A key criterion for evaluating policies to expand health insurance coverage is weighing the costs of such policies against the willingness of the public to pay for coverage expansions. We use new panel survey data from New York State to estimate residents' willingness to pay (WTP) to expand public insurance coverage. Using a nonparametric double-bounded contingent valuation (CV) approach, we specifically ask residents about their WTP to reduce the rate of uninsurance in the state. Our results imply an aggregate lower-bound WTP of over $2,800 per year to cover one person. We also analyze heterogeneity in WTP by sub-group and changes in individual WTP over time between 2008 and 2010. We find that a large majority of residents are willing to pay additional taxes to reduce the number of uninsured in the state, and that average WTP remained remarkably stable despite the economic downturn and the politically polarized discussions surrounding the Affordable Care Act. Decomposing the changes in individual WTP, we find that economic factors related to the recession, including changes in income and employment status, cannot explain changes in individual WTP, whereas individual changes in political opinions about health insurance reform between 2008 and 2010 are strongly correlated with changes in WTP. (JEL H20, H42, H51, H75,113)

I. INTRODUCTION

Public expansion of health insurance coverage has been a subject of intense debate in the United States at both national and state levels, especially in recent years (Epstein et al. 2009; Gruber 2008). The 2010 Affordable Care Act (ACA) puts

into motion substantial further increases in public health insurance, most of which take effect in 2014. One reason for the controversy surrounding the issue is that the costs of reform are substantial and salient, (1) while the benefits are difficult to estimate, and may include direct components from improving access as well as indirect benefits associated with insurance against future risk and altruistic concerns about fairness and equity.

The high costs of reform make it important to assess not only whether people support reforms in principal, through polling or election results, but also to estimate their willingness to pay (WTP) higher taxes to finance reforms. That is, as Kessler and Brady (2009) put it, it is important to examine the willingness of the public to "put their money where their mouth is." In addition, there has been recent interest among policymakers and academics in examining how economic recessions affect support for tax-funded social safety nets (Russell Sage Foundation 2011), which can be crucial for the long-run viability of such policies. Finally, we also test whether public policy debates prior to the passage of the ACA may have changed WTP directly by affecting opinions about public insurance expansions.

Despite these concerns, there has been relatively little systematic analysis of WTP for health reform. A handful of opinion polls have asked whether respondents support increasing taxes to fund health insurance reform, although none have been longitudinal studies of the same individuals. Apart from one poll by Kessler and Brady (2009), these polls do not specify the size of tax increases and do not elicit WTP specifically. (2)

In their analyses, Kessler and Brady (2009) use a contingent valuation (CV) approach and national survey data to measure support for selected reforms. Specifically, they examine expansions of coverage that would be financed through proportional increases in individual income tax payments. Using individual-specific estimates of required tax increases based on reported income, each respondent is asked whether they would support a coverage expansion. Each respondent is only asked a single "yes/no" question about their proposed tax increase. However, the survey is repeated for three selected reforms--a subsidy for private insurance, an expansion of Medicaid, or coverage of individuals with chronic illness. In each case, some individuals are asked if they would be willing to pay amounts that would cover 50% of the estimated cost of reforms, while others are asked about amounts that would cover 100% or 150%. Using this approach, Kessler and Brady compute the percentage of respondents answering "yes" for each reform/payment option. They do not, however, attempt to use their data to make any dollar estimates of WTP either per respondent or per covered beneficiary.

On the basis of their analysis of the percentages of respondents supporting reforms, Kessler and Brady (2009) conclude that the lower the estimated cost of reforms, the higher the percentage of respondents willing to pay for reforms, but that except for coverage of the chronically ill, only a minority of individuals are willing to pay the increases in their taxes Kessler and Brady estimate will be required to fund the programs. They also find that the percentage of respondents willing to support reforms is lower at higher incomes. However, the implication of this finding for the relationship between income and absolute WTP is difficult to evaluate because the amounts asked in Kessler and Brady's WTP questions are dependent on both income and the progressivity of average federal income tax rates at the time (higher income individuals are asked about their WTP larger amounts) and the authors do not report any dollar estimates of WTP.

No study to date has estimated individual WTP by administering a set of double-bounded questions. Nor has any study looked directly at relationships between income and WTP holding constant the amounts asked, or estimated how other demographic factors may affect heterogeneity in individual WTPs. There also has been no attempt to validate concordance between WTPs and stated opinions regarding health reform.

The passage of the ACA in 2010 shifted attention towards national reforms, whereas during the prior two decades states had been very active in health policy. However, attention has recently shifted back towards the states, which, since the U.S. Supreme Court ruling in 2012 regarding Medicaid expansions, face decisions about the willingness of voters to support state-specific public insurance expansions (KFF 2013a). New York State in particular is of interest because it is a large, wealthy state and has taken a leadership role in establishing a state insurance exchange as well as in expanding Medicaid coverage (KFF 2013b).

In this paper, we use a CV approach to estimate willingness of New York State residents to pay higher taxes to fund public policies to reduce uninsurance in the state. Specifically, we consider three questions. First, what is the WTP higher general taxes for hypothetical reforms to reduce the rate of uninsurance in the state, both overall and specifically for low-income individuals? Second, how does WTP vary with individual sociodemographic characteristics and with views on specific types of health reform proposals? Third, how does WTP change longitudinally, and what factors, including changes in economic conditions and opinions about reform, may be correlated with those changes?

In exploring these questions, we seek to contribute to the literature on WTP for reforms in several ways. First, we use a double-bounded CV approach to provide lower-bound estimates of aggregate WTP for public health insurance expansion. Second, in addition to looking at relationships between WTP and sociodemographic characteristics, we consider how WTP corresponds with individuals' self-reported views on reform. Third, we use standardized longitudinal data to consider changes in WTP for reforms and their correlates. To our knowledge, all three aspects of this research are new to the literature.

We find that there is broad-based WTP to reduce uninsurance among New York State residents. The lower-bound WTP amounts are substantial when judged against several possible cost estimates of covering the uninsured. For example, in 2008, for both our overall and low-income proposals, more than 70% of respondents indicated a WTP at least $25 in higher taxes annually to reduce uninsurance in the state by a quarter. Our lower-bound estimates of the average WTP higher taxes are about $95 and $90 annually per resident for the low-income and overall referenda, respectively. Taking into account the number of residents and the number of individuals who are uninsured in New York State, these estimates imply average lower-bound WTPs of about $2,881 (3) and $3,465, (4) respectively, to insure one additional person for one year. (5)

There are several different cost benchmarks that one could compare with these estimated benefits. One approach is to compare our estimates of WTP against Medicaid costs. Medicaid costs are relevant not only because they provide a general indicator of the possible costs of expanding coverage, but because state decisions about implementing Medicaid expansions has emerged as a major policy issue under the ACA. (6) The national average cost of Medicaid coverage in 2009 was $2,900 for an adult and $2,305 for a child, and the New York State average costs were $4,277 and $2,505, respectively. (7) Another possible benchmark is the marginal cost of Medicaid expansion under the ACA, rather than the average cost per beneficiary prior to expansion. Estimates based on analyses by the Urban Institute (2012) suggest that marginal costs of expanding Medicaid in New York State would be about $1,080 per beneficiary coverage-year. (8) An alternative benchmark is the cost of obtaining coverage through a state exchange. Under approved rates for individual coverage effective 2014 for the New York State Benefits Exchange, the cost of the most affordable standardized "Bronze" plan will be $3,029 per year. (9) Thus, our 2008 lower-bound WTP estimates of $2,881-$3,465 suggest that New York State residents would be willing to pay amounts to expand coverage that are comparable to the cost of either traditional Medicaid expansions or expansions through private insurance purchased on the state exchange.

The national average premium for private group coverage was higher however, at $4,669 in 2009, as reported in the Medical Expenditure Panel Survey (AHRQ 2009).

In addition, we find WTP (measured in absolute dollars) is higher when indicators of socioeconomic status (SES) are higher and that individual WTP is correlated in expected ways with opinions about support for health insurance reform policies, consistent with individuals being willing to "put their money where their mouths are." As noted, we also find that average WTP remained remarkably stable between 2008 and 2010, despite both the onset of a major recession in late 2008 that peaked in 2010, and an intense debate over the ACA in 2009 through early 2010 that could have shifted attention towards national reforms. There is, however, considerable variation within individuals in their WTP over time. Interestingly, this individual variation is not explained by changes in economic factors, such as employment and health insurance coverage, and is more strongly correlated with changes in individual opinions about specific health insurance reform mechanisms.

We briefly describe our conceptual framework for analyzing WTP in Section II. Section III describes the data, while Section IV describes our estimation methods and findings. Section V discusses study limitations, and Section VI concludes. II.

II. CONCEPTUAL FRAMEWORK

The primary assumption in this research is that an individual's reported WTP for the proposed policy captures information about their underlying preferences for the policy that would not necessarily be captured by measures of the direct impact of policies, for example by estimates of the impact of an expansion in health insurance coverage on recipient health outcomes. A person's preferences for expanding coverage may be influenced not only by direct self-interest, but also by altruism. Thus, individuals may be motivated by "warm-glow" or altruistic benefits related to their beliefs about the appropriate roles of government, fairness and equity, and this may be reflected in their WTPs. In addition, of course, preferences may reflect individuals' own self-interest and their current and anticipated future coverage. Individuals who are currently uninsured may have a direct interest in expanding coverage. For example, Krueger and Kuziemko (2013) use a similar survey elicitation strategy to estimate the willingness of the uninsured to pay for their own coverage and find that over 60% of the uninsured would be willing to pay $2,000 per year to purchase insurance if that option were available to them immediately. Likewise, individuals who currently have private coverage may have a preference for expanding coverage as insurance against the risk of becoming uninsured in the future. In the case of individuals with public coverage, it is possible they may support expanding coverage as they have already benefited from such coverage, but it is also possible that they could oppose it if they believe that an expansion could be accompanied by cutbacks in existing benefits.

