Posts

In this post we discuss what is perhaps the most elegant technique for doing causal inference: regression discontinuity (RD). Although it is known today as a toolkit in the econometric toolbox, it was originally developed by an education researcher in 1960.

So far, we first saw that fixed effects was able to get rid of confounders that did not change over time. We then noted that difference in differences designs could do better, as they got rid of confounders that changed over time parallel to the whole control group.

In the previous post we explored the fixed effects approach to causal inference. Here we discuss the difference in differences approach, which is less widely applicable, but can make a stronger claim as to uncovering a cause.

Over the past few decades, economists have been the main drivers of causal analysis in social science. The most influential text has probably been Angrist and Pischke’s Mostly Harmless Econometrics.

In the previous blogpost, we used a basic decision tree and logistic regression to predict who would graduate high school among a bunch of kindergarten kids. In this post, we bring the random forest and XgBoost algorithms to bear on the data set, to see if they can improve the predictions in the test data set.

In ever more domains of life experts have to compete with algorithms when they make predictions. Almost always, the experts mistrust algorithmic predictions. However, the work of Meehl, Kahneman and others suggests that it is hard to find evidence of expert judgments trumping even simple algorithms.

We continue to look at the results of the observational study at a secondary school in Coevorden, where we this time focus on the relation between chronotype and grades.

We continue our journey of fitting count models to explain absenteeism based on the chronotype of students. We discussed binomial, quasi-binomial, poisson and quasi-poisson models and now move on to negative binomial, hurdle and zero inflated models.

A widely shared belief is that some people naturally rise early while others prosper during the night. To be precise, the belief is that this behavior is not induced by the environment, but is the expression of an innate biological clock.

In the previous post, we investigated the claim by Doug Rohrer and colleagues that mixing up math questions has an effect size of about 0.8 on a math test. In the analysis, we assumed that the points scored on the test could simply be added up.