Data sources
Our methodological approach is similar to that of Tait et al. (2019) [
11]. We merged three sources of data on country-level characteristics: The United Nations Treaty Series Database, which provides information on treaty ratification, the World Bank World Development Indicators, which provides information on child health and economic conditions and, the Polity IV database, which provides information on political democratization. Supplementary Table
1 provides weblinks to these datasources. Of the 193 countries for which ratification data were available, 192 (99%) had ratified the CRC. As the sole non-ratifying country, the United States was excluded from the sample. The final analytic sample consisted of 192 countries contributing information over the period 1990–2015.
Measures
Many of our measures have also been previously described in Tait (2019) [
11]. Our primary outcome measure was child mortality, which was measured as deaths of children under the age of five per 1000 live births. We wanted to account for a broad range of social, political, and economic factors, which may confound or modify the association between CRC ratification and child mortality rates in a country. However, we were constrained in our ability to use many of the standard approaches for doing so. For example, because almost all countries have ratified the CRC, our sample lacked a control or comparison group of non-ratifying countries that would enable multivariable regression. Because our sample size was relatively small, it precluded conducting an extensive stratified analyses, as we would quickly run out of a sufficient sample size of countries within each stratum of each confounding or moderating variable of interest.
We thus reduced the set of covariates of interest to two variables, which perhaps best comprehensively characterize the conditions of a country. The first was a country’s overall economic conditions at year of CRC ratification, which we measured using per capita gross national income (GNI). GNI represents income acquired through all domestic and foreign sources, which is likely a more accurate reflection of the economic conditions of a country than is gross domestic product (GDP), given today’s more global marketplace. GNI was adjusted for purchasing power parity using the World Bank Atlas Method, which adjusts GNI to reduce changes in the exchange rate that are attributable to inflation. We then divided countries by the World Bank’s income-country groupings: low, lower-middle, upper-middle, and high and, due to sample size constraints, we collapsed the two middle groups into one category. We also examined the split of countries amongst the World health Organizations 6 regions: Africa, Americas, Eastern Mediterranean, Europe, South-East Asia, and Western Pacific.
The second variable represented overall political conditions at the year of CRC ratification for each country, which we measured by the extent to which a country’s political systems were democratized. We measured this using the Polity IV composite index, which ranges from − 10 to 10, with higher values representing greater levels of democratization. Due to sample size constraints, we dichotomized countries between those whose index score was less than 6 (which we termed non-democratic countries) and those whose score was 6 or greater (which we termed democratic countries).
Statistical analyses
Our analytic strategy is based on that of Tait et al. (2019, 11]. In order to account for differences in the year of ratification, we first standardized the time scale of ratification by ‘zeroing’ at year of ratification for each country. For example, as described in Tait et al. (2019), if a country ratified the CRC in 1980, 1980 was converted to year-zero, and 1985, 5 years post ratification, was converted to year 5 [
11].
First, we conducted a series of descriptive statistics to provide a sense of the sample distribution across the variables in our study. Next, we plotted the unadjusted trends in child mortality rates. We then used t-tests to get a sense of the difference between child mortality rates at year of CRC ratification and 5-years and 10-years post ratification. We also used joinpoint regression to determine where there were ‘inflection points’ in child mortality trends; years at which trends in child mortality had changed in ways that are statistically differentiable. Inflection points are fit when the joinpoint regression model determines there is a significant change in the linear trend in the variable (child mortality rate) over time.
These techniques still fall short of testing our central research question, because while they interrogate the trajectory of child mortality rates, they do not tell us enough about how this trajectory may have been affected by CRC ratification. Many gold-standard quasi-experimental methods (e.g. difference-in-differences models) provide strong tests of how a discreet shift in societal conditions, such as a policy change, or indeed ratification of a human rights treaty is associated with an outcome, such as child mortality rates [
12]. However, these techniques are reliant on the presence of a ‘control’ group, a group that did not experience the policy or other societal shift, to which we can compare the ‘treatment’ group that was exposed to the shift. The widespread ratification of the CRC (with the exception of the United States) means that there is no control group available to conduct such a model.
Instead, we drew on an alternative quasi-experimental method, interrupted times series analysis (ITSA), which is considered the best alternative in these circumstances [
13]. ITSA examines whether the trend in an outcome (child mortality rates) changes after a policy shift (CRC ratification) occurs. ITSA does so by using the pre-shift trend to mathematically construct a projected counterfactual trend: the trend in child mortality, had CRC ratification not occurred. Put differently, it attempts to construct a control group from the treatment group itself. The difference between the actual, observed child mortality trend and the mathematically derived counterfactual trend is the estimated association between CRC ratification and child mortality rates. We analyzed the association with shorter-term trends (5 years post ratification) and longer-term trends (20 years post ratification).
ITSA relies on having multiple observations available in the pre- and post- ratification periods, in order to have a stable indication of trends. However, longer periods of data may then contaminate the analyses by introducing other policy shifts that have influenced trend changes. With this balance in mind, we structured our data to include 5 years of pre-ratification data and up to 20 years of post-ratification data (recent ratifiers contributed less post-ratification data).
Another assumption of ITSA is that the countries in the sample are sufficiently similar, so that we can be fairly confident that any trend changes can be attributable to changes in CRC ratification. In order to account for these confounding and effect-modifying issues, we stratified our ITSA analyses by GNI and democratization. Thus, the country-subgroups in which we conducted separate ITSA analyses were: low-income non-democratic, low-income democratic, middle-income non-democratic, middle-income non-democratic, high-income non-democratic and high-income democratic.
We were also concerned that static measurement of confounders may be problematic and induce misclassification. As Supplementary Table
2 suggests, some, but not many countries moved across democratization categories during the years of data included in the study. Slightly more moved across income groups. An alternative categorization to account for this then might be economic growth rate, rather than economic level itself, since it better characterizes the economic trajectory. However, as Supplemental Table
3 suggests, the set of countries within GNI growth rate categories then becomes even more dissimilar in other respects than when they are categorized by GNI. In other words, the available options are all imperfect, but we believe year of ratification as the basis for categorizing countries by GNI and democratization level is the least problematic of these.
Because there is little prior literature in this area, we are unsure about lag effects. For this reason, and reasons of analytic constraints, we did not test lag effects. All analyses were conducted using Stata/SE version 14 (College Station, TX).