In order to gain a deeper understanding of preferences, and to verify that responses plausibly capture preferences, we also estimate how WTPs vary with sociodemographic characteristics such as age, gender, race, and educational attainment, as well as with general political opinions and specific opinions about several health insurance reform proposals.

A natural starting point is to question how WTP changes with income. Our discussion of preferences for reforms to expand coverage suggests two possible hypotheses. The first is that WTP may increase with income for altruistic reasons. Consistent with this is research on the relationship between altruism and income such as Andreoni (1990) and Feldstein and Taylor (1976), which finds that among high-income groups, donations to public goods increase with income. However, offsetting this effect, WTP for self-interested reasons may decrease with income if higher income individuals are less likely to be uninsured or are less likely to anticipate potentially benefiting from expanded public coverage; although we cannot disentangle the two types of effects, we can estimate their net impact. In addition to income, we also examine how other components of SES (including insurance coverage) relate to WTP, conditional on income.

In the face of concerns that CV analysis elicits respondents' true preferences, responses to opinion questions about individuals' political convictions and support for particular types of reforms provide possible indicators of their concordance. Although these opinion measures are often highly correlated with income, education, and other SES measures, opinion and political ideology polls are widely believed to track closely with actual behavior in voting, which is the relationship we are interested in validating. In general, we expect that self-identified liberals will be more likely to support a publicly funded expansion of coverage than conservatives (KFF). Following Bundorf and Fuchs (2008), we hypothesize that those with more favorable views of government intervention may be more likely to support universal coverage, suggesting that WTP will be higher among individuals who support reform options such as universal Medicare or universal mandates. (10)

Over time, WTP may be affected by changes in respondents' economic status and insurance coverage and by changes in their opinions about reform. We collected data from our respondents in two periods: the first quarter of 2008 just prior to the onset of the 2008 recession, and the first quarter of 2010, just prior to the passage of the ACA, a period of intense policy debate about reform. We hypothesize that economic factors may impact WTP directly through effects on employment, income, and insurance coverage. If personal resource availability falls, this may reduce support for expanding coverage for altruistic reasons, but support may increase for self-interested reasons to the extent individuals become or expect to become uninsured so that the predicted net impact is ambiguous.

At the macro level, the economic downturn may have the increased perceived need for reform, but also raised concerns about the ability to pay for it. Concurrently, the policy debate about the ACA may have increased awareness of issues surrounding health reform, increasing intensity of opinions pro and con, while shifting attention away from state-level reforms towards national reforms. While we lack a way to disentangle these various effects, we can test the extent to which WTP is associated with changes in employment, income, insurance status, and opinions about reforms at the individual level.

When attempting to generalize results from one state, it is important to keep in mind that WTP estimates may be affected by factors specific to New York State. For example, the state has a highly regulated insurance market, which may cause respondents to believe that high WTP is needed to afford policies in that state. It is also possible that respondents rarely interact with the individual health insurance market and are not aware of specific costs of coverage. New York is also an outlier state in tax policy, and this could lead to either lower or higher WTP for coverage reforms, relative to other states. These factors should

all be kept in mind when considering the results of a single state study. However, New York State's high level of policy activity also makes it important to investigate in light of the ACA implementation process.

III. DATA

Our analyses use data from three surveys of New York State residents. The first two are state-representative cross-sectional surveys, each with 800 respondents, conducted in the first quarters of 2008 and 2010 as parts of the Cornell University Empire State Polls (ESPs). (11) The sampling frames were split equally between upstate and downstate residents to allow comparisons between these geographic sub-regions, and sample weights allow us to produce estimates that are representative of the state as a whole. Within the sample frames, phone numbers were randomly selected, giving each household an equal chance of being included. Within each household, the surveyor asked to speak to the adult resident whose birth date was closest to a randomly selected date, giving each adult within the household an equal chance of being selected, thus making it an individual-level survey. (12) In addition to the 2008 and 2010 cross-sectional surveys, we commissioned a special follow-up survey in the first quarter of 2010 in which the 2008 ESP sample was resurveyed, creating a longitudinal sample of 411 of the original 800 respondents.

The survey was targeted at individuals, rather than at households. As a result there was a large proportion of individuals who were randomly selected to participate in the survey, but whom we were unable to successfully contact to request their participation. Of those whom we successfully contacted, 69.3% completed the survey in 2008. (13) Similarly, for the 2010 ESP and 2010 follow-up survey, 72.9% and 96%, respectively, of those who were contacted completed the survey. The higher completion rate of the 2010 follow-up sample is likely due to the fact that they had already completed the 2008 survey. The vast majority of nonresponses to each survey was due to bad or disconnected telephone numbers, or because after at least six attempts there was no successful contact. Including potential respondents that we were never able to contact to request participation as nonresponders, the response rates were 23.3%, 34.6%, and 51%, respectively.

The first round of the survey was conducted during the first quarter of 2008, just after the peak of the business cycle in December 2007, when the unemployment rate was under 5% in New York State and the share of nonelderly state residents without health insurance at some point during the year was 13.8%. The second round was conducted during the first quarter of 2010, just before the ACA was signed into law. By this time the state unemployment rate was at a peak of 8.9%, while the share of nonelderly state residents without health insurance had risen to 16.7% (BLS 2012; U.S. Census 2012).

As our main outcome measure, respondents to the ESP were asked a set of double-bounded questions about their WTP for a hypothetical policy that would reduce the rate of uninsurance in the state by one-quarter. In order to isolate support only for an increase in the prevalence of health insurance, our WTP survey question intentionally avoided discussion of the specific reform policy that might be used to potentially achieve such results. Two slightly different variants of this WTP question were asked. In 2008, half the respondents were asked if they would support a referendum to reduce the overall rate of uninsurance in the state by one-quarter ("overall referendum"). The other half were asked if they would support a referendum to reduce the rate of uninsurance by one-quarter among those in the state who earn less than 300% of the FPL. We refer to this as a targeted "low-income referendum" in the rest of the text. In 2010, to minimize respondent burden, and upon seeing data from 2008 suggesting that the difference in the wording of the question appeared to have very little effect on responses, respondents to the 2010 cross-sectional survey were only asked about the "overall referendum." However, the 411 followup respondents in 2010 were asked the same version of the question that they answered in 2008. (14)

Our surveys also include questions about demographics, education, employment, income, health insurance coverage, and political philosophy (liberal, middle of the road, or conservative). Respondents were also asked a series of questions eliciting opinions regarding health reform, based on a similar survey conducted in New Jersey by Cantor et al. (2007) and modified for New York State by Simon and White (2008). Opinion questions that we use in our analyses include: how important it was for state government leaders to work to reduce the costs of health insurance ("cost important") and the number of uninsured ("# uninsured important"); whether respondents support mandates requiring large employers to provide health insurance to their employees ("large employer mandate"); whether they support a universal health insurance mandate ("universal mandate"); and, in 2008 only, we asked whether they support allowing high-deductible plans to be offered in New York State. (15) Some opinion questions, such as "cost important" and questions about particular proposals, use a 4-point Likert scale. In these cases responses are coded as 1 if respondents indicated that an issue was "extremely important" or "very important" or they "strongly favor" or "somewhat favor" a proposal and are coded as 0 if respondents indicated that the issue was "somewhat important" or "not important" or "somewhat oppose" or "strongly oppose" a proposal. (16)

IV. METHODS AND FINDINGS

We use these survey data to answer three questions: First, in each survey, how much are respondents willing to pay to reduce uninsurance? Second, using cross-sectional data, what determines an individual's WTP? Third, utilizing longitudinal data, what factors are associated with changes in an individual's WTP over time?

A. CV Method

We assess WTP for health insurance reform by using CV, an approach developed to estimate the value of goods for which no markets exist, a common feature of most public goods. CV has been used extensively since its validation during litigation surrounding the Exxon Valdez oil spill in 1989, for example in Donaldson (1990), Philips et al. (1997), and Cawley (2008). The CV method provides an estimate of the Hicksian compensating variation in monetary welfare associated with the referendum in question. In the context of health insurance expansion, compensating variation measures how much a respondent would pay and remain indifferent between the current state and the state in which taxes are higher and public health insurance is expanded.

During the initial work in 1989 that validated the use of CV in economics, the National Oceanic and Atmospheric Administration appointed a panel of expert economists, including Kenneth Arrow, Robert Solow, Paul Portney, Edward Learner, Roy Radner, and Howard Schuman, to study whether CV methods can be used reliably to estimate the values of nonmarket, passively used goods. The panel issued a report including recommended practices for CV estimation, which proposed the following guidelines: (1) use probability sampling; (2) use survey approaches that allow high response rates; (3) conduct interviews personally or over the telephone, as opposed to eliciting values through the mail or in today's terms, the Internet; (4) pretest the survey; (5) report sample statistics and WTP by respondent characteristics; (6) format the question as one of WTP for a referendum, since this is a more conservative design; (7) provide respondents with the option of abstaining from voting, in addition to the "yes" and "no" voting options; (8) remind respondents that their WTP for a referendum would reduce their remaining budget for other goods (Arrow et al. 1993). A recent review of the literature on CV analysis by Kling, Phaneuf, and Zhao (2012) and work by Carson (2012) suggests that this consensus on best practices for CV analysis continues to hold; we follow all of these eight recommendations. Specifically, our surveys adhered to recommendations (1)--(4) and (6)-(8), and the analysis we report in the paper follows recommendation (5). Further, while a common criticism of CV methods is that they are used to value things that are difficult to conceptualize, our analysis focuses on a policy change that provides a good that is relatively easy to conceptualize and value (Kling, Phaneuf, and Zhao 2012). (17)

In all three surveys (2008 ESP, 2010 ESP, and 2010 follow-up), for the overall referendum version of our WTP question, we began by asking respondents the question below. The low-income referendum added the italicized text in brackets.

