estimates compared to the total population.
Next, given that there are several ways of measuring both predictors and outcomes, we wanted to ascertain that the main results were not an artifact of specific operationalizations.For operator ownership, which can be subject to change over time, we excluded a small number of LTCFs that experienced a change in operator ownership during the study period between November 2019 and 2020.We also employed a more narrowly specified indicator of death from COVID-19.Reassuringly, none of these alterations in the sample yielded results substantively different to those reported in the main analysis.
Further, given that staff composition can plausibly be defined in different ways, we also used alternative indicators for facility size, nurse presence, and care staff.Notably, switching from the broader category of "care staff" to the more specific and, implicitly, better-trained category of assistant nurses produces virtually no change to the main results.This indicates that, at least within the context of crisis, all members of the regular staff appear to have mattered to a comparable degree.By contrast, when exchanging the binary measure of nurses with a measure of the nurse to resident ratio, its significance disappears.It should be noted that this variable derives from an alternate survey, issued by the NBHW, which means a loss of observations.Still, this observation, together with the null result for number of care staff, points toward a situation in which, in the specific context of the pandemic, quality trumped quantity, as stability and on-site competence among the staff was beneficial while higher staff density was not.
Finally, in order to check whether our results are dependent on our substantive focus on COVID-19 mortality during the full year of 2020, while accounting for community spread at the county level, we, in turn, used the risk of SARS-Cov-2 infection as the outcome variable, and limited analysis to the first wave (spring of 2020).Further, as mentioned in the discussion on empirical strategy, we supplanted the county-level stratification approach with the more fine-grained municipal level in order to set more refined, and thus reasonably tougher, constraints on higher-order confounding.These alterations yielded only a few substantive differences compared to the main analysis.A difference to note is that in the specification with municipal strata, the coefficient for the facility size is not statistically significant.
In sum, the original results, as presented in Table 1, are unlikely to be affected by the limited match rate.Concerns about idiosyncratic operationalization decisions inherent in the central variables of interest, with some minor exceptions, appear neither model-dependent nor contingent upon the specific study period or even the COVID-related outcome defined here.

Discussion
Similarly to evidence in the extant literature, we found no systematic relationship between the form of ownership and a risk of dying from COVID-19 in nursing homes, and a mixed picture with regard to the effect of staffing variables.However, we also found that those staffing variables that do affect the chances of survival from COVID-19 for elder citizens-staff turnover and having a nurse on the site-are systematically associated with ownership.Specifically, public LTCFs in Sweden during the COVID-19 pandemic tended to have lower staff turnover compared to the rest of LTCFs, and, in turn, this factor was found to lower the risk of dying from COVID-19 in Swedish nursing homes.This is in line with the previous literature that found staff turnover to be crucial for the quality of care at nursing homes in general (Shen et al., 2023) and for infection control in particular (Castle & Engberg, 2005;Loomer et al., 2022).High turnover creates organizational and operational challenges that may disrupt the care process, and undermine the integrity of the infection control measures, potentially leading to SARS-Cov-2 outbreaks.Previous literature points to low pay, poor working conditions, and few opportunities for advancement as potential predictors of high turnover in for-profit facilities (Gandhi et al., 2021;Shen et al., 2023).
On the other hand, we found a link between private ownership and the presence of nursing personnel on site, which, in turn, emerged as a potent predictor of lower COVID-19 attributed mortality in Swedish LTCFs.This is a novel factor that thus far has not been examined in the literature.The quality gains from the presence of nurses on site could be related to the soundness of infection control (Davidson & Szanton, 2020), the adequacy of response to medical emergencies (Ewert et al., 2023, p. 70), and the higher continuation of care, compared to a situation where nursing services are provided externally.The large effect size that this variable exerts in our analysis is a clarion call to both researchers and policymakers to examine the causes and effects of the organization of nursing services in LTCFs.
Our finding that larger facilities tend to have higher COVID-19 related mortality is congruent with previous research, particularly more recent studies that underlined that facility size one of most consistent predictors of COVID-19 cases and deaths (Konetzka et al., 2021).This finding also aligns with the results from the literature on the quality of care in Swedish nursing homes that found that smaller facilities provide higher quality (Spangler et al., 2019).However, our analysis reveals that, unlike the above discussed staffing characteristics, the ownership type does not drive the difference in the facility size.
Taken together, these results provide support to our theoretical argument that ownership matters for the COVID-19 epidemic, and that the effect runs through the "staffing channel."However, we do not find any direct relationship between different forms of ownership and COVID-19 mortality.Speculatively, this could be because different ownership forms have different staffing advantages.Moreover, our theoretical emphasis on the heterogeneity of non-private forms of ownership does not find support in the data, as the main difference seems to run between public and nonpublic forms of ownership.This is in contrast to recent research from the same setting that found quality differences between different types of nonpublic nursing homes (Broms et al., 2024).One reason why we find no difference in COVID-19 outcomes among different private providers is the fact that COVID-19 outbreaks and deaths in nursing homes were exceptionally strongly affected by the SARS-Cov-2 infection rates in the surrounding area (Konetzka et al., 2021;Levin et al., 2022).In line with this conjuncture, the differences in COVID-19 mortality between nonpublic providers that we detect in the analysis with the individual-level controls (model 1 in Table 1) disappear once we control for a proxy for the community spread (model 2 in Table 1).The null result regarding the heterogeneity of private forms of ownership could also be due to the fact that at the time of the pandemic, private equity, which previous research has shown to engage in quality shading (Broms et al., 2024), was not present in the Swedish residential care market.

