ArticlePDF Available

Abstract and Figures

The literature provides a wide range of techniques to assess and improve the quality of data. Due to the diversity and complexity of these techniques, research has recently focused on defining methodologies that help the selection, customization, and application of data quality assessment and improvement techniques. The goal of this article is to provide a systematic and comparative description of such methodologies. Methodologies are compared along several dimensions, including the methodological phases and steps, the strategies and techniques, the data quality dimensions, the types of data, and, finally, the types of information systems addressed by each methodology. The article concludes with a summary description of each methodology.
The architecture of the DaQuinCIS framework. Fusionplex [Motro and Anokhin 2005], iFuice [Rahm et al. 2005], and HumMer [Bilke et al. 2005]. We now comment on the methodologies that are strongly focused on specific types of information systems, namely DWQ and IQM. DWQ is specifically oriented to data warehouses. In Jarke et al. [1995], it is observed that most researches have studied DW in their role as buffers of materialized views, mediating between update-intensive OLTP systems and query-intensive decision support. This neglects the organizational role of data warehousing as a means of obtaining a centralized control of information flows. As a consequence, a large number of quality issues relevant for DW cannot be expressed with traditional DW meta models. DWQ makes two contributions towards solving these problems. First, the metadata about DW architectures are enriched with explicit enterprise models. Second, mathematical techniques for measuring or optimizing several aspects of DW quality are proposed. The Artkos tool also contributes to address the DQ issues of data warehouses [Vassiliadis et al. 2001]. Artkos proposes a metamodel made of several entities, among them: (1) activities, atomic units of data processing work; (2) scenarios, sets of activities to be executed together; and (3) quality factors, defined for each activity, corresponding to the dimensions and metrics described in Section 2.3. IQM is strongly focused on Web information systems, as it considers a wide set of existing tools to evaluate information quality in the Web context, namely site analyzers, traffic analyzers, port scanners, performance monitoring systems, Web mining tools and survey tools to generate opinion-based user feedback. Several information quality criteria (in our terminology, dimensions and metrics, as seen in Section 3.3) can be measured with the help of these tools. IQM provides systematic sequential steps to match information quality criteria with measurement tools.
Content may be subject to copyright.
Methodologies for Data Quality Assessment and Improvement
Universit`a di Milano - Bicocca
Politecnico di Milano
Politecnico di Milano
Universit`a di Milano - Bicocca
The literature provides a wide range of techniques to assess and improve the quality of data. Due to the
diversity and complexity of these techniques, research has recently focused on defining methodologies that
help the selection, customization, and application of data quality assessment and improvement techniques.
The goal of this article is to provide a systematic and comparative description of such methodologies. Method-
ologies are compared along several dimensions, including the methodological phases and steps, the strategies
and techniques, the data quality dimensions, the types of data, and, finally, the types of information systems
addressed by each methodology. The article concludes with a summary description of each methodology.
Categories and Subject Descriptors: A.1 [Introductory and Survey]; H.2.m [Database Management]:
General Terms: Management, Measurement
Additional Key Words and Phrases: Data quality, data quality measurement, data quality assessment, data
quality improvement, methodology, information system, quality dimension
ACM Reference Format:
Batini, C., Cappiello, C., Francalanci, C., and Maurino, A. 2009. Methodologies for data quality assessment
and improvement. ACM Comput. Surv. 41, 3, Article 16 (July 2009), 52 pages.
DOI =10.1145/1541880.1541883
This article has been partially supported by the European IST project n. 27347 SEEMP - Single European
Employment Market-Place, the European IST project WSDiamond - Web Services DIAgnosability, Monitoring
and Diagnosis and the Italian FIRB project Eg4M - eGovernment for Meditteranean countries.
Author’s address: C. Batini, Dipartimento di Informatica, Sistemistica e Comunicazioni (DISCO), Universita’
degli Studi di Milano Bicocca, Viale Sarca 336, 20126, Milano, Italy; email:
Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted
without fee provided that copies are not made or distributed for profit or commercial advantage and that
copies show this notice on the first page or initial screen of a display along with the full citation. Copyrights for
components of this work owned by others than ACM must be honored. Abstracting with credit is permitted.
To copy otherwise, to republish, to post on servers, to redistribute to lists, or to use any component of this
work in other works requires prior specific permission and/or a fee. Permissions may be requested from
Publications Dept., ACM, Inc., 2 Penn Plaza, Suite 701, New York, NY 10121-0701 USA, fax +1 (212) 869-
0481, or
2009 ACM 0360-0300/2009/07-ART16 $10.00
DOI 10.1145/1541880.1541883
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:2 C. Batini et al.
Because electronic data are so pervasive, the quality of data plays a critical role in
all business and governmental applications. The quality of data is recognized as a
relevant performance issue of operating processes [Data Warehousing Institute 2006],
of decision-making activities [Chengalur-Smith et al. 1999], and of interorganizational
cooperation requirements [Batini and Scannapieco 2006]. Several initiatives have been
launched in the public and private sectors, with data quality having a leading role, such
as the Data Quality Act enacted by the United States government in 2002 [Office of
Management and Budget 2006] and the Data Quality Initiative Framework enacted
by the government of Wales in 2004 to improve the information quality of all general
medical practices [DQI 2004].
At the same time, information systems have been migrating from a hierarchical/
monolithic to a network-based structure, where the set of potential data sources that
organizations can use has dramatically increased in size and scope. The issue of data
quality has become more complex and controversial as a consequence of this evolu-
tion. In networked information systems, processes are involved in complex information
exchanges and often operate on input obtained from external sources, which are fre-
quently unknown a priori. As a consequence, the overall quality of the data that flows
across information systems can rapidly degrade over time if the quality of both processes
and information inputs is not controlled. On the other hand, networked information
systems offer new opportunities for data quality management, including the availabil-
ity of a broader range of data sources and the ability to select and compare data from
different sources to detect and correct errors, and, thus, improve the overall quality of
The literature provides a wide range of techniques to assess and improve the qual-
ity of data, such as record linkage, business rules, and similarity measures. Over
time, these techniques have evolved to cope with the increasing complexity of data
quality in networked information systems. Due to the diversity and complexity of
these techniques, research has recently focused on defining methodologies that help
select, customize, and apply data quality assessment and improvement techniques.
This article defines a data quality methodology as a set of guidelines and techniques
that, starting from input information describing a given application context, defines
a rational process to assess and improve the quality of data. The goal of this arti-
cle is to provide a systematic and comparative description of existing data quality
The article is organized as follows. Section 2 introduces the basic data quality issues
common to all methodologies, which represent the perspectives used in this article for
comparative analysis such as: (1) the methodological phases and steps, (2) the strate-
gies and techniques, (3) the data quality dimensions, (4) the types of data, and, finally,
(5) the types of information systems. Section 3, the core of the article, compares ex-
isting methodologies along the coordinates introduced in Section 2. The comparison
is performed with: synoptic tables, that highlight at a glance groups of methodologies
with similar approaches, in-depth comments, and qualitative evaluations. Section 4
describes ongoing research and future research directions in the field of data qual-
ity methodologies. Finally, the article concludes with a summary description of each
methodology in Appendix A describing: (1) the phases of each methodology and their
mutual dependencies and critical decisions, (2) a general description highlighting the
focus of each methodology and original contribution to the data quality assessment and
improvement process, and (3) detailed comments discussing the applicability of each
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:3
There exist several perspectives that can be used to analyze and compare data quality
(DQ) methodologies:
(1) the phases and steps that compose the methodology;
(2) the strategies and techniques that are adopted in the methodology for assessing and
improving data quality levels;
(3) the dimensions and metrics that are chosen in the methodology to assess data qual-
ity levels;
(4) the types of costs that are associated with data quality issues including:
(a) costs associated with poor data quality, that is process costs caused by data
errors and opportunity costs due to lost and missed revenues; these costs are
also referred to as indirect costs;
(b) costs of assessment and improvement activities, also referred as direct costs;
(5) the types of data that are considered in the methodology;
(6) the types of information systems that use, modify, and manage the data that are
considered in the methodology;
(7) the organizations involved in the processes that create or update the data that are
considered in the methodology, with their structure and norms;
(8) the processes that create or update data with the goal of producing services required
by users that are considered in the methodology;
(9) the services that are produced by the processes that are considered in the
Methodologies differ in how they consider all of these perspectives. In the remainder
of this article the last three perspectives—organization, process, and service—will not
be investigated, as they are rarely mentioned in methodologies.
2.1. Common Phases and Steps
In the most general case, the sequence of activities of a data quality methodology is
composed of three phases:
(1) State reconstruction, which is aimed at collecting contextual information on orga-
nizational processes and services, data collections and related management proce-
dures, quality issues and corresponding costs; this phase can be skipped if contextual
information is available from previous analyses.
(2) Assessment/measurement, which measures the quality of data collections along rel-
evant quality dimensions; the term measurement is used to address the issue of
measuring the value of a set of data quality dimensions. The term assessment is
used when such measurements are compared to reference values, in order to enable
a diagnosis of quality. The term assessment is adopted in this article, consistent
with the majority of methodologies, which stress the importance of the causes of
poor data quality.
(3) Improvement concerns the selection of the steps, strategies, and techniques for
reaching new data quality targets.