"Suppose there is a new voter referendum in the state. The referendum is a proposal to fund policies that will reduce uninsurance in the state by a quarter [among families earning less than three times the poverty level, which is about $60,000 for a family of 4 (low-income version)]. Set aside for now how it will reduce uninsurance, but assume it will do so with certainty. If the referendum passes, you and everyone else will have to pay $50 more in taxes every year. Given your current budget, would you vote for or against this referendum?"

Possible responses of the hypothetical taxpayers were to vote, for the referendum, vote against it, not vote, or choose a "don't know" option. To more narrowly estimate each respondent's WTP, respondents were then asked up to two additional dichotomous choice questions that depended on responses to the initial question. If respondents indicated that they would vote for such a referendum, they were then asked if they would be willing to pay $100, and then $250 if they responded affirmatively to the $100 question. If respondents were unwilling to vote for the referendum at a cost of $50, they were asked if they would be willing to pay $25, and then $5. These amounts are the same for all respondents, and do not vary by the income of the respondent.

An important methodological issue is how best to estimate WTP from responses to these types of questions. The concern in particular is that there are infinitely many continuous distributions of underlying preferences that are consistent with the sets of discrete responses observed in the data, and different distributions can imply different mean values of the WTP. Our approach is to estimate a lower bound on respondents' WTP based on responses to these questions in the most conservative way possible, and without imposing distributional assumptions. Specifically, we assign to each respondent a WTP equal to the highest cost at which they respond "yes" to the referendum question. For example, if a respondent says "yes" to $50 and "no" to $100, the lower-bound estimate is $50. Individuals who respond "no" to all questions are given WTP equal to zero. This estimation approach is known commonly as the Turnbull distribution-free estimator (e.g., see Haab and McConnell 2003). Using this approach, mean and median estimates of WTP do not require any information other than that contained in survey responses. (18)

B. Descriptive Statistics

Table 1 presents summary statistics for responses to questions about health reform, sociodemographic characteristics, and political preferences in each of the three sub-samples described in the data section. The columns on the right of Table 1 indicate significant differences in mean values between pairs of surveys. The vast

The average characteristics of the two crosssectional samples are fairly similar. As expected, the share of the sample that was employed dropped from 65% to 60%, consistent with the macroeconomic downturn. The panel respondents also look fairly similar to the full 2008 sample, with the exceptions that the follow-up respondents are slightly more likely to be married and white. Comparing the responses of the panel sample in 2008 and 2010 (Columns B and C) for time-varying questions shows that fewer respondents thought that healthcare costs and the number of uninsured in the state should be top priorities for lawmakers, and that support for the universal mandate declined, consistent with the cross-sectional patterns.

C. The Distribution of WTP

Figure 1 shows the distribution of lower bounds on respondents' WTP for each of the samples. In the full 2008 cross-sectional surveys, out of 800 respondents, 432 were asked whether they would support a referendum to reduce the overall rate of uninsurance by one-quarter (overall referendum) and 368 were asked whether they would support a referendum to reduce by one-quarter the rate of uninsurance among those who earn less than 300% of the FPL (low-income referendum). About 86.4% of overall referendum respondents and 88.3% of low-income referendum respondents indicated willingness to support such a referendum at a cost of at least $5 per respondent. These percentages are both substantially higher than the fractions of respondents who reported being willing to pay higher taxes to expand health insurance in earlier polls in which the size of tax increases was not specified. Moreover, the share of respondents who would vote in support of the referenda remains substantial at higher levels of tax increases. 76.37% and 80.91% of the overall and low-income respondents indicated they were willing to support paying a cost of at least $25, while 64.3% and 62.8% indicated, respectively, a willingness to support paying at least $50. In the 2010 survey we did not ask the low-income referendum question, but results were similar for the overall referendum. Out of 800 respondents in 2010, 88% indicated a WTP at least $5, 76.20% indicated a WTP at least $25, and 60% a WTP at least $50. In both cross-sectional surveys the modal lower-bound WTP was $250, and the second most prevalent range of WTPs was $100 to $250 for all of the 2010 survey groups, as shown in Figure 1. WTP was also noticeably more polarized in 2010 than it was in 2008, as the percentage of respondents with WTP below $5 jumped from 13.3% in the 2008 full sample to 17.5% in the 2010 full sample. (19)

Tables 2 and 3 show the mean lower-bound estimates of WTP, which are relatively stable across the two referenda questions and over time. The mean lower-bound WTP based on the full 2008 cross-sectional sample was $95.09 per resident-year for the overall referendum, and $90.35 for the low-income referendum. (20) In 2010, respondents indicated a similar mean lower-bound WTP of $93.33 for the overall referendum.

The primary purpose of our WTP analyses is to provide information that could be of use in cost-benefit analyses of public policies that expand insurance coverage. An intuitive way to express our WTP estimates is to normalize them to a beneficiary-year level and compare them to typical costs for health insurance. For illustration, in 2008 there were about 19.4 million people in New York State. Using our mean lower-bound WTP estimates from 2008, and assuming our samples are representative of state residents, yields a lower-bound estimate of total WTP of about $1.84 billion for the overall proposal, and $1.75 billion for the low-income proposal. At the same time there were 2.56 million (21) uninsured residents in New York, about 2.02 million (22) of whom earned less than 300% of the FPL. This implies an average WTP per beneficiary coverage-year of at least $2,881 for the overall proposal, and at least $3,465 for the low-income proposal, about 20% more per beneficiary coverage-year.

The difference in WTP per beneficiary between the overall and low-income proposals arises because, even though the WTP per resident is fairly similar for both proposals, implying a similar total aggregate WTP, including nonlowincome uninsured substantially increases the total population that would need to be covered. One possible explanation for the similarity is that respondents generally believed that most of the uninsured were in the low-income category (<300% FPL), even when they were asked about the uninsured in general. That is, respondents may have been unaware that responding with a similar WTP for the overall and the low-income referenda implies a lower level of funding per beneficiary per year for the overall referendum, since there are substantially more uninsured residents than there are residents who are both uninsured and low-income. A second possible explanation is that respondents were unwilling to support as high a level of funding per beneficiary when both high-income and low-income people were covered. We cannot distinguish between these two cases in our data.

The cost-benefit analysis of interest to policymakers is to compare these measures of benefits to the actual costs of public insurance expansions. As previously discussed, there are several different types of health insurance packages that could be used to increase coverage, but our lower-bound estimates of the benefits of coverage expansion for low-income residents are higher than estimated costs of expansion through either Medicaid or private insurance based on approved rates for individual coverage that will become available in the state exchange in 2014, suggesting that the costs and benefits justify low-income expansions, and possibly even income-neutral expansions.

We also estimated alternative parametric models using these data, but such models require additional structural assumptions. (23) Our estimates are highly sensitive to these distributional assumptions on the upper tail because a large share of the respondents to our survey fell into the highest WTP category, so we prefer to avoid distributional assumptions. Although the large share in the highest WTP category is a drawback to estimating an upper bound WTP and to using parametric estimates, this does not affect the main conclusion of our work as the conservative lower-bound estimates of WTP are in the vicinity of actual costs for Medicaid expansions. We discuss this point further in Section VI.

D. Validation of Cross-Sectional CV Estimates

An important component of CV analyses is validating that estimates could plausibly reflect respondents' preferences by carefully examining whether variation in WTP reflects expected patterns. We test hypotheses from our conceptual framework by first reporting how WTP changes with income. We next estimate correlations between a full set of respondent characteristics, including demographics and opinions, and support for the proposed referenda using a probit model.

Tables 2 and 3 report the average lower-bound WTP estimates by household income for each survey sample. As shown, there is broad support for paying higher taxes to fund the referenda at all income levels. For the full sample in 2008, among respondents with household incomes above $200,000 per year, WTP was $133.52, compared to $59.99 for respondents in households with incomes below $30,000 per year. In the full 2010 sample the respective mean WTPs were $136.51 and $81.97. Figure 2 shows the full distributions of WTP within each income group for the general referendum sample in 2008. A notable feature of this figure is that the share supporting successively higher proposed tax amounts drops at a much slower rate among individuals from higher income households, relative to those from lower-income households. For households with income below $30,000, the share supporting the overall referendum drops from over 85% at a cost of $5 to under 10% at a cost of $250, while for households earning over $200,000 it only falls from 80% to 45%.

The income results in Tables 2 and 3 and in Figure 2 are in the direction consistent with altruism being a normal good, but the income gradients are not as high as one might expect, perhaps because of the countervailing effect exerted by the decreased need for public coverage as income increases. While these findings may seem at odds with Kessler and Brady's (2009) finding that support is higher at lower-income levels, they reflect quite different methods and are not necessarily contradictory.

As noted earlier, in our survey the amounts in the WTP questions are uniform, whereas WTP questions in Kessler and Brady change according to income, with reference amounts being substantially higher at higher incomes. This would mean, for example, that someone with an income of $125,000 may be asked a dollar amount many times the amount asked of someone with an income of $30,000, so that even if a high-income individual says "no" to large estimated tax increase, their maximum WTP could still exceed that of a low-income individual who says "yes" to a small increase. Consequently, although Kessler and Brady do not express their results in the form of dollar estimates of WTP, it is possible that their data could imply comparable aggregate WTP estimates. At the same time, it is still possible that because our WTP lower bounds for the highest income group (>$200,000) is about the same as that for the next highest group (>80,000 to <200,000), we too might find that WTP may not increase as rapidly with income as it would need to if progressive or even proportional taxation were used to finance the expansion.

