Background: Infant mortality rates (IMRs) vary dramatically across US states. A potential explanation centers on compositional differences in births from sociodemographic groups with a high risk of infant mortality. Objective: I seek to identify the contribution of key compositional factors to state-level disparities in IMRs using a series of Kitagawa–Blinder–Oaxaca decompositions. Methods: Drawing on linked birth–death records for US infants born between 2015 and 2017, I decompose cross-state disparities in IMRs into two components: (1) disparities attributable to differences in the distribution of maternal education, race/ethnicity, and age; and (2) disparities attributable to differences in the association between these sociodemographic characteristics and infant mortality (plus unmeasured compositional differences). I apply this approach to analyze disparities between the US IMR and 27 state IMRs. I then decompose IMR gaps between 630 pairs of states. I use linear regression to explore state-level predictors of variation in the second decomposition component. Results: In 7 of the 18 sample states with IMRs higher than the rest of the United States, led by Louisiana, South Carolina, and Georgia, more than 50% of this disparity can be attributed to the proportion of births from high-risk sociodemographic groups. In 11 high-IMR states, including Oklahoma, Indiana, and Missouri, more than 50% of the disparity is unexplained by the distribution of observed sociodemographic characteristics. The sample also includes nine states with IMRs lower than the rest of the United States. In Colorado, Oregon, and Minnesota, more than 50% of this advantage can be attributed to sociodemographic composition. Conversely, in six states, including New York, New Jersey, and California, the contribution of sociodemographic factors is outweighed by the unexplained decomposition component. Regression analyses show that variation in this component is associated with state differences in contextual predictors. Contribution: Decomposing cross-state differences in IMRs reveals considerable heterogeneity in the contribution of sociodemographic composition. This highlights variability in the social processes that produce disparities in infant mortality across populations.

中文翻译：

---

美国 27 个州婴儿死亡率差异的分解分析（作者：Benjamin Sosnaud）


背景：美国各州的婴儿死亡率 (IMR) 差异很大。一个可能的解释集中在婴儿死亡率高风险的社会人口群体的出生构成差异上。目标：我试图使用一系列 Kitakawa-Blinder-Oaxaca 分解来确定关键构成因素对 IMR 州级差异的贡献。方法：根据 2015 年至 2017 年出生的美国婴儿的相关出生死亡记录，我将 IMR 的跨州差异分解为两个组成部分：（1）由于母亲教育、种族/族裔和年龄分布的差异而导致的差异; (2) 差异归因于这些社会人口特征与婴儿死亡率之间的关联差异（加上未测量的构成差异）。我应用这种方法来分析美国 IMR 和 27 个州 IMR 之间的差异。然后我分解 630 对状态之间的 IMR 差距。我使用线性回归来探索第二个分解组件中变化的状态级预测因子。结果：在 IMR 高于美国其他地区的 18 个样本州中，有 7 个州（以路易斯安那州、南卡罗来纳州和佐治亚州为首），超过 50% 的这种差异可归因于高风险社会人口出生比例组。在 11 个高 IMR 州，包括俄克拉荷马州、印第安纳州和密苏里州，超过 50% 的差异无法通过观察到的社会人口特征的分布来解释。该样本还包括 IMR 低于美国其他地区的 9 个州。在科罗拉多州、俄勒冈州和明尼苏达州，超过 50% 的优势可归因于社会人口结构。 相反，在纽约、新泽西和加利福尼亚等六个州，社会人口因素的贡献被无法解释的分解成分所抵消。回归分析表明，该组件的变化与上下文预测变量的状态差异相关。贡献：分解 IMR 的跨州差异揭示了社会人口构成的贡献存在相当大的异质性。这凸显了社会进程的可变性，导致不同人群的婴儿死亡率存在差异。