The Spiritual Anthropologist

Where Alcohol and Anthropology Meet

Religion and Infant Mortality in the United States: A Flawed Study

By 0

Vaccine Banner

In a recent article, posted on Atheist Republic, there was a suggestion that fundamentalist religion was positively associated with PNIMRs. However, the analysis seems to be an interpretation, of an interpretation, of maybe even one more level of interpretation before getting to the actual study. Furthemore, the study itself faces significant methodological errors. 

The study, “Religion and Infant Mortality in the United States: A Community-Level Investigation of Denominational Variations in Postneonatal Deaths,” conducted by Garcia, Bartkowski, and Xu, and published in May of 2018 attempts to test the following hypotheses.

H1: Counties with a greater proportion of Catholic adherents will exhibit reduced PNIMRs.
H2: Counties with a greater proportion of conservative Protestants will exhibit higher PNIMRs.
H3: The effects of religious ecology on PNIMRs will be more pronounced in 2010 than in 1990 because of medical advancements and public awareness campaigns implemented after 1990.

There are a number of issues with the study. First off, an ordinary least squares regression was used, without any real justification. The model was assumed to be linear, without any justification. And the variables were assumed to be independent, even when there is evidence rejecting that assumption. Admittedly, it is an easy to use model, so it’s not too surprising, but I don’t see any initial analysis of the data which justifies a specific model, at all. The adjusted R-squared also suggests a very poor fit for the model. That being said, sometimes even with a poor fit, we can glean some information out of such a model. But in most cases, the regressed variables were not statistically significant.
The study therefore does not reasonably justify the claim that religious affiliation is related to PNIMRs. There are a few exceptions, but even in those cases, the impact on PNIMRs seems to be very small: orders of magnitude less than some of the control variables, and they may be due to an inability to properly capture how poverty relates to religious affiliation: poverty and religious affiliation are not independent variables as religious affiliation tends to be used as a way to gain access to a social network, counseling resources, etc and it seems to provide some level of stress reduction (APA).
Another interesting point is that the theory being tested suggests that technological advancements would reduce PNIMRs and make the effects of religious opposition to medical treatment more pronounced. However, even limiting the analysis to conservative protestants, the regression model for 2000 gives a coefficient of 0.003 for the relationship between adherence to conservative Protestantism and PNIMRs and a coefficient of 0.001 for 2010. So there is an apparent drop in the relationship between conservative Protestant affiliation and PNIMRs. H3 does specifically look at the change from 1990 to 2000, but there should be a similar change between 2000 and 2010, as medical technology continues to evolve.
The analysis for H3 is problematic. The issue is that no test for significance was used. Generally, when a researcher wants to test to see if there is a change over time, the researcher would employ some statistical test for difference. Instead, a simple observation between the two statistics was made. Moreover, the comparison was made between a value which did not reach statistical significance, and a variable which did reach statistical significance.
I understand the desire to show results consistent with the initial theory being tested. There is a great deal of positive publication bias. A peer reviewed journal is much less likely to publish a paper if there isn’t an apparent result which is “interesting.” However, the following errors are glaring:
  • Failure to justify the model used, including justification for linearity and independence.
  • Weak R-squared
  • Dwarfing of the relationship between religiosity and PNIMRs, by control variables
  • Apparent reduction in the effect that conservative Protestant affiliation has on PNIMRs, in the years between 2000 and 2010 , along with a lack of use of actual statistical testing for differences, and relying on data which was not statistically significant, there is no reasonable justification for the conclusion that was made.

For these reasons, the study should be thrown out as being incredibly flawed. The conclusion of the paper is not justified based on the actual findings of the paper.