A body of literature has demonstrated that users' mental health conditions,such as depression and anxiety, can be predicted from their social medialanguage. There is still a gap in the scientific understanding of howpsychological stress is expressed on social media. Stress is one of the primaryunderlying causes and correlates of chronic physical illnesses and mentalhealth conditions. In this paper, we explore the language of psychologicalstress with a dataset of 601 social media users, who answered the PerceivedStress Scale questionnaire and also consented to share their Facebook andTwitter data. Firstly, we find that stressed users post about exhaustion,losing control, increased self-focus and physical pain as compared to postsabout breakfast, family-time, and travel by users who are not stressed.Secondly, we find that Facebook language is more predictive of stress thanTwitter language. Thirdly, we demonstrate how the language based models thusdeveloped can be adapted and be scaled to measure county-level trends. Sincecounty-level language is easily available on Twitter using the Streaming API,we explore multiple domain adaptation algorithms to adapt user-level Facebookmodels to Twitter language. We find that domain-adapted and scaled socialmedia-based measurements of stress outperform sociodemographic variables (age,gender, race, education, and income), against ground-truth survey-based stressmeasurements, both at the user- and the county-level in the U.S. Twitterlanguage that scores higher in stress is also predictive of poorer health, lessaccess to facilities and lower socioeconomic status in counties. We concludewith a discussion of the implications of using social media as a new tool formonitoring stress levels of both individuals and counties.