I spent years as an executive compensation consultant both creating and analyzing salary surveys for the world's largest companies.

What it teaches you is that: 1. Never use averages, always use medians 2. Sample sizes are critical, and anything below 5 samples you just can't trust (ideally 30, as the Central Limit Theorem suggests) 3. Garbage in, garbage out: if you don't know the population of the sample, then you can't trust the output (i.e., how many employees, what's the revenue run rate, what's the user growth look like, what's the valuation and how many funding rounds, etc).

I love the spirit of this data, but based on the core principles outlined above, you can't hang your hat on this data.

In software engineering land, not knowing these three principles is like not knowing the principles of MVC.