The positive relationship between income and health is well established. However, the direction of causality remains unclear: do economic resources influence health, or vice versa? Exploiting a new source of exogenous income variation, this study examines the impact of the Alaska Permanent Fund Dividend (APFD) on newborns' health outcomes. The results show that income has a significantly positive, but modest effect on birth weight. We find that an additional $1,000 ($2,331 in 2011 dollars) increases birth weight by 17.7 g and substantially decreases the likelihood of a low birth weight (a decrease of around 14% of the sample mean). Furthermore, the income effect is higher for less‐educated mothers. Based on a gestation‐weight profile in the sample, increased gestation owing to the APFD could explain a maximum of 34%–57% of the measured weight increase, although we are unable to examine all the potential mechanisms. ( JEL I10, I18, I12)

ABBREVIATIONS AFDC Aid to Families with Dependent Children APFD Alaska Permanent Fund Dividend CES Consumer Expenditure Survey EITC Earned‐Income Tax Credit OLS Ordinary Least Squares WIC Women, Infants, and Children

I. INTRODUCTION The positive relationship between income and health is well established (Deaton 2003; Smith 2004, 2005). However, the direction of causality remains unclear: do economic resources influence health, or vice versa (Smith 2004)? To distinguish causal relationships from correlations, previous studies have used exogenous income shocks, such as lottery winnings, Germany's reunification, 2001 tax rebates, the Alaska Permanent Fund Dividend (APFD), and inheritances (Evans and Moore 2011; Frijters, Haisken‐DeNew, and Shields 2005; Kim and Ruhm 2012; Lindahl 2005; Meer, Miller, and Rosen 2003). However, these previous studies focus on adult health. An income shock during adulthood is unlikely to have an instant effect on health, because health has a stock nature, accumulating and/or depreciating over a person's lifetime. Driven by this concern, other studies have shifted their focus to the relationship between economic resources and children's health. These studies assume that it is more likely that children's health is influenced by family income than vice versa (Case, Lubotsky, and Paxson 2002; Currie and Stabile 2003).1 Furthermore, a newborn's health accumulates quickly during pregnancy, which is a much shorter period than an adult's full lifespan, making it less susceptible to confounding effects. Moreover, birth weight is a reliable measure of children's health, with considerable influence on health and socioeconomic outcomes later in life (Almond, Chay, and Lee 2005; Case, Lubotsky, and Paxson 2002; Corman, Joyce, and Grossman 1987; Rosenzweig and Schultz 1983; Rosenzweig and Wolpin 1991). Taking this shift in focus a step further, recent studies have explored the link between income and infant health using various exogenous income shocks. Currie and Cole (1993) used Aid to Families with Dependent Children (AFDC) income receipts and found no impact on birth weight. Hoynes, Page, and Stevens (2011), Almond, Hoynes, and Schanzenbach (2011), and Hoynes, Miller, and Simon (2015) used other social programs—the Supplemental Nutrition Program for Women, Infants, and Children (WIC), the Food Stamp Program, and the Earned Income Tax Credit, respectively—and found a positive impact on birth weight. Then, Lindo (2011) used a husband's displacement, Burlando (2011) used a prolonged blackout in Tanzania, and Mocan, Raschke, and Unel (2013) used skill‐biased technology shocks to study this relationship, all of which found a positive income effect. This study exploits the APFD, a new source of income shock to Alaskan residents, to obtain a consistent estimate of the effect of income on the health of their newborns. This approach has several advantages over those of previous studies. First, the APFD is a random income shock and, by design, its eligibility and amount is not related to the recipient's education or income. Second, the APFD is large enough to provide a significant income shock. The first dividend was $1,000 per person in nominal dollars ($2,331 in 2011 dollars). Therefore, given an average household size of 2.83 individuals, the first dividend per household was 10.4% of the median family income in Alaska ($27.3 K in 1981).2 Third, the APFD applies to the general population rather than only to the poor. Therefore, our results are more generalizable. The results show that the income increase owing to the APFD modestly improves newborns' health at birth. The APFD increases birth weight by 34.8 g (from the sample mean of 3,454 g), reduces the likelihood of low birth weight by approximately 0.7 percentage points (14% of the sample mean, 0.05), and improves the 5‐minute APGAR score by 0.063 (from the sample mean of 8.87). Based on a gestation‐weight profile in the sample, increased gestation owing to the APFD could explain a maximum of 34%–57% of the measured weight increase, although we are unable to examine all the potential mechanisms. The rest of this paper is organized as follows. Section II reviews the Alaska Permanent Fund, and Section III discusses the empirical methodology. Section IV presents the results. Lastly, Section V concludes the paper.

