report-consumer-friendly-scoring
Dieses Dokument ist Teil der Anfrage „Gutachten des Sachverständigenrats für Verbraucherfragen“
4
Executive Summary –
Recommended actions for consumer-
friendly scoring
in receiving information. At the same time, the trade
1. Making scoring secret of how a scoring system has been developed
and programmed would be maintained.
comprehensible for
consumers 3. However, disclosure alone will not necessarily give
consumers a better understanding of how scoring
1. The Advisory Council for Consumer Affairs recom- works. This will require a variety of measures, which
mends that data protection authorities operational- include: providing examples of consumer scores
ise the comprehensibility requirements set out in the and how they are tiered according to different vari
GDPR (cf. Article 15 para. 1 letter h) for scoring and ables; the production of visual teaching aids (e. g.
score-based business processes. Comprehensibility by consumer organisations); general efforts to raise
should be measured according to the standards scoring-related competence among consumers. Any
relevant to the average consumer. Where scoring assessments of how comprehensible scores are to
entails a level of complexity that is no longer com- consumers should be based not only on expert
prehensible to the individual consumer, measures opinion but on empirical evidence.
should be taken to ensure that scoring processes
can be understood not only by supervisory authori- 4. Consumers already have a right to tailored and
ties, but, at the very least, by consumer bodies and meaningful written information whenever they are
non-state actors as well. scored (see Article 13 para. 2 letter f, 15 para. 1 letter
h GDPR). However, this right has not yet been set
2. Scoring services should release clear and com- out in more concrete terms. Companies, superviso-
prehensible information for consumers about ry authorities and consumer organisations should
the main criteria used to score them and, pref- work together to develop standards for scoring ser-
erably, how these variables are weighted. Trade vices, which would help guarantee relevance and
secrets, of course, must remain inviolable. The comprehensibility. The Advisory Council further
definition of which variables are considered cru- recommends informing consumers of how their
cial for consumers cannot be left exclusively to personal score is to be interpreted against the dis-
lawmakers: this task should additionally fall with- tribution of score values among the population as
in the remit of consumer organisations, or, alter- a whole (e. g. does my score put me in the “upper
natively, the “market watchdogs” of Germany’s third”?).
consumer advice centres. At any rate, full disclo-
sure to supervisory authorities of scoring sys- 5. Prompt, free-of-charge notification should be pro-
tems and their attributes is a must (see page 5 of vided – or at least offered as an option for consum-
the Advisory Council’s Digital Sovereignty report). ers – in the event of major changes to a person’s
score (e. g. if the person slips into a lower category).
Some members of the Advisory Council advocate Naturally, there are certain limitations to this: in
further-reaching transparency. They believe that order to register a change in score, scoring services
all scoring variables should be disclosed to the would have to retain historical score values. There
consumer and that the relative weighting of each are many practical applications (such as fraud rec-
component should be indicated in the calculation ognition or determining possible payment modali-
of the score. To this extent, any interests on the part ties) for which this option will not be available. At
of scoring services and users in maintaining secrecy banks and insurance companies, scores are calcu-
would take second place to the consumer’s interest lated on an ad-hoc basis. This means that no score
5
history is maintained, and potential changes are not
apparent at the time the next “event” is registered. 3. Identifying and revealing
This proposal can therefore be implemented only
at institutions where data collection is ongoing,
discrimination
e. g. credit scoring services and the Federal Motor
Transport Authority in Flensburg (with its “Register 1. The Advisory Council for Consumer Affairs recom-
of Driver Fitness”, which already sends out such no- mends that consumer information rights, as set out
tifications). in Article 15 para. 1 letter h of the GDPR, be strength-
ened. In particular, consumers should be able to as-
certain how scores are distributed among different
groups with different protected attributes (to the
extent that this can be established by the services
themselves). This will allow consumers to provide
2. Fostering knowledge evidence of algorithmic discrimination.
and competence 2. The Advisory Council also recommends strength-
ening the position of supervisory authorities (see
As recommended in the Advisory Council’s Digital Sover- recommendation 7).
eignty report, NGOs, consumer protection organisations
and consumer protection projects should provide edu- 3. Furthermore, it recommends that associations be
cation on basic issues related to scoring in all its man- given the right to pursue representative actions in
ifestations, as well as on the use of scoring in specific cases of discrimination through scoring.
fields of business.