To further examine the correlates of referendum support, we turn to multivariate results from a probit regression model. Since the data include dichotomous choice questions with follow-up questions, there are multiple responses per individual. The latent model is expressed as:

[[??].sub.i] = [x.sub.i][beta] + [[epsilon].sub.i]

where [[??].sub.i] is a 5 x 1 vector of continuous values indicating support for the referendum in question at each possible cost ($5, $25, $50, $100, and $250), [x.sub.i], is a 5 XK matrix that includes demographic characteristics and opinions about health insurance reform and referendum costs, and [[epsilon].sub.i] N(0, [[SIGMA].sub.i]), allowing for within-person correlation. We observe:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

This model is estimated separately for the 2008 and 2010 rounds of the survey, and for each version of the referendum question.

Table 4 reports the probit estimates of support based on data from the two proposed referenda questions in the 2008 sample, and the general referendum question in the 2010 sample. Reported marginal effects suggest support for all referenda decreases with the cost of the reform, as expected. A $ 100 per respondent increase in taxes decreases support for the referenda by about 30 percentage points.

We find that responses to the opinion questions about health care reform are also highly correlated with willingness to support the referenda. Respondents who indicated support for a universal mandate were willing to pay higher taxes to support the referenda, and were between 9 and 25 percentage points more likely to support the referenda. Respondents who thought that the number of uninsured people in the state was an important issue to address were not necessarily more likely to support the referenda raising taxes in 2008, but by 2010 they were 16 percentage points more likely to support it. Respondents to the low-income referendum question in 2008 and to the overall referendum question in 2010 who indicated support for a universal Medicare-style program were 18 and 24 percentage points, respectively, more likely to support the referenda. These findings suggest that when it comes to paying for reforms, New Yorkers broadly are prepared to put their "money where their mouth is," although amounts may not necessarily be proportional to the costs of reforms.

Turning to sociodemographic variables, respondents with less than high school educations were less likely to support the referenda, suggesting that this demographic group responds more to the burden aspect of the tax than to the fact that they themselves are statistically more likely to benefit from the reforms. Self-identified liberals were slightly more likely than people who classified themselves as "middle of the road" to support the referenda, while conservatives were slightly less likely to support the overall referendum, consistent with findings in Bundorf and Fuchs (2008) that those with more favorable views of government intervention were more likely to support universal coverage. Although the marginal effect of income in the model is negligible, income is strongly correlated with other included covariates, such as employment and education. The positive income and WTP correlation is presumably absorbed by these two sets of variables, which tend in the same direction as the bivariate income comparison. (24) In the Appendix, we present results from additional specifications that test whether WTP results are affected by the inclusion of detailed measures of an individual's current source of coverage; we find little evidence that source of coverage affects WTP. Table 4 also indicates that those who live in upstate New York have lower WTP, which may reflect the realization that upstate residents would cross-subsidize downstate residents, where the cost of care is higher, and/or that upstate residents have relatively less to gain from a coverage expansion since the rate of uninsurance is higher in the downstate area.

E. Changes in WTP During the Great Recession

The second part of the analysis focuses on factors that lead individuals to change their stated WTP over time. Panel data analysis of CV is relatively novel to the literature, and the only other paper we know of that uses this method is Gartoulla, Liabsuetrakul, and Pradhan (2010), who estimate changes in the WTP for obstetric care in Nepal. Both within-person intertemporal variation in WTP and the repeated cross-sectional variation provide valuable information about the stability of preferences over time. The specific time period of February-March 2008 to February-March 2010 is of particular interest for this analysis because of the deep recession that took place in between the two dates. Specifically, we are interested in whether people prefer governments to increase the size and generosity of health insurance social safety nets during economic downturns or whether they more strongly prefer the government to reduce the tax burden on households, and how these changes in preferences depend upon changing circumstances. At the same time, as noted, the latter part of our study period was marked by intense debate over the ACA which may have influenced opinions about reform as well. As discussed, while we cannot fully disentangle the combined effects of the recession and the ACA debates, we can examine the extent to which individuals' WTPs have been associated with changes in their economic circumstances and in their opinions about reform.

Of the 411 respondents who were surveyed in 2008 and again in 2010, 192 were asked both times about an overall referendum and 219 were asked both times about the low-income referendum. As reported in Tables 2 and 3, average WTP remained remarkably stable during the study period. The mean lower-bound WTP for the overall referendum fell slightly from $99.87 to $85.14 in 2010, and rose slightly from $91.60 to $93.35 for the low-income referendum. However, despite the stability of mean WTPs, over time there were substantial within-person changes.

Figure 3 shows the lower-bound WTP estimates in 2010 by each respondent's 2008 WTP, focusing on low-income referendum respondents. Although not shown, the figure looks very similar for the overall referendum. Only 37% of respondents reported no change in WTP between 2008 and 2010. Further, people whose WTPs were in the tails of the distribution in 2008 were the most likely to report the same WTP in 2010. Conversely, those who were in the middle in 2008 were the most likely to report a different WTP in 2010. Since the mean lower-bound WTP changed very little between 2008 and 2010, by implication the 63% of respondents who changed their WTP reported changes that largely counteracted each other.

Using the panel sample, we estimate how changes in individual economic conditions, such as employment status, income, and insurance status, and changes in opinions about health reform, affect changes in WTP. The empirical model is:

[DELTA] WT[P.sub.i] = [DELTA][x.sub.i], [beta] + [[epsilon].sub.i]

where the independent variables that change over time include income, employment status, insurance status, and opinions about health insurance reform questions. (25) Each covariate except for income is included asymmetrically to test, for example, whether gaining a job has a different effect than losing a job.

Table 5 presents the estimates from this model. Results based on the overall referendum question suggest that losing one's own insurance tends to reduce WTP significantly, by about $208. Consistent with the cross-sectional findings, this suggests that those who are most likely to gain insurance personally are less willing to pay for the referenda.

In the low-income sample, those who became employed between 2008 and 2010 were less willing to pay for reform, with a mean decrease of $74. This could be due to the fact that being employed is positively correlated with having health insurance, so gaining employment reduces demand for insurance, and the demand for reform that provides insurance. However, in the cross-sectional probit model this effect was positive, suggesting that those who have been employed continuously are more willing to pay than those who are newly employed, all else equal. The results also highlight the importance of changing political opinions. Those who changed their minds between 2008 and 2010 and became supporters of a universal Medicare system increased their WTP on average by about $59, while a change in opinion in the opposite direction was associated with a reduction in WTP of about $30. Those who became supporters of a large employer mandate decreased their WTP taxes to support alternative government-led reform described in the referendum proposals by about $51. These findings are informative in two ways. The correlations between changes in opinions and changes in WTP further validate that the WTP measure is correlated with preferences in each period. Second, while the opinion questions have no causal interpretation, and simply reflect preferences that are not captured by economic variables alone, they are informative in clarifying that changes in WTP were driven heavily by factors other than changes in personal economic outcomes during the recession.

A concise and informative way to convey how economic changes and opinion changes affected WTP is to decompose the aggregate change into components due to each type of variable. To do this we estimate an Oaxaca-Blinder decomposition model (26):

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

The first term in this decomposition tells us how changes in observable characteristics between 2008 and 2010 led to changes in WTP holding constant the effects of characteristics at their 2008 levels. The second component tells us how changes in the effects of characteristics on WTP contributed to the overall change. To put these components in context, one question we are interested in investigating is whether and how WTP changes with individual conditions, such as unemployment, the prevalence of which rose in New York State from 4.9% in the first quarter of 2008 to 8.9% in the first quarter of 2010 (BLS 2012). The model decomposes changes in reported WTP into a component due to changes in the employment status of respondents, and a separate component due to changes in the effect on WTP of being employed in 2010 versus being employed in 2008.

The results of this decomposition are shown in Table 6. There are two sets of variables included in the decomposition: those related to changes in opinion, and those related to changes in socioeconomic wellbeing. The socioeconomic changes include changes in income, employment status, and health insurance coverage (either gaining or losing insurance). Regarding the overall referendum, the change in the mean lower-bound WTP predicted by the change in the explanatory variables is a decrease of $12.66. This decrease is composed of an $8.65 decrease attributable in a mechanical sense to observed changes in SES and opinions about insurance reform, and a $4.03 decrease due to a change in the effect that these variables have on WTP. (27) The second row of the table shows estimates from a model that includes only socioeconomic characteristics as explanatory variables, and the explanatory power of changes in these characteristics is nearly zero. Estimates in the third row are from a model that only includes opinion variables. Results show that changes in opinions alone explain nearly the entire decrease in lower-bound WTP estimated in the full model. This implies that changes in individual economic outcomes explain a very small share of the total change in WTP, while other factors that are correlated with opinions about health care reform had much larger effects.

One possible interpretation of these findings is that the debates leading to the passage of the ACA increased the level of information about health insurance reform, or changed opinions in other ways that directly affected preferences over WTP for reforms. However, changes in opinions themselves are not causal mechanisms, and could instead pick up changes in other characteristics that are also correlated with our measure of WTP, including changes in economic conditions at both individual and macroeconomic levels. In any case, these findings suggest two broad take-a-ways. First, regardless of the interpretation of the impacts of changes in opinion, consistent with our cross-section findings, views on reform are strongly correlated with WTP for it. Second and perhaps more strikingly, changes in individual economic conditions appear to have had very little direct effect on WTP despite the magnitude of the economic downturn during the study period.

Table 6 also shows there is less total change in WTP for the low-income referendum question, so the decomposition is somewhat less informative. The mean change in WTP was positive $6.69, but the components had opposite signs. Changes in characteristics explain a $3.22 decrease in WTP, which was more than offset by a $7.29 increase in WTP caused by changes in coefficients and a positive interaction term. Again, when we include only economic variables the included variables explain almost none of the mean change in WTP, whereas when we include only changes in opinions the results closely match those of the full model.