II. ALASKA PERMANENT FUND The Alaska Permanent Fund dates back to a financial windfall ($900 million) from the 1969 auctioning off of drilling rights in Prudhoe Bay. However, within a few years, concerns began to develop that the money had been squandered on basic community needs such as water, sewage, roads, schools, airports, and so on.3 With another boom in the mid‐1970s from oil‐related businesses and a growing consensus that Alaskans should receive permanent benefits from the development of the country's natural resources, Alaska's Constitution was amended in 1976 to create the Alaska Permanent Fund. At least twenty‐five percent of all mineral lease rentals, royalties, royalty sale proceeds, federal mineral revenue sharing payments, and bonuses received by the State shall be placed in a permanent fund, the principal of which shall be used only for those income‐producing investments specifically designated by law as eligible for permanent fund investments. All income from the permanent fund shall be deposited in the general fund unless otherwise provided by law. (Alaska Constitution Article IX, Section 15) The fund received its first deposit of dedicated oil revenues (valued at $734,000) in 1977, when the first barrel of oil from Prudhoe Bay arrived in Valdez through the Trans‐Alaska pipeline. Soon thereafter, extensive discussions took place on how to manage the fund and use its earnings, with policymakers taking 4 years to make any decisions (Groh and Erickson 2012). Some wanted to manage the fund as a development bank and use it for public infrastructure (as in 1969), while others wanted to manage it like a savings account and distribute its earnings directly to residents. In the latter case, the assumption was that residents would know how best to spend the money. In April 1980, overcoming diverse views, Governor Jay Hammond signed a bill supporting the management of the fund as a savings account. According to the bill, the dividend's earnings would be distributed to all Alaskan residents aged 18 years or older, depending on the number of years of residence since 1959.4 Each year of residence since 1959, when Alaska became a state, meant an additional $50. Therefore, the maximum possible payout was $1,050. However, the dividend could not be distributed until 1982 because of legal challenges against the cumulative residency feature—that is, “the longer you are here, the more you get.” Ron and Patricia “Penny” Zobel, both attorneys, filed a lawsuit that the cumulative residency provisions were unconstitutional. The litigation took 2 years to be resolved (Groh and Erickson 2012).5 In July 1980, Alaska's Superior Court ruled that the cumulative residency‐based dividend was unconstitutional. However, in October 1980, the State's Supreme Court judged that it was constitutional. The Zobels then appealed to the U.S. Supreme Court, and the final ruling was made on June 14, 1982. The U.S. Supreme Court judged the cumulative residency‐based dividend to be unconstitutional because it violated the U.S. Constitution's equal protection guarantee. Less than 2 weeks before this ruling, the Alaskan legislature had passed a standby bill whereby, if the U.S. Supreme Court invalidated the original plan, then every individual residing in Alaska for more than 6 months would receive an unconditional and equal annual dividend. On June 16, 1982, Governor Hammond signed the bill. The first dividend of $1,000 ($2,331 in 2011 dollars) per individual was distributed from June to December 1982. Our choice of study period is based on the criteria that the period should be short enough to avoid other confounding effects, but long enough to capture the dividend effect. The 1982 dividend was distributed from June to December 1982, while the 1983 dividend of $386.15 ($872 in 2011 dollars) was distributed from September to November 1983.6 Unfortunately, we do not know precisely when each individual received a payment. Furthermore, we do not know exactly how long it takes the dividend to affect birth weight during a 40‐week pregnancy. Therefore, we selected June 1984 as the end of the study period, as it is 40 weeks after September 1983, and seems a reasonable choice considering all factors.7 Alaskans had expected to start receiving the dividend in April 1980, but in the end, had to wait until June 1982. We divide the treatment period into two distinct periods: one anticipatory period, after the legislation of the APFD in April 1980, but before distribution; and a realized period, after the distribution of the APFD began in June 1982. We use this unusual opportunity to analyze the effects of income on health. Moreover, the amount of the dividend was uniform, regardless of demographic characteristics such as age, race, and education. An infant born in the previous calendar year, for example, was entitled to the same amount as his or her grandfather. Therefore, there is no need to be concerned with selection problem.

