1. University of Pittsburgh at Johnstown
Explaining Differences in Infant Mortality Rates across Wealthy Countries
Dr. Jeremiah Coldsmith
John J. Richard
University of Pittsburgh at Johnstown

2. Background
Globally poverty is the primary predictor of infant mortality rates (IMR). However, given the relatively low levels of poverty in developed countries, does poverty maintain its predictive power? No. In developed countries inequality – rather than poverty – predicts IMR.

3. Motivating Question
Seven Possible Mechanisms Linking Inequality and IMR:
A Direct Effect of Inequality:
Top earners may price lower income individuals out of healthcare markets.
Inequality may change buying behavior of low income individuals from basic “non-visible” goods to “visible” status goods.
A Mediating Effect of the Status of Women:
Countries with policies that are more supportive of women, particularly of family friendly workplace policies, are both more egalitarian AND are have more female friendly healthcare policies.
A Mediating Effect of Basic Population Health:
More egalitarian countries have better health outcomes which are likely to improve IMR.
A Mediating Effect of the Strength of the Healthcare System:
More egalitarian countries have stronger healthcare systems which are likely to improve IMR.
A Prior Effect of the Strength of the Welfare State:
Countries with stronger welfare states are more egalitarian AND have stronger healthcare systems.
A Prior Effect of Uneven Development:
Underdeveloped rural areas are likely to create national level inequality AND are likely to have poorer health and weaker healthcare systems.
A Prior Effect of Social Cohesion:
Countries with higher levels of social cohesion have greater support for welfare policies and social support which both improve health outcomes.

4. Problems in the Existing Literature
There are two major theoretical problems in the existing literature:
While they claim to be interested in studying the relationship between inequality and IMR in wealthy or developed countries, the sample sizes betray this claim.
Additionally, no study to this point has included measures of all six factors which could potentially explain away the relationship between inequality and IMR.
There are two additional methodological issues:
To this point, the vast majority of these studies rely on OLS regression, which is not entirely appropriate for a dependent variable that is a rate, like IMR.
Also, study authors rarely use the same indicators or measures of their concepts, making it hard to tell if inconsistent results are due to the variables included (or left out) of an analysis, the measures used to indicate the concepts, or differences in the sample of cases chosen for the analysis.

5. End Goal
Our end goal is to construct a structural equation model (SEM) which will allow the complex interrelationships between all eight theoretically derived variables to be fully taken into account. SEM also permits the use of multiple indicators, allowing us to include multiple measures of the concepts into a single study. SEM can be computed using a variety of estimation techniques, allowing for the structure of the dependent variable to be better taken into account. However, at this point, our full SEM refuses to run. Parts of it will run, but not the full model. SEM is notoriously finicky, so such issues are relatively common.

6. Intermediate Analysis
Table 1: Standardized OLS Regression of Infant Mortality Rates within Developed Countries
Variables
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6
Model 7
Model 8
Model 9
Poverty
.544***
-.272
-.276
-.177
-.227
-.284
-.211
-.274
-.257
GINI
.948***
.965***
.795***
.696**
.669**
.708**
.726**
Unemployment Rate
-.135
-.151
Life Expectancy
-.363**
-.603***
-.562***
-.624***
-.643***
-.577**
Cancer Mortality
-.353*
-.356*
-.338*
-.389*
-.302†
Percent of the Labor Force Female
-.154
-.136
Generalized Trust
.070
.039
Population Density
.082
-.003 n 34 R2
.30
.53
.54
.65
.71
.73
.72
.75
Note: *** p ≤ .001, ** p ≤ .01, * p ≤ .05, † p ≤ .1