The state reconstruction phase is optional if the assessment phase can be based on
existing documentation. Since methodologies typically make this assumption, we will
not further discuss the state reconstruction phase. Although adopting different names,
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:4 C. Batini et al.
methodologies organize the assessment and improvement phases in terms of a common
set of basic steps. The steps of the assessment phase are:
data analysis, which examines data schemas and performs interviews to reach a
complete understanding of data and related architectural and management rules;
DQ requirements analysis, which surveys the opinion of data users and administra-
tors to identify quality issues and set new quality targets;
identification of critical areas, which selects the most relevant databases and data
flows to be assessed quantitatively;
process modeling, which provides a model of the processes producing or updating
measurement of quality, which selects the quality dimensions affected by the quality
issues identified in the DQ requirements analysis step and defines corresponding
metrics; measurement can be objective when it is based on quantitative metrics, or
subjective, when it is based on qualitative evaluations by data administrators and
Note that in all the steps of the assessment phase, a relevant role is played by meta-
data that store complementary information on data for a variety of purposes, including
data quality. Metadata often provide the information necessary to understand data
and/or evaluate them.
The steps of the improvement phase are:
evaluation of costs, which estimates the direct and indirect costs of data quality;
assignment of process responsibilities, which identifies the process owners and defines
their responsibilities on data production and management activities;
assignment of data responsibilities, which identifies the data owners and defines their
data management responsibilities;
identification of the causes of errors, which identifies the causes of quality problems;
selection of strategies and techniques, which identifies all the data improvement
strategies and corresponding techniques, that comply with contextual knowledge,
quality objectives, and budget constraints;
design of data improvement solutions, which selects the most effective and efficient
strategy and related set of techniques and tools to improve data quality;
process control, which defines check points in the data production processes, to mon-
itor quality during process execution;
process redesign, which defines the process improvement actions that can deliver
corresponding DQ improvements;
improvement management, which defines new organizational rules for data quality;
improvement monitoring, which establishes periodic monitoring activities that pro-
vide feedback on the results of the improvement process and enables its dynamic
In Section 3.1, methodologies are compared in their assessment and improvement ca-
pabilities by evaluating their completeness along the set of phases and steps introduced
in this section. Note that, usually, each methodology refers to a specific assessment
or improvement functionality by using different terms. In the appendix, we describe
methodologies by adopting the original terms, but we provide the correspondence be-
tween such terms and the classification presented here.
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:5
2.2. Strategies and Techniques
In their improvement steps, methodologies adopt two general types of strategies, namely
data-driven and process-driven.Data-driven strategies improve the quality of data by
directly modifying the value of data. For example, obsolete data values are updated by
refreshing a database with data from a more current database. Process-driven strate-
gies improve quality by redesigning the processes that create or modify data. As an
example, a process can be redesigned by including an activity that controls the format
of data before storage.
Strategies, both data- and process-driven, apply a variety of techniques: algorithms,
heuristics, and knowledge-based activities, whose goal is to improve data quality. An
open-ended list of the improvement techniques applied by data-driven strategies is:
(1) acquisition of new data, which improves data by acquiring higher-quality data to
replace the values that raise quality problems;
(2) standardization (or normalization), which replaces or complements nonstandard
data values with corresponding values that comply with the standard. For example,
nicknames are replaced with corresponding names, for example, Bob with Robert,
and abbreviations are replaced with corresponding full names, for example, Channel
Str. with Channel Street.
(3) Record linkage, which identifies that data representations in two (or multiple) tables
that might refer to the same real-world object;
(4) data and schema integration, which define a unified view of the data provided
by heterogeneous data sources. Integration has the main purpose of allowing a
user to access the data stored by heterogeneous data sources through a unified
view of these data. In distributed, cooperative, and P2P information systems (see
Section 2.6), data sources are characterized by various kinds of heterogeneities that
can be generally classified into (1) technological heterogeneities, (2) schema het-
erogeneities, and (3) instance-level heterogeneities. Technological heterogeneities
are due to the use of products by different vendors, employed at various lay-
ers of an information and communication infrastructure. Schema heterogeneities
are primarily caused by the use of (1) different data models, as in the case of a
source that adopts the relational data model and a different source that adopts
the XML data model, and (2) different representations for the same object, such
as two relational sources that represent an object as a table and an attribute.
Instance-level heterogeneities are caused by different, conflicting data values pro-
vided by distinct sources for the same objects. For instance, this type of hetero-
geneity can be caused by independent and poorly coordinated processes that feed
the different data sources. Data integration must face all the types of these listed
(5) Source trustworthiness, which selects data sources on the basis of the quality of
their data;
(6) error localization and correction, which identify and eliminate data quality errors
by detecting the records that do not satisfy a given set of quality rules. These tech-
niques are mainly studied in the statistical domain. Compared to elementary data,
aggregate statistical data, such as average, sum, max, and so forth are less sensitive
to possibly erroneous probabilistic localization and correction of values. Techniques
for error localization and correction have been proposed for inconsistencies, incom-
plete data, and outliers [Dasu and Johnson 2003]; [Batini and Scannapieco 2006].
(7) Cost optimization, defines quality improvement actions along a set of dimensions
by minimizing costs.
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:6 C. Batini et al.
Two main techniques characterize process-driven strategies:
Process control inserts checks and control procedures in the data production process
when: (1) new data are created, (2) data sets are updated, or (3) new data sets are
accessed by the process. In this way, a reactive strategy is applied to data modification
events, thus avoiding data degradation and error propagation.
Process redesign redesigns processes in order to remove the causes of poor quality
and introduces new activities that produce data of higher quality. If process redesign
is radical, this technique is referred to as business process reengineering [Hammer
and Champy 2001]; [Stoica et al. 2003].
Several techniques typical of data- and process- driven strategies are compared in
Redman [1996] by discussing the improvement that each technique can achieve along
different quality dimensions and the implementation cost of each technique. This com-
parison is performed from both a short-term and a long-term perspective. The compar-
ison focuses on: (1) acquisition of new data, (2) record linkage, (3) error localization
and correction, (4) process control, and (5) process redesign techniques. In general,
in the long term, process-driven techniques are found to outperform data-driven tech-
niques, since they eliminate the root causes of quality problems. However, from a short-
term perspective, process redesign can be extremely expensive [Redman 1996][English
1999]. On the contrary, data-driven strategies are reported to be cost efficient in the
short term, but expensive in the long term. They are suitable for one-time application
and, thus, they are recommended for static data.
2.3. Dimensions
In all methodologies, the definition of the qualities, dimensions, and metrics to assess
data is a critical activity. In general, multiple metrics can be associated with each
quality dimension. In some cases, the metric is unique and the theoretical definition of
a dimension coincides with the operational definition of the corresponding metric. For
this reason, in the following we make a distinction between theoretical and operational
definitions of dimensions only when the literature provides multiple metrics.
Quality dimensions can be referred either to the extension of data—to data values, or
to their intension—to their schema. Although the quality of conceptual and logical data
schemas is recognized to be a relevant research area [IWCMQ 2003], most definitions of
data quality dimensions and metrics are referred to data values as opposed to schemas.
This article focuses mainly on quality dimensions and metrics referred to data values.
The data quality literature provides a thorough classification of data quality dimen-
sions; however, there are a number of discrepancies in the definition of most dimensions
due to the contextual nature of quality. The six most important classifications of quality
dimensions are provided by Wand and Wang [1996]; Wang and Strong [1996]; Redman
[1996]; Jarke et al. [1995]; Bovee et al. [2001]; and Naumann [2002]. By analyzing these
classifications, it is possible to define a basic set of data quality dimensions, including
accuracy,completeness,consistency, and timeliness, which constitute the focus of the
majority of authors [Catarci and Scannapieco 2002].
However, no general agreement exists either on which set of dimensions defines the
quality of data, or on the exact meaning of each dimension. The different definitions
provided in the literature are discussed in the following.
Accuracy. Several definitions are provided for the term accuracy. Wang and Strong
[1996] define accuracy as “the extent to which data are correct, reliable and certified.”
Ballou and Pazer [1985] specify that data are accurate when the data values stored in
the database correspond to real-world values. In Redman [1996], accuracy is defined
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:7
Tab le I. Different Definitions Provided for Completeness
Reference Definition
Wand and Wang 1996 Ability of an information system to represent every meaningful
state of a real world system
Wang and Wand 1996 Extent to which data are of sufficient breadth, depth, and scope
for the task at hand
Redman 1996 Degree to which values are included in a data collection
Jarke et al. 1995 Percentage of real-world information entered in data sources
and/or data warehouse
Bovee et al. 2001 Information having all required parts of an entity’s description
Naumann 2002 Ratio between the number of non-null values in a source and the
size of the universal relation
Liu and Chi 2002 All values that are supposed to be collected as per a collection
as a measure of the proximity of a data value, v, to some other value, v’, that is con-
sidered correct. In general, two types of accuracy can be distinguished, syntactic and
semantic. Data quality methodologies only consider syntactic accuracy and define it as
the closeness of a value, v, to the elements of the corresponding definition domain, D.
In syntactic accuracy, we are not interested in comparing vwith its real-world value v’;
rather, we are interested in checking whether vis any one of the values in D, or how
close it is to values in D. For example, v=‘Jean’ is considered syntactically accurate
even if v’ =‘John’.
Completeness. Completeness is defined as the degree to which a given data collection
includes data describing the corresponding set of real-world objects.
Table I reports the research contributions that provide a definition of completeness.
By comparing such definitions, it can be observed that there is a substantial agreement
on the abstract definition of completeness. Definitions differ in the context to which they
refer, for example, information system in Wand and Wang [1996], data warehouse in
Jarke et al. [1995], entity in Bovee et al. [2001].
In the research area of relational databases, completeness is often related to the
meaning of null values. A null value has the general meaning of missing value,a
value that exists in the real world but is not available in a data collection. In order to
characterize completeness, it is important to understand why the value is missing. A
value can be missing either because it exists, but is not known, or because it does not
exist, or because it is not known whether it exists (see Atzeni and Antonellis [1993]).