In summary, we use a double-bounded CV approach to estimate the nonparametric lower-bound WTP in order to answer the three questions relevant to health policy towards the uninsured, finding that (1) there is a high prevalence of positive individual WTP, where the estimated aggregate WTP is at least as high as $2,800 to cover one uninsured person per year, which is close to the actual average cost nationally for Medicaid; (2) correlates of WTP are largely as expected, for example, WTP increases with income; (3) average WTP stayed remarkably stable over time, despite the economic and political changes between 2008 and 2010. In addition, at the individual level, we find that changes in political opinions about health insurance reform are strongly correlated with changes in WTP, while economic factors related to the recession, such as changes in income and employment, are not. These findings move the literature forward in several dimensions. Prior polls, apart from one, have not elicited specific responses about WTP, and no prior study has used a set of double-bounded questions or examined how demographics affect WTP. Nor has any prior study examined whether there is concordance between WTPs and stated opinions regarding health reform. We next discuss limitations of our work, before concluding with the lessons to be drawn for public policy.

V. CAVEATS

There are several important limitations to the analyses in this paper. One limitation, as with all CV analyses, is that hypothetical answers are not necessarily representative of actual behavior, or how voters in other states or other periods of time would respond to the same hypothetical questions (e.g., the WTP of New York State residents may be different from the national WTP). Arrow et al. (1993) recommend using independently drawn samples at different points to assess the reliability of responses to CV questions, which we do in this paper. A second limitation is that responses to the ESP poll are affected by the numerical prompts in the question, potentially causing psychometric anchoring bias (Green et al. 1998). Anchoring has been found to be problematic in other polls, but without variation in numerical prompts across respondents, this is not testable using our data. A third limitation is that a large proportion of our sample (more than 20%) falls into the highest WTP category of supporting a $250 tax increase. Although this is not problematic per se, it suggests that our lower-bound estimate may substantially understate the true mean WTP as many individuals may have been willing to pay substantially more than $250. Given our lower-bound estimates show that aggregate WTP is comparable to the actual costs of covering the uninsured under public health insurance currently, for the purposes of the present analysis a smaller lower-bound is not especially problematic. However, future estimates of WTP could estimate mean WTP ranges more accurately using a larger set of double-bounded questions.

Other possible problems are that our longitudinal analysis only considers an economic downturn. Results from a downturn may not necessarily be symmetric for a period of macroeconomic growth. Future survey data could build upon our results to address this issue. In addition, as discussed, debate surrounding the ACA may have influenced responses to state-specific questions over time. Finally, as highlighted earlier, survey results may, of course, be sensitive to the particular framing of questions, and the choice of the sample population (New York State). Our survey results indicate substantially broader support for insurance reform than is suggested by other polls cited earlier in this paper. Potential explanations include differences in the way questions are framed, the use of a CV approach and targeting of different geographic populations (e.g., a national sample or other states). However, questions about generalizability remain.

VI. DISCUSSION

Despite the fact that uninsurance is a highly policy-relevant and timely problem, cost-benefit analyses of reforms have been conspicuously absent from discussions because of a lack of estimates of the benefits of such reforms. In part this is due to the complicated nature of calculating the benefits of providing public health insurance, which could include direct short-term benefits, insurance against future risk of becoming uninsured, and altruism. However, there is a history in economics of using methods to elicit and measure benefits associated with goods for which there are no observable markets. We use one such method, CV, to estimate the WTP higher taxes to fund a policy that would reduce the rate of uninsurance.

Combining CV methods with more carefully phrased questions that directly specify cost yields estimates of the share of respondents willing to pay any increased taxes to support increased coverage that are substantially greater than previously reported estimates for national and state-specific samples. Moreover, estimates of WTP indicate broad-based and deep support that remained remarkably stable between 2008 and 2010, despite a major economic downturn and passage of national reforms, which may have diluted support for state-specific reforms. We find that in both 2008 and 2010 more than 80% of respondents were prepared to pay some additional taxes to reduce the rate of uninsurance, and the average nonparametric lower-bound estimate of WTP by New York State residents exceeded $90 per respondent in both years. This implies a lower-bound WTP of about $2,880 per beneficiary-year for an expansion of coverage that is independent of income, and $3,465 for a policy that targets low-income residents, which as discussed, is comparable to potential costs associated with a Medicaid expansion or offering coverage through a state exchange. Moreover, as discussed above, there is reason to believe that our lower-bound estimates substantially understate the average WTP.

One of the innovations in this paper is the use of panel data covering 2008-2010 to conduct longitudinal CV analyses to address these key questions about how respondents' views of the role of government change when they face changing circumstances during recessionary times. Our results reveal that despite the large economic downturn, on balance the majority of people were still willing to pay higher taxes for health insurance reform that would expand social safety nets. This finding has important policy implications in gauging the extent to which economic downturns could affect the long-run viability of health insurance reform. Despite the fact that many respondents changed their WTP over time, on net the changes balanced out. Moreover, an Oaxaca-Blinder decomposition shows that these changes in WTP were not driven by changes in individual economic outcomes, such as changes in income and employment status. Instead we find that, conditional on economic outcomes, changes in opinions about specific health insurance reform proposals explain nearly all of the variation. Possible explanations are that noneconomic variables correlated with opinions were the primary drivers of changes in WTPs, or that changes in opinions themselves, perhaps due to debates prior to the passage of the ACA, causally affected WTP. In any case, our analysis finds a strong relation between support for reforms and WTP for them.

ABBREVIATIONS

AAPOR: American Association for Public Opinion Research

ACA: Affordable Care Act

CATI: Computer-Assisted Telephone Interviewing

CV: Contingent Valuation

ESP: Empire State Poll

FPL: Federal Poverty Line

HSAs: Health Savings Accounts

SES: Socioeconomic Status

SRI: Survey Research Institute

WTP: Willingness to Pay

doi: 10.1111/ecin.12083

REFERENCES

Agency for Healthcare Research and Quality. "Center for Financing, Access and Cost Trends." Medical Expenditure Panel, Table I.C., 2009. Accessed March 12, 2014. http://meps.ahrq.gov/mepsweb/data_stats/ summ_tables/insr/national/series_l/2009/ticl.pdf.

Andreoni, J. "Impure Altruism and Donation to Public Goods: A Theory of Warm-Glow Giving." The Economic Journal 100(401), 1990,464-77.

Arrow, K., R. Solow, P. Portney, E. Learner, R. Radner, and H. Schuman. "Report of the NOAA Panel on Contingent Valuation." Federal Register, 58(10), 1993, 4601-14.

Blinder, A. S. "Wage Discrimination: Reduced Form and Structural Variables." Journal of Human Resources, 8, 1973,436-55.

Bundorf, M. K., and V. R. Fuchs. "Public Support for National Health Insurance: the Roles of Attitudes and Beliefs." Forum for Health Economics and Policy, Frontiers in Health Policy Research, 10(1), 2008, 1-33.

Bureau of Labor Statistics. "Local Area Unemployment Statistics, New York State (Series Id: LASST360000000000003)." 2012. Accessed March 12,2014. http://data.bls.gOv/timeseries/LASST360000000000003.

Cantor, J., S. Brownlee, C. Zukin, and M. Roller. "New Jersey Health Care Opinion Poll: Public Concerns and Support for Reform," 2007. Accessed March 12, 2014. http://www.cshp.rutgers.edu/Downloads/7530.pdf.

Carson, R. "Contingent Valuation: A Practical Alternative When Prices Aren't Available." Journal of Economic Perspectives, 26(4), 2012, 27-42.

Cawley, J. "Contingent Valuation Analysis of Willingness to Pay to Reduce Childhood Obesity." Economics and Human Biology, 6, 2008, 281-92.

Congressional Budget Office. "Updated Estimates for the Insurance Coverage Provisions of the Affordable Care Act." March 2012. Accessed March 12, 2014. http://cbo.gov/sites/default/files/ cbofiles/attachments/03-13-Coverage%20Estimates.pdf (Tables 1 and 2).

Donaldson, C. "Willingness to Pay for Publicly-Provided Goods: A Possible Measure of Benefit?" Journal of Health Economics, 9(1), 1990, 103-18.

Epstein, A., H. Aaron, K. Baicker, J. Hacker, and M. Pauly. "Perspective Roundtable: Health Care Reform in Perspective." New England Journal of Medicine, October 15, 2009. Accessed March 12, 2014. http://www.nejm.org/ doi/full/10.1056/NEJMp0909384#t=article.

Feldstein. M., and A. Taylor. "The Income Tax and Charitable Contributions." Econometrica, 44(6), 1976, 1201-22.

Gartoulla, P. T. Liabsuetrakul, and N. Pradhan. "Change in Willingness to Pay for Normal and Caesarean Section During Pregnancy and After Delivery in Kathmandu." Tropical Medicine and International Health, 15(10), 2010, 1227-34.

Green, D., K. Jacowitz. D. Kahneman, and D. McFadden. "Referendum Contingent Valuation, Anchoring, and Willingness to Pay for Public Goods." Resource and Energy Economics, 20(2). 1998. 85-116.

Gruber, J. "Covering the Uninsured in the United States." Journal of Economic Literature, 46(3), 2008, 571-606.

Haab, T. C., and K. E. McConnell. Valuing Environmental and Natural Resources: The Econometrics of Non-Market Valuation. Northampton, MA: Edward Elgar Publishing. 2003.

Howard, Paul. "Building a Market-Based Health-Insurance Exchange in New York." Manhattan Institute Report, 2011. Accessed February 18, 2014. http://www. manhattan-institute.org/html/mpr_13.htm.

Jost, T. S., and M. Hall. "The Role of State Regulation in Consumer-Driven Health Care." American Journal of Law and Medicine, 31, 2005, 395-418.

Kaiser Family Foundation (KFF). "Kaiser Health Tracking Poll: Public Opinion on Health Care Issues." June 2009. Publication #7925. Accessed February 18, 2014. http://www.kff.org/kaiserpolls/upload/7925.pdf.

--. "Kaiser Public Opinion, Data Note: A Snapshot of Public Opinion on the Individual Mandate." March 2012. Publication #8296. Accessed February 18, 2014. http://kaiserfamilyfoundation.files.wordpress.com/2013 /01/8296.pdf.

