Do health vouchers broaden the choices of citizens with low socioeconomic status? An analysis of Medicaid in Brooklyn

ABSTRACT Governments worldwide have committed to extending choices in public management to deliver services effectively; yet, how these programmes ensure equality remain unclear. This study investigated whether such programmes widened choices among different groups of citizens, focusing on Medicaid in Brooklyn, New York, in the 2000s. Information on patient admissions from the Statewide Planning Research and Cooperative System and hospitals from the American Hospital Association was analysed using a difference-in-difference-in-differences approach. Findings indicate that Medicaid programme failed to broaden the spatially confined choices of hospitals to patients with low socioeconomic status compared to non-Medicaid or uninsured groups.


Introduction
Since public service delivery systems have been increasingly fragmented and implemented across sectors, many governments focus on the effectiveness of public management (Andrews and Entwistle 2010;Osborne, Radnor, and Nasi 2012;Kwon and Park 2018). This is an ongoing issue as they deliver services using a market-oriented approach (e.g. vouchers, privatization, and contracting out). This approach aims to improve the efficiency as well as the responsiveness to users via a market that can be modified to better satisfy the requirements of consumer sovereignty, though not guaranteed (Lowery 1998).
The approach, derived from variants of public choice, rational choice, and the new institutional economics perspective from the 1960s, was widely cited in the 1980s as the new public management concept (Hood 1991;Osborne 1993;Dawson and Dargie 1999). Subsequently, this is now framed as the market-based governance (Donahue and Nye 2004;Warner and Hefetz 2008) or the new public governance, which emphasizes the governance of inter-organizational relationships (Rhodes 1996;Peters and Pierre 1998;Osborne 2006;Ansell and Gash 2008). This motivated us to explore how public service organizations effectively deliver services to beneficiaries in a market of public services and whether the public service delivery system is well designed for the targeted beneficiaries.
To that end, this study attempts to address the managerial challenges posed by service delivery inequality using the specific case of a voucher-based assistance programme. A voucher programme is a scheme that facilitates consumer choices by offering demand-driven purchasing power to citizens at subsidized prices or at no additional charge involving issues for the service redistribution (Friedman 1955;Steuerle, Peterson, and Reischauer 2010). Such programmes have been central to recent reforms and have pervasively been implemented by many countries' public services sectors; examples include British healthcare and social services (Dixon and Le Grand 2007;Rodrigues and Glendinning 2015), US health and education services (Johnston and Romzek 1999;Kisida and Wolf 2015;Lee, Jilke, and James 2020), Swedish homecare and healthcare services (Gingrich 2011;Feltenius and Wide 2021), and Korean social services (Jung, Choi, and Jang 2009;Kim, Baek, and Kum 2012;Lee and Park 2018).
Since the natural motivation for voucher-based programmes is to improve choices regarding accessing services, this study explored whether a particular case of Medicaid has broadened the hospital choice of patients by giving them purchasing power to choose hospitals they otherwise would not have chosen. Specifically, this article aims to demonstrate the relevance of Medicaid outcomes by investigating the differences between patients of low socioeconomic status (SES) and others, in exercising their hospital choices, which could result from choice restriction disparity. This article also determines how Medicaid has helped ameliorate health inequalities among citizens regarding extension of citizen choices in the 2000s when Medicaid was negatively modified by the Deficit Reduction Act (DRA) of 2005 and not expanded by the Patient Protection and Affordable Care Act (ACA) of 2010.
The findings contribute to the public management literature in several ways. First, this study contributes to the literature theoretically by classifying the programme concerning extending choice as either demand-driven (e.g. voucher, reimbursement, or direct cash programme) or supply-driven (privatization or contracting out). Many scholars have conducted deeper research on the impacts of a supply-side programme at the provider or area level in health markets (Johnston and Romzek 1999;Cookson et al. 2010;Roh, Moon, and Park 2011). In contrast, relatively few studies on demand-driven programmes exist, and little attention has been paid to the impacts of the programmes on citizen attitudes at an individual level. This article fills the void by addressing the challenges of the programmes in an analytical way. It generates new theoretical insights into how to manage demand-driven programmes in public service delivery and what should be done when they fail to deliver services to beneficiaries in need.
Second, this study provides practical implications by suggesting performance indicators for public service management such as choice behaviours in relation to service accessibility. Given the likelihood of service performance misperception, especially in a health sector characterized by high levels of information asymmetry, both subjective and objective performance indicators are important (Cheon et al. 2019). Therefore, whether vouchers impact individual choice behaviours will offer complementary indicators to measure other aspects of performance.
Moreover, this study addresses the question of whether the programmes positively impact service beneficiaries' behaviours as part of the diversity of performance management. This is significant because previous research has primarily investigated the performance of organisations operating within voucher programmes (Ford and Andersson 2017). Therefore, this is an attempt to understand public service management in the context of individual behaviours to demonstrate not only how such programmes evolved over time but also how they changed people's behaviours according to their SES. The findings would be of interest to public administrators or policymakers, those researching public service management theories, and those launching and managing choice programmes for vulnerable citizens.