Let us consider the table Person reported in Figure 1, with attributes Name,Surname,
BirthDate, and Email. If the person represented by tuple 2 has no email, tuple 2 is
complete. If it is not known whether the person represented by tuple 4 has an email,
incompleteness may or may not occur. During quality assessment, a Boolean value
(complete or not complete) should be associated with each field to calculate completeness
as the ratio between complete values and the total number of values, both at the tuple
and at the source level.
Consistency. The consistency dimension refers to the violation of semantic rules de-
fined over a set of data items. With reference to the relational theory, integrity con-
straints are a type of such semantic rules. In the statistical field, data edits are typical
semantic rules that allow for consistency checks.
In the relational theory, two fundamental categories of integrity constraints can be
distinguished, namely: intra-relation constraints and inter-relation constraints. Intra-
relation constraints define the range of admissible values for an attribute’s domain.
Examples are “Age must range between 0 and 120,” or “If WorkingYears is lower than
3, then Salary cannot be higher than 25.000 euros per year.” Inter-relation integrity
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:8 C. Batini et al.
Fig. 1. Null values and data completeness.
Table II. Existing Definitions of Time-Related Dimensions
Reference Definition
Wand and Wang 1996 Timeliness refers only to the delay between a change of a real
world state and the resulting modification of the information
system state
Wang and Wand 1996 Timeliness is the extent to which the age of data is appropriate
for the task at hand
Redman 1996 Currency is the degree to which a datum is up-to-date. A datum
value is upto- date if it is correct in spite of possible
discrepancies caused by timere-lated changes to the correct
Jarke et al. 1995 Currency describes when the information was entered in the
sources and/or the data warehouse.
Volatility describes the time period for which information is valid
in the real world
Bovee et al. 2001 Timeliness has two components: age and volatility. Age or
currency is a measure of how old the information is, based on
how long ago it was recorded. Volatility is a measure of
information instability, the frequency of change of the value for
an entity attribute
Naumann 2002 Timeliness is the average age of the data in a source
Liu and Chi 2002 Timeliness is the extent to which data are sufficiently up-to-date
for a task
constraints involve attributes from different relations. As an example, let us con-
sider a Movies relation that includes the Title,Director, and Year attributes and
an OscarAwards relation, specifying the MovieTitle and the Year when the award was
won. An inter-relation constraint could state that for each movie appearing in both rela-
tions, “Movies.Year must be equal to OscarAwards.Year.” There is an extensive literature
on consistent databases. For example, Arenas et al. [1999], considers the problem of
the logical characterization of the notion of consistent answer in a relational database,
which may violate given integrity constraints. The authors propose a method for com-
puting consistent answers, by proving their soundness and completeness. Integrity
constraints have also been studied as enablers of data integration [Cal`
ı et al. 2004].
In the statistical area, data from census questionnaires have a structure correspond-
ing to the questionnaire schema. Semantic rules, called edits, can be defined on the
questionnaire schema to specify the correct set of answers. Such rules typically de-
note error conditions. For example, an edit could be: if MaritalStatus is “married,” Age
must not be lower than 14. After the detection of erroneous records, the act of restoring
correct values is called imputation [Fellegi and Holt 1976].
Time-related Dimensions: Currency, Volatility, and Timeliness. An important aspect of data
is their update over time. The main time-related dimensions proposed in the literature
are currency, volatility, and timeliness. Table II compares the definitions provided in
the literature for these three time dimensions. Wand and Wang [1996] and Redman
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:9
[1996] provide very similar definitions for timeliness and currency. Wang and Strong
[1996] and Liu and Chi [2002] assume the same meaning for timeliness, while Bovee
et al. [2001] provides a definition for timeliness in terms of currency and volatility.
The definition of currency expressed in Bovee et al. [2001] corresponds to timeliness as
defined by Wang and Strong [1996] and Liu and Chi [2002]. This comparison shows that
there is no agreement on the abstract definition of time-related dimensions; typically,
currency and timeliness are often used to refer to the same concept.
2.4. Costs
Costs are a relevant perspective considered in methodologies, due to the effects of low
quality data on resource consuming activities. The cost of data quality is the sum of the
cost of data quality assessment and improvement activities, also referred to as the cost
of the data quality program and the cost associated with poor data quality. The cost of
poor quality can be reduced by implementing a more effective data quality program,
which is typically more expensive. Therefore, by increasing the cost of the data quality
program, the cost of poor data quality is reduced. This reduction can be seen as the
benefit of a data quality program.
The cost of a data quality program can be considered a preventive cost that is in-
curred by organizations to reduce data errors. This cost category includes the cost of
all phases and steps that compose a data quality assessment and improvement process
(see Section 2.1).
The costs of poor quality can be classified as follows [English 1999]:
(1) process costs, such as the costs associated with the re-execution of the whole process
due to data errors;
(2) opportunity costs due to lost and missed revenues.
The cost of poor data quality is strongly context-dependent as opposed to the cost of
a data quality program. This makes its evaluation particularly difficult, as the same
data value and corresponding level of quality has a different impact depending on the
recipient. For example, an active trader receiving obsolete information on a stock may
incur considerable economic losses as a consequence of wrong investment decisions.
In contrast, a newspaper receiving the same obsolete information to publish monthly
trading reports may not experience any economic loss.
2.5. Types of Data
The ultimate goal of a DQ methodology is the analysis of data that, in general, describe
real world objects in a format that can be stored, retrieved, and processed by a software
procedure, and communicated through a network. In the field of data quality, most
authors either implicitly or explicitly distinguish three types of data:
(1) Structured data, is aggregations or generalizations of items described by elementary
attributes defined within a domain. Domains represent the range of values that
can be assigned to attributes and usually correspond to elementary data types of
programming languages, such as numeric values or text strings. Relational tables
and statistical data represent the most common type of structured data.
(2) Unstructured data, is a generic sequence of symbols, typically coded in natural
language. Typical examples of unstructured data are a questionnaire containing
free text answering open questions or the body of an e-mail.
(3) Semistructured data, is data that have a structure which has some degree of flex-
ibility. Semistructured data are also referred to as schemaless or self-describing
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:10 C. Batini et al.
Fig. 2. Different representations of the same real-world object.
[Abiteboul et al. 2000; Buneman 1997; Calvanese et al. 1999]. XML is the markup
language commonly used to represent semistructured data. Some common charac-
teristics are: (1) data can contain fields not known at design time; for instance, an
XML file does not have an associated XML schema file; (2) the same kind of data
may be represented in multiple ways; for example, a date might be represented by
one field or by multiple fields, even within a single data set; and (3) among the fields
known at design time, many fields will not have values.
Data quality techniques become increasingly complex as data lose structure. For ex-
ample, let us consider a registry describing personal information such as Name,Surname,
Region, and StateOfBirth. Figure 2 shows the representation of Mr. Patrick Metzisi,
born in the Masai Mara region in Kenya, by using a structured (Figure 2(a)), unstruc-
tured (Figure 2(b)), and semistructured (Figure 2(c)) type of data. The same quality
dimension will have different metrics according to the type of data. For instance, syn-
tactic accuracy is measured as described in Section 2.3 in the case of structured data.
With semistructured data, the distance function should consider a global distance re-
lated to the shape of the XML tree in addition to the local distance of fields.
The large majority of research contributions in the data quality literature focuses
on structured and semistructured data. For this reason, although we acknowledge the
relevance of unstructured data, this article focuses on structured and semistructured
An orthogonal classification of data in the data quality literature is based on viewing
data as a manufacturing product [Shankaranarayan et al. 2000]. From this perspective,
three types of data are distinguished:
raw data items, defined as data that have not undergone any processing since their
creation and first storage—they can be stored for long periods of time;
information products, which are the result of a manufacturing activity performed on
component data items, which are generated every time the corresponding information
product is required and are stored temporarily until the final product is manufac-
As will be discussed in Section 3, this classification allows the application to data of
quality techniques traditionally used for quality assurance in manufacturing processes.
2.6. Types of Information Systems
DQ methodologies are influenced by the type of information system they refer to both
in assessment and in improvement activities. The literature provides the concept
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:11
of information system architecture (IS architecture) to describe the coordination
model supported by a company’s information system [Zachman 2006]. Different IS
architectures or, simply, types of information systems are distinguished on the basis
of the degree of data, process and management integration supported by a technical
system. As the degree of integration of data, process, and management decreases, the
data quality assessment and improvement techniques that can be applied become
more sophisticated. At the same time, data quality assessment and improvement is
more challenging. The following types of information systems can be distinguished
based on their degree of integration:
—In a monolithic information system, applications are single-tier and do not provide
data access services. Although data are usually stored in a database that can be
queried, separate applications do not share data. This can cause data duplication,
possibly affecting all quality dimensions.
—A data warehouse (DW) is a centralized collection of data retrieved from multiple
databases. Data warehouses are periodically refreshed with updated data from the
original databases by procedures automatically extracting and aligning data. Data
are physically integrated, since they are reformatted according to the data warehouse
schema, merged, and finally stored, in the data warehouse.
—A distributed information system is a collection of application modules coordinated by
a workflow. Applications are typically divided in tiers, such as presentation, applica-
tion logic, and data management, and export data access functionalities at different
tiers. Data can be stored in different databases, but interoperability is guaranteed
by the logical integration of their schemas.
—A cooperative information system (CIS) can be defined as a large-scale information
system that interconnects multiple systems of different and autonomous organiza-
tions sharing common objectives [De Michelis et al. 1997]. Cooperation with other
information systems requires the ability to exchange information. In CISs, data are
not logically integrated, since they are stored in separate databases according to dif-
ferent schemas. However, applications incorporate data transformation and exchange
procedures that allow interoperability and cooperation among common processes. In
other words, integration is realized at a process level.
