Read on Twitter
Much like credit scoring, the history of job evaluation systems seems super relevant for contemporary discussions of algorithmic bias. (From a 1980 EEOC hearing on job segregation and wage discrimination.)
What's interesting about these cases is that they were long enough ago that you can see the whole system change as different methods are proposed, challenged, replaced, etc. My sense is that much contemporary algorithmic fairness research zooms too narrowly in on the algo itself.
In part that's because the algorithms being examined haven't been used for very long, and sometimes are in quickly shifting systems. So it's hard to zoom out.
Link to the hearing: https://babel.hathitrust.org/cgi/pt?id=mdp.39015005297109&view=1up&seq=658&q1=gap
Related: Does anyone have a good summary of trends in the use of job evaluation systems in the US?
Interestingly, at least according to this source, their use in the UK declined in the 1980s-1990s but rose again in the 2000s in part as a way to help shield companies from claims of gender discrimination: https://www.employment-studies.co.uk/system/files/resources/files/mp58.pdf
... which is kind of fascinating given that the Comparable Worth movement hinged a lot of its arguments on those exact same systems, which at least in the 1970s US showed that employers were underpaying women's jobs according to the employers' own evaluation system.
Some of the proponents of comparable worth recognized the limits of the job evaluation approach - they argued it was a conservative lower bound on the devaluation of women's work, precisely because the systems baked in various sexist biases in terms of what factors to evaluate.
But in the US, it seems, most employers now rely on "market-based" systems of pay setting rather than job evaluation methods.
In some ways, these market-based pay setting systems are an even clearer example than job evaluations of how algorithmic decision-making involves replacing your biases with someone else's biases. https://scatter.wordpress.com/2017/06/13/algorithmic-decisionmaking-replaces-your-biases-with-someone-elses-biases/
There’s not one clear takeaway from the story. The 1970s systems demonstrated how employers undervalued women’s jobs, but were still biased, and the result of challenging them was not raising wages as much as forcing employers to find new tools to justify the existing system.
There were exceptions - wins in state governments for example that led to a lot of women getting raises - but the movement floundered on legal technicalities, the shifting politics winds of the Reagan era, and the ideological strength of the “free market” as arbiter of worth.
And the threat of this whole class of claim - discrimination, civil rights, all that - is part of the incentive for employers to try to abandon the employment relationship entirely. It became a political/legal risk.