--. "State Decisions on Health Insurance Exchanges and the Medicaid Expansion, as of June 20 2013." 2013a. Accessed July 11, 2013. http://kff.org/healthreform/state-indicator/state-decisions-for-creating- health-insurance-exchanges-and-expanding-medicaid/.

--. "State Exchange Profiles: New York." January 28, 2013. 2013b. Accessed July 10, 2013. http://kff.org/ health-reform/state-profile/state-exchange-profilesnew-york/.

Kessler, D., and D. W. Brady. "DataWatch: Putting the Public's Money Where Its Mouth Is." Health Affairs, 28(5), 2009, w917-25, online issue.

Kling, C., D. Phaneuf, and J. Zhao. "From Exxon to BP: Has Some Number Become Better Than No Number?" Journal of Economic Perspectives, 26(4), 2012, 3-26.

Krueger, A. B., and I. Kuziemko. "The Demand for Health Insurance Among Uninsured Americans: Results of a Survey Experiment and Implications for Policy." Journal of Health Economics, 32(5), 2013, 780-93.

Musumeci, Mary Beth. "Implementing the ACA's Medicaid Related Health Reform Provisions After the Supreme Court's Decision." Focus on Health Reform: Report, August 2012, The Henry J. Kaiser Family Foundation, 2012. Accessed July 15, 2013. http:// kaiserfamilyfoundation.files.wordpress.com/2013/01/ 8348.pdf.

New York Times. "Support for a Public Health Plan." 2009. Accessed March 12, 2014. http://www.nytimes.com/ imagepages/2009/06/21 /health/policy/21 poli_graphic_ ready.html.

Oaxaca, R. "Male-Female Wage Differentials in Urban Labor Markets." International Economic Review, 14(October), 1973, 693-709.

Parente, S., and T. Bragdon. "Healthier Choice: An Examination of Market-Based Reforms for New York's Uninsured." 2009. Accessed March 12, 2014. http://www. manhattan-institute.org/html/mpr_10.htm.

Phillips, K. A., R. K. Homan, H. S. Luft, P. H. Hiatt, K. R. Olson, T. E. Kearney, and S. E. Heard. "Willingness to Pay for Poison Control Centers." Journal of Health Economics, 16(3), 1997, 343-57.

Russell Sage Foundation. "Special Initiative on the Social and Economic Effects of the Great Recession," 2011. Accessed March 12, 2014. http://www. russellsage.org/research/social-effects-great-recessiondescription.

Simon, K., and W. White. "Informing Health Care Reform Options for New York State." Cornell University report written for New York State Health Foundation, 2008. Accessed March 12, 2014.http://nyshealthfoundation. org/resources-and-reports/resource/informing-healthcare-reform-options-for-new-york-state.

Supreme Court. National Federation of Independent Business et al. v. Sebelius, Secretary of Health and Human Services, et al., 132 Supreme Court of the United States 2566. 2012. Accessed February 18, 2014.http://www. supremecourt.gov/opinions/11pdf/11-393c3a2.pdf.

Survey Research Center. "Empire State Poll Introduction and Method, Cornell University." 2008. Accessed February 18, 2014.http://www.sri.cornell.edu/sri/files/esp/ 2008/Report_1_2008-Intro_Method.pdf.

--. "Empire State Poll Introduction and Method, Cornell University," 2010.http://sri.cornell.edu/sri/files/esp /2010/Report%201 %20-%202010%20-%20Introduction%20and%20Methodology.pdf.

Urban Institute. "The Cost and Coverage Implications of the ACA Medicaid Expansion: National and Stateby-State Analysis," 2012. Accessed July 2013. http:// www.urban.org/UploadedPDF/412707-The-Costand-Coverage-Implications-of-the-ACA-Medicaid- Expansion.pdf.

U.S. Census Bureau. "Small Area Health Insurance Estimates," 2012. Accessed March 12, 2104. http://www. census.gov/did/www/sahie/data/interactive/.

SUPPORTING INFORMATION

Additional Supporting Information may be found in the online version of this article:

Appendix 1: Alternative Specifications of Probit Model of Referendum Support Including Detailed Insurance Coverage*

(1.) For example, at the federal level, it has been estimated that under the ACA, the total gross cost of expanding coverage to 30 million uninsured Americans over the years 2012-2021 will be $1,496 billion (CBO 2012).

(2.) The 2008 and 2009 Kaiser Family Foundation (KFF 2008,2009) Health Tracking Polls and a 2009 New York Times and CBS News poll (New York Times, 2009) found the shares of respondents willing to pay higher taxes to increase health insurance coverage ranged from 34% and 41% in Kaiser polls to 57% in the New York Times/CBS poll. Kessler and Brady (2009) found that 51.3% of respondents were willing to support a 5% increase in federal income taxes to expand Medicaid to cover half the uninsured and Cantor et al. (2007) found that 47.9% of certain New Jersey residents were willing to pay more than $500 per year in taxes so that "everyone in New Jersey has health insurance they can't lose, no matter what," and another 9.2% were willing to accept a "more modest tax hike."

(3.) $2,881 equals $95.09 WTP per person, times 19.38 million residents (source: http://quickfacts.census. gov/qfd/states/36000.html), divided by 25% of 2.56 million uninsured residents (source: http://www.statehealthfacts. org/profileind.jsp?sub=40&rgn=34&cat=3).

(4.) $3,465 equals $90.35 WTP per person, times 19.38 million residents, divided by 25% of 2.02 million uninsured residents below 300% of the federal poverty line (FPL) (source: http://www. nyhealthcarecommission.org/docs/uninsured_in_new_york. pdf).

(5.) Further, in response to a repeated question about willingness to pay for an overall expansion of coverage in 2010, the average WTP remained remarkably stable despite both the onset of a major recession in late 2008 that peaked in 2010, and an intense debate over the ACA in 2009 through early 2010 that could have shifted attention towards national reforms.

(6.) Participation in Medicaid expansions specifically became optional following a ruling by the U.S. Supreme Court in 2012. Participation for selected eligible groups is initially at no cost to states, and state cost shares increase to 10% by 2020 and remain fixed thereafter (Musumeci 2012). An interesting question is how individual taxpayers may view the costs of expanding state Medicaid coverage. While the costs at the state level are modest, taxpayers may also consider the possible implications for their federal taxes.

(7.) The cost of a child Medicaid beneficiary in 2009 was $2,305 for the United States on average and $2,505 for New York. See http://www.statehealthfacts.org/comparetable.jsp? ind=183&cat=4 (last accessed December 2012).

(8.) In Table 8 (Urban Institute 2012), the authors estimate that over 10 years the total cost (federal and state) of the Medicaid expansion will be $3,456 billion, and that 320,000 new people will be eligible in the state (Table 9). Combining these estimates leads to an estimate of $ 1,080 per beneficiary coverage-year for marginal Medicaid enrollees (non-parents in the 100%--133% of FPL range).

(9.) See http://www.govemor.ny.gov/assets/documents/ Approved2014HeaIthInsuranceRates.pdf

(10.) Asking both opinion and WTP questions in the same survey raises a question of whether the responses are independent of each other. If respondents feel the need to justify their answers to initial questions, this may affect subsequent responses. In particular, this seems an important concern if opinion questions are asked before WTP questions, but in our survey, WTP questions come first. However, it is still possible that responses to WTP questions may affect respondents' answers to subsequent qualitative questions about their opinions about reforms, and lead to answers that look more similar to each other than would otherwise be the case.

(11.) The ESP is a random telephone survey of New York State residents aged 18 and over conducted annually in the first quarter (February and March) by the Cornell Survey Research Institute (SRI) since 2003. Responses are collected through a computer-assisted telephone interviewing (CATI) system. ESP data have been used in prior CV research studies, such as Cawley (2008) who studies WTP to reduce childhood obesity through taxes.

(12.) If the resident was unavailable, the interview was terminated and the Cornell SRI attempted to follow up later. If these attempts were unsuccessful (called unsuccessfully 5+ times) or the respondent refused, additional households were added to the sample and this process continued until a total of 800 respondents was obtained. The survey design thus ensured that subjects who are selected are not more likely to be those out of the labor force simply by virtue of being more likely to answer the phone. It is expected that chance variation in the sample should not cause results in either the 2008 or the 2010 ESP to vary by more than 3.5 percentage points in no more than 1 in 20 cases from the answers that would have been obtained if all New York State residents were interviewed. More details of the survey can be found at the ESP website at http://www.sri.cornell.edu/sri/esp.introduction.cfm and from SRI (2008, 2010).

(13.) These rate calculations follow American Association for Public Opinion Research (AAPOR) definitions: the response rate is computed by dividing the number of completed surveys by the total eligible sample, while the cooperation rate is computed by dividing the number of completed surveys by the number of potential interviews (which includes all instances where the properly selected individual was reached but refused to complete the survey, but not due to language or physical limitations). SRI notes that these rates compare favorably with telephone surveys conducted by other research organizations such as the Pew Research Center (SRI, 2008, 2010). We note that these rates compare favor ably to other health-related telephone polls too. For example, the cooperation rate of the Political Typology Survey conducted by The Pew Research Center used in Bundorf and Fuchs (2008) is 45% (p. 3).

(14.) Because the regular ESP does not re-sample individuals included in previous ESPs, there is no overlap between the 2010 follow-up survey of 2008 ESP respondents and the 2010 ESP cross-section sample of new respondents.

(15.) The high-deductible plan question was only asked in the 2008 ESP. High-deductible plans were not allowed in the individual market in New York State by law at the time of the 2008 survey (Jost and Hall 2005; Parente and Bragdon 2009). Although Health Savings Accounts (HSAs) remained unavailable in the individual market in New York State (Howard 2011), HSA compatible high-deductible plans have been available on a limited basis under the state-subsidized Healthy New York program available to low-income working families from 2007 onwards (11 NYCRR 362. available at http://www.dfs.ny.gov/insurance/r_misc/rhlthny 171a3.htm).