—In the literature, the term Web Information System (WIS) [Isakowitz et al. 1998]
is used to indicate any type of information adopting Web technologies. From a
technical perspective a WIS is a client/server application. Such systems typically
use structured, semi structured, and unstructured data, and are supported by de-
velopment and management tools based on techniques specific to each type of
—In a peer-to-peer information system (P2P), there is no distinction between clients
and servers. The system is constituted by a set of identical nodes that share data
and application services in order to satisfy given user requirements collectively. P2P
systems are characterized by a number of properties: no central coordination, no
central database, no peer has a global view of the system, Peers are autonomous and
can dynamically connect or disconnect from the system. However, peers typically
share common management procedures.
This section compares methodologies based on the classification criteria discussed in
the previous section. Table III shows the list of methodologies considered in this paper
identified by acronyms together with the extended name of the methodology and the
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:12 C. Batini et al.
Table III. Methodologies Considered in the Article
Acronym Extended Name Main Reference
TDQM Total Data Quality Management Wang 1998
DWQ The Datawarehouse Quality Methodology Jeusfeld et al. 1998
TIQM Total Information Quality Management English 1999
AIMQ A methodology for information quality assessment Lee et al. 2002
CIHI Canadian Institute for Health Information methodology Long and Seko 2005
DQA Data Quality Assessment Pipino et al. 2002
IQM Information Quality Measurement Eppler and M ¨
ISTAT ISTAT methodology Falorsi et al 2003
AMEQ Activity-based Measuring and Evaluating of product
information Quality (AMEQ) methodology
Su and Jin 2004
COLDQ Loshin Methodology (Cost-effect Of Low Data Quality Loshin 2004
DaQuinCIS Data Quality in Cooperative Information Systems Scannapieco et al. 2004
QAFD Methodology for the Quality Assessment of Financial Data De Amicis and Batini 2004
CDQ Comprehensive methodology for Data Quality
Batini and Scannapieco
Tab le IV. Methodologies and Assessment Steps
Extensible to
Step/Meth Data DQ Requirement Identification of Process Measurement Other Dimensions
Acronym Analysis Analysis Critical Areas Modeling of Quality and Metrics
TDQM + + + + Fixed
DWQ + + + + Open
TIQM + + + + + Fixed
AIMQ + + + Fixed
CIHI + + Fixed
DQA + + + Open
IQM + + Open
ISTAT + + Fixed
AMEQ + + + + Open
COLDQ + + + + + Fixed
DaQuinCIS + + + + Open
QAFD + + + + Fixed
CDQ + + + + + Open
main reference. The acronym will be used to identify each methodology in the remainder
of this article.
Costs, dimensions, and phases represent the most discriminating criteria, leading to
the identification of four types of methodologies, which are discussed in Section 3.7,
and highlighting the fundamental differences among them. In the rest of the section,
methodologies are compared in depth along the perspectives identified in Section 2.
3.1. Methodologies, Phases, and Steps
Tables IV, V, and VI show the phases and steps addressed by each methodology. A
methodology has been considered to include a phase or a step if it provides at least a
discussion of the corresponding phase or step, and possibly, methodological guidelines
and original techniques. For example, DWQ generically refers to a process modeling
step, but does not provide execution details. For the same reason, the measurement of
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:13
Tab le V. Methodologies and Improvement Steps-Part 1
Assignment Assignment Selection Identification
Step/Meth. Evaluation of Process of Data Strategies the Causes
Acronym of Costs Responsibilities Responsibilities and Techniques of Errors
TDQM + + + + +
DWQ + + + +
TIQM + + + + +
COLDQ + + +
DaQuinCIS + +
CDQ + + + + +
Table VI. Methodologies and Improvement Steps-Part 2
Step/Meth. Process Design of data Process Improvement Improvement
Acronym Control Improvement Solutions Redesign Management Monitoring
TDQM + + +
DWQ + +
TIQM + + +
COLDQ + + + +
CDQ + + +
quality step is not associated with CIHI in Table IV. Assessment and improvement are
discussed separately in the next two sections.
3.1.1. The Assessment Phase. Table IV compares the steps followed by different
methodologies in the assessment phase. In general, methodologies refer to the steps
classified in Section 2.1, although with different names. However, it is not difficult to
recognize a name correspondence by analyzing the objectives of the step. For example,
CIHI discusses a methodological step that identifies the databases with a quality level
below a given acceptability threshold. This step has a clear correspondence with the
find critical areas step of Section 2.1.
The most commonly addressed steps of the assessment phase are data analysis and
measurement of quality. However, they are performed according to different approaches.
For example, the measurement of quality step is performed with questionnaires in
AIMQ, with a combination of subjective and objective metrics in DQA, or with statistical
analyses in QAFD. Different measurement approaches meet the specific requirements
of different organizational contexts, processes, users or services. Only a few methodolo-
gies consider the DQ requirements analysis step, identifying DQ issues and collecting
new target quality levels from users. This step is particularly relevant for evaluat-
ing and solving conflicts in target DQ levels from different stakeholders. For example,
QAFD recommends the collection of target quality levels from different types of experts,
including business experts and financial operators, but does not help the reconciliation
of incompatible DQ levels. A few methodologies support process modeling. Note that
with the exception of AMEQ, the methodologies supporting process modeling also adopt
a process-driven strategy for the improvement phase (see next section).
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:14 C. Batini et al.
Table VII. Methodologies and Types of Strategies
Strategy/Meth. Acronym Data-driven Process-driven
TDQM Process Redesign
DWQ Data and schema integration
TIQM Data cleansing
Error localization and correction
Process Redesign
ISTAT Normalization
Record linkage
Process Redesign
COLDQ Cost optimization Process Control
Process Redesign
DaQuinCIS Source trustworthiness
Record Linkage
CDQ Normalization
Record Linkage
Data and schema integration
Error localization and correction
Process Control
Process Redesign
The last column of Table IV specifies whether the methodology allows extensibility
to dimensions (and metrics) other than those explicitly dealt with in the methodology.
For example, CDQ explicitly mentions dimensions among those that will be described
in Section 3.3, but the approach can be easily generalized to other dimensions. On the
contrary, ISTAT provides detailed measurement and improvement procedures for ac-
curacy, completeness, and consistency, and consequently, the whole approach is strictly
hardwired to such dimensions.
Note that the methodologies that address both the process modeling and measurement
of quality steps, are based on the fitness for use approach. They evaluate the quality of
data along the processes in which they are used and, thus mainly provide subjective
3.1.2. The Improvement Phase. Tables V and VI compare the improvement steps of
different methodologies.
The identification of the causes of errors is the most widely addressed improvement
step. DQA emphasizes the importance of the identification of the causes of errors step,
but it does not discuss its execution. Similarly, DWQ refers to a mathematical model
based on the concept of dependency to support the identification of the causes of er-
rors step, but the definition of the model is presented as ongoing work and is not
Only six methodologies address multiple improvement steps, as confirmed by
Table VII. Improvement activities are mostly based on process redesign, with the ex-
ception of the DWQ methodology, which provides an extension of the Goal Question
Metric [Basili et al. 1994] initially proposed in the software engineering field. The cost
evaluation step is usually mandatory in DQ methodologies. This step is considered crit-
ical for measuring the economic advantage of improvement solutions and to choose the
most efficient improvement techniques. In contrast, the management of the improve-
ment solution step is explicitly performed only by TDQM. Other methodologies refer
to the broad range of management techniques and best practices available from the
change management field [Kettinger and Grover 1995]. Furthermore, it is possible to
repeat the assessment phase of the methodology in order to evaluate the results of
the improvement phase. As an example, DQA explicitly recommends the application of
previous methodological steps to evaluate the effectiveness of improvement.
Finally, the relationship among data quality, process, and organization is considered
by TIQM, TDQM, and CDQ. These methodologies thoroughly discuss the assignment
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:15
of responsibilities on processes and data. These steps are supported by the results of
the state reconstruction phase. CDQ discusses a set of matrices to represent the re-
lationship among processes, organizational units, and databases, which are produced
during the state reconstruction phase and are subsequently used in the assignment of
responsibilities steps.
3.2. Methodologies, Strategies, and Techniques
Table VII shows the strategies and techniques adopted by different methodologies. A
methodology is associated with a strategy if it provides guidelines to select and design
corresponding techniques.
Notice that the column labelled Process-driven in Table VII provides the same in-
formation as columns, Process control and Process redesign of Table VI. The column
labelled Data-driven explicitly mentions the data-driven techniques implicitly consid-
ered in Tables V and VI.
Table VII shows that five DQ methodologies adopt mixed strategies, variously com-
bining data-driven and process-driven techniques. The methodology applying the wider
range of data- and process-driven techniques is TIQM. Conversely, TDQM provides
guidelines to apply process-driven strategies by using the Information Manufacturing
Analysis Matrix [Ballou et al. 1998], which suggests when and how to improve data.
It is worth noting that a methodology exclusively adopting either a data- (as for DWQ
and DaQuinCIS) or a process-driven strategy, may not be flexible for organizations
that have DQ practices. The only methodology that explicitely addresses this issue is
CDQ, which jointly selects data- and process-driven techniques. The selection of the
most suitable strategy and technique is based on domain-dependent decision variables
[Batini et al. 2008].
Normalization, record linkage, data and schema integration, represent the data-
driven techniques most widely adopted in DQ methodologies, while process redesign,
as discussed in previous section, is most relevant in process-driven methodologies. We
now discuss specific contributions related to the data- and process-driven techniques
considered in Section 2.2.
3.2.1. Data-Driven Techniques. Normalization techniques have been proposed in sev-
eral domains, including census and territorial data domains. Both ISTAT and CDQ
provide normalization techniques improving DQ by comparing data with look-up ta-
bles and defining a common metaschema. For example, the ISTAT methodology uses
the national street registry as a lookup table for territorial data.