1. For this purpose, the Federal Government should
develop information and discussion materials as
part of its digitalisation strategy for the current par-
liamentary term, with the aim of improving skills 4. Ensuring that non-
on the part of consumers, multipliers and decision-
makers. The underlying principles and quality as-
telematics based options
pects of scoring, as well as forms and causes of un- remain available
equal treatment are just as much part of this basic
knowledge as the rights enjoyed by those scored. 1. The Advisory Council for Consumer Affairs recom-
mends the introduction of legal guarantees to main-
2. Measures should be taken to foster the competence tain telematics-free options for those seeking insur-
people require in order to take informed decisions ance (especially motor vehicle liability insurance
concerning their participation in a scoring process. and health insurance). In particular:
This includes having the skills to identify scoring
services and seek alternatives, as well as to verify, 2. Policyholders who do not use telematics-based tariffs
assess (e. g. is the information relevant to the con- may not suffer substantial disadvantage compared to
sumer disclosed?) and utilise such services. the holders of telematics-based policies.
3. Most members of the Advisory Council for Con
sumer Affairs believe that telematics policies should
6
be self-financing and should not be offered at the 4. The use of proxy variables, as for example in geo-
expense (even indirectly) of policyholders who do scoring, requires special justification (there must
use telematics. Since solidarity objectives are rele- be a causal connection!) and must be subject to the
vant particularly in health insurance, steps would scrutiny of the relevant supervisory authority. The
need to be taken to prohibit cheaper telematics use of proxy variables should be minimised. Where
tariffs that exist only because they attract policy- proxy variables are used, plausible reasoning must
holders with above-average health and do not sig- be given as to their substantive connection with the
nificantly reduce the expenses incurred by insurers. target variable.
5. Ensuring score quality 6. Ensuring data quality
1. The Advisory Council for Consumer Affairs recom- 1. When developing scores, a sufficient level of data
mends that ambitious quality principles be devel- quality must be ensured and documented for
oped on the basis of best practices. This should supervisory authorities.
be based on existing quality assurance initiatives
for algorithmic processes. These quality principles 2. Scoring services and users should enter into volun-
should be developed and updated (drafted, imple- tary commitments to improve their data govern-
mented, monitored) on a collaborative basis by ance, in particular their data quality management,
industry, supervisory authorities, consumer organ- in accordance with the standards set in the quality
isations and the market watchdogs of Germany’s principles.
consumer advice centres.
3. In applying the procedure, measures must be taken
2. Scoring services operating in sensitive fields should to ensure that data is accurate, complete and up-
be obliged to file information with supervisory au- to-date.
thorities that is verifiable in detail and reveals the
high quality of their procedures. Only then will it be 4. In its report on Digital Sovereignty, the Advisory
possible to test scores for consumer fairness. This Council for Consumer Affairs already outlined the
obligation would apply to scores which use statis- option of a data dashboard, which would allow
tical measures to predict behaviour (e. g. false pos- consumers to scrutinise their own data. This would
itive rates, hit rate, gini coefficient, area under the facilitate consumer-oriented data management.