(16.) "Favor" is coded as 1 and "oppose" as 0 where only these two responses were offered. In all cases, responses of "don't know" or "refused" are treated as missing in our analysis of these and other questions.

(17.) We would like to thank an anonymous referee for this point.

(18.) The only potential constraint that the lower bound model can impose is monotonicity, but responses in our data are monotonic without imposing any constraints, so our estimates are free of distributional assumptions. See Haab and McConnell (2003) for more on distributional assumptions. majority of respondents felt that it was important to reduce the number of uninsured residents, although the share decreases over time from 88% to 81% in the cross-sectional samples, and by a similar amount in the panel sample. In 2008 only 39% favored increasing coverage by imposing a large employer mandate; this share dropped slightly to 36% in 2010. Support for a universal insurance mandate also fell from 76% in 2008 to 71% in 2010. Similarly, the share supporting universal Medicare fell from 70% in 2008 to 63% in 2010.

(19.) A way to improve the accuracy of hypothetical WTP responses suggested by Arrow et al. (1993) in their report to the NOAA on CV is to offer respondents the option to say that they would not vote for the referendum. This option was offered for all questions, and these responses were conservatively coded as a lack of support for the referendum. On average, 2.8% of responses to all questions were "Would not Vote," with a range for specific questions from 0.8% to 5.7%.

(20.) We report findings on a per respondent basis, but it is possible respondents in households with multiple earners who file jointly may misinterpret the question to refer to total tax payments by the household.

(21.) See http://www.statehealthfacts.org/profileind. jsp?sub=40&rgn=34&cat=3

(22.) See http://www.nyhealthcarecoimnission.org/docs/ uninsured_in_new_york.pdf

(23.) To explore the sensitivity of our estimates, we also considered several alternative specifications for 2008 results in which we made structural assumptions. Among the specifications tested, the best-fitting parametric model was a loglinear WTP model, which suggested a mean WTP of $449.99 per household for referendum version 1 in 2008 and $286.03 for version 2. The bootstrapped 95% confidence intervals of the mean estimator were ($291.76, $860.25) and ($201.28, $423.69), respectively.

(24.) In alternative model specification that include only opinion variables, income, age, gender, and children, the coefficient on income is significantly positive in every sample, but small in magnitude, consistent with the bivariate pattern. For example, in the 2008 overall referendum sample, the coefficient on income suggests that a $100,000 increase in annual income increases the conditional probability of supporting the referendum by about 8.9 percentage points.

(25.) We also ran a separate model to test whether changes in WTP could have been predicted by characteristics in 2008. We found that none of the 2008 characteristics had significant explanatory power in predicting subsequent changes in WTP.

(26.) See Oaxaca (1973) and Blinder (1973).

(27.) These decreases do not add exactly to $ 12.66 because they are partially offset by small increases in WTP from the simultaneous interaction effect of the two changes.

KURT LAVETTI, KOSALI SIMON and WILLIAM D. WHITE*

* This research was supported in part by Cornell University Agricultural Experiment Station federal formula funds, Project No. NYC-3247412 received from the National Institutes for Food and Agriculture (NIFA) U.S. Department of Agriculture, and by funding from the New York State Health Foundation, Grant No 2007-2055189. Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the view of the U.S. Department of Agriculture or the New York State Health Foundation. We are grateful to Anna Hill for excellent research assistance and to Asako Moriya, two anonymous referees, and participants at ASHEcon and APPAM annual research conferences for helpful comments. Lavetti: Assistant Professor, Department of Economics, Ohio State University, Columbus, OH 43210. Phone 1-301-752-0895, Fax 1-614-292-3906. E-mail lavetti. 1 @osu.edu

Simon: Professor, School of Public and Environmental Affairs, Indiana University, Bloomington, IN 47405. Phone 1-812-856-3850, Fax 1-812-855-7802, E-mail simonkos@ indiana.edu

White: Professor, Department of Policy Analysis and Management, Cornell University, Ithaca, NY 14853. Phone 1-607-254-6476, Fax 1-607-255-4071, E-mail wdw8@Cornell.edu

TABLE 1
Descriptive Statistics, 2008-2010

                                Mean Comparisons

                                              B
                                 A       2008 Panel
                             Full 2008   Respondents
                              Sample       in 2008

Reform opinion questions
Cost important                  0.92         0.91
#Uninsured important            0.88         0.87
Large employer mandate          0.39         0.40
Universal mandate               0.76         0.71
Support high deductible         0.68         0.67
Universal medicare              0.70         0.66
Sociodemographic questions
Age                            48.51        51.99
Male                            0.50         0.51
Married                         0.53         0.60
Children (in household)         0.41         0.36
<High school                    0.09         0.06
High school diploma             0.22         0.19
Some college                    0.28         0.27
White (non-Hispanic)            0.66         0.75
Black (non-Hispanic)            0.15         0.13
Hispanic                        0.13         0.08
Insured                         0.92         0.94
Employed                        0.65         0.62
Income                        80,483       84,027
Liberal                         0.34         0.36
Conservative                    0.30         0.31
Upstate                         0.50         0.53
N                             800          411

                                  Mean Comparisons

                                  C             D
                             2008 Panel        New
                             Respondents   Respondents
                               in 2010       in 2010

Reform opinion questions
Cost important                  0.85          0.85
#Uninsured important            0.80          0.81
Large employer mandate          0.38          0.36
Universal mandate               0.63          0.71
Support high deductible         0.65
Universal medicare              0.61          0.63
Sociodemographic questions
Age                              --           51.21
Male                             --           0.49
Married                          --           0.53
Children (in household)          --           0.39
<High school                     --           0.06
High school diploma              --           0.24
Some college                     --           0.25
White (non-Hispanic)             --           0.66
Black (non-Hispanic)             --           0.21
Hispanic                         --           0.11
Insured                         0.93          0.92
Employed                        0.58          0.60
Income                         81,040        80,206
Liberal                          --           0.33
Conservative                     --           0.32
Upstate                          --           0.50
N                              411           800

                                 Tests of Differences in Means

                             A vs. D   A vs. B   B vs. C   C vs. D

Reform opinion questions
Cost important               ***                 **
#Uninsured important         ***                 ***
Large employer mandate
Universal mandate            **                  **        ***
Support high deductible
Universal medicare           ***
Sociodemographic questions
Age                          ***       ***
Male
Married                                **
Children (in household)                *
<High school                 **        *
High school diploma
Some college
White (non-Hispanic)                   ***
Black (non-Hispanic)         ***
Hispanic                               ***
Insured
Employed                     *
Income
Liberal
Conservative
Upstate
N

Notes: The values in this table show the fraction of the relevant
sample that reported "yes" to the specific health reform questions
in the first panel, and who have the relevant demographic
characteristics indicated in the second panel. Column A corresponds
to the 800 respondents in the 2008 ESP, column B corresponds to the
2008 answers of the subset of the 2008 ESP who also responded in
2010, column C corresponds to the 2010 answers of the subset of the
2008 ESP who also responded in 2010, and column D corresponds to
the 2010 ESP (new) sample. The second set of columns test whether
the indicated columns represent values that are statistically
different from each other in a two tailed f-test. See Appendix for
wording of survey questions. "--" indicates question was not asked
in survey round.

* Significant difference at 10% level; ** significant at 5% level;
*** significant at 1% level.

TABLE 2
Lower-Bound WTP Estimates for Overall
Referendum

                   2008      2008      2010      2010
                   Full      Panel     Panel      New
                  Sample    Sample    Sample    Sample

Mean lower        $95.09    $99.87    $85.14    $93.33
  bound WTP       (7.63)    (10.51)   (10.80)   (5.20)
                   [368]     [192]     [192]     [800]
WTP, income       $133.52   $113.09   $135.34   $136.51
  $200,000+       (26.39)   (33.76)   (38.62)   (15.28)
                   [26]      [15]      [11]      [85]
WTP, income       $130.01   $144.29   $127.86   $115.98
  $80,000 to      (13.28)   (16.75)   (17.66)   (10.31)
  $199,999         [103]     [59]      [58]      [184]
WTP, income       $90.64    $85.16    $77.47    $80.81
  $30,000 to      (13.11)   (18.21)   (18.41)   (8.00)
  $79,999          [126]     [67]      [68]      [352]
WTP, income       $59.99    $66.05    $48.76    $81.97
  below $30,000   (17.96)   (26.25)   (24.30)   (13.21)
                   [75]      [35]      [43]      [144]
WTP, missing      $60.14    $62.92     $7.60    $54.18
  income data     (25.75)   (38.71)   (14.02)   (22.91)
                    [38]     [16]      [12]      [35]

Notes: Mean lower bound estimates of WTP reported overall and by
income group for each survey sample. Standard errors in
parentheses, and number of observations in brackets. "Overall
Referendum" refers to the first version of the survey question in
which respondents are asked their WTP to reduce uninsurance by one
quarter.

TABLE 3
Lower-Bound WTP Estimates for Low-Income
Referendum

                    Low-Income Referendum

                   2008      2008      2010
                   Full      Panel     Panel
                  Sample    Sample    Sample

Mean lower        $90.35    $91.60    $93.35
  bound WTP       (7.08)    (9.90)    (9.93)
                   [432]     [219]     [219]
WTP, income       $135.35   $89.98    $68.58
  $200,000+       (24.10)   (35.15)   (41.37)
                   [31]      [18]      [13]
WTP, income       $94.00    $96.55    $112.56
  $80,000 to      (14.51)   (19.90)   (19.52)
  $199,999         [101]     [53]      [52]
WTP, income       $95.39    $97.96    $103.00
  $30,000 to      (11.26)   (15.41)   (13.97)
  $79,999          [166]     [88]      [105]
WTP, income       $59.35    $50.91    $50.46
  below $30,000   (16.27)   (30.32)   (28.01)
                   [78]      [27]      [32]
WTP, missing      $86.27    $103.83   $85.58
  income data     (20.02)   (25.49)   (37.26)
                   [56]      [33]      [17]

Notes: Mean lower bound estimates of WTP reported overall and by
income group for each survey sample. Standard errors in parentheses,
and number of observations in brackets. "Low-Income Referendum" refers
to the second version of the survey question in which respondents are
asked their WTP to reduce uninsurance by one quarter among those who
earn less than 300% of the FPL.