Record linkage has been investigated in the database research since the ’50s and has
been applied in many contexts such as healthcare, administrative, and census appli-
cations. In such contexts, it is crucial to produce efficient computer-assisted matching
procedures that can reduce the use of clerical resources, and at the same time, minimize
matching errors. CDQ discusses three types of record linkage techniques:
(1) Probabilistic techniques, based on the broad set of methods developed over the past
two centuries within statistics and probability theory, ranging from Bayesian net-
works to data mining.
(2) Empirical techniques that make use of algorithmic techniques such as sorting, tree
analysis, neighbor comparison, and pruning.
(3) Knowledge-based techniques, extracting knowledge from files and applying reason-
ing strategies.
Criteria for choosing among these three types of techniques are discussed within
the CDQ methodology. The DaQuinCIS project has developed a specific record linkage
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:16 C. Batini et al.
technique [Bertolazzi et al. 2003]. In the DaQuinCIS platform, record linkage is per-
formed in two phases: (1) first, record linkage aligns different copies of the same entities
in different data sources; (2) second, record linkage also supports the query process-
ing phase by identifying the same instances in the query results returned by each
data source. The record linkage method is based on the Sorted Neighborhood method
[Hernandez and Stolfo 1998], but some new features are introduced:
—the matching key is automatically selected instead of being selected by the key
—the matching algorithm is based on a function that normalizes a classic edit distance
function upon string lengths.
Data and schema integration [Lenzerini 2002] is a broad area of research that par-
tially overlaps with data quality. Data-driven improvement techniques applied in the
methodologies are often based on the use of new data to improve the quality of a given
data collection. As a consequence, DQ improvement techniques focus primarily on
instance-level heterogeneities, in order to identify similar records, detect conflicting
values, and select a single final instance.
Methodologies that address instance-level heterogeneities are DWQ, ISTAT, DaQuin-
CIS, and CDQ. In DWQ, heterogeneous information sources are first made accessible in
a uniform way through extraction mechanisms called wrappers, then mediators take on
the task of information integration and conflict resolution. The resulting standardized
and integrated data are stored as materialized views in the data warehouse.
ISTAT suggests how to resolve heterogeneities among data managed by different
public agencies by adopting a common model for representing the format of exchanged
data, based on the XML markup language. In this way, the comprehension of hetero-
geneities among agencies is made easier, while the solution of such heterogeneities is
left to bilateral or multilateral agreements.
In DaQuinCIS, instance-level heterogeneities among different data sources are dealt
with by the DQ broker. Different copies of the same data received as responses to the
request are reconciled by the DQ broker, and a best-quality value is selected.
CDQ follows an approach similar to ISTAT, with more emphasis on the autonomy of
organizations in the cooperative system. In fact, the resolution of heterogeneities in the
case studies, proposed as best practices, is performed through record linkage on a very
thin layer of data, namely the identifiers. All other data are reconcilied only in case of
autonomous decisions of the agencies involved.
3.2.2. Process-Driven Techniques. Methodologies addressing the process redesign step
tend to borrow corresponding techniques from the literature on business process reengi-
neering (BPR) [Muthu et al. 1999; Hammer 1990]. TDQM represents an exception in
this respect, as it proposes an original process redesign control approach that is referred
to as an “information manufacturing system for the Information Product” [Ballou et al.
1998]. This methodology proposes the Information Production Map (IP-MAP) model
[Shankaranarayan et al. 2000] that is used to model the information products managed
by the manufacturing processes. An information production map is a graphical model
designed to help analysts to visualize the information production process, identify the
ownership of process phases, understand information and organizational boundaries,
and estimate the time and quality metrics associated with the current production pro-
cess. The description of processes is a mandatory activity, consistent with the general
orientation of process-driven strategies. After modelling and assessing the information
production process, new process control activities are identified and/or process redesign
decisions are taken.
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:17
Complex solutions such as IP-MAP cannot always be adopted due to their high costs
and, in some cases, the practical unfeasibility of a thorough process modeling step. For
this reason, other methodologies adopt less formal, but more feasible solutions. For
example, CDQ is based on a set of matrices that describe the main relationships among
data, information flows, processes, and organizational units. The relationship between
organizational units and processes has also been modeled in extensions of IP-MAP
proposed in the literature [Scannapieco et al. 2002].
3.3. Methodologies and Dimensions
Table VIII shows the quality dimensions considered by the methodologies surveyed
in this article. In Table VIII, a dimension is associated with a methodology, if the
methodology provides a corresponding definition. For each methodology’s dimensions,
we address the corresponding references (see Table III).
Notice the large variety of dimensions defined in the methodologies, which confirms
the complexity of the data quality concept. This is not surprising, since nowadays a
large number of phenomena can be described in terms of data. Multiple classifications
of quality dimensions are proposed by the methodologies. TIQM classifies dimensions
as inherent and pragmatic. COLDQ distinguishes among schema,data,presentation,
and information policy dimensions. CIHI provides a two-level classification in terms of
dimensions and related characteristics. CDQ proposes schema and data dimensions.
Table IX shows the metrics provided for quality dimensions by different methodolo-
gies. We do not include metrics for semantic accuracy because the two methodologies
addressing it, namely QAFD and CDQ, do not provide specific measurement methods.
In general, multiple metrics are defined for each dimension, and each dimension ac-
cordingly has multiple entries in the table. Note that subjective metrics such as user
surveys have been defined for almost all quality dimensions. Different metrics for the
same dimension are identified by acronyms, which are used in Table X to associate
them with the methodologies in which they are used and/or defined.
The last column of Table X provides for each dimension and each metric associated
with the dimension, (1) the number of methodologies that use the metrics, and (2) the
total number of methodologies that mention the corresponding dimension. The ratio
between these values measures the degree of consensus on dimension metrics among
methodologies. Such consensus is high for accuracy, completeness, and consistency,
while it is significantly lower for two of the time-related dimensions, timeliness and
currency, and almost all other dimensions.
The majority of metrics with only one occurrence in methodologies are mentioned
in IQM, which analyzes the quality of Web information. Such metrics are defined by
considering the measurement tools that are available in the specific Web context. For
example, using a site analyzer, it is possible to assess dimensions such as accessibil-
ity, consistency, timeliness, conciseness, and maintainability. Traffic analyzers can be
used to assess applicability and convenience, while port scanners are useful to assess
security. The high number of measurement tools in the Web context results in a high
number of metrics specific to IQM.
AIMQ has several specific metrics and dimensions. This is due to the top-down ap-
proach adopted in AIMQ in the definition of dimensions and metrics, which uses two
different classifications (not represented in Table VIII): (1) product vs. service quality,
and (2) conforms to specification vs. meets or exceeds customer expectations, leading
to a widely scattered set of related dimensions/metrics.
Note that the AIMQ methodology uses only subjective metrics to assess quality di-
mensions. In AIMQ, data quality is mainly assessed by means of questionnaires that
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:18 C. Batini et al.
Table VIII. Methodologies and Quality Dimensions
Acronym Data Quality Dimension
TDQM Accessibility, Appropriateness, Believability, Completeness, Concise/Consistent
representation, Ease of manipulation, Value added, Free of error, Interpretability,
Objectivity, Relevance, Reputation, Security, Timeliness, Understandability
DWQ Correctness, Completeness, Minimality, Traceability, Interpretability, Metadata
Evolution, Accessibility (System, Transactional, Security), Usefulness
(Interpretability), Timeliness (Currency, Volatility), Responsiveness, Completeness,
Credibility, Accuracy, Consistency, Interpretability
TIQM Inherent dimensions: Definition conformance (consistency), Completeness, Business
rules conformance, Accuracy (to surrogate source), Accuracy (to reality), Precision,
Nonduplication, Equivalence of redundant data, Concurrency of redundant data,
Pragmatic dimensions: accessibility, timeliness, contextual clarity, Derivation
integrity, Usability, Rightness (fact completeness), cost.
AIMQ Accessibility, Appropriateness, Believability, Completeness, Concise/Consistent
representation, Ease of operation, Freedom from errors, Interpretability, Objectivity,
Relevancy, Reputation, Security, Timeliness, Understandability
CIHI Dimensions: Accuracy, Timeliness Comparability, Usability, Relevance
Characteristics: Over-coverage, Under-coverage, Simple/correlated response
variance, Reliability, Collection and capture, Unit/Item non response, Edit and
imputation, Processing, Estimation, Timeliness, Comprehensiveness, Integration,
Standardization, Equivalence, Linkage ability, Product/Historical comparability,
Accessibility, Documentation, Interpretability, Adaptability, Value.
DQA Accessibility, Appropriate amount of data, Believability, Completeness, Freedom
from errors, Consistency, Concise Representation, Relevance, Ease of manipulation,
Interpretability, Objectivity, Reputation, Security, Timeliness, Understandability,
Value added.
IQM Accessibility, Consistency, Timeliness, Conciseness, Maintainability, Currency,
Applicability, Convenience, Speed, Comprehensiveness, Clarity, Accuracy,
Traceability, Security, Correctness, Interactivity.
ISTAT Accuracy, Completeness, Consistency
AMEQ Consistent representation, Interpretability, Case of understanding, Concise
representation, Timeliness, Completeness Value added, Relevance, Appropriateness,
Meaningfulness, Lack of confusion, Arrangement, Readable, Reasonability,
Precision, Reliability, Freedom from bias, Data Deficiency, Design Deficiency,
Operation, Deficiencies, Accuracy, Cost, Objectivity, Believability, Reputation,
Accessibility, Correctness, Unambiguity, Consistency
COLDQ Schema: Clarity of definition, Comprehensiveness, Flexibility, Robustness,
Essentialness, Attribute granularity, Precision of domains, Homogeneity,
Identifiability, Obtainability, Relevance, Simplicity/Complexity, Semantic
consistency, Syntactic consistency.
