This is actually a dangerous source to draw conclusions from. Even ignoring the qualitative bias of H1-B workers having less leverage than an average worker with easy mobility which likely causes relative wage depression in that data set, the key issue is that a significant chunk of a high salary software engineer is in equity and it rarely shows up in the labor certificate application. So H1-B data is mostly base salary and can be misleading to use when drawing a conclusion on total comp. Extra care need to be taken.
Both good points. All of the data mentioned has its role. The key with H1B data is that the biases are better known. With self reported datasets, who knows what the poster's motivation is? Especially with companies that don't have many data points!
The risk to an employee is asking for too little and H1B data can help put of floor on that number.