III. EMPIRICAL METHODOLOGY A. Data As the principal source of data, we employ the Natality Detail File, which includes all live births in the United States. This dataset provides a complete picture of newborns in the United States. For each birth, the dataset contains detailed demographic information on the mother (e.g., age, education, marital status, and race) and the baby (e.g., gender and siblings). It also contains detailed pregnancy records, including pregnancy characteristics (e.g., parity and plurality), prenatal care (e.g., the number of prenatal visits and the timing of the first prenatal visit), and pregnancy outcomes (e.g., birth weight and the APGAR score). The full analysis period of our study is 78 months, observing births from January 1978 to June 1984. From these 78 months, we obtain the following three observable periods: 27 months for the pretreatment period (January 1978–March 1980); 26 months for the anticipatory period, when individuals expected to receive the APFD (April 1980–May 1982); and 25 months for the realized period, when individuals began to receive the APFD (June 1982–June 1984).8 The timeline of the study period, including the major events, is presented in Figure 1. Figure 1 Open in figure viewerPowerPoint Timeline of Study Period Note: Digits following a year represent the month of that year—e.g., 1978.1 is January of 1978. B. Outcome Measurements We use birth weight as a reliable measure of children's health, as it is widely known to have considerable influence on health and socioeconomic outcomes later in life (Almond, Chay, and Lee 2005; Corman, Joyce, and Grossman 1987; Rosenzweig and Schultz 1983; Rosenzweig and Wolpin 1991). We measure birth weight in grams and define a low birth weight as less than or equal to 2,500 g. This outcome is significantly related to high medical costs and early death. Another outcome measure, the APGAR score, indicates the general condition of a newborn based on their appearance, pulse, response to stimulation, muscle tone, and breathing. The score ranges from zero to two for each of these five categories. Thus, the total APGAR score ranges from a minimum of zero to a maximum of ten. A higher score means the newborn is healthier.9 In addition, we use a low APGAR score—measured as less than or equal to seven—as a binary outcome. Furthermore, we want to examine possible mechanisms underlying the causal relationship between income and birth weight. In the literature, nutritional intake, prenatal care, cigarette/alcohol consumption, and parent/neighborhood characteristics—such as education and the availability of medical care—have been explored. Here, we examine the month prenatal care begins, the number of prenatal care visits, and the gestation period, all of which is available from our dataset. C. Econometric Model Difference‐in‐Difference Estimations We have data from one pretreatment period and two treatment periods. The best identification strategy fits naturally into a difference‐in‐difference estimation: (1) Y is the health outcome, measured by birth weight or APGAR score; i denotes the individual; s denotes the state; and m denotes the month of the year. Anticipatory indicates those babies born between April 1980 and May 1982 (as discussed in the previous section), and Realized indicates those born between June 1982 and June 1984. Importantly, the difference‐in‐difference estimate b captures the anticipatory effect and c captures the realized effect of the APFD on the health outcome. whereis the health outcome, measured by birth weight or APGAR score;denotes the individual;denotes the state; anddenotes the month of the year.indicates those babies born between April 1980 and May 1982 (as discussed in the previous section), andindicates those born between June 1982 and June 1984. Importantly, the difference‐in‐difference estimatecaptures the anticipatory effect andcaptures the realized effect of the APFD on the health outcome. State is a dummy for each state, and X contains the usual demographic information, including the mother's age, race (white, black, and Native American—including Aleuts, Eskimos, and others), and education level (no high school degree, high school degree, some college, and college degree or higher). Other variables include the child's gender, birth order, and the month of birth. The first child tends to show a lower birth weight. Thus, we include a binary variable for the first child (Wilcox, Chang, and Johnson 1996). In addition, we include the plurality of the pregnancy as a binary variable because such newborns are more likely to show a lower birth weight. However, given that the treatment is a random shock, there should not be much difference in the results, whether or not we include these control variables. We estimate each case separately and report both results. If people's beliefs about the APFD amount to be distributed were accurate, the permanent income hypothesis suggests that the effects of the anticipatory and realized periods will be the same. However, some people might not change their behavior, owing to liquidity constraints and an imperfect capital market. Therefore, even though the permanent income hypothesis predicts otherwise, a smaller income effect can be expected during the anticipatory period. Similarly, Evans and Moore (2011) found evidence of a sharp change in consumption immediately following an exogenous income shock. In addition, those who had not yet received the dividend might have felt deprived, which might have had temporary negative effects on their health until the dividend was received (Eibner and Evans 2005). Control Group A difference‐in‐difference model provides consistent estimates of the treatment effect only if the outcomes are the same for both the treatment state and the control states without the intervention. However, this can never be guaranteed (Evans and Lien 2005). The rest of the states in the United States did not benefit from the APFD and, thus, can be regarded as potential control states. The Natality Detail File does not provide complete data for five states during our study period. Data collection began in 1979 in Connecticut, Hawaii, Mississippi, and New Jersey, and in 1980 in Georgia. Therefore, we decided to ignore these states.10 To determine the best control group, we run the following regression model, based on Evans and Lien (2005), using data from Alaska and each potential control state for the 27‐month pretreatment period (January 1979–March 1980): (2) Equation 2 includes the same set of covariates as Equation 1, and U and V are dummy variables for the state and the month of birth, respectively. In addition, i denotes the individual; s denotes the state; and m indicates each of the 27 months over the pretreatment period. We use these 27 months to determine whether other states show the same birth weight pattern as Alaska. The key term in this regression is the coefficient λ m , which allows the dummy variables for the month to vary between the treated state, Alaska, and a potential control state (Evans and Lien 2005). According to Equation 2, if we cannot reject the hypothesis that λ 1 = λ 2 = λ 3 = … = λ 27 = 0, then Alaska and the control states share the same monthly birth weight pattern, conditional on X. Table 1 presents the F‐test results as p values. Based on the results, Montana and 12 other states are rejected at the 90% confidence level. Some small states might not have a large enough sample to reject the F‐test when selecting control states. Therefore, we sort the potential control states in ascending order of number of observations. The smallest rejected state is Montana. Ten states have fewer observations than Montana (Delaware, the District of Columbia, North Dakota, South Dakota, Vermont, Wyoming, Rhode Island, Nevada, New Mexico, and New Hampshire). We might not be able to reject the null hypothesis for these states because of a Type II error. Therefore, we exclude these additional ten states. These states are identified shown in italics and control states are boldfaced in Table 1. Table 1. Control States States Count p value (F‐test) States Count p value (F‐test) Delaware 10,322 .515 South Carolina 111,200 .016 District of Columbia 10,975 .536 Maryland 117,737 .117 North Dakota 13,473 .760 Kentucky 130,558 .047 South Dakota 16,330 .358 Alabama 136,457 .108 Vermont 16,394 .102 Washington 139,226 .365 Wyoming 16,744 .260 Minnesota 140,723 .154 Rhode Island 26,246 .088 Tennessee 151,115 .064 Nevada 26,260 .473 Massachusetts 156,501 .033 New Mexico 27,162 .823 Wisconsin 160,206 .074 New Hampshire 28,473 .247 Virginia 166,225 .304 Montana 30,777 .098 Missouri 167,708 .372 Maine 36,102 .167 Louisiana 172,950 .036 Idaho 44,150 .133 North Carolina 186,655 .108 Arkansas 47,100 .005 Indiana 191,667 .278 Arizona 50,979 .469 Florida 264,983 .187 Nebraska 57,666 .191 Pennsylvania 272,612 .194 West Virginia 66,281 .516 Michigan 316,459 .075 Kansas 85,378 .023 Ohio 367,909 .189 Utah 89,753 .196 Illinois 404,328 .113 Oregon 90,935 .037 California 415,017 .137 Iowa 102,727 .164 New York 525,921 .098 Colorado 102,775 .258 Texas 555,211 .209 Oklahoma 106,276 .087