Data: Accuracy, Null Values, Completeness, Consistency, Currency, Timeliness,
Agreement of Usage, Stewardship, Ubiquity, Presentation: Appropriateness, Correct
Interpretation, Flexibility, Format precision, Portability, Consistency, Use of storage,
Information policy: Accessibility, Metadata, Privacy, Security, Redundancy, Cost.
DaQuinCIS Accuracy, Completeness, Consistency, Currency, Trustworthiness
QAFD Syntactic/Semantic accuracy, Internal/External consistency, Completeness,
Currency, Uniqueness.
CDQ Schema: Correctness with respect to the model, Correctness with respect to
Requirements, Completeness, Pertinence, Readability, Normalization, Data:
Syntactic/Semantic Accuracy, Semantic Accuracy, Completeness, Consistency,
Currency, Timeliness, Volatility, Completability, Reputation, Accessibility, Cost.
include 46 independent items for each quality dimension. Items have the following
general structure: “This information is (attribute or phrase).” For example, complete-
ness is associated with six items, including: (1) this information provides all necessary
values, (2) this information is sufficiently complete for our needs, (3) this information
fulfils the needs of our tasks.
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:19
Table IX. Dimensions and Metrics
Dimensions Name Metrics Definition
Accuracy Acc1 Syntactic accuracy: it is measured as the distance between the value stored in the
database and the correct
Syntactic Accuracy=Number of correct values/number of total values
Acc2 Number of delivered accurate tuples
Acc3 User Survey - Questionnaire
Completeness Compl1 Completeness =Number of not null values/total number of values
Compl2 Completeness =Number of tuples delivered/Expected number
Compl3 Completeness of Web data =(Tmax-T
current)(CompletenessMax -
Compl4 User Survey - Questionnaire
Consistency Cons1 Consistency =Number of consistent values/number of total values
Cons2 Number of tuples violating constraints, number of coding differences
Cons3 Number of pages with style guide deviation
Cons4 User Survey - Questionnaire
Timeliness Time1 Timeliness = (max (0; 1-Currency/Volatility))s
Time2 Percentage of process executions able to be performed within the required time
Time3 User Survey - Questionnaire
Currency Curr1 Currency =Time in which data are stored in the system - time in which data are
updated in the real world
Curr2 Time of last update
Curr3 Currency =Request time- last update
Curr4 Currency =Age +(Delivery time- Input time)
Curr5 User Survey - Questionnaire
Volatility Vol1 Time length for which data remain valid
Uniqueness Uni1 Number of duplicates
Appropriate amount of
Appr1 Appropriate Amount of data = Min ((Number of data units provided/Number of
data units needed); (Number of data units needed/Number of data units
Appr2 User Survey - Questionnaire
Accessibility Access1 Accessibility =max (0; 1-(Delivery time - Request time)/(Deadline time - Request
Access2 Number of broken links - Number of broken anchors
Access3 User Survey - Questionnaire
Credibility Cred1 Number of tuples with default values
Cred2 User Survey - Questionnaire
Interpretability Inter1 Number of tuples with interpretable data, documentation for key values
Inter2 User Survey - Questionnaire
Usability Usa1 User Survey - Questionnaire
Derivation Integr1 Percentage of correct calculations of derived data according to the
Integrity derivation formula or calculation definition
Conciseness Conc1 Number of deep (highly hierarchic) pages
Conc2 User Survey - Questionnaire
Maintainability Main1 Number of pages with missing meta-information
Applicability App1 Number of orphaned pages
App2 User Survey - Questionnaire
Convenience Conv1 Difficult navigation paths: number of lost/interrupted navigation trails
Speed Speed1 Server and network response time
Comprehensiveness Comp1 User Survey - Questionnaire
Clarity Clar1 User Survey - Questionnaire
Traceability Trac1 Number of pages without author or source
Security Sec1 Number of weak log-ins
Sec2 User Survey - Questionnaire
Correctness Corr1 User Survey - Questionnaire
Objectivity Obj1 User Survey - Questionnaire
Relevancy Rel1 User Survey - Questionnaire
Reputation Rep1 User Survey - Questionnaire
Ease of operation Ease1 User Survey - Questionnaire
Interactivity Interact1 Number of forms - Number of personalizable pages
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:20 C. Batini et al.
Tab le X. Methodologies and Quality Metrics
Acc1 X X X X X X X X X 9/13
Acc2 X 1/13
Acc3 X X X 2/13
Compl1 X X X X X X x x 7/12
Compl2 X X 2/12
Compl3 X 1/12
Compl4 X 2/12
Cons1 X X 6/10
Cons2 X 1/10
Cons3 X 1/10
Cons4 X X 2/10
Time1 X X X 3/7
Time2 X X 2/7
Time3 X X 2/7
Curr1 X X 2/8
Curr2 X X X 2/8
Curr3 X X 1/8
Curr4 X 1/8
Curr5 X X 2/8
Vol1 X X 2/2
Uni1 X X 1/2
Appr1 X 1/2
Appr2 X 1/2
Access1 X 1/4
Access2 X 1/4
Access3 X X 2/4
Cred1 X 1/2
Cred2 X 1/2
Inter1 X 1/2
Inter2 X 1/2
Usa1 X 1/1
Integr1 X 1/1
Conc1 1/2
Conc2 X 1/ 2
Main1 X 1/1
App1 1/1
App2 X 1/1
Conv1 X 1/1
Speed1 X 1/1
Comp1 X X X 3/3
Clar1 X X X 3/3
Trac1 X 1/1
Sec1 X 1/1
Sec2 X 1/1
Corr1 X 1/1
Obj1 X 1/1
Rel1 X 1/1
Rep1 X 1/1
Ease1 X 1/1
Interact1 X 1/1
Finally, the metrics provided by the DWQ methodology have a tight relation with the
processes using data. All dimensions are evaluated along the expectations of the users
of a particular process. For example, completeness is defined as the ratio of not null
values to the number of values required by a specific process, as opposed to the total
number of values stored in a database (see Section 2.3).
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:21
Table XI. Comparison Between English and Loshin Classifications
Cost Categories TIQM COLDQ
Cost of assessment Assessment or inspection costs: Detection costs
DQ activities Information quality analysis software costs
People time in the assessment process
Cost of Process improvement and defect prevention cost Data correction costs
improvement DQ Data maintenance costs:
activities Acquisition overhead costs
Decay costs
Infrastructure costs
Process improvement costs
Process costs of Process failure costs: Operational impacts:
poor data quality Unrecoverable costs Rollback costs
Liability and exposure costs Rework costs
Recovery costs of unhappy customers Prevention costs
Information scrap and rework: Warranty costs
Redundant data handling and support costs Tactical and strategic impacts:
Costs of hunting or chasing missing information Delay costs
Business rework costs Preemption costs
Workaround costs and decreased productivity Idling costs
Data verification costs Increased difficulty costs
Software rewrite costs Lost difficulty costs
Data cleaning and correction costs Organizational mistrust costs
Data cleaning software costs Misalignment costs
Opportunity costs Lost and missed opportunity costs: Lost revenue costs:
of poor data Lost opportunity costs Spin costs
quality Missed opportunity costs Reduction costs
Lost shareholder value Attrition costs
Blockading costs
3.4. Methodologies and Costs
The cost dimension is considered only in TIQM, COLDQ, and CDQ. In this section we
analyze costs from two different points of view: (1) cost classifications, and (2) criteria
provided for cost quantification.
3.4.1. Cost Classifications. Both TIQM [English 1999] and COLDQ [Loshin 2004] pro-
vide detailed classifications for costs. A third classification is provided by Eppler and
Helfert [2004].
Table XI compares the TIQM and COLDQ classifications. In TIQM, data quality costs
correspond to the costs of business processes and data management processes due to
poor data quality. Costs for information quality assessment or inspection measure data
quality dimensions to verify that processes are performing properly. Finally, process
improvement and defect prevention costs involve activities to improve the quality of
data, with the goal of eliminating, or reducing, the costs of poor data quality. Costs due
to poor data quality are analyzed in depth in the TIQM approach, and are subdivided
into three categories:
(1) Process failure costs are incurred when poor quality data causes a process not to per-
form properly. As an example, inaccurate mailing addresses cause correspondence
to be misdelivered.
(2) Information scrap and rework. When data is of poor quality, they involve several
types of defect management activities, such as reworking, cleaning, or rejecting.
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:22 C. Batini et al.
(3) Loss and missed opportunity costs correspond to the revenues and profits lost be-
cause of poor data quality. For example, due to low accuracy of customer e-mail
addresses, a percentage of customers already acquired cannot be reached by peri-
odic advertising campaigns, resulting in lower revenues, roughly proportional to
the decrease of the accuracy of addresses.
COLDQ analyzes the costs of low data quality, classifying them according to their
domain impact, namely:
—the operational domain, which includes the components of the information processing
system and the operating costs of the system;
—the tactical domain, which attempts to address and solve problems before they arise;
—the strategic domain, which stresses long-term decisions.
For both the operational and tactical/strategic impact several cost categories are
introduced. Here, we describe the operational impact costs:
rollback costs are incurred when work that has been performed needs to be undone;
rework costs are incurred when a processing stage must be repeated;
prevention costs arise when a new activity is implemented to take the actions neces-
sary to prevent operational failure due to a data quality problem;
warranty costs are related to guarantees against losses.
Finally, we mention that CDQ proposes a classification that reconciles the hetero-
geneities among TIQM, COLDQ, and Eppler and Helfert [2004]. For details, see Batini
and Scannapieco [2006].
