In Chapter One, I investigate the compensation channels through which firm-specific pay premia are constructed using a unique data set of compensation records at over 1600 firms. I find that high paying firms increase both base and incentive pay relative to their competitors. To understand why firms treat these two premia as complements, I embed an optimal contract model into a matching framework where workers of heterogeneous ability pair with firms of differing productivity. My model reveals that higher ability workers sorting to high productivity firms can explain why high paying firms simultaneously increase base and incentive pay relative to their competitors. In Chapter Two, which is based on joint work with Jeanna Kenney, I investigate gender differences in negotiation strategies. Although gender differences in negotiation are widely documented, the effect of agents (e.g., talent agents, real estate agents, lawyers, etc.) on these differences remains under-explored. Using a bargaining game, we find that, absent agents, men make more aggressive demands than women. Introducing agents who negotiate on behalf of their clients entirely closes this gender gap. Belief elicitations suggest that this is because agents do not anticipate that there are gender differences in a client’s propensity to reward their agent for adopting aggressive strategies. In Chapter Three, which is based on joint work with Judd Kessler, Corrine Low, and Xiaoyue Shan, I investigate how employers review job applicant resumes. Under plausible assumptions, signaling models predict that applicants who omit information from their resume (e.g., GPA) will be seen by employers as belonging to the lowest type for that category. Using data from a two-year field experiment, however, we find that employers do not always treat information omitted from a resume as proof that a candidate is a lowest type. As a result, low- or even medium-type applicants can benefit by omitting information such as GPA from their resumes. Our results suggest that existing signaling models can be enriched to better describe settings where senders have discretion in choosing which signals are observable.