Conceptualizing extending choice
Theoretically, in public management, extending choice is valued as an instrument for public managers to place competitive pressure on public service providers to increase both allocative and operational efficiency and to raise providers' responsiveness to citizens.
From the perspective of a supply-driven choice policy, offering choice generally means increasing the number of alternatives of providers or services, thus inducing provider competition (Johansen and Zhu 2013;Lee 2019). Contracting out or privatization facilitates the choice from the supply side. However, measuring choice simply by counting the number of alternatives may not be effective because a smaller set of better or more diverse alternatives can provide more choice than a set that is simply larger (Dowding and John 2004;Botti and Iyengar 2006).
From the perspective of a demand-driven choice policy, offering choice means giving citizens the ability to choose. To offer a choice, public managers consider providing financial support for citizens to take advantage of existing providers analogously in private markets. A classic example of this type of programme is a voucher scheme such as Medicaid, a US government health insurance programme that provides healthcare vouchers to low-income families and individuals (Currie and Fahr 2005;Le Grand 2006). A hospital will redeem a voucher with the health department of the relevant government and, in return, receive payment from public funds. Increased financial resources may allow patients to cover their medical costs or mitigate distance costs, information costs, or switching costs associated with leaving their current provider and picking a new, better alternative when deciding between choices they would not have had access to previously. Accordingly, Medicaid entitlement allows patients to access services they desire and to which they are entitled.
In the context of public management, service programme performance has been discussed regarding subjective measures such as citizen satisfaction (Charbonneau and Van Ryzin 2012;Grosso and Van Ryzin 2012;James and Van Ryzin 2017). However, the extent to which such programmes extend choice is not readily monitored from the perspective of the users' satisfaction with the services provided. Researchers point to the problems of measuring users' perceived service quality due to cognitive and desirability biases (James 2009;Marvel 2016). Thus, conceptualizing and measuring how public services are delivered in terms of 'citizen choice' remain problematic in both theory and practice.
A growing body of individual behavioural changes is important to extend policyrelated discussions and classical public administration theories (Schneider and Ingram 1990;Grimmelikhuijsen et al. 2017;Barnes 2020). Although extant research repeatedly notes that such evidence provides insights into the practice of public management, it remains scant in the field of public management. Therefore, how much citizens change how they exercise public programme-related choices over time can be discussed in the context of neighbouring disciplines such as economics and public health service. This can complement the subjective measures of government performance and connect them with theory and practice within public management.
In line with this, this article derived insights based on 'bypassing' hospital choices concerning potential service accessibility. 'Bypassing' refers to patients not choosing a nearby hospital in favour of one farther away to seek healthcare services outside their local community. This is consistent with the concept discussed in the public administration literature regarding 'exit response', a reactive replacement of one service provider with another due to dissatisfaction with public services (Gofen 2012;Jilke, Van Ryzin, and Van de Walle 2016).
Extant studies on health policies examined bypassing choices in many countries such as Europe (VrangbAEk et al. 2007) and the US (Bronstein and Morrisey 1991), in different regions such as rural (Tai, Porell, and Adams 2004;Liu, Bellamy, and McCormick 2007) and urban areas (Haynes, Lovett, and Sünnenberg 2003). Overall, they found that a higher bypass tendency demonstrates patients' choice to seek healthcare they perceived to be better despite spatial access barriers. Hence, patients whose medical expenses are reimbursed through voucher-based assistance programmes can choose to travel long distances to receive better services.

