Since 1974 the EEOC has collected labor force data from state and local governments with 100 or more employees within the 50 U.S. states and the District of Columbia. The reporting agencies provide information on their employment totals, employees' job category, and salary by sex and race/ethnic groups. Since 1993 the EEO-4 survey is conducted biennially in odd-numbered years. You can see a list of the annual tables from the EEO-4 dataset.
Larger government entities also provide data at the department or function level. These data have been used periodically to describe equal opportunity progress (e.g. Henderson 1978) and pay gaps (e.g. Reese and Warner 2011) in government employment. Fidan Kurtulus (project co-Principal Investigator), Margaret Reid, Will Miller (project advisory committee member) and Brinck Kerr have the most sustained contemporary research program using these data. They have explored sex and race based gender gaps in response to law changes (Kurtulus, 2016), sex and race based glass ceilings (Reid, Miller and Kerr 2004; Miller, Kerr and Reid 1999), job level sex segregation (Kerr, Miller and Reid 2002), and labor market ethnic competition (Kerr, Miller, and Reid 2000). Researchers have less often added contextual data to EEO-4 data, with Census data (e.g. Riccucci and Saidel 1997), state-level laws on integration (Kurtulus 2016), and court ordered integration (McCrary 2007, Amalia Miller 2012) being the only examples we have found to date.
As a panel of departments within governments, there are 7 million observations with more than 250,000 new observations a year. As a panel of government units it is smaller (370,170 observations, 14,000 new observations a year), but still a substantial data resource. The panel nature of these data have been barely explored (only Kurtulus 2016, Amalia Miller 2012, and McCrary 2007). In addition, these data could easily be combined with political data, EEO-1 private sector reports, EEO-5 school reports and Census based data in potentially creative and scientifically generative ways.