A more difficult methodological challenge to address is selection bias. It may be that there are unmeasured personal characteristics that are associated both with greater levels of religiosity and positive behaviors such that estimated effects of religiosity would be biased when these potential selection factors are not accounted for. For example, if individuals who are motivated to engage in positive health behaviors participate in religious activities and exhibit better mental health outcomes, without measuring motivation to engage in health practices, the effect of religiosity on better mental health would be overestimated.
While it is difficult to separate the causal effects of religiosity and spirituality from other variables that are correlated with behavioral outcomes, researchers have developed novel approaches to reduce selection bias by using proxy measures for religiosity (instrumental variables approach), such as the geographic density of ethnicity (Gruber, 2005) and historical religiosity in counties (Heaton, 2006) or using statistical techniques such as propensity score matching (Lillard & Price, 2007). Despite advances in the estimation of causal effects of religiosity on positive outcomes, most researchers agree that drawing conclusions about the causal impact of religiosity is difficult because of data limitations, such as the lack of religiosity measures in national data sets (Lillard & Price, 2007).