基于世界银行数据20个低收入国家婴儿死亡率影响因素的岭回归分析

Ridge Regression Analysis of Factors Influencing Infant Mortality Rates in 20 Low-Income Countries Based on Downloaded Data from the World Bank

摘要: 目的：论证低收入国家是否存在人均GDP与婴儿死亡率关系变化的转折，判断人均GDP是否是婴儿死亡率的关键性影响因素，寻找不同阶段婴儿死亡率的影响因素。方法：以20个低收入国家婴儿死亡率及经济社会指标为研究对象，以测算的婴儿死亡率与人均GDP相关分析最小样本量为基础，进行阶段性分组。通过相关分析判断婴儿死亡率与人均GDP从非负相关向负相关的关系转变，再以婴儿死亡率为因变量，各经济社会指标为自变量，分别进行阶段性岭回归分析，进一步判断人均GDP对婴儿死亡率从无积极影响向有积极影响的关系转变。自变量的回归系数t检验差异有统计学意义的自变量为婴儿死亡率的影响因素，比较标准化回归系数绝对值大小判断对婴儿死亡率的影响程度。结果：20个低收入国家婴儿死亡率、人均GDP处于全球低层次水平，婴儿死亡率总体上呈下降趋势，但平均下降量小、下降速度缓慢，人均GDP总体上呈增长趋势，但平均增长量小、增长速度缓慢。在78个阶段中，婴儿死亡率与人均GDP有45个(57.69%)阶段呈不同程度的相关关系、33个(42.31%)阶段无相关关系。尼日尔1981年前1个阶段二者无相关关系，后两个阶段呈负相关系；卢旺达1980年前2个阶段二者无相关关系，后3个阶段呈负相关系。78个阶段的岭回归模型F检验差异均有统计学意义(P < 0.05)，k值均较小，R2均接近1。在45个相关阶段中有20个阶段、在33个不相关阶段中有28个阶段人均GDP回归系数t检验差异均无统计学意义(P > 0.05)，另外在5个国家的6个阶段中出现了总生育率、城镇人口增长率、谷类产量、农业用地面积、人均居民最终消费支出、CO2排放量等6个自变量的回归系数t检验差异均无统计学意义(P > 0.05)，除此之处，其他自变量的回归系数t检验差异均有统计学意义(P < 0.05)。尼日尔在2000年前的2个阶段中，1个阶段的人均GDP回归系数t检验差异无统计学意义(P > 0.05)、一个有统计学意义(P < 0.05)但标准回归系数为正值，后1个阶段有统计学意义(P < 0.05)且标准回归系数为负值；卢旺达在2000年前的4个阶段中，3个阶段人均GDP回归系数t检验差异无统计学意义(P > 0.05)、1个阶段有统计学意义(P < 0.05)但标准回归系数为正值，后2个阶段差异有统计学意义(P < 0.05)且标准回归系数为负值。岭回归模型自变量回归系数t检验差异有统计学意义的经济社会因素指标共24个，累积频数为412例次。其中，人口相关因素包括人口密度、总生育率、超百万城市群的人口、人口增长、城镇人口增长率(53.16%)；农业相关因素包括农业用地、耕地、农业用地面积、农业增加值、谷类产量(16.50%)；环境与能源相关因素包括CO2排放量、一氧化氮排放量、甲烷排放量、通电率、耗电量、能源使用量(13.11%)；经济相关因素包括人均GDP、人均居民最终消费支出、人均当前卫生支出、卫生支出占比(10.68%)；教育相关因素包括小学生毕业率、中小学女生与男生的入学比例、高等院校入学率(5.10%)；卫生相关因素仅感染HIV的成年女性(1.46%)。前十位影响因素依次为人口密度、总生育率、超百万城市群的人口、人均GDP、农业用地、人口增长、城镇人口增长率、耕地、一氧化氮排放量、甲烷排放量(79.37%)。人均GDP在每个阶段的婴儿死亡率第一、第二影响因素中均位居第5，第三影响因素中位居第17，第四影响因素中位居第2，第五影响因素中位居第4。结论：低收入国家婴儿死亡率下降态势不容乐观。人均GDP与婴儿死亡率关系缺乏稳定性，仅尼日尔、卢旺达出现了从无积极影响向有积极影响的关系变化。婴儿死亡率影响因素复杂，人口相关因素占主导位置，农业、环境与能源、经济、教育、卫生等相关因素处于次要位置，人均GDP并不总是婴儿死亡率的影响因素。政府高度重视儿童生存发展，政府主导落实并适时调整防控策略，是取得实效的关键所在。