TABLE 4
Probit Model of Referendum Support

Dependent Variable: Support for Referendum

                                   2008                 2010

                                            Low-
                           Overall        Income       Overall

Referendum cost            -0.003 ***   -0.003 ***   -0.003 ***
Think cost is               0.063        0.014       -0.008
  important
Think # uninsured is       -0.044       #REF!         0.163 ***
  important
Support large               0.055        0.135 ***    0.002
  employer mandate
Support universal           0.189 ***    0.090 **     0.251 ***
  mandate
Support                     0.049        0.100 *
  high-deductible
  plans
Support universal           0.066        0.182 ***    0.236 ***
  medicare
Age                         0.000        0.001        0.002 *
Male                        0.030        0.165 ***    0.031
Married                     0.083       -0.037        0.042
Children                   -0.031        0.079        0.068 *
<High school               -0.172 *     -0.344 ***   -0.038
HS diploma                 -0.116       -0.106       -0.154 ***
Some college               -0.029       -0.139 **    -0.087 *
White (Non-Hispanic)        0.026       -0.002       -0.095
Black (Non-Hispanic)       -0.202 *     -0.025       -0.249 **
Hispanic                   -0.097       -0.092       -0.190 *
Uninsured                  -0.092        0.012       -0.008
Employed                    0.036       -0.003        0.141 ***
Income ($ 10,000s)          0.000        0.004        0.004
Liberal                     0.070        0.118 *      0.028 ***
Conservative               -0.114 *     -0.012       -0.084 *
Upstate                    -0.125 **     0.006       -0.032
Introduction               -0.049        0.001
N obs.                        1405         1540         3200
N clusters                    281          308          640
[R.sup.2]                   0.230        0.279        0.274

Notes: Dependent variable equals 1 if respondent indicated that they
would vote for the referendum at the specified price. Data are from
full samples in respective years. Reported coefficients are marginal
effects. Income is in tens of thousands of dollars.

* Coefficient is significant at 10% level based on standard errors
clustered by individual; ** significant at 5% level; *** significant
at 1% level.

TABLE 5
Change in WTP vs. Change in Characteristics and Opinions, Panel
Sample

Overall Referendum                             Coefficient (SE)

Gained insurance                                -15.89 (50.29)
Lost insurance                                 -207.75 (68.12) ***
Became employed                                  17.56 (26.64)
Became unemployed                                -6.46 (37.52)
Change in income                                 -0.63 (2.18)
Changed opinion up: cost important              128.34 (61.92) **
Changed opinion down: cost important             -7.53 (24.53)
Changed opinion up: # uninsured important       -67.04 (47.11)
Changed opinion down: # uninsured important     -42.90 (27.46)
Changed opinion up: large employer mandate      -22.20 (21.31)
Changed opinion down: large employer mandate    -33.27 (24.39)
Changed opinion up: universal mandate           -12.74 (27.39)
Changed opinion down: universal mandate           3.79 (23.53)
Changed opinion up: high deductible               3.21 (27.82)
Changed opinion down: high deductible           -12.56 (20.98)
Changed opinion up: universal medicare           15.54 (28.13)
Changed opinion down: universal medicare        -36.75 (23.22)
N                                                      126
[R.sup.2]                                             0.07

Overall Referendum                             Coefficient (SE)

Gained insurance                               -1582.4835
Lost insurance                                 -87.73 (8.62) ***
Became employed                                -21.22 (27.72)
Became unemployed                              -21.01 (33.95)
Change in income                                 0.22 (1.72)
Changed opinion up: cost important
Changed opinion down: cost important
Changed opinion up: # uninsured important
Changed opinion down: # uninsured important
Changed opinion up: large employer mandate
Changed opinion down: large employer mandate
Changed opinion up: universal mandate
Changed opinion down: universal mandate
Changed opinion up: high deductible
Changed opinion down: high deductible
Changed opinion up: universal medicare
Changed opinion down: universal medicare
N                                                     159
[R.sup.2]                                             0.02

Overall Referendum                             Coefficient (SE)

Gained insurance
Lost insurance
Became employed
Became unemployed
Change in income
Changed opinion up: cost important              66.12 (46.99)
Changed opinion down: cost important            -8.50 (22.62)
Changed opinion up: # uninsured important      -50.72 (47.80)
Changed opinion down: # uninsured important    -44.47 (27.13)
Changed opinion up: large employer mandate     -25.60 (20.09)
Changed opinion down: large employer mandate   -26.73 (22.46)
Changed opinion up: universal mandate           10.06 (22.10)
Changed opinion down: universal mandate          3.18 (22.84)
Changed opinion up: high deductible              4.73 (24.53)
Changed opinion down: high deductible           -6.96 (20.26)
Changed opinion up: universal medicare          11.57 (26.80)
Changed opinion down: universal medicare       -36.12 (21.24) *
N                                                     146
[R.sup.2]                                             0.04

Low-Income Referendum                          Coefficient (SE)

Gained insurance                               -28.54 (46.11)
Lost insurance                                  74.17 (46.05)
Became employed                                -74.02 (30.94) **
Became unemployed                              -45.59 (37.48)
Change in income                                -1.03 (0.99)
Changed opinion up: cost important             -39.77 (88.53)
Changed opinion down: cost important           -46.15 (45.98)
Changed opinion up: # uninsured important       18.21 (41.70)
Changed opinion down: # uninsured important     -4.67 (17.04)
Changed opinion up: large employer mandate     -51.15 (19.49) ***
Changed opinion down: large employer mandate   -22.07 (26.51)
Changed opinion up: universal mandate          -46.09 (37.83)
Changed opinion down: universal mandate          6.15 (23.00)
Changed opinion up: high deductible             13.85 (21.70)
Changed opinion down: high deductible            8.88 (21.85)
Changed opinion up: universal medicare          59.34 (23.63) **
Changed opinion down: universal medicare       -30.15 (16.27) *
N                                                      132
[R.sup.2]                                             0.20

Low-Income Referendum                          Coefficient (SE)

Gained insurance                               -41.02 (42.61)
Lost insurance                                  80.58 (46.47) *
Became employed                                -74.37 (27.58) ***
Became unemployed                               -9.98 (32.96)
Change in income                                -1.29 (1.18)
Changed opinion up: cost important
Changed opinion down: cost important
Changed opinion up: # uninsured important
Changed opinion down: # uninsured important
Changed opinion up: large employer mandate
Changed opinion down: large employer mandate
Changed opinion up: universal mandate
Changed opinion down: universal mandate
Changed opinion up: high deductible
Changed opinion down: high deductible
Changed opinion up: universal medicare
Changed opinion down: universal medicare
N                                                      179
[R.sup.2]                                             0.07

Low-Income Referendum                          Coefficient (SE)

Gained insurance
Lost insurance
Became employed
Became unemployed
Change in income
Changed opinion up: cost important             -20.23 (48.75)
Changed opinion down: cost important            -8.77 (28.80)
Changed opinion up: # uninsured important       18.82 (38.81)
Changed opinion down: # uninsured important    -12.53 (17.95)
Changed opinion up: large employer mandate     -50.13 (16.49) ***
Changed opinion down: large employer mandate   -29.88 (23.76)
Changed opinion up: universal mandate          -33.60 (29.97)
Changed opinion down: universal mandate          4.16 (19.26)
Changed opinion up: high deductible              5.74 (23.01)
Changed opinion down: high deductible           -2.14 (18.31)
Changed opinion up: universal medicare          58.12 (21.29) ***
Changed opinion down: universal medicare       -41.46 (17.55) **
N                                                      155
[R.sup.2]                                             0.08

Notes: Changed opinion up indicates that the respondent disagreed
with the reform proposal in 2008 and agreed with the proposal in
2010. Changed opinion down indicates that the respondent changed
their opinion in the opposite direction. Change in income is measured
in tens of thousands of dollars.

* Coefficient is significant at 10% level; ** significant at 5%
level; *** significant at 1% level.

TABLE 6 Change in WTP vs. Changes in Characteristics: Oaxaca-Blinder
Decomposition

                        Explanatory Variables

                                     Insurance
Change    N    Income   Employment   Coverage

Overall referendum
-12.66   310     X          X            X
-13.20   355     X          X            X
-10.23   329

Low-income referendum
  6.69   332     X          X            X
  4.80   388     X          X            X
  3.87   367

                       Decomposition

         Opinions   Due to Change in      Due to
          About       Explanatory       Change in
Change    Reform       Variables       Coefficients

Overall referendum
-12.66      X           -8.65 *           -4.03
-13.20                   -0.58            -12.91
-10.23      X           -8.73 **          -4.17

Low-income referendum
  6.69      X            -3.22             7.29
  4.80                   -1.06             4.49
  3.87      X            -3.85             6.66

Notes: Changes in opinions about reform includes changes in "cost
important," "# uninsured important," "large employer mandate,"
"universal mandate," "high deductible," and "universal medicare." The
column labeled "change" is the predicted change based on the included
explanatory variables, marked with "X." The "decomposition" columns
show the contributions of changes in explanatory variables and
changes in estimated coefficients to the total change. Data are from
the panel sample. The interaction term of the decomposition is not
shown.

* Decomposition component is significantly different from zero at 10%
level; ** significant at 5% level; *** significant at 1% level.