Inequality in exercising choice of public services
There are concerns that policies to extend choices may worsen or at least not alleviate existing inequality in service access. Much of the controversy surrounding choice policies stems from concerns about the inequality across citizens.
Therefore, this section elaborates healthcare inequality since inequality can be defined differently depending on which dimension is emphasized and many interpretations of inequality are possible in the programme contexts. In health services, wealthier, white, or female patients have a greater likelihood of bypassing or being given broader spatial access to service providers and older patients strongly prefer closer hospitals (Tai-Seale, Freund, and LoSasso 2001;Allard, Tolman, and Rosen 2003). This study reflects those that consider healthcare inequality as deeply intertwined in the differences between various SES groups concerning income, age, race, and gender; possibly because SES embodies an array of resources such as money, knowledge, prestige, power, and beneficial social connections. In this vein, extant research on health services claims that there may be a link between choice-related polices and SES inequality (Cookson et al. 2010).
Accordingly, this study argues that the health voucher programme beneficiaries with low SES may reap fewer benefits since they experience more problems in exercising their choices. In the process of programme implementation, Herd et al. (2013) found that administrative burden on individuals can lead to problems of taking up Medicaid in Wisconsin, United States. This can occur because an individual action depends on not only direct (e.g. medical and distance costs) but also indirect (e.g. information or switching costs) costs. Indirect costs are associated with beneficiaries' ability to handle administrative burdens such as learning, psychological, or compliance costs (Moynihan, Herd, and Harvey 2015;Barnes 2020).
However, only a handful of empirical studies on this issue exist in public management and show inconclusive results. For example, in European public infrastructure services, a market-oriented policy was less effective for the disadvantaged in their complaint behaviours, as they may be less able to exercise their right to choose (Clifton et al. 2011;Jilke and Van de Walle 2013); however, in British national health services, an extant study found that competition did not exacerbate SES inequality during the English National Health reform in the 1990s at an area level (Cookson et al. 2010).
Nevertheless, in the context of public health management, there exists little microlevel evidence of how such programmes lead to actual individual behavioural changes. The aforementioned literature suggests that patients with low SES are more likely to confront access barriers to exercise their choices of hospitals, which may not be alleviated or even worsen when they benefit from Medicaid, compared to non-Medicaid or uninsured patients.

Medicaid in Brooklyn in the 2000s
This article employed a retrospective analysis of data from the 2000s for several reasons. First, the Medicaid programme in the United States is historically the longestlasting consumer-driven purchasing policy, having operated and evolved since 1965 (Gruber 2003). It is the mechanism through which state and local governments provide services to enhance low-income patients' choices and to improve the purchasing relationship between the government, providers, and patients. In the programme, patients are given a Medicaid entitlement that they can present at a hospital of their choice to receive government-provided health insurance funds that align with their choice (Daniels and Trebilcock 2005;Le Grand 2011). Consequently, the programme offers a unique opportunity to observe the evolution of a classic example of this type of voucher-based assistance over a few decades.
Second, the 2000s are a particularly reasonable period to observe how Medicaid has evolved to its current status regarding citizens' access to care. The programme was modified by the DRA of 2005 in ways that negatively affected citizens' access to care (Rosenbaum 2006). Furthermore, few people were required to purchase health insurance, and the number of people eligible for Medicaid had not yet grown before the ACA of 2010 was passed. Thus, patients with low SES had not yet secured the insurance that the ACA made possible, allowing both inpatient and outpatient care more affordable. Moreover, substantial inequalities existed within the US healthcare markets in the 2000s (OECD 2015).
Additionally, Brooklyn, New York, during the 2000s experienced a movement of change that possibly affected service accessibility, making it a good case. The state of New York has a history of being relatively liberal in expanding Medicaid eligibility compared with other states, partly because of the eagerness of state officials to avail themselves of federal funds (Gusmano, Burke, and Thompson 2012). Nonetheless, this commitment to providing generous health and social welfare benefits has had a decisive effect on the scope and structure of the Medicaid programme as well as the regulation of private health insurance (Ward 2008;Gusmano, Burke, and Thompson 2012).
Moreover, among all of New York City's boroughs, Brooklyn had the highest number of residents eligible for, but not covered by Medicaid, during the time analysed (New York State Department of Health [NYSDH] 2003). More importantly, Brooklyn is particularly interesting because there was little change in the large number of hospital facilities in the 2000s, meaning that a given set of hospitals driven by the supply was relatively homogenous for patients.
Therefore, Brooklyn was utilised as a rough proxy for a local market, consistent with (McLafferty and Grady's 2005) study using Brooklyn as a single analysis unit.