3.4.2. Criteria for Cost Quantification. The assessment of the total cost of data qual-
ity supports the selection of the types of data quality activities to be performed (see
Section 2.2) and their prioritization. TIQM, COLDQ, and CDQ are the only method-
ologies providing criteria for this activity. In TIQM, selection and prioritization are
achieved with the following steps:
—identify current users and uses of data;
—list the errors that negatively affect data quality;
—identify the business units most often impacted by poor quality data;
—estimate the direct cost of the current data quality program;
—estimate the costs of data errors for all users and uses of data, grouped by business
—use costs to justify and prioritize data quality initiatives, including the institution-
alization of a continuous quality improvement program.
Each type of error occurs with a given frequency and involves a cost. Note that the
cost of different error categories is a contingent value that varies with the process that
makes use of the data. Models for process representation allow the identification of the
activities affected by data errors. Since activities are typically associated with a total
organizational cost, the cost of rework can provide quantitative and objective estimates.
COLDQ focuses on the evaluation of the cost associated with poor data quality, as an
argument for supporting the investment in a knowledge management program. The
evaluation is achieved with the following steps:
—map the information chain to understand how information flows within the organi-
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:23
Table XII. Methodologies and Types of Data
Type of data/Meth. acronym Structured Semistructured
TDQM x x
TIQM x Implicitly considered
AIMQ x Implicitly considered
CIHI x x
IQM x x
AMEQ x Implicitly considered
COLDQ x Implicitly considered
DaQuinCIS x x
CDQ x x
—identify current users and uses of data;
—identify the impact of the quality of data on customers;
—isolate flawed data by locating critical areas;
—identify the impact domain associated with each instance of poor data quality;
—characterize the economic impact based on the ultimate effects of bad data;
—aggregate the totals to evaluate the overall economic impact;
—identify opportunities for improvement.
The result is called the data quality scorecard. It summarizes the cost associated with
poor data quality and can be used as a tool to find the best solutions for improvement.
In CDQ the minimization of the cost of the data quality program is the main criterion
for choosing among alternative improvement processes. First, different improvement
processes are identified as paths of data- and process-driven techniques applied to the
data bases, data flows, and document bases involved in the improvement. Then, the
costs of the different processes are evaluated and compared, and the minimum-cost
process is selected.
3.5. Methodologies and Types of Data
We observed in Section 2.5 that the types of data influence the DQ dimensions and
the assessment and improvement techniques. Table XII associates the types of data
classified in Section 2.5 and DQ methodologies. Most methodologies address structured
data, while only a few also address semistructured data. In Table XII we have imputed
implicitely considered when the methodology does not explicitely mention the type of
data, but phases and steps can be applied to it. For example, AIMQ uses the generic
term information, and performs qualitative evaluation through questions that apply to
structured data, but may refer to any type of data, including unstructured data.
Concerning semistructured data, ISTAT considers the standardization of address
data formats and their expression in a common XML schema. This schema is imple-
mented to minimize discrepancies across agencies and allow interoperability. In the
DaQuinCIS methodology [Scannapieco et al. 2004], a model associating quality values
with XML documents is proposed. The model, called Data and Data Quality (D2Q),
is intended to be used in the context of data flows exchanged by different organiza-
tions in a cooperative information system. In the exchange of data, the quality of data
flows becomes critical to avoid error propagation. D2Qcan be used in order to certify
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:24 C. Batini et al.
Table XIII. Methodologies and Information Systems
Type of inf.syst.
/Meth. acronym Monolithic Distributed DataWarehouse Cooperative Web
TDQM focused implicitly considered
DWQ strongly focused
TIQM focused focused
AIMQ focused implicitly considered
CIHI focused focused
DQA focused implicitly considered
IQM strongly focused
ISTAT focused focused strongly focused
AMEQ focused
COLDQ focused
DaQuinCIS focused focused strongly focused
QAFD focused
CDQ focused focused strongly focused
the accuracy, completeness, currency, and consistency of data. The model is semistruc-
tured, thus allowing each organization to export the quality of its data with a certain
degree of flexibility. The quality values can be associated with various elements of the
data model, ranging from individual data values to the whole data source.
CDQ attempts an extension to semistructured data of steps and techniques origi-
nally developed for structured data. For example, all data types, both structured and
semistructured, are surveyed in the state reconstruction phase. A specific data-driven
improvement technique is also proposed for unstructured data, called data profiling.
This technique is used to relate a text file to a database schema by discovering recurring
patterns inside text [Aiken 1996].
3.6. Methodologies and Types of Information Systems
Table XIII shows to what extent the different methodologies deal with the types of
information systems introduced in Section 2.6. The table adopts a four-value scale,
where: (1) strongly focused means that the whole organization of the methodology is
conceived and tailored to the corresponding type of information system, while providing
generic guidelines for other types of information systems, (2) focused means that the
methodology provides detailed guidelines and techniques for the corresponding type of
information system, (3) implicitly considered has the same meaning as in Table XII,
and (4) a missing value indicates that the methodology either provides generic guide-
lines, or does not explicitly address the corresponding type of information system. Web
information systems are included in the table, since one methodology is strongly fo-
cused on them; further issues related to such systems will be discussed in Section 4.
No methodology mentions P2P systems. They will be considered in the conclusions and
open issues (Section 4).
It can be observed that AMEQ, COLDQ, and QAFD focus on monolithic information
systems. They typically consider structured data sets within a single system, and ignore
DQ issues raised by data exchanges among separate applications or organizations.
TIQM can be applied to both monolithic and distributed systems, because when the
data architecture is analyzed, both a centralized and a distributed system are pro-
vided as case studies. CIHI is considered focused on distributed systems since national
databases are mentioned in the description of the methodology.
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:25
Fig. 3. General view of the Istat methodology.
Other methodologies, ISTAT, DaQuinCIS, and CDQ, can be applied to multiple types
of systems. Each methodology provides detailed guidelines for the most complex type
of system, the cooperative information system, and can therefore also be applied to
monolithic and distributed systems. The methodologies follow different approaches for
dealing with multiple data flows and heterogeneous databases.
ISTAT has a strong interorganizational approach, as it has been conceived for the
Italian public administration, which is characterized by a distributed structure with
highly autonomous administrations (this is typical of many other countries). The three
main phases of the methodology are shown in Figure 3.
The assessment phase (Phase 1) identifies the most relevant activities to be performed
in the improvement phase (Phases 2 and 3):
(1) Phase 2 acts on local databases owned and managed by different administrations.
Tools are distributed to perform DQ activities autonomously, and courses are offered
to improve local DQ skills.
(2) Phase 3 concerns the overall cooperative information system of a set of administra-
tions, in terms of exchanged data flows and central databases set up for coordination
purposes. These activities are centrally planned and coordinated.
DaQuinCIS provides a framework that offers several services and tools for coop-
erative information systems (see Figure 4). The data quality broker, described in
Section 3.2, is the core of the architecture, and is responsible for the reconciliation of het-
erogeneous responses to queries. The quality notification service is a publish/subscribe
engine used as a general message bus between the architectural components of different
cooperating organizations [Scannapieco et al. 2004]. It allows quality-based subscrip-
tions for organizations to be notified on changes of the quality of data. The best-quality
value selected by the broker, mentioned in Section 3.2, is proposed to requesting or-
ganizations, which can choose to replace their data with higher quality data (quality
improvement function). The quality factory component (see Figure 4) is responsible for
evaluating the quality of the internal data of each organization (the interested reader
can refer to Cappiello et al. [2003b]). Requests from external users or information sys-
tems are processed in the quality factory by the quality analyzer, which performs a
static analysis of the values of the data quality dimensions associated with requested
data, and compares them with benchmark quality parameters contained in the quality
repository. The rating service will be addressed in Section 4.
Other contributions in the literature that provide tools for quality-based query pro-
cessing and instance-level conflict resolution for cooperative information systems are
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:26 C. Batini et al.
Fig. 4. The architecture of the DaQuinCIS framework.
Fusionplex [Motro and Anokhin 2005], iFuice [Rahm et al. 2005], and HumMer [Bilke
et al. 2005].
We now comment on the methodologies that are strongly focused on specific types
of information systems, namely DWQ and IQM. DWQ is specifically oriented to data
warehouses. In Jarke et al. [1995], it is observed that most researches have studied
DW in their role as buffers of materialized views, mediating between update-intensive
OLTP systems and query-intensive decision support. This neglects the organizational
role of data warehousing as a means of obtaining a centralized control of information
flows. As a consequence, a large number of quality issues relevant for DW cannot be
expressed with traditional DW meta models. DWQ makes two contributions towards
solving these problems. First, the metadata about DW architectures are enriched with
explicit enterprise models. Second, mathematical techniques for measuring or optimiz-
ing several aspects of DW quality are proposed.
The Artkos tool also contributes to address the DQ issues of data warehouses
[Vassiliadis et al. 2001]. Artkos proposes a metamodel made of several entities, among
them: (1) activities, atomic units of data processing work; (2) scenarios, sets of activities
to be executed together; and (3) quality factors, defined for each activity, corresponding
to the dimensions and metrics described in Section 2.3.
IQM is strongly focused on Web information systems, as it considers a wide set of
existing tools to evaluate information quality in the Web context, namely site analyzers,
traffic analyzers, port scanners, performance monitoring systems, Web mining tools
and survey tools to generate opinion-based user feedback. Several information quality
criteria (in our terminology, dimensions and metrics, as seen in Section 3.3) can be
measured with the help of these tools. IQM provides systematic sequential steps to
match information quality criteria with measurement tools.
3.7. Summary Comparison of Methodologies
The detailed comparison of methodologies discussed in the previous sections clearly in-
dicates that methodologies tend to focus on a subset of DQ issues. The broad differences
in focus across methodologies can be recognized at a glance by classifying methodologies
into four categories, as shown in Figure 5:
complete methodologies, which provide support to both the assessment and improve-
ment phases, and address both technical and economic issues;
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:27
Fig. 5. A classification of methodologies.
audit methodologies, which focus on the assessment phase and provide limited sup-
port to the improvement phase;
operational methodologies, which focus on the technical issues of both the assessment
and improvement phases, but do not address economic issues.
economic methodologies, which focus on the evaluation of costs.