Conclusion
As in many countries around the globe, the residents of Sweden's nursing homes made up a large proportion of those who died during the COVID-19 pandemic.While the spread of the virus throughout society as a whole was the most important cause for this tragic statistic, the level of SARS-CoV-2 infection and death from COVID-19 varied considerably between the country's nursing homes.
To answer the question as to what factors may have driven this variation, we devised a study in which we focused on facility-operator ownership as a causal factor of the first order, nuancing the concept of ownership beyond a standard public-private distinction.We then developed a novel theoretical framework linking the type of ownership with the extent of provider profit-maximization incentives, which we hypothesized may affect COVID-19 outcomes directly, but also indirectly via second-order factors, such as staffing and the design characteristics of nursing homes.Finally, we examined the resulting testable propositions utilizing a novel high-quality data set, in which we matched data on almost all residents of Swedish nursing homes with the characteristics of the homes they resided during the first year of the pandemic.
Our findings revealed, first of all, no direct association between ownership type and COVID-19 mortality.Instead we found that different types of ownership affect the structural characteristics of nursing homes differently.Specifically, public facilities have, on average, lower personnel turnover and fewer personnel who work somewhere else besides the facility, but privately run LTCFs more often have a nurse employed on site.No association between ownership and facility size was discerned.
Furthermore, when it comes to the association between second-order facility-level factors and COVID-19 outcomes, our analysis presents heterogeneous findings.Specifically, lower risk of death from COVID-19 is systematically associated with facilities with lower staff turnover, the presence of an on-site nurse, and smaller facility size, but these life-saving features are not systematically linked with any specific type of ownership.
Although ownership has no direct association with COVID-19 outcomes, we found that public facilities have lower staff turnover and fewer staff with a job "on the side," while it was more common for private facilities to have a nurses on site.This suggests that public and private facilities are good at different things.One can speculate that these are causal mediating factors, as we suggest in our theory, and the fact that they pull in different directions might be the reason why we find no association between ownership and COVID-19 outcomes.
Nevertheless, our results hold implications for policymakers and practitioners.First of all, our findings speak against the wholesale rejection of market provision of social services that strengthened its position in the popular discourse in the wake of the coronavirus pandemic (Visontay et al., 2020).Neither do they support the notions that "public sectors aren't built to handle pandemics" (McConnell & Stark, 2021).Instead, our study revealed that both public and private providers have organizational features that matter for saving lives in a pandemic and these should be examined and harnessed for the public benefit.For example, our finding that a nurse employed on site is systematically associated with lower COVID-19 mortality suggests several distinct causal mechanisms through which the presence of nurses may affect COVID-19-induced death, which should be further examined.Furthermore, our finding regarding staff turnover adds to the growing recognition of the importance of the issue of inadequate employment conditions and security, particularly in elder-care's private sector, and calls for a considered regulatory response.For example, restricting employment during the pandemic to just one workplace may reduce the spread of infection in facilities.Finally, the lack of the association between facility size and death from COVID-19 speaks against the rationalization of turning residential elder care into large-scale facilities.
Although our article has made several important contributions of both theoretical and empirical nature, it is not without limitations, and as such these present avenues for further research.One way to extend this research is to test our theoretical argument using different methodological approaches, by, for example, applying methods within the mediation analysis framework or qualitatively through, for example, process tracing.Developing a more finegrained theory-with a focus on, for example, staffing features-that links ownership with other facility-level characteristics, and ultimately service quality is a another clear direction in which this research may be taken.Finally, as indicated earlier by our finding on nurse employment on site, there are several distinct causal mechanisms, including mere physical presence, nurse density vis-� a-vis residents, and better integration of nurses into facility management, through which the presence of nurses may affect COVID-19 induced death.Understanding which of the specific mechanisms matter most presents itself as an important avenue for further research.This article encourages future research in all of these directions.