ROC) for the population as a whole and for relevant The Advisory Council reaffirms its recommenda-
population groups (by sex, age, education etc.). This tion that this option be explored. Such explorations
would also make it possible to identify discrimina- should cover current developments in the area of
tion and cases of questionable score quality. secure identity management via blockchain-based
systems, which allow consumers to manage their
3. As the situation currently stands, scoring proce- own identity data securely and definitively.
dures that pursue objectives which have not been
appropriately identified to the public are prohib- 5. The Advisory Council recommends that research
ited by law. In addition to the role of supervisory be conducted promptly to appraise and, where ap-
authorities (see recommendation 7), consumer or- plicable, improve the quality of data used in rele-
ganisations or the market watchdogs of Germany’s vant scoring processes, with a particular focus on
consumer advice centres could also apply their entity recognition. Where necessary, improvements
expertise and contribute to uncovering “falsely la- should be made via statutory provisions. Measures
belled” scores as well. must be taken to ensure that a score calculated for
7
a certain person is correctly assigned to that person. at BaFin). This task force should be set up immedi-
The duty for providers to inform individuals that they ately after the Data Ethics Commission has finished
are being scored (see recommendation for action 1) its work.
will serve to minimise the risk of identity mix-ups.
In this regard there is clearly a conflict between the
interests of scoring services and users, on the one
hand, and data protection interests on the other.
For this reason the Advisory Council recommends 8. Preventing “super scores”
that the Federal Government’s Data Ethics Commis-
sion discuss ways of improving entity recognition The Advisory Council for Consumer Affairs recommends
and develop concrete recommendations. that developments in China and in other countries
which are experimenting with “super scoring” are close-
ly followed and analysed. In particular, public debate is
required on the change in social values and structures
that such systems entail.
7. Improving oversight The development of “super scores” by international
commercial actors may also have an impact on Germany.
1. The Advisory Council for Consumer Affairs recom- Lawmakers and supervisory authorities should prepare
mends that the Federal Government explore whether for an examination of whether measures can and should
a digital agency (see the Advisory Council’s report on be taken to ensure that “super scores” cannot be offered
“Consumer Law 2.0”) could act as a competence cen- commercially in Germany.
tre to assist supervisory authorities in exercising their
mandates. This might consist, for example, in setting The Advisory Council recommends that an examination
up a federal institute as a centre of method expertise be carried out into the extent to which existing instru-
for quality assurance, which could also be used for ments (especially purpose limitation and the “no tie-
“non-digital” purposes. ins” rule) contained in the GDPR may also be used to
prevent “super scores”.
2. The responsible supervisory authorities should be
put in the position (both structurally and in part
through salary improvements for specialists, espe-
cially in statistics and IT) to perform the aforemen-
tioned tasks. Developments at the Federal Financial
Supervisory Authority (BaFin) over the last few years
could serve as good practice. The responsible super-
visory authorities should be granted the considera-
ble financial resources required for them to perform
the aforementioned additional tasks and test con-
crete scoring services.
3. To ensure that the present recommendations are
promptly implemented, the Advisory Council for
Consumer Affairs proposes the creation of a task
force at the level of the Federal Government (for
example at the Federal Chancellery) in order to
develop guidelines for the elaboration of quality
principles on the basis of existing procedures (e. g.
8
Members and staff of the SVRV
Members of the SVRV
Professor Lucia Reisch (Chair) Professor Hans-Wolfgang Micklitz
Professor of Intercultural Consumer Research and Professor of Economic Law at the European Universi-
European Consumer Policy at Copenhagen Business ty Institute in Florence
School
Professor Andreas Oehler
Dr Daniela Büchel (Vice-Chair) Professor of Finance at the University of Bamberg
Member of the Trade Germany Board, REWE Group, and Director of the University’s Research Centre for
Managing Director of REWE Markt GmbH and of Household Finance and Financial Literacy
Penny-Markt GmbH
Professor Kirsten Schlegel-Matthies
Professor Gerd Gigerenzer Professor of Home Economics at the University of
Director of the Harding Centre for Risk Literacy at Paderborn
the Max Planck Institute for Human Development in
Berlin Professor Gert G. Wagner
Max Planck Fellow at the Max Planck Institute for
Helga Zander-Hayat Human Development in Berlin, Research Associate at
Member of the Board of Management of North the Alexander von Humboldt Institute for Internet and
Rhine-Westphalia Consumer Advice Centre Society, Berlin, and Senior Research Fellow for at the
German Socio-Economic Panel Study at the German
Professor Gesche Joost Institute for Economic Research (DIW Berlin)
Professor of Design Research at the University of Fine
Arts, Berlin
Staff of the SVRV
Head of the Bureau:
Thomas Fischer, M.A.