From a historical perspective, there exists a correlation between quality dimensions
and the evolution of ICT technologies. First-generation information systems (in the
’50s and ’60s of the past century) were monolithic, as their technological architecture
consisted of a single mainframe and a single database. Information flows were simple
and repetitive and most errors were caused by incorrect data entry. The main large-scale
applications were census management and medical data analysis, and data quality
focused on accuracy, consistency, completeness, and time-related dimensions, consistent
with our definitions in Section 2.3 and also Catarci and Scannapieco [2002]. The most
critical issues with data quality management were error localization and correction
in data sources, and record linkage between new data sources and pre-existing data
The evolution of information systems from monolithic to network-based has caused a
growth of the number of data sources in both size and scope and, consequently has signif-
icantly increased the complexity of data quality management. DQ methodologies have
started to focus on new quality dimensions, such as the completeness of the data source,
the currency of data, and the consistency of the new data sources compared to the enter-
prise database. With the advent of the Web, data sources have become difficult to assess
and control over time. At the same time, searching and navigating through the Web
is potentially unlimited. As a consequence of this fast evolution, methodologies have
started to address new quality dimensions, such as accessibility and reputation.Acces-
sibility measures the ability of users to access data, given their culture, physical status
and available technologies, and is important in cooperative and network-based informa-
tion systems. Reputation (or trustworthiness) is a property of data sources measuring
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:28 C. Batini et al.
their ability to provide correct information and is particulary relevant in Web-based
and peer-to-peer systems, as further discussed in Section 4.
This evolution of ICT and the consequent growing complexity of DQ is a fundamen-
tal reason why methodologies have specialized on a subset of DQ issues, as shown in
Figure 5. The majority of methodologies belong to the audit category and focus on the
technical issues of the assessment phase. To some extent, this is related to the need
for assessing data quality as part of improvement activities in order to evaluate the
effectiveness of the improvement techniques. However, the assessment of quality also
raises novel and complex DQ issues that are particularly interesting from a scientific
perspective. For example, the definition of metrics for dimensions such as completeness
and accuracy is heavily grounded on database theory and broadens the scope of tra-
ditional database design issues, such as view integration and schema normalization.
In contrast, the economic issues of assessment and improvement require an empiri-
cal approach. The experimentation of techniques in current real-world contexts can be
challenging and only a few methodologies focus on providing empirical evidence on the
economics of DQ.
It should be noted that audit methodologies are more accurate than both com-
plete and operational methodologies in the assessment phase. First of all, they are
more detailed as to how to select appropriate assessment techniques, by providing
more examples and related contextual knowledge. Second, they identify all types of
issues, irrespective of the improvement techniques that can or should be applied.
AIMQ and QAFD methodologies, for instance, describe in detail how objective and
subjective assessments can be performed and provide guidelines to interpret results.
DQA discusses the operating definitions that can be used to measure the different DQ
dimensions, to evaluate aggregate measures of DQ for databases and, more recently,
data sources.
Operational methodologies focus DQ assessment on identifying the issues for which
their improvement approach works best. One of the main contributions is the identifica-
tion of a set of relevant dimensions to improve and the description of a few straightfor-
ward methods to assess them. For example, TDQM is a general-purpose methodology
and suggests a complete set of relevant dimensions and improvement methods that can
be applied in different contexts. The completeness decreases as methodologies focus on
particular contexts. For example, DWQ analyzes the data warehouse context and de-
fines new quality dimensions tailored to the architecture of a data warehouse. The list
of relevant dimensions represents an important starting point for the improvement
process, since it supports companies in the identifications of the DQ issues affecting
their datawareouse. Note that the assessment procedures are described more precisely
in operational methodologies that focus on a specific context, rather than in general-
purpose methodologies. Thus, the specialization of operational methodologies reduces
their completeness and applicability if compared with complete methodologies, but in-
creases the efficiency of the proposed techniques.
Furthermore, operational methodologies are more efficient when they focus on a par-
ticular DQ issue, especially in the application of data-oriented techniques. For example,
the ISTAT methodology focuses on localization data, that relate the addresses within
an administrative territory with the personal data of resident people. Such data are
most frequently exchanged across agencies. In this domain, the ISTAT methodology
compares the different heterogeneous formats adopted for localization data, the data
owners of the different domain attributes, such as the ZIP code, and proposes a unified
format and a new approach to data exchange based on the XML markup language.
As a second example, if DQ issues are related to the accuracy and completeness of
personal data, improvement methodologies can be more straightforward in targeting
record linkage techniques; this is the case of the DaQuinCIS and ISTAT methodologies,
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:29
that use record linkage to integrate different sources, by providing domain-specific
similarity algorithms for the discovery of duplicate records. For example, deduplication
of names of streets is performed in bilingual regions, such as the Alto Adige region in
Italy, adopting similarity functions specialized to paradigmatic errors such as imputing
“u” instead of “ ¨u”.
Another issue addressed in operational methodologies is the governance of the DQ
process and related risk management and feasibility problems. In this respect, the orig-
inal contribution of the ISTAT methodology versus all other methodologies, is its focus
on the large-scale application to all the central administrations of a country, includ-
ing peripheral organizational units, distributed over the territory, but hierarchically
dependent on central agencies. Usually, such a group of administrations has critical
characteristics, such as (1) a high complexity, in terms of interrelations, processes, and
services in which different agencies are involved, due to the fragmentation of compe-
tencies; (2) a high level of autonomy, which makes it difficult to enforce common rules;
(3) a high heterogeneity of meanings and representations of data and data flows; and
(4) large overlaps among heterogeneous records and objects. Improving DQ in such a
complex structure is usually a very risky and costly activity. This is the reason why the
main goal of the ISTAT methodology is to achieve feasibility. Therefore, attention is pri-
marily focused on the most common type of data exchanged between agencies, namely,
address data. In order to reduce the complexity of the problem, an assessment is first
performed on the most relevant data bases and data flows to discover the most critical
areas. The methodology mentions an experience where the less accurate records among
address data concern regions where street names are bilingual. In a second phase, more
in-depth assessment activities are performed on the local databases owned by different
administrations under their responsibility, using common techniques and tools that are
centrally provided. Finally, data exchanges among agencies are centrally planned and
Complete methodologies are extremely helpful in providing a comprehensive frame-
work to guide large DQ programs in organizations that process critical data and at-
tribute to DQ a high strategic priority, such as banks and insurance companies. On the
other hand, they show the classical tradeoff between the applicability of the method-
ology and the lack of personalization to specific application domains or technological
contexts. Being high-level and rather context independent, complete methodologies are
only marginally affected by the evolution of ICT technologies and, over time have been
revised to encompass the variety of data types, sources, and flows that are part of
modern information systems. For example, the IP-MAP model of TDQM has evolved to
IP-UML in order to manage the growing complexity of systems in terms of processes
and actors. However, its role and use within the overall framework of TDQM has not
changed significantly.
Economic methodologies complement other methodologies and can be easily posi-
tioned within the overall framework provided by any complete methodology. Most audit
and improvement methodologies have a cost evaluation step (see Table XI). However,
they mostly focus on the cost of DQ initiatives, while a complete cost-benefit analysis
should also consider the cost of doing nothing—the cost of poor data quality, which is
typically of an organizational nature. Economic methodologies focus on both aspects.
In particular, COLDQ focuses on the evaluation of the cost associated with poor data
quality, characterizing the economic impact based on the ultimate effects of bad data.
The result is called the data quality scorecard, and can be used as a tool to find the best
solutions for improvement. In CDQ, the overall evaluation of the cost of poor quality is
further developed to take into account the fact that the same quality improvement can
be obtained with different priorities and paths and the minimization of the cost of the
data quality program is the main criterion to choose among alternative improvement
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:30 C. Batini et al.
processes. The methodology suggests that different improvement processes can iden-
tified as paths of data- and process-driven techniques applied to the data bases, data
flows and document bases involved in the improvement. Then, the costs of the different
processes should be evaluated and compared, to select the minimum-cost process.
Concerning the validation of methodologies in real application contexts, we describe
in the following, the most significant experiences reported in the literature. The de-
scription proceeds case by case due to the dispersed nature of the experiences reported.
In fact, the empirical validation of methodologies is often missing or based on case
studies and is seldom generalized with large scale scientific experimentations, since
it typically takes place in industrial contexts and is part of companies’ core compe-
tencies for consulting in the field. As a consequence, process testing typically remains
Experiences of use of early TDQM versions are reported in several U.S.A. Depart-
ment of Defence (DoD) documents (see US Department of Defense [1994]). Specifically,
the use of DQ tools developed over SQL scripts and programming approaches to check
data quality are supported. Recently, the methodology has been the basis for a law
enforcement tool [Sessions 2007]. Other applications of TDQM performed at the DoD
Medical Command for Military Treatment Facilities (MTF) are reported in Wang [1998]
and Corey et al. [1996]. A claimed advantage of TDQM is that based on target payoffs,
critical DQ issues, and the corresponding types of data, one can effectively evaluate
how representative and comprehensive the DQ metrics are and whether this is the
right set of metrics. This is the reason for its extensive application in different con-
texts, such as insurance companies, as described in Nadkarni [2006] and Bettschen
[2005]. Nowadays, there are also many contributions extending TDQM, by improving
IP-MAP [Scannapieco et al. 2005; Shankaranarayanan and Wang 2007] or by propos-
ing a TDQM based Capability Maturity Model [Baskarada et al. 2006