Notes
information is unsuitable for such a measure.However, we complement our registerbased measure with a variable from an alternate data source for robustness tests-a survey from the NBHW, where representatives of facilities themselves evaluate the ratio of nurses to residents.Because the coverage of facilities is lower in this data and the closest point of measurement is a year before the pandemic, we still rely on the register-based information in the main specifications.4. In the Online Appendix, we also report results from stratification at the lower, municipal level.The fact that many smaller municipalities only have a very small number, sometimes only one, of LTCFs in the sample makes the county-level stratification our preferred strategy.5.The proportional hazard assumption was met for all models (see section C5 in the Online Appendix).

Notes on contributors
Rasmus Broms is associate professor of political science at the University of Gothenburg, Sweden.His research focuses on institutional quality and local politics.
Carl Dahlstr€ om is professor of political science and research fellow at the Quality of Government Institute, both at the University of Gothenburg, Sweden.His research is concerned with bureaucratic politics, welfare state policymaking and marketization of public services.
Jenna Najar is a postdoc at the Institute of Neuroscience and Physiology at the University of Gothenburg, Sweden, and a medical doctor, resident in psychiatry, at the clinic Cognition, Psychiatry, and old age psychiatry at Sahlgrenska University Hospital.Her research focuses on dementia epidemiology.
Marina Nistotskaya is professor of political science and research fellow at the Quality of Government Institute, both at the University of Gothenburg, Sweden.Her research focuses on the causes and effects of organizational structures of public bureaucracies, including the provision of public services.

Figure 3 .
Figure 3. Facility-level associations between operator ownership, size, and staffing.

Table 1 .
LTCF characteristics and risk of dying from COVID-19: Results from adjusted Cox proportional hazard regression.Coefficients are hazard ratios from Cox regressions with 95% level confidence intervals in brackets.
� p < 0.05, �� p < 0.01, ��� p < 0.001.Care staff, full-year, and Care staff w/other income range between 0 and 1.The output for the individual-level controls-age (years); sex; cardiovascular, lung and kidney disease; dementia; hypertension; diabetes; region of birth (Sweden, Scandinavia, Europe, Asia/the Soviet Union, America [North and South]/Oceania, and Africa), and education (compulsory or less [reference category] or secondary or higher)-is omitted.