Rothman is right on the mark (source / read for free):
Today it is typical to train epidemiologists to use regression models as their ﬁrst approach to data analysis. Such training fosters the idea that regression modeling is the primary, and perhaps the only approach to use in analyzing epidemiologic data. As a result, the rift between epidemiologists and their data, and more so between readers and the data, is growing. We should reverse this trend, one paper at a time.
Data science has the same problem: jumping to fancy algorithms obscures basic exploratory data analysis, and this leads to bogus results.