Data and measures
Data were utilized from the following years : 2003, 2004, 2008, and 2009. They were primarily obtained from the New York State Bureau's Statewide Planning Research and Cooperative System inpatient data, which contain information about all admissions that occurred in Brooklyn, including patients' attributes of the admissions and an identifier for the hospitals to which they were admitted. However, given that it does not collect information regarding income, this study retrieved income-related information at the zip code level from the US Census Bureau and US Zip Code Database. The information on attributes of the selected hospitals is from the American Hospital Association, which includes hospital-level data.
Next, a particular diagnosis is specified by diagnosis-related groups (DRGs) since a certain diagnosis can restrict which doctors and providers a patient can see. This makes it possible to determine whether hospital choice corresponds to the patient and to lessen the influence of professional beliefs and attitudes (e.g. a patient's cardiologist preferring a particular hospital) and other hospital-physician affiliations. Patients with a particular diagnosis similarly need a doctor's referral or an insurance company's pre-approval to seek care at a specific hospital. Among the diagnoses, cardiac diagnosis is identified as these patients must make a prudent choice to ensure they select a well-equipped hospital. Further, hospitals are generally not allowed to refuse care to patients during a medical emergency. Moreover, coronary disease and stroke are known to be higher among people and communities with low SES (NYSDH 2003;Northwell-Health 2016).

Dependent variable
The 'bypassing choice of a provider' variable is a dichotomous identifier variable assigned a value of 1 when an admission of a patient to a hospital occurs outside the borough and 0 when an admission is from within the borough. Bypassing behaviour is measured by the choice made from the set of providers patients can choose rather than the distance from their selected provider.
Bypassing represents how individuals experience or access services since choosing a provider means selecting one of all possible alternatives from which patients can choose. In the context of the health service markets with highly differentiated services, markets tend to be local. Therefore, it is important to observe that patients travel from their residential areas to the areas where the required services are provided (Zwanziger, Mukamel, and Indridason 2002).

Explanatory variables
The explanatory variable, Medicaid programme, is assigned a value of 1 if an admission specifies Medicaid as the primary or sole source of medical insurance and 0 otherwise. Two counterpart samples are created. The first is designated as a non-Medicaid group that includes dummy variables for each of 5 insurance types, Medicare, private insurance, Blue Cross, Workers' compensation (which covers employment-related injuries), and Others including federal programmes, and the Civilian Health and Medical programe of the Uniformed Services (CHAMPUS). The second includes only the uninsured (self-pay).
Next, to measure patients' potential vulnerability, SES variables are included such as income, age, race, and gender, which play an important role in the health market. Low SES was measured as follows: for the income variable, annual household income was used; for the age variable, admission information was collected from patients older than 17 years because they, not their parents, make their own choices; for the race variable, 1 indicated admissions of African American patients and 0 admissions of others, and dummy variables are included for each of 3 races (Caucasian, Hispanic, Asian or others); for the gender variable, 1 indicated admissions of female patients and 0 of male patients. These factors are only acceptable for assessing SES when considering bypassing behaviours in the health market. The variables and measures used are presented in Table 1.

Control variables
Control variables were divided into demand and supply-side factors affecting bypassing choices.
First, regarding the demand-side factors, whether the admission required medical or surgical DRGs and the duration of hospitalization were included to capture the patients' level of medical severity. It is because severely ill patients would require a hospital with more resources and care facilities, potentially influencing their bypassing behaviour. In addition, the nature of the admittance-related situations encountered was investigated. For example, in an emergency, people have little time to research the best option or to assess who has the best hospital care; consequently, they may choose what the closest hospital offers. The emergency is measured by the admissions of cardiac with acute myocardial infraction (DRGs 222-223) and of acute myocardial infraction (DRGs 280-285) in the total cardiac diagnoses admissions (DRGs 215-316) analysed.
Second, regarding the supply-side factors, information regarding hospital attributes was collected such as hospital ownership to control for risk selection or cream skimming by providers concerning ownership (Norton and Staiger 1994;Boonen, Donkers, and Schut 2011;Kang, Kim, and Jung 2020), and the number of beds and outpatients (Roh and Moon 2005).