IV. RESULTS A. Main Findings Table 2 summarizes the basic statistics for the births in Alaska from January 1978 to June 1984. We exclude observations from the noncontrol states, as listed in the previous section. Some observations with missing information on the variables are also excluded. As a result, we obtain 7.7 million births for the analysis, of which 52,346 occurred in Alaska. Table 2. Descriptive Statistics: Births between January 1978 and June 1984a Alaska Control Mean SD Mean SD Outcomes Birth weight (g) 3,453.62 574.39 3,361.44 585.51 Low birth weight (≤2,500 g) 0.05 0.21 0.06 0.24 Five‐minute APGAR 8.87 0.94 9.05 0.96 Low 5‐minute APGAR (≤7) 0.02 0.15 0.02 0.14 Month prenatal care begins 2.89 1.51 2.75 1.48 Number of prenatal care visits 10.24 3.81 10.68 3.63 Demographic variables Male 0.51 0.50 0.51 0.50 Mother's age 25.64 5.03 25.19 5.19 White 0.75 0.43 0.84 0.37 Black 0.03 0.18 0.14 0.35 Native American 0.19 0.39 0.01 0.07 Other 0.03 0.17 0.01 0.12 No high school diploma 0.14 0.35 0.21 0.41 High school diploma 0.44 0.50 0.42 0.49 Some college 0.24 0.42 0.18 0.39 College diploma or higher 0.16 0.37 0.14 0.35 Married 0.85 0.35 0.83 0.38 Pregnancy conditions Plurality 0.02 0.14 0.02 0.14 Gestation period 39.68 2.70 39.51 2.79 Number of children 1.93 1.00 1.91 0.97 First child 0.42 0.49 0.43 0.49 Second child 0.31 0.46 0.33 0.47 Third child 0.15 0.36 0.15 0.36 ≥Fourth child 0.10 0.30 0.09 0.29 Obs. 52,346 7,650,406 Mothers in Alaska are slightly older and more likely to be Native American. They are also more likely to be educated and married. However, we find no difference in the distribution of newborns' gender and plurality between Alaska and the control states. With regard to birth order, approximately 42% and 43% of the births were the first child in Alaska and in the control states, respectively. Newborns in Alaska show slightly more siblings and longer gestation than in the control states. Table 3 shows the effects of the APFD on the birth weight of newborns as difference‐in‐difference estimates. We measure the APFD effect using the interaction between the Alaska dummy variable and the dummy variables Anticipatory and Realized. Table 3. The Effects of the Alaska Permanent Fund Dividend on Birth Outcomes Dependent Variable Birth Weight Low Birth Weight 5‐Minute APGAR Low 5‐Minute APGAR (1) (2) (3) (4) (5) (6) (7) (8) Alaska*Anticipatory 17.488** 19.866** −0.004** −0.005** 0.029** 0.043** −0.003** −0.003** (5.194) (1.144) (0.001) (0.0004) (0.011) (0.004) (0.0003) (0.0003) Alaska*Realized 30.924** 34.833** −0.007** −0.007** 0.050** 0.063** −0.004** −0.004** (6.569) (2.185) (0.001) (0.0005) (0.011) (0.006) (0.0004) (0.0004) Anticipatory 2.442 1.118 0.001 0.001* −0.070** −0.081** 0.0003 0.0001 (5.194) (1.197) (0.001) (0.0004) (0.011) (0.004) (0.0003) (0.0003) Realized 2.072 2.271 0.002 0.001* −0.132** −0.142** −0.0002 −0.0004 (6.569) (2.262) (0.001) (0.001) (0.011) (0.006) (0.0004) (0.0004) Including Independent Variables N Y N Y N Y N Y R2 0.0002 0.116 0.00004 0.078 0.003 0.035 0.00001 0.008 Obs. 7,702,752 7,702,752 7,702,752 7,702,752 7,702,752 7,702,752 7,702,752 7,702,752 The first column shows the results based on model (1), excluding other independent variables X. Birth weight, the outcome measure used here, increases by 17.5 g when the APFD is anticipated and by 30.9 g when the APFD is received. Standard errors are clustered at the state level, based on the assumption that births within a state might be correlated, even when time‐invariant and state‐specific characteristics are controlled for. The estimates are statistically significant at the 5% level. In the second column, we add the other independent variables—listed in the previous section—to examine any change in the results. The R2 increases substantially, and the key coefficients change from 17.5 to 19.9 g for the anticipatory period and from 30.9 to 34.8 g for the realized period.11 The difference in outcomes when other independent variables are included seems small enough to assure the exogeneity of the APFD income. On average, Alaskan households with a newborn received $1,961 ($4,500 in 2011 dollars) annually for the first two dividends.12 Thus, an additional $1,000 ($2,331 in 2011 dollars) increases birth weight by 17.7 g. Birth weight is a good composite indicator of newborns' health. However, it remains unclear whether an increase in birth weight means improved health outcomes across the whole range of birth weight. In this regard, some studies have focused on low birth weight, defined as less than or equal to 2,500 g. It is well known that a low birth weight increases infant mortality (McCormick 1985). The third column of Table 3 shows the results using low birth weight as an outcome. According to the data, approximately 5% of all newborns in Alaska have a low birth weight. The APFD diminishes the incidence of low birth weights by approximately 0.4 percentage points for the anticipatory period and 0.7 percentage points for the realized period. Both of these findings are statistically significant and substantial in magnitude (8% and 14% of the sample mean). When we add other control variables in the fourth column, the effect of the APFD increases from 0.4 to 0.5 percentage points for the anticipatory period and shows no change for the realized period.13 The next two columns present the results using the 5‐minute APGAR score as an alternative health measure. The higher the APGAR score, the better is the newborn's health. The average 5‐minute APGAR score is 8.87, and the APFD increases the APGAR score by 0.043 for the anticipatory period and 0.063 for the realized period, when including other control variables. This increase in the APGAR score due to the APFD is significant, but small in magnitude. The 5‐minute APGAR scores show a slight downturn over time, as shown in the coefficients of Anticipatory and Realized, which are statistically significant, but small in magnitude. In the literature, a low APGAR score (i.e., less than or equal to seven) is also sometimes used as a health outcome (Nelson and Ellenberg 1981; Thorngren‐Jerneck and Herbst 2001). When we examine that binary outcome here, we find that the APFD diminishes the incidence of low 5‐minute APGAR scores by approximately 0.3 percentage points for the anticipatory period and 0.4 percentage points for the realized period. Both are statistically significant and substantial in magnitude. Then, there might be heterogeneous effects of the APFD, depending on income level. That is, those with a lower income might exhibit bigger effects, even with the same level of income shock.14 Although we do not have information on income, we can use education level as a proxy. We present this subgroup analysis in Table 4, using both birth weight and low birth weight indicator as the dependent variables. The results show that those who are less educated (i.e., a high school diploma or lower) experience greater effects, as expected. In the last column, those with a college diploma or higher show a decrease of 14 g in birth weight or a positive probability of a low birth weight during the anticipatory period, which is counterintuitive. For less‐educated mothers, the income effect is greater during the realized period rather than during the anticipatory period, possibly owing to liquidity constraints. Table 4. The Effects of the Alaska Permanent Fund Dividend on Birth Weight by Educational Level Dependent Variable Birth Weight Subsamples by Mother's Education No High School Diploma High School Diploma Some College College Diploma or Higher Alaska*Anticipatory 30.876** 41.529** −2.026 −14.268** (2.059) (1.187) (1.454) (1.447) Alaska*Realized 43.669** 49.782** 28.687** 1.185 (2.254) (2.437) (2.453) (2.404) R2 0.094 0.106 0.110 0.110 Obs. 1,596,747 3,273,657 1,405,571 1,106,203 Dependent Variable Low Birth Weight Subsamples by Mother's Education No High School Diploma High School Diploma Some College College Diploma or Higher Alaska*Anticipatory −0.006** −0.010** −0.004** 0.006** (0.001) (0.0003) (0.001) (0.001) Alaska*Realized −0.010** −0.013** −0.003** 0.002** (0.001) (0.0005) (0.001) (0.001) Mean of Dependent 0.092 0.061 0.052 0.044 R2 0.064 0.078 0.082 0.084 Obs. 1,596,747 3,273,657 1,405,571 1,106,203 B. Mechanisms In this section, we examine the potential mechanisms underlying the causal relationship between income and birth weight. Typical inputs in a birth weight production function are the following: nutritional intake, prenatal care, cigarette/alcohol consumption, and parent/neighborhood characteristics, such as education and the availability of medical care (Currie and Cole 1993). We examine some of these channels to identify the underlying mechanisms in the relationship. First, we determine whether the increase in birth weight is driven by any increase in the use of prenatal care that might, in turn, be the result of the APFD. Prenatal care is measured by the month of pregnancy in which prenatal cares begins (e.g., first month, third month, etc.) and the number of prenatal care visits. As shown in the first and second columns of Table 5, prenatal care begins earlier when the APFD is anticipated and when it is realized, after controlling for the common downward trend over time.15 For the anticipatory period, prenatal care begins 2.46 days (0.082 × 30) earlier, which is statistically significant. For the realized period, prenatal care begins 2.25 days earlier. However, the earlier prenatal care did not change the number of prenatal care visits during pregnancy in a statistically significant way. This finding is not surprising given that a 2‐day earlier start of prenatal care would not give enough time for a full additional visit during the pregnancy, although prenatal care