Research staff of the Bureau:
Johannes Gerberding
Dr Christian Gross
Dr Ariane Keitel
Sarah Sommer, M.A.
TABLE OF CONTENTS 9
Table of contents
A
About this report 13
I. Introduction 14
II. Scores and scoring 16
III. Objectives of the report 20
Objective 1: Improve the information base and increase knowledge of scoring 20
Objective 2: Broaden the empirical basis and address legal issues 21
Objective 3: Suggest rules for consumer-friendly scoring 21
B
Areas for action:
the state of research 25
I. Transparency and comprehensibility 26
1. Transparency in predictive scoring 26
2. Transparency in behavioural scoring 27
3. Keeping transparency and comprehensibility of scoring systems on the agenda 28
4. Scoring transparency as a special form of algorithm transparency 30
5. Transparency as a condition for a social debate on scoring 32
II. Non-discrimination and equal treatment 34
1. What is discrimination? 34
2. Discrimination through scoring input 35
3. Score quality and non discrimination 36
4. Undesirable unequal scoring-based treatment beyond discrimination 39
III. Enforcement of rights 40
IV. Score quality 41
1. Quality of the algorithm underlying a score 41
2. The utility of newer and more complex algorithms 45
V. Baseline data 46
1. Accuracy, currency and completeness 46
2. Use of proxy variables 47
3. Weighting of input variables 48
VI. Competing fairness criteria 50
10 TABLE OF CONTENTS
VII. Consumers and society: e
xpectations, knowledge,
competence and implications 52
1. Consumers’ expectations and acceptance of scoring 52
2. Knowledge and competence 54
3. Social implications 57
VIII. The danger of a super score 61
1. Scoring models abroad 61
2. Data accumulation and data trading 65
3. Repersonalisation of anonymised data 68
4. Aggregation of data into a super score 69
C
Market survey: credit reference
agencies, motor i nsurance
telematics and health insurance
policies71
I. Introduction and key issues 72
II. Survey design 73
1. Overview of providers 74
2. The questionnaires 75
III. Discussion of findings and highlighted consumer problems 76
1. Diffusion of scoring in the market segments under examination 76
2. Transparency 78
3. Score calculation and statistical quality 80
4. Behavioural effects 84
5. Discrimination 85
6. Aggregation of data and inclusion of new consumer attributes 87
7. Supervision 88
TABLE OF CONTENTS 11
D
Public knowledge and acceptance
of scoring 91
I. Preliminary study, 2017 92
II. Representative survey, 2018 93
1. Analysis of the findings 94
2. Multivariate regression analyses: presentation and discussion of findings 106
3. Population survey findings: general summary and conclusions 109
E
The legal framework for scoring 111
I. The basis in data privacy law 113
1. Profiling (Article 4(4) GDPR) 113
2. Automated individual decision-making (Article 22 GDPR) 115
3. Scoring of probability values (section 31 of the Federal Data Protection Act) 118
II. Rules for specific areas of activity 124
1. The law governing standard business terms 124
2. The law governing insurance contracts and insurance supervision 125
3. Social insurance law and statutory health insurance 128
III. Building blocks for a scoring regime 129
1. Regulating the ‘how’ of scoring versus regulating the ‘whether’ 129
2. Scoring regulation and algorithm regulation 130
3. Guaranteeing a defined score quality 130
4. Guaranteeing transparency and comprehensibility 132
5. Guaranteeing non-discrimination 135
IV. Supervision 138
12 TABLE OF CONTENTS
About this report 13
A
About this
report