Methods
A triple difference (difference-in-difference-in-differences, DDD) approach was utilized. The standard difference-in-difference (DD) compares two groups (difference estimator), programme (treated) and non-programme (untreated) in this case. However, because no observations were available from the pre-Medicaid period, this study specifies the changes in outcome measures and compared treatment and control group outcomes for the earlier and later time periods along with before and after the DRA. The treatment group consists of admissions from Medicaid patients, for whom the programme influences the outcome, whereas the control group consists of others from non-Medicaid or uninsured patients, for whom the policy has little effect. The DDD allows for changes of time trends in outcomes across different SES levels within the treatment group to be compared to their counterparts in the control group over time.
To draw conclusions that are as robust as possible, some assumptions were made. First, the treatment and control groups were assumed to differ only according to when the programme operates, meaning that no contemporaneous shock affect the outcomes, and the impact of treatment at the starting point was invariant. Thus, This can demonstrate how outcomes evolved in multiple periods. Second, the model could control for potentially confounding trends such as changes in outcome trends across different SES levels. Thus, the affected SES levels had the same attitude towards bypass but were immune to different attitudes across treatment or control groups. The identifying assumption was fairly strong in this approach.
The model combined two strategies to account for the effects of Medicaid and low SES-specific trends in earlier and later periods. The first strategy was to compare the Medicaid and non-Medicaid groups' outcomes and examine the DD between the outcome measures for the earlier and later periods, along with before or after the DRA. In equation (1) of the DD estimator γ (Δ 2 T ), T denotes Medicaid, and NT denotes non-Medicaid admissions. I T is a binary variable that equals 1 for a Medicaid admission. I T y represents a Medicaid admission i in year y; yn and y0 denote the later and initial years, respectively. The binary variable used equals 1 for a bypassing admission and 0 otherwise: The next strategy was to compare the outcomes of low SES with non-low SES admission groups, according to the earlier and later periods. Changes in the bypassing admissions that did not belong to a low SES group were also considered, resulting in a DDD estimator α M (Δ M 3 ). In equation (2), MT denotes low SES, and NMT denotes non-low SES. The first differences (Δ 2 MT ) are identified by the difference between the Medicaid and non-Medicaid bypassing admissions of low SES groups over time. The second differences (Δ 2 NMT ) include the differences between the Medicaid and non-Medicaid bypassing admissions of non-low SES groups. Then, the outcomes were compared for the different SES levels within the same treatment group to those within the same control group.
Equation (3) was estimated using linear models based on admission-level data. The DD coefficients γ 1 indicate the specific Medicaid time effect on the low SES group's propensity towards bypassing, with the reference year being 2003. M iy is a binary variable that equals 1 for a Medicaid admission, and S iy is a full set of SES variables for an admission i in year y, Y iy are year dummy variables (earlier and later Medicaid, before and after DRA). The DDD coefficients α M account for the time changes of gaps in outcomes across different SES levels in the treatment group relative to their counterparts in the control group over time with the reference year being 2003.
Here, z iy are the admission attributes of the medical condition from patient at time y (care types, length of stay, or emergency), and p ijy are the admission attributes from hospital j at time y (ownership, size, and capability), and ε ijy is the idiosyncratic error term. probðI iy j �Þ ¼α 0 þα 1 M iy þα 2 S iy þα 3 Y iy þγ 1 ðM � SÞ iy þγ 2 ðM � YÞ iy þγ 3 ðS � YÞ iy þα M ðM � S � YÞ iy þβ 0 z iy þβ 1 p ijy þε ijy (3) Next, as an example of patient medical condition attributes, Table 3 depicts the numbers of bypassing choices of providers in emergency situations. As expected, it shows that patients with more urgent needs were less likely to bypass nearby hospitals than those with less urgent needs.  In addition, as supply-side factors bypassing choices, Table 4 presents the attributes of hospitals. A total of 130 hospitals were available to patients, comprising 89 hospitals selected by bypassing and 41 by non-bypassing during the period considered. The yearly number of hospitals by non-bypassing was 10 in 2003, 2004, and 2008 and 11 in 2009, indicating that choice driven by the supply side was fairly stable in the analysed market. The hospitals chosen by bypassing comprised more non-profit than public hospitals, larger hospitals, and hospitals with more capacity. This coincides with previous findings in which patients were more likely to bypass hospitals if those hospitals were publicly owned, since patients perceive public hospitals as smaller, less technologically advanced, with fewer specialists, and providing only basic healthcare services (Bronstein and Morrisey 1991;Escarce and Kapur 2009). The descriptive statistics of all the variables used in the analyses are presented in Table 5. Table 4. Hospitals' attributes (Years: 2003, 2009  Bypassing trends Figure 1 illustrates the time trends in bypassing rates regarding low SES (solid line) and other groups (dashed line) over time. The figure reveals how bypassing rates changed in the Medicaid (right graph) versus the other group (left graph) with connected-line plots over the period. The trends in the uninsured low SES group confirm the expected results for the low SES Medicaid group. Figure 1(a) illustrates the bypassing rate trends between admissions of lowincome (< median) and others in the Medicaid group, which remains the lowest of all the included groups over the period except for the year 2004. Specifically, the bypassing rates of low-income groups continued to increase in both Medicaid and uninsured groups before the DRA of 2005 when Medicaid eligibility was restricted, while they appear to stop increasing in both groups after 2005. The gaps in time trends of bypassing rates between admissions of low income and other individuals gradually increased in the Medicaid group after 2005, whereas those between admissions of low income and others in the uninsured group somewhat decreased after 2005.  Figure 1(b) shows that the bypassing rate trends of older adult (aged > 65 years) admissions in Medicaid were the lowest of all the groups, and this low trend remained steady over the years. The trends of older adults in Medicaid did not increase considerably compared to those in the uninsured group regardless of the DRA. Figure 1(c) demonstrates that the bypassing rate trends of African American admissions were lower than those of other admissions both in the uninsured and Medicaid groups. Interestingly, the bypassing rate gaps between African Americans and others appeared considerably stable and slightly increased in the Medicaid group. However, the gaps decreased in the uninsured right after 2005, which presents a contrast with those in the Medicaid group.