visits happen weekly in the last month of pregnancy. Table 5. The Effects of the Alaska Permanent Fund Dividend on Medical Care Use Dependent Variable Month Prenatal Care Begins Number of Prenatal Care Visits Gestation Period (in weeks) Alaska*Anticipatory −0.082** −0.023 0.089** (0.009) (0.028) (0.006) Alaska*Realized −0.075** −0.104 0.107** (0.015) (0.053) (0.008) Anticipatory −0.038** 0.137** −0.055** (0.009) (0.027) (0.006) Realized −0.040* 0.276** −0.113** (0.015) (0.054) (0.007) Including Independent Variables Y Y Y R2 0.111 0.102 0.042 Obs. 7,702,752 7,702,752 7,702,752 When we examine the gestation period, presented in the third column, we find a statistically significant increase of 0.1 (one‐tenth of a week) for both the anticipatory period and the realized period. The birth weight in Alaska increases between 170 g (at the mean) and 196 g (at the 25th percentile) with an additional week of gestation at 38 weeks of pregnancy, depending on the weight distribution of newborns. At 39 weeks, the weight gain is smaller, but still 120 g for newborns of median weight. Therefore, a longer gestation of even one‐tenth of a week would explain between 12 and 20 g of the increase in birth weight, which could explain between 34% and 57% of the estimated birth weight increase.16 However, some of the results in Table 5 should be interpreted with caution. There is some evidence of varied reliability in the data of the birth certificates. The APGAR score, which is subjective in nature, might differ depending on who measures it (Montgomery 2000). Prenatal care variables in the birth certificate are also poorly measured, and measurement errors in the gestation period could be significant as a result of recall bias or misclassification by postconceptional bleeding (Clark, Fu, and Burnett 1997; Schoendorf and Branum 2006). Other changes might help explain the positive APFD effect on newborns' health. For example, the earned‐income tax credit (EITC) became available in several states in the late 1980s and 1990s (Hoynes, Miller, and Simon 2015). However, there was no change in the EITC in Alaska during the study period. In addition, no states used in the control group experienced a change in the EITC during our study period.17 Carrington (1996) examined the effects of pipeline construction on the labor market in Alaska, but the pipeline was completed in 1977, before the introduction of the APFD. Therefore, our results are not subject to these confounding factors. We cannot check other potential mechanisms, such as grocery consumption, drinking, and smoking because of data limitations. Although the Consumer Expenditure Survey (CES) facilitates the measurement of recurring monthly expenses, such as house and car payments, reasonably well, it is somewhat limited in the measurement of goods such as alcohol and food outside the home (Meyer and Sullivan 2012). Moreover, medical expenditure in the CES is not available for the entire study period.18 C. Robustness Checks First, we examine whether the first‐year APFD effect differs from the 2‐year APFD effect investigated above. The different size and distribution timing of the first‐year APFD could yield results that differ from the observed effects. Moreover, we limit the study period to the pretreatment period, and split the period in several ways to enable falsification testing. This examines whether our estimated effect is genuine or whether it is potentially spurious because of the large sample size and any time trends in birth weight. Then, we examine the compositional change of newborns owing to endogenous fertility decisions by Alaskans after the introduction of the APFD. Finally, we use a synthetic control approach to check the robustness of the original results against a different identification methodology. Falsification Test The left section of Table 6 shows the results using the first‐year distribution of the APFD only. Babies born after September 1983, the month that the 1983 dividend payment began, are excluded from the estimation. We expect that the health improvement could be bigger because of the bigger dividend amount in the first year. Note that the first‐year distribution of the APFD occurred over a longer monthly period, making it more difficult to match the timing of receiving the APFD. The results show that the birth weight and low birth weight outcomes increase slightly. However, the 5‐minute APGAR scores become slightly smaller.19 Table 6. Robustness Check Dependent Variable Receiving 1982 Check Only (Born Between January 1978 and July 1983) Pretreatment Period Only (Born Between January 1978 and March 1980) Birth Weight Low Birth Weight 5‐Minute APGAR Birth Weight Low Birth Weight 5‐Minute APGAR Alaska*Anticipatory 19.576** −0.005** 0.043** −15.612** 0.006** −0.085** (1.134) (0.0004) (0.003) (1.459) (0.0005) (0.004) Alaska*Realized 35.861** −0.008** 0.061** −24.621** 0.012** −0.039** (2.172) (0.001) (0.006) (1.513) (0.001) (0.007) Anticipatory 1.381 0.001 −0.081** 4.116** −0.001 −0.021** (1.162) (0.0004) (0.004) (1.389) (0.0005) (0.005) Realized 2.279 0.001 −0.131** 6.027** −0.001 −0.055** (2.262) (0.0005) (0.006) (1.556) (0.001) (0.007) Including Independent Variables Y Y Y Y Y Y R2 0.116 0.078 0.035 0.111 0.076 0.032 Obs. 6,457,260 6,457,260 6,457,260 2,052,279 2,052,279 2,052,279 Given our large sample size and any trends in birth weight over time, our estimated effect of the APFD may be spurious. Therefore, we perform falsification tests by limiting the study period to the pretreatment period only (January 1978–March 1980) and splitting the period in several ways to run identical models. The results reported here use October 1978–June 1979 as the placebo anticipatory period and July 1979–March 1980 as the placebo realized period. These show no effect over time.20 Another possible concern is a compositional change of newborns owing to endogenous fertility decisions by Alaskans after the introduction of the APFD.21 We examine the observable characteristics of newborns such as mother's age, race, marital status, and education level in Table 7. The number of observations decreases for education level because of missing observations. No coefficients show statistically significant results at the 95% confidence level, except marital status. Therefore, we feel comfortable that the observable characteristics of mothers did not change during our study period. Table 7. Compositional Change of Birth Dependent Variable Mother's Age White Married Graduated High School Alaska*Anticipatory 0.037 −0.002 −0.015* −0.011 (0.043) (0.007) (0.005) (0.007) Alaska*Realized 0.017 −0.002 −0.010 −0.009 (0.053) (0.008) (0.006) (0.008) Anticipatory 0.209** −0.010 −0.020** 0.013 (0.043) (0.007) (0.005) (0.007) Realized 0.539** −0.015 −0.038** 0.023 (0.053) (0.008) (0.006) (0.008) Including Independent Variables N N N N R2 0.002 0.001 0.002 0.001 Obs. 7,702,752 7,702,752 7,702,752 7,382,178 Synthetic Control Approach Another possible identification strategy for the APFD is the synthetic control approach. Abadie, Diamond, and Hainmueller (2010) determined a policy effect on an entire population without concern for measurement errors, as they assumed that the aggregate is measured with very little error. Here, to measure the effects of California's tobacco control program on statewide smoking rates, the authors constructed a synthetic California as a weighted average of the 38 states, which best reproduce the values of a set of predictors of cigarette consumption in California before the intervention. Of the 50 possible candidate states, 12 states with cigarette tax hikes or formal tobacco control programs are excluded. The difference in the cigarette consumption level between California and the synthetic California during the intervention period is the effect of California's tobacco control program. The synthetic control approach assumes that the observable characteristics of possible controls are time‐invariant. In our study, we exclude five states because they violate this assumption by showing sharp changes in the composition of mothers: California, Minnesota, Oregon, Utah, and Washington. Therefore, we use 45 states as possible controls for Alaska. However, our results are qualitatively robust to the inclusion of the discarded states. Ultimately, the following three states are selected as the synthetic Alaska, with weights shown in the parentheses: South Dakota (46.9%), Maine (44.9%), and Nevada (8.2%). We use a quarterly period for smooth presentation.22 During the pretreatment period, we find no difference between Alaska and the synthetic Alaska, as shown in Figure 2A. However, the APFD increased the birth weight by 28 g during the anticipatory period and 35 g during the realized period, on average. Note that these values are surprisingly similar to those obtained using the difference‐in‐difference estimation (i.e., 20 g and 34 g, respectively). Then, we perform a placebo test using South Dakota as the treated state, but find no intervention effect, as shown in Figure 2B. Figure 2 Open in figure viewerPowerPoint (A) Effects of the Alaska Permanent Fund Dividend on Birth Weight: Synthetic Control Approach and (B) Synthetic Control Approach: Placebo Test (South Dakota) Notes: (A) The first line indicates the beginning of the anticipatory period and the second line indicates the beginning of the realized period. We used a synthetic control approach following Abadie et al. A synthetic control was decided from among 45 possible control states. State weights in the synthetic Alaska are South Dakota (46.9%), Maine (44.9%), and Nevada (8.2%). (B) The first line indicates the beginning of the anticipatory period and the second line indicates the beginning of the realized period.

