For a software policy to calculate gender inequality

Vincent-Arnaud Chappe

The government has just decided to generalize the use of software to calculate the wage gap between women and men in companies, in order to force them to reduce wage inequality by 2022. This software is part of a history of statistical equipment of companies in favor of professional equality.

Initially, in 1983, with the great Roudy law on professional equality, companies with more than 300 employees are required to produce an annual “Comparative Status Report on the Situation between Women and Men” (RSC). It is expected to objectify the gap between their respective situations (especially in terms of salaries but also in terms of hiring, conditions of employment, training, etc.) with the idea that revealing inequalities and the awareness that will follow will be sufficient to move the management and the social partners. But the tool remained largely undefined by the law, was based on a model of trust in the goodwill of the actors, and is actually very seldom used. Several reports focus on its improvement and identifying ways of use and action, without this leading to visible results.

A change of rationale

A turning point took place in 2001 with the passing of the “Génisson law”, as the law established an obligation for companies with more than 50 employees to negotiate on professional equality with a view to signing an agreement. In this perspective, the RSC was “renovated”, taken over by the State, which set a large number of compulsory indicators. Even if the obligation is strengthened, it remains fundamentally in a model of trust where companies are constrained to an obligation of means rather than result. In fact, RSC remains little used.

In 2006, the situation changed once again: a law on equal pay – the Ameline law – obliges companies to reduce wage gaps between women and men by 2010. The deadline was eventually removed, but it introduced a change of rationale: the transition from a single obligation of means – to negotiate with the social partners – to an obligation of result – to eliminate actual differences. The RSC was no longer used to establish an upstream observation, which would have open the way to awareness, but became a tool for monitoring policies to reduce inequalities.

With this in mind, a committee chaired by Anne de Ravaran was set up to make the RSC more effective and more suitable to companies, and lead to the creation of a standardized Excel spreadsheet for the presentation of data. The RSC was gradually invested by various actors, especially from 2013 onwards, when sanctions against companies failing to negotiate were instituted. Unions created training courses and started to use new indicators. Consulting firms investing in the market and helping companies produce and analyze their quantified diagnosis.

The National Agency for the Improvement of Working Conditions (ANACT) launched an automated Excel program to automatically produce the RSC from a staff database. However, a recent study of professional equality agreements shows that the RSC is far from being systematically used or is ambiguously used (it is for instance more often used to legitimize the pay gap than to establish a policy aiming at reducing this gap).

Opacity of computer code

The software promoted by the government is part of this logic of elaboration and automation of the calculation of inequalities, linked to a logic of obligation of result. But there is also another trend: the development in large companies of sophisticated statistical analysis methods, known as “multilinear regression”, used to calculate the wage gap between women and men “all else being equal”, that is, by neutralizing differences in age, degree, position, etc., the list depending on both the choice of users and the data available upstream.

These methods reduce the observed gap and are supposed to extract a “pure” differential. However, they are contested by economists and sociologists, as well as by trade unions such as the CGT, who believe that gender inequalities are measured “all else being unequal” because they are systemic and evolving (the “position” of employees, for example, not possibly being isolated from their gender).

The use of such software, if it can trigger a dynamic of action important for equal pay, raises several questions: how do we parameter the tool? Which variables are taken into account? What criteria of statistical reliability are we keeping? The importance of the gap varies depending on how the variables of a salary equation are neutralized, one does not arrive at the same difference: 26% gross wage gap? 16% gap in hourly wages? 13% in equivalent occupations? 9% pure discrimination? All these figures are “true” – and there are other possible ones –, according to the reference system and the political reading grid. The issue of software configuration – and its uses – is therefore fundamentally political.

Using software means taking the risk of creating a black box, where the social implications of the algorithm are hidden behind the technicality and opacity of the computer code. So that the social partners can judge, and possibly criticize, they must be able to have access to the equation used, as well as to the data being processed, or we decide to abandon the negotiated equality model on which the occupational equality and, in France the RSC, are based (which is a strong political choice).  

A « hackaton »

It should also be noted that such a “multilinear regression” model also requires such considerable technical skills, that it is not immediately definable and understandable, and even less questionable, without possessing the analytical skills relevant to the social sciences. To generalize such a model, that everyone does not master, would require at least a rise in technical competence of the social partners (directors or trade unions) on the subject. Without these skills, the black box of the software produces meaningless results for the actors who use it, and it is hard to see how these actors can then fight against gender inequalities that that take root in the organizational functioning of the company.

Indeed, by focusing attention on wage indicators, such software encourages only catch-up policies in the form salary increase for women, without taking into account the systemic and organizational dimensions of inequalities, and the need to address them at this level.

Catching up with women is necessary and compulsory, but if we do not fight against the mechanisms of production of inequalities (gender segregation of jobs and undervaluation of predominantly female jobs related to maternity, glass ceilings and sticky floors, the standard of full temporal and mental availability, etc.), the gaps will not fail to occur. The attention paid to the indicators must therefore be only a means and cannot replace a genuine policy of professional equality.

These remarks are not so much criticisms but invitations to take into account the political dimensions of this calculation software. There is not “one” objectivity, but different sociotechnical constructs of objectivity: the way in which this software is parameterized, the actors who have control over it and the way it is used and included in other modes of action will shape the equality policies between women and men The government is talking about organizing a hackaton to produce this software: the idea is stimulating, but professional equality is too serious a problem to entrust it to computer specialists alone: it must also involve the employees, especially women, and their union representatives.

The IT equipment of occupational equality is probably necessary, desirable and inescapable, but it should not lead to a purely algorithmic government of occupational equality that would overlook its political dimensions. Or we risk to see the problem by the small end of the telescope and to promote a very narrow vision of equality.

Published in Le Monde Economie on March 9, 2018


Photo#1 – Manif 8 mars 2017 Paris. Photo credit: Jeanne Menjoulet

Photo#2 – 7novembre16h34. Paris, November 7,2016. Photo credit: Elide