Descriptive statistics
In Figure 1(d), the trend for rates of female admissions contrasts with those of male patients: the rates of female admissions appeared to decrease after 2005 in the Medicaid group while those of male admissions in the same group continued to increase. These trends contrast with those for the uninsured, as uninsured female admissions continued to increase before 2005.
In general, the bypassing rates of the Medicaid group were found to be lower than those of the insured during the period, and those of low SES individuals were observed to be lower than those of the other group. Overall, all graphs demonstrate that extending choices was less effective for low SES admissions than for other admissions in the Medicaid group when compared with the number of bypassing admissions in the uninsured group. Table 6 depicts the results of whether Medicaid influences the disparities in hospital choice between Medicaid and the non-Medicaid or uninsured groups over time, accounting for the control variables including their medical condition, or hospital attributes. The impacts were estimated with the samples of Medicaid and non-Medicaid groups with patient attributes in column (1) and with both patient and hospital attributes in column (2). Next, the impacts were estimated in columns (3) and (4) in the same models with the samples of the Medicaid and uninsured group, an alternative counterpart.

Effects of Medicaid on bypassing choices
The DD estimators γ signs indicated the Medicaid bypassing admission trends in 2004, 2008, and 2009, compared with that in 2003. The findings revealed that no positive trend appeared in any year. Therefore, it is unclear from the admission trends whether low SES Medicaid patients benefited from the wider range of choices driven by the increased purchasing power of the Medicaid programme.
However, the first DDD estimators α M1 positive signs indicated that bypassing admission is more common for relatively wealthier Medicaid admissions than for other non-Medicaid (or uninsured) admissions. These were observed in all columns, but some in columns (1), (2), and (4) are significant. The choice to bypass was less likely to occur with low-income Medicaid admissions than with uninsured admissions during the period.
The second DDD estimators, α M2 in columns (1), (2), and (4) are negatively associated with the dependent variable, bypassing the choice of provider, during the period analysed. The negative signs in columns indicated a smaller time trend for older Medicaid bypassing admissions than for other Medicaid, non-Medicaid, or uninsured admissions. Interestingly, this was more likely to happen in 2008 after the DRA of 2005 was enacted.
Similarly, the negative α M3 signs in columns (1) through (4) indicated that the time trend for African American Medicaid bypassing admissions is smaller than for other Medicaid or African American non-Medicaid (or uninsured) admissions. As expected, it turned out to be significant in 2008 after the DRA of 2005. This suggests that programme eligibility substantively and negatively influences African American Medicaid bypassing admission.
In addition, the negative α M4 signs in columns (1) and (2) indicated that female Medicaid bypassing admissions are less likely to occur than other admissions; this result remained in columns (3) and (4) but is insignificant.
The findings show that low SES Medicaid patients are less likely to display bypassing admissions over time, indicating that they still face a narrower hospital choice than others. This supports the hypothesis that Medicaid may not broaden low SES inpatients' hospital choice compared to non-Medicaid or uninsured low SES inpatients over time. Notably, the findings also reveal that these trends held over time. It was expected that the affordability offered by the programme would be effective in broadening hospital choice for low SES patients; however, this is not apparent. Some Medicaid beneficiaries still face geographically narrow choices of hospitals, which may represent their limited choices. This is an interesting finding because it confirms that Medicaid may be insufficient to alleviate existing SES inequalities that could limit patients' access to better hospitals compared to other coverages. Further, these results elucidate whether Medicaid may reproduce inequalities in accessing health services that are embedded in factors such as income, race, or gender over time. Although Medicaid attempts to improve low SES patients' healthcare choices, this programme requires some support to be more inclusive to allow patients to enjoy the potential benefits provided by choosing more accessible hospitals.

