A lot of physics experiments are very indirect, and you're not actually measuring the 'thing' you're looking for. As an example, in particle collision experiments, you're often looking at how the resulting collisions behave, and then applying our current understanding and models to reason about why the resulting collisions behave the way they do. Basically, it's like saying, 'This particle X had Y behavior, which is explained by our current model that says there must be a Z interacting with X to produce Y.' This is, of course, an extreme simplification, but it illustrates why we can claim with a high level of confidence that certain 'unmeasurable' things must exist.