V. CONCLUSION The positive relationship between income and health has been widely examined. In the literature, some works using exogenous income shocks have found little or no impact when focusing on adulthood (Kim and Ruhm 2012; Lindahl 2005; Meer, Miller, and Rosen 2003). This result might be due to the stock nature of health. In contrast, other works focusing on children have exploited social programs, husband displacement, a prolonged blackout, and skill‐biased technology shocks as alternative sources of exogenous income shocks (Almond, Hoynes, and Schanzenbach 2011; Burlando 2011; Currie and Cole 1993; Hoynes, Page, and Stevens 2011; Hoynes, Miller, and Simon 2015; Lindo 2011; Mocan, Raschke, and Unel 2013). In this study, we use the APFD as a new source of an exogenous income shock. The APFD applies to the overall population, rather than a specific part of the population (e.g., a low‐income group), and measures the health outcome of newborns with minimum confounding factors. We find that the APFD increases birth weight by a modest 34.8 g. The APFD leads to a substantial decrease in cases of low birth weight (14% of the sample mean) and a small increase in 5‐minute APGAR scores. It is difficult to compare this study's results with the findings of previous studies. Lindahl (2005) found that a 10% increase in income generates a health outcome improvement with a 0.01–0.02 standard deviation. In our study, an increase of about 7.2% (after combining 10.4% and 4% for the first and the second dividend, respectively) in income generates an increase with a 0.06 standard deviation. Our result is larger than that of Lindahl, and in line with previous findings on the stock nature of health. Focusing on infant health, Hoynes, Miller, and Simon (2015) found that an increase of $1,000 in income from the EITC is associated with a 19 g increase in birth weight. The cash transfer program in Mexico showed birth weight improvements of 127.3 g with monthly cash transfers of $90 to $160, or approximately 17%–20% of household consumption (Barber and Gertler 2008). Our results show a 17.7 g increase for the $1000 APFD (i.e., 47 g for a 10% income shock). We explain this smaller effect by the fact that the APFD affects the overall population rather than just the poor, which contrasts with the programs adopted in other studies. Nonetheless, it is notable that the income effect for less‐educated mothers (i.e., high school diploma or less) is closer to the effect identified in prior studies. We also examined the mechanisms underlying the causal relationship between income and birth outcome. Here, our results indicate that earlier prenatal care and longer gestation could explain some of the mechanism.

