The kinderopvangtoeslag (childcare allowance) is a benefit provided by the Dutch state through a special department of the tax authority (Belastingdienst/Toeslagen, hereafter BD/T). It is meant to allow parents to combine the care for their children with employment and other activities by providing them an allowance to purchase childcare from registered child care centres. The amount of the allowance is related to a number of factors and can be quite substantial. 
Strict legislation was therefore put into place to combat fraud, harshly enforced by the BD/T. This in turn created a situation of (in retrospect) overenforcement which resulted in significant problems for parents, who were in some cases forced to pay back large amounts of allowances. A parliamentary inquiry concluded in 2020 that the way the parents were treated amounted to an ‘unprecedented injustice’, which resulted in the resignation of the government.
The Netherlands Institute for Human Rights decided to conduct an own-initiative investigation into the BD/T after having received several dozen complaints in the context of its quasi-judicial procedure in the beginning of 2021. The complaints alleged that the enforcement of BD/T had not just been harsh, but also discriminatory.
At the time, investigatory reports conducted by the State Auditor Service and the Data Protection Authority contained evidence of (incidental) discriminatory treatment in the BD/T enforcement processes and practices. They found, among other things, that one of the self-learning algorithms used to select applicants for investigation used ‘foreign nationality’ as a (negative) risk factor. In addition, BD/T illegally registered whether persons next to the Dutch nationality also possessed another nationality, and made use of nationality data in its fraud investigations.
The issue that the Netherlands Institute for Human Rights sought to address was whether the findings of these organisations were evidence of incidental missteps or whether they were rather reflective of processes of institutional discrimination. In addition, the Institute wanted to gain further insight into the potential discriminatory effects of the application of AI to select persons for investigative action. As such, the Institute sought to investigate whether the enforcement practices and processes of BD/T in practice affected persons with a foreign descent disproportionately, and if so, whether this gave rise to a presumption of discrimination on grounds of race/ethnicity in the area of social protection.
For the purposes of the investigation, the Netherlands Institute for Human Rights requested data from the BD/T on all persons who received childcare allowances in 2014 and 2018. 2014 was selected because it featured significantly in the complaints submitted to the Institute, and because the year was also considered in the investigations by other authorities. 2018 was selected as a ‘control year’, as the last full year before the Dutch supreme administrative court rulings which sought to limit the harsh enforcement practice of BD/T (inter alia by reinstating the principle of proportionality).
The Institute then defined two groups to be compared based on the data available at BD/T to broadly match the groups at issue in the judgments of the European Court of Human Rights in Biao v Denmark and Court of Justice of the EU in Feryn:
With respect to these groups, a statistical analysis was conducted to answer the following questions:
In order to conduct this analysis, for each individual who received childcare allowance in 2014 and 2018, the following data was requested.  Whereas the BD/T cooperated in providing the data, all told the administrative processes leading up to the first set of data being made available took some five months. The second request for data, relating in particular to the blacklist, took a similar amount of time.
The following data was requested, namely:
See Table 1 for an explanation and evaluation of each type of data.
Based on this data, for each process the following contingency table was constructed.
|Table with frequencies
|Negative outcome (C)
|Neutral/Positive outcome (D)
|Hypothetically disadvantaged group (A)
|Number of people who have characteristic X and experienced the negative outcome
|Number of people who have characteristic X and experienced the neutral/positive outcome
|Comparison group (B)
|Number of people who don’t have characteristic X and experienced the negative outcome
|Number of people who don’t have characteristic X and experienced the neutral/positive outcome
The chi-square statistical test was used to test whether the variables (descent and outcome) were associated. The odds ratio was used to express the strength of the differences between the groups.
As an example of the way this was applied, reference can be made to the numbers in the CJEU case C-389/20, CJ v Tesorería General de la Seguridad Social (TGSS). This case revolved around the question as to whether the exclusion of domestic workers from unemployment benefits was indirectly discriminatory on grounds of sex. In essence, there were two social security regimes: a general one applying to most Spanish workers, and the special social security regime applying to domestic workers. Within each social security regime, the number of men and women can be tabulated as follows:
|Special regime (negative outcome)
|General regime (positive outcome)
|7 770 798
|8 101 899
A chi-square test applied to these proportions is statistically significant,  suggesting an association between sex and the distribution between the social security regimes. Looking at the residuals of the chi-square test, we can intuit that women are overrepresented in the special regime/negative outcome. Next, an odds ratio can tell us something about the strength of this association. For the numbers above, it can be calculated as follows:
This means that, relatively speaking, women are 22 times more likely to fall within the special regime than men. The Court of Justice strongly suggested that this was sufficient for a presumption of indirect discrimination.