Abstract: Objectives: To demonstrate whether there is a turning point in the relationship between infant mortality rate and per capita GDP in low-income countries, judge whether per capita GDP is the key influencing factor of infant mortality rate, and find out the influencing factors of infant mortality rate at different stages. Methods: The infant mortality rate and economic and social indicators of 20 low-income countries were taken as the research objects, and were divided into stages based on the minimum sample size of the correlation analysis between the infant mortality rate and the per cap-ita GDP. Through correlation analysis, it was judged that the relationship between infant mortality and per capita GDP changed from non-negative correlation to negative correlation. Then, with infant mortality as the dependent variable and various economic and social indicators as the independent variable, the stage ridge regression analysis was carried out respectively to further judge that the relationship between per capita GDP and infant mortality changed from no positive impact to posi-tive impact. The independent variable with statistically significant difference in the regression coef-ficient t test of the independent variable is considered as the influencing factor of infant mortality. The absolute value of the standardized regression coefficient is compared to determine the degree of influence on infant mortality. Results: The infant mortality rate and per capita GDP of the 20 low-income countries were at the low level in the world. The infant mortality rate generally showed a downward trend, but the average decline was small and the decline rate was slow. The per capita GDP generally showed a growth trend, but the average growth was small and the growth rate was slow. Among the 78 stages, there were 45 (57.69%) stages of correlation between infant mortality and per capita GDP, and 33 (42.31%) stages had no correlation. There was no correlation between the two in the first stage of the Niger in 1981, while the latter two stages showed a negative correla-tion; in Rwanda, there was no correlation between the two in the first two stages of 1980, and there was a negative correlation in the last three stages. The difference of F test of ridge regression model in 78 stages was statistically significant (P < 0.05), k value was small, and R2 was close to 1. There was no statistically significant difference in the regression coefficient t test of GDP per capita in 20 of 45 related stages and 28 of 33 unrelated stages (P > 0.05), in six stages of five countries, the regres-sion coefficient t-test difference of six independent variables, including total fertility rate, urban population growth rate, grain yield, agricultural land area, per capita final consumption expendi-ture of residents, and CO2 emissions, was not statistically significant (P > 0.05), in addition, the re-gression coefficient t-test difference of other independent variables was statistically significant (P < 0.05). In the two stages of Niger before 2000, the regression coefficient of GDP per capita in one stage had no statistical significance (P > 0.05), while the other had statistical significance (P < 0.05), but the standard regression coefficient was positive, and the latter stage had statistical significance (P < 0.05), and the standard regression coefficient was negative; in Rwanda, in the four stages be-fore 2000, there was no statistically significant difference in the regression coefficient t test of per capita GDP in the three stages (P > 0.05), and there was statistical significance in the one stage (P < 0.05), but the standard regression coefficient was positive, and the difference in the last two stages was statistically significant (P < 0.05), and the standard regression coefficient was negative. There were 24 indicators of economic and social factors with statistically significant differences in the re-gression coefficient t test of independent variables of ridge regression model, and the cumulative frequency was 412 cases. Among them, population related factors included population density, total fertility rate, population of more than one million urban agglomeration, population growth, and ur-ban population growth rate (53.16%); agricultural related factors included agricultural land, culti-vated land, agricultural land area, agricultural added value and grain yield (16.50%); environmen-tal and energy-related factors included CO2 emissions, nitric oxide emissions, methane emissions, power-on rate, power consumption, and energy use (13.11%); economic related factors included per capita GDP, per capita final consumption expenditure, per capita current health expenditure, and the proportion of health expenditure (10.68%); education related factors included the gradua-tion rate of primary school students, the enrollment ratio of girls and boys in primary and second-ary schools, and the enrollment rate of colleges and universities (5.10%); health related factors only affect adult women infected with HIV (1.46%). The top ten influencing factors were population den-sity, total fertility rate, population of more than one million urban agglomeration, per capita GDP, agricultural land, population growth, urban population growth rate, cultivated land, nitric oxide emissions, methane emissions (79.37%). Per capita GDP ranked fifth among the first influencing factors of infant mortality at each stage, fifth among the second influencing factors, seventeenth among the third influencing factors, second among the fourth influencing factors and fourth among the fifth influencing factors. Conclusions: The decline of infant mortality in low-income countries is not optimistic. The relationship between per capita GDP and infant mortality is unstable. Only Niger and Rwanda have changed from no positive impact to positive impact. The influencing factors of in-fant mortality are complex, and the population-related factors occupy the leading position, while the related factors such as agriculture, environment and energy, economy, education and health are in the secondary position. The per capita GDP is not always the influencing factor of infant mortality. The government must attach great importance to the survival and development of children, and the government should lead the implementation and timely adjustment of prevention and control strategies, which is the key to achieving results.