Academic journal article International Journal of Child Health and Human Development

A Study of Time Trend in Global Infant Mortality Rates: Regression with Autocorrelated Data

Academic journal article International Journal of Child Health and Human Development

A Study of Time Trend in Global Infant Mortality Rates: Regression with Autocorrelated Data

Article excerpt

(ProQuest: ... denotes formulae omitted.)


Infant mortality rate (IMR) is defined as the number of deaths within one year of age per 1,000 live births (1). The IMR is considered to be one of the important indicators of a population's health and well-being along with life expectancy (LE) and under-five mortality rate (U5MR) (2, 3). In the year of 2010, the IMR of the United States was 6.6 per 1000, compared with 66.9 per 1000 in Sudan (4). This tenfold difference in IMR reveals a large gap in health conditions between these two nations. As IMRs from developed countries are generally lower than those in developing countries, the IMR is also a reflection of general socioeconomic conditions across different countries.

With improvement in socioeconomic conditions, improvement in health care for deliveries and newborns, and with recent efforts made by the World Health Organization (WHO), the global IMR has been decreasing in most countries of the world (5). According to the Millennium Development Goal 4 (MDG 4), the global mortality rates for children who are younger than five years old are supposed to be reduced by two-thirds from 1990 to 2015 (6, 7).

Although the IMR has declined to a relatively low level in some countries in recent decades, there are still many infants, especially those from the developing countries, who are suffering from many risk factors that lead to high IMR. Our study of time trend of IMR over the past a few decades will provide the health organizations and the general public a clear insight into this global health issue. In our article, we have investigated recent achievements regarding reductions in the IMR, globally with a special attention on specific nations or regions. We modeled the current trend of IMRs and measured gaps from the goal as set in the MDG4 (8), etc. The results from this study will help us to better understand the trend of IMR to formulate strategies for improvement in the IMRs and to decide where to focus our attention and efforts and how to allocate the health resources around the world more effectively (9).


In this study we used countries as the study units (2). For each of the countries, annual IMR data were collected and the earliest time point in records starts from 1932. In our regression analysis model, the response variable is the IMR and the independent variable is the time (year).

Our data set includes annual infant mortality rate for 195 countries/regions around the world between the times of May 1932 to May 2010. The data set is mainly collected from the Child Mortality Estimates (CME) website: (4), compiled with those from the World Bank website (7). A closer examination of the data set showed that for most of the countries/regions, the data were mostly missing before 1950. To deal with this issue, we set the time interval starting from 1950 for our study. Countries with sparse data were not included in the study. Also we took consideration to cover countries from different continents and at different level of socioeconomic status in order to investigate the influence of those factors on the trend of the IMR. To make it representative and comprehensive, we selected the countries and regions that are listed in Table 1.

Statistical method

The linear regression model is formulated as ynxi = Xnxpßpxi + snxi, where n indicates the number of observations, p indicates the number of covariate. We assume the error term £ ~ Nn(0, a2In), which leads to the ordinary-least-squares (OLS) estimator of ß (10):

... (1)

with variance-covariance matrix:

... (2)

However, in our study this ideal assumption of error terms cannot always be satisfied-there may be some correlations among the error terms, which can be modeled with a more general assumption of e ~ Nn(0, X), where the error-covariance matrix X is symmetric and positive-definite. Different diagonal entries in X represent non-constant error variances and nonzero off-diagonal entries correspond to correlated errors. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.