Advanced Course in Multiple Regression and Causal Inference Using Stata

Multiple regression and causal inference provide researchers with the needed assurance that observed associations are valid. After using a bivariate test (such as chi-squared test, t-test, ANOVA, etc.) to demonstrate association between two variables of interest, the next critical question is whether those associations are valid, i.e., whether they can in fact be explained by other variables (confounding) or study methodology (bias). In other words, to strengthen causal inference, it is vital to eliminate confounding and bias.

Multiple regression models remain the most widely known approach for controlling for confounding, identifying risk factors and estimating independent effects. They are also commonly used to build predictive models, and the principles underlying them have influenced many modern artificial intelligence methods. Therefore, all quantitative researchers should be able to understand, build and interpret multiple regression models. Yet, many researchers who are comfortable with bivariate tests have never built or fully interpreted a regression model, leaving a critical gap between the analyses they conduct and the conclusions they need to draw. This course closes that gap.

This course is designed for quantitative researchers who are already proficient in descriptive statistics. Being comfortable conducting hypothesis tests, such as chi-squared tests, t-tests, and ANOVA, will be a plus. Using Stata as the analytical platform, participants will move from the bivariate tests to the regression models that extend and formalise them. By the end of the course, participants will be equipped to build, interpret and critically evaluate linear and logistic regression models, and to apply them with confidence in their own research. They will also be able to extend their knowledge and skills into other models in the generalised linear model family, including Poisson and binomial models.