Discussion
This study provides evidence of how the impacts of the Medicaid programme on low SES and other patients evolved from 2003 and 2004 to 2008 and 2009, including the period before and after the DRA of 2005, compared with the non-Medicaid or uninsured groups. The findings revealed that the disparities between hospital choices of low SES and other admissions in the Medicaid group have widened over the analysed period compared to those in non-Medicaid insurance or uninsured groups. This indicates that the programme's intended goal regarding extending beneficiaries' healthcare choices may not have been achieved, since low SES beneficiaries benefited less because the care needed to prevent or maintain certain health conditions is spatially confined for them.
The findings have key implications. First, they provide individual-level micro evidence on whether the programme beneficiaries' access to services improved. Extant research on voucher programmes found overall positive impacts on individual behaviours; for example, mobility to a better place in the case of housing vouchers or higher test scores in the case of educational vouchers. However, scant research explains where the programme benefits lie among beneficiaries. Findings show that the programmes do not necessarily guarantee greater equality in accessing services among beneficiaries, and they may leave low SES beneficiaries with narrower hospital choices when retrenched. Although this is not inherent in the voucher scheme per se, as voucher programmes come in various designs, it is possible to appropriately evaluate the effectiveness of public service management at the time when programmes are operating. Therefore, public managers should closely observe the service management process to understand whether the beneficiaries of a service are those who need them most, which user attributes regarding limiting choices of providers are exacerbated by such public programmes, or whether there exist administrative burdens related to the programmes. These key elements could play a critical role in assessing the programme management performance from the beneficiary' point of view. More attention should be paid to the nature of these costs when considering citizens' needs for services and the utilization of more adequate services.
Second, along with the abovementioned efforts to improve such programmes, additional management strategies must be constructed from the supply side. Public managers should internalize the administrative inconvenience as well as the costs pertaining to the existence of inequality in a market of public services, and not just focus on reducing the operational costs by unleashing market forces for public service organizations to be efficient. A higher level of governance of organizational relationships is needed in the implementation process. For example, governments should develop well-matched performance indicators for public service organizations such as inducements for those who provide appropriate services for the beneficiaries most in need or enforcement of strict penalties for those who refuse services to them. This is intended to secure citizens' spatial access to public service providers before putting them in a position where they make choices autonomously.
However, the findings may be affected by several limitations. First, other SES variables may also influence hospital bypassing admissions, such as social class or education. Although samples collected over long periods allow us to relax the common trend assumption by introducing a degree of non-parallel evolution in the outcomes between Medicaid and non-Medicaid (in the absence of a specific treatment effect), the fact remains that the samples were not the same initially. Second, the supply-side factors that affect provider choices were not fully captured due to apprehensions regarding potential insurance payment issues (e.g. provider network's refusal to include a particular hospital or to accept a particular type of insurance), or whether Medicaid patients used traditional fee-for-service or managed-care Medicaid plans as their payment method was not differentiated due to data unavailability. Third, bypassing admissions were measured by the relationship between the location of a patient's residence and that of the selected hospital. However, a patient's work location, physician's location and relatives location may influence hospital choice, particularly if those locations are not close to the patient's home.
However, to the best of my knowledge, the results reported herein are the first to offer evidence of how a voucher-based assistance programme has influenced citizens' hospital choices over the years. Further, there exist interactive effects between the programme and strata of citizens before and after the programme shrank. The results can aid public management scholars in making theoretically and empirically grounded contributions to the practice of such programmes in public management. Further research, including one with a longer period of analysis after the ACA, is necessary to verify the impact of such programmes and generate new and better theories of how such programmes decrease choice restriction disparities.