For each of the contingency tables constructed, the chi-square test was applied and resulted in being statistically significant. The results indicated an association between descent and the various measures (e.g. selected for investigative action under the risk algorithm), with persons with a foreign descent being overrepresented in the negative outcome. The following table summarises the odds ratio’s that were calculated.
|High allowance high risk
|Intent/gross negligence 
|Odds ratio 2014
|Odds ratio 2018
This means that for high allowance, high risk in 2018, persons with a foreign descent were 3.52 more likely, relatively speaking, to be selected for further investigation by a civil servant. For CAF in 2014 this was almost 16 times more likely, relatively speaking.
Visually, these results can be represented with a waffle chart. If we were to represent the population of persons who received childcare allowance in 2018 with 100 squares, the proportion of persons with a Dutch or a foreign descent would be as follows.
Next, a waffle chart showing the proportion of people selected by the High Allowance, High Risk-rule in 2018 for further investigation:
The Institute took the view that if a significantly greater proportion of individuals  with a foreign descent when compared to persons with a Dutch descent were selected for investigation/negative outcome, it was sufficient for a presumption of indirect discrimination on the grounds of race/ethnicity.
It was considered that the two groups compared in the investigation (persons with a foreign and with a Dutch descent) broadly matched the groups compared in Biao v Denmark (ethnic and non-ethnic Danes) and Feryn (foreigners versus non-foreigners) in which the ECtHR and the CJEU respectively concluded that discrimination on grounds of race/ethnicity was at issue. It was further considered that the recent CJEU cases of Jyske Finans or KV did not alter that conclusion, since these cases – in essence – concerned a difference in treatment between different groups than which were examined here: EU/EEA citizens versus third country-nationals. Finally, the Dutch Supreme Court in recent criminal cases upheld the court of appeal judgment that hate speech broadly directed at persons of foreign descent constituted discrimination on grounds of race.
The next question was what a ‘significantly greater proportion of individuals’ meant. The case law gives little guidance in this respect. The Institute reasoned as follows. If we take the hypothetical/standardised situation where there is a 50/50 division of persons with a Dutch and a foreign descent respectively, a 3.52 odds ratio (the 2018 High Allowance High Risk ratio) means that if a selection of 100 persons took place for further investigation the division of persons so selected would be 22/78 (Dutch/foreign descent).
This was considered a significant shift (>25 points). Thus, the Institute found that all of the processes and practices under investigation with the exception of the ‘calculator’ met this standard and thus affected a significantly greater proportion of persons with a foreign descent when compared with persons with a Dutch descent. For the ‘calculator’, the Institute has taken the view that while the differences are smaller, a consistent disadvantage still exists for persons with a foreign descent for multiple years.
Taken together the Institute considers there to be sufficient evidence for there to be a presumption of indirect discrimination on grounds of race/ethnicity. In reaching this conclusion the Institute weighed in, apart from its own investigation, the findings of the earlier investigations by other authorities as well the fact that the Dutch Central Bureau for Statistics meanwhile found that of the persons recognised as having been a victim of the ‘unprecedented injustice’, about half were born abroad.
The onus will thus be on BD/T to prove that there has been no breach of the principle of equal treatment in each of the individual cases that will be investigated in the context of the quasi-judicial procedure that the Institute operates.
 As an example, in 2021, the benefit was provided for some one million children in childcare, averaging some € 5000 per recipient.
 Due to data protection rules, the data was not made available outside the BD/T technical infrastructure. The investigative team was instead provided with access to this infrastructure.
 X² (1, N= 16256859) = 325434, p < 2.2e-16.
 There was no systematic registration of this qualification. As such the data on which the analysis was conducted was created by conducting a manual search of the notes attached to each individual case by civil servants.
 The results here should be treated with caution. The persons who received childcare allowance in 2014 and 2018 were used as a sample to estimate the descent of all people on the FSV/blacklist. The sample was large, however.
 This programme was cancelled in 2018.
 Case C-274/18, Schuch-Ghannadan, ECLI:EU:C:2019:828, para. 45.
 The formula for calculating this is Odds ratio/Odds ratio + 1.
 Case C-167/97, Seymour-Smith, ECLI:EU:C:1999:60, para. 61.
The views on this blog are always the authors’ and they do not necessarily reflect Equinet’s position.