1 In particular, Case, Lubotsky, and Paxson ( 2002

In particular, Case, Lubotsky, and Paxson ( 2 The second dividend of $386.15 was 4% of the median family income.

The second dividend of $386.15 was 4% of the median family income. 3 http://www.apfc.org/home/Media/publications/2009AlaskansGuide.pdf, accessed September 17, 2013.

http://www.apfc.org/home/Media/publications/2009AlaskansGuide.pdf, accessed September 17, 2013. 4 Alaska State Legislature Chapter 21: Session Laws of Alaska. 1980 (http://news.google.com/newspapers?id=a44qAAAAIBAJ&sjid=SVwEAAAAIBAJ&pg=5776,629859&dq=alaska+permanent+fund+dividend&hl=en), accessed September 17, 2013.

Alaska State Legislature Chapter 21: Session Laws of Alaska. 1980 (http://news.google.com/newspapers?id=a44qAAAAIBAJ&sjid=SVwEAAAAIBAJ&pg=5776,629859&dq=alaska+permanent+fund+dividend&hl=en), accessed September 17, 2013. 5 Their lawsuit included another issue related to abolishing income tax for those who had lived in Alaska for more than 3 years. Alaska's Supreme Court ruled in favor of the Zobels regarding the income tax, and as a result, Alaska's income tax was abolished completely in 1980.

Their lawsuit included another issue related to abolishing income tax for those who had lived in Alaska for more than 3 years. Alaska's Supreme Court ruled in favor of the Zobels regarding the income tax, and as a result, Alaska's income tax was abolished completely in 1980. 6 However, starting in 1984, all dividend payments were distributed regularly in the last quarter of the year. In addition, the size of the dividend could be anticipated based on 10.5% of the fund's total realized earnings over the previous five fiscal years.

However, starting in 1984, all dividend payments were distributed regularly in the last quarter of the year. In addition, the size of the dividend could be anticipated based on 10.5% of the fund's total realized earnings over the previous five fiscal years. 7 This end limit choice is made to ensure all babies had received the dividend during pregnancy. For example, babies born in July 1984, 40 weeks from October 1983, might not have received the payment during pregnancy if the dividend had been distributed in September 1983. Nonetheless, when the end of the study period is extended to August 1984, 40 weeks from November 1983, the results remain much the same.

