I am a PhD student in engineering right now. I've found that most of what I would have liked to measure was already available in the literature, but the papers of interest took some digging to find. I will be doing some experiments, but they will be very targeted at filling in gaps. Most researchers seem to think they are targeting gaps, but (in my field at least) literature reviews are typically not detailed enough to truly know where the cutting edge is. Researchers often end up reinventing the wheel.
I also have extended existing data by taking advantage of the properties of a special case to get some information most would consider very difficult to obtain. (Basically, I use a proxy which has a known strong correlation to the quantity of interest in the special case. It doesn't work outside of the special case.) Most avoid getting this information because either the researcher chooses a bad proxy, or knows how to get the information more directly but dislikes the method they know (because it would be a lot of work).
I supplement existing data with sanity checks on existing theories too. In fact, I will be presenting a paper at a conference in a few weeks where the entire motivation of the paper was to criticize an earlier theory which fails many sanity checks. (Wrong in the limits, wrong trend, lacks a necessary functional dependency, etc.)
The remainder of my time is spent developing mathematical models. Math is cheap and I think most researchers in the physical sciences would benefit from more of it.