Disentangling statistical and taste-based discrimination using observational data poses significant challenges. We develop a framework to define and estimate both types of discrimination in terms of potential outcomes by combining two bodies of literature, on discrimination and causal inference. Statistical discrimination can be seen a rational response to limited information when the decision-maker cannot observe some relevant attribute -- e.g. an officer's suspicions about crime -- and instead uses race as a proxy. In contrast, taste discrimination arises purely from prejudice.
We situate discrimination within a causal mediation framework: statistical discrimination is that indirect effect of a certain group-belonging which operates via a specific type of mediator: the decision-maker's beliefs about an unobserved attribute. Taste discrimination is the Natural Direct Effect of a group-belonging (Pearl 2001), holding all mediators fixed. We discuss the identification assumptions required to estimate both types of discrimination within a generalized, multiple-mediation framework, and suggest an estimation approach based on g-computation using Monte-Carlo simulation.
Using data on stop-and-frisk actions by the New York Police Department, we show how to measure discrimination in practice. Police officers decide to search or frisk based on the crime that they suspect a civilian has committed or is about to commit, and statistical discrimination can arise if these beliefs vary by race. We find that blacks and hispanics face taste as well as statistical discrimination relative to whites. Finally, we extend our framework to consider the accuracy of officers’ suspicions, and demonstrate how to measure discrimination when beliefs reflect accurate priors based on previous experiences of weapon discovery.
The usual assumption is that sport is good for your health but is it also good for longevity – will you live longer? However, the evidence base linking to sporting prowess on the playing field, in the ring or on the race track to longevity in the wider population is missing. This research investigates the longevity of sporting legends who have reached the pinnacle of their profession in seven different popular sports. The time frame covers 180 years from 1841 to the present – it includes the inception and record-keeping of the sports concerned and their subsequent development. If it can be shown that sport increases longevity, this strengthens the case for participation throughout one’s life and for promoting the health and longevity benefits of physical exercise in any chosen field. For example, it can be a route out of poverty if played to a professional level and have a positive influence on the development of younger generations. A key finding is that most sport is generally associated with greater longevity but some, such as tennis and golf, stand out more than others. The ability to compete in tennis or golf well into old age but at a less competitive level is an important factor. The amount of uplift varies by sport with the risk of injury also being important in for example boxing. This risk has changed over time and in some cases for the better e.g. in the case of horse racing. Differences also point to wider social determinants such as educational attainment, socio-economic factors but also leadership which are important in sports like cricket, rugby and football. The approach uses novel cohort based statistical techniques to adjust background rises in longevity over time so that comparisons are conducted on a fair basis.
Background: Experimental studies remain the gold standard in making causal inference. However, using experimental studies to estimate the effect of education on attitude towards domestic violence (ATDV) was not feasible due to ethical issues. Propensity Score Methodology (PSM) can be used to overcome this challenge. Therefore, PSM was used to investigate the effect of education on ATDV among men and women in Nigeria.
Methods: A total of 14,495 and 33,419 records were extracted for men and women respectively from the 2016-2017 Multiple Indicator Cluster Survey (MICS) in Nigeria. The outcome variable was ATDV. The treatment variable was education while the covariates were age, residence, geopolitical zones, marital status, ethnicity, parity, wealth index, alcohol use and media exposure (use of television or radio). For the PSM analyses, selection bias was checked among the levels of education using the multinomial logit regression. Propensity scores (PS) and PS weights were generated for the treatment variable and average treatment effects of ATDV were estimated using logistic regression that combined regression adjustment and inverse-probability weighting. Descriptive statistics, odds ratios and 95%CI were presented.
Results: Mean age of men and women were 30.8±10.2 and 29±9.4 years respectively. About 16% men had tertiary education while lower proportion (14%) of women had tertiary education. Twenty-two percent men and 34.5% women justified domestic violence (DV). Selection bias was effectively corrected (SD diff ≈ 0, Variance ratio ≈ 1). Men (AOR = 0.84, 95% CI: 0.78, 0.92) and women (AOR=0.94, 95%CI: 0.80, 2.22) who attained tertiary education were less likely to justify DV in comparison to their uneducated counterparts.
Conclusion: Education played a crucial role in ATDV among men and women in Nigeria. Embracing the use of PSM will enable researchers make causal inference from non-experimental/ cross-sectional studies in situations where randomized control trials are not feasible.