This end limit choice is made to ensure all babies had received the dividend during pregnancy. For example, babies born in July 1984, 40 weeks from October 1983, might not have received the payment during pregnancy if the dividend had been distributed in September 1983. Nonetheless, when the end of the study period is extended to August 1984, 40 weeks from November 1983, the results remain much the same. 8 We cannot use a regression discontinuity model to compare outcomes just before and after the income shock for the following reasons. First, we do not know from the data precisely when individuals received the APFD. Second, even if we did have this information, we do not know how long it took for the shock to influence fetus health.

We cannot use a regression discontinuity model to compare outcomes just before and after the income shock for the following reasons. First, we do not know from the data precisely when individuals received the APFD. Second, even if we did have this information, we do not know how long it took for the shock to influence fetus health. 9 The APGAR score is measured 5 minutes after birth and recorded on the birth certificate. It is regarded as critically low if it is below three, fairly low if it is between four and six, and normal if it is between seven and ten. For more information, see Casey, McIntire, and Leveno ( 2001

The APGAR score is measured 5 minutes after birth and recorded on the birth certificate. It is regarded as critically low if it is below three, fairly low if it is between four and six, and normal if it is between seven and ten. For more information, see Casey, McIntire, and Leveno ( 10 The other possibility is to include these five states only for the years available. We tried this approach as well, and found our results to be robust to the change.

The other possibility is to include these five states only for the years available. We tried this approach as well, and found our results to be robust to the change. 11 When we use all 50 states as the control the coefficients decrease to 17.3 ( SE 1.42) and 31.3 ( SE 2.31) for the anticipatory and realized periods.

When we use all 50 states as the control the coefficients decrease to 17.3 ( 1.42) and 31.3 ( 2.31) for the anticipatory and realized periods. 12 The typical household size was 2.83, and the APFD payments per recipient were $1,000 for 1982 and $386.15 for 1983. Over the 2 years, the average household received $3,923, which is $1,961 annually. We assume that the newborn is exposed to this $1,961 income shock.

The typical household size was 2.83, and the APFD payments per recipient were $1,000 for 1982 and $386.15 for 1983. Over the 2 years, the average household received $3,923, which is $1,961 annually. We assume that the newborn is exposed to this $1,961 income shock. 13 When a quantile regression is used to explore any heterogeneous APFD effect across the birth weight distribution, the greatest and statistically significant effect is found at approximately 2,500 g. For example, the APFD effect is 47.33 ( SE 18.44) during the anticipatory period and 74.04 ( SE 17.76) during the realized period at the fifth quantile (2,410 g). However, when we use lower cutoffs for low birth weight, we do not find any monotonically increasing effect below 2,500 g.

When a quantile regression is used to explore any heterogeneous APFD effect across the birth weight distribution, the greatest and statistically significant effect is found at approximately 2,500 g. For example, the APFD effect is 47.33 ( 18.44) during the anticipatory period and 74.04 ( 17.76) during the realized period at the fifth quantile (2,410 g). However, when we use lower cutoffs for low birth weight, we do not find any monotonically increasing effect below 2,500 g. 14 Birth weight shows a different trend over time, depending on education level: there is a decreasing trend for those who are less educated (i.e., less than high school diploma), but an increasing trend for those who are better educated (i.e., some college or above).

Birth weight shows a different trend over time, depending on education level: there is a decreasing trend for those who are less educated (i.e., less than high school diploma), but an increasing trend for those who are better educated (i.e., some college or above). 15 This trend is estimated using the coefficients of Anticipatory and Realized . Here, we estimate significant trends of an increase in the number of prenatal care visits and a decrease in gestation periods.

This trend is estimated using the coefficients of and . Here, we estimate significant trends of an increase in the number of prenatal care visits and a decrease in gestation periods. 16 Arguably, gestation is an outcome itself and also a factor related to birth weight. Low birth weight is caused by either short gestation or slow growth per gestation, or both. The literature (Hendler et al. 2005 2005

Arguably, gestation is an outcome itself and also a factor related to birth weight. Low birth weight is caused by either short gestation or slow growth per gestation, or both. The literature (Hendler et al. 17 Only Maryland (1987), Vermont (1988), and Wisconsin (1989) had EITC changes in the 1980s, and those years were not within our study period (Llobrera and Zahradnik 2004 www.stateEITC.info).

Only Maryland (1987), Vermont (1988), and Wisconsin (1989) had EITC changes in the 1980s, and those years were not within our study period (Llobrera and Zahradnik www.stateEITC.info). 18 During our study period, data on medical expenditure is available for 1980 and 1981 in the CES. Unfortunately, other data sources such as the birth certificates and the Behavioral Risk Surveillance System provide no data on health behaviors and consumption during our study period.

During our study period, data on medical expenditure is available for 1980 and 1981 in the CES. Unfortunately, other data sources such as the birth certificates and the Behavioral Risk Surveillance System provide no data on health behaviors and consumption during our study period. 19 As an alternative, to ensure all residents had received the payment, we changed the realized period to January 1983–June 1984, and found similar results.

As an alternative, to ensure all residents had received the payment, we changed the realized period to January 1983–June 1984, and found similar results. 20 As seen in Figure 2A, there is a relative downward trend during the pretreatment period. The magnitude varies from −8 g to −26 g for the realized period, depending on the respective artificial cutoff.

As seen in Figure 2A, there is a relative downward trend during the pretreatment period. The magnitude varies from −8 g to −26 g for the realized period, depending on the respective artificial cutoff. 21 Existing literature suggests no or very little effect of income on fertility (Moffitt 1998 2011 2011

Existing literature suggests no or very little effect of income on fertility (Moffitt 22 The results are qualitatively similar to those obtained using a monthly period.