I know there are many questions that ask for book recommendations but I'm wondering is there any text that teaches entirely through the analysis of real data sets. Doesn't matter if it's time series, panel, cross sectional etc. In my experience an awful lot of what you learn theoretically is based on well behaved idealised data, and once you sit down to a real data set you're suddenly on your own trawling through sites like this looking for answers. Something like the two Josh Angrist books on econometrics but written more in the style of a textbook.
Tintle, et al., "Introduction to Statistical Investigations* (Wiley 2016) does exactly this. The book's strategy comprises six elements (as laid out in the preface):
Spiral approach to statistical process. This is a six-step method beginning with "ask a research question," progressing to "look back and ahead," and then iterating.
Randomization-based introduction to statistical inference. Students "first learn how to make inferences using simulations of chance models."
Focus on logic and scope of inference.
Integration of exposition, examples, and explorations.
Easy-to-use technology integrated throughout (using Web applets).
Real data from genuine studies. "We utilize real data from genuine research studies throughout the book ... from a variety of fields of application and popular culture."
In a workshop earlier this summer, co-author Beth Chance commented that "all the examples are real. We worked hard to find them!"