This is an interesting paper but I have concerns about the approach.

First, the definition of belief in Heaven and belief in Hell. These are often co-incident, I'd expect far more often co-incident than occurring in isolation. So that by looking at the metric in Table 1, the proportion of belief in Heaven minus the proportion of belief in Hell is really the proportion who believe in Heaven ALONE minus the proportion who believe in Hell ALONE. This would explain why belief in Heaven so often outweighs belief in Hell.

In other words, as we proceed to the higher horizontal scale points we're not looking at the effect of those who believe in Heaven so much as at the effect of those who DO NOT believe in Hell. This offers a divergent interpretation from that posited in the paper - namely that lack of belief in Hell is correlated with higher crime rates, not that belief in Heaven itself does so.

More seriously, the data suffer from time differences that call into question their linkage. Your survey of religious beliefs includes data from as far back as 1981, but you take dominant religion from 2011, 30 years later. Meanwhile, your crime data is an average from between 2003 and 2008, unless there are fewer years, so the data are collected unevenly by country.

You also have a substantial endogeneity problem by including the imprisonment rate as a predictor of the crime rate.

Because data are collected so unevenly by country (waves of the belief study, as well as crime data), it seems wise to control by country rather than aggregate all countries together.

So this is an interesting question, but a proper exploration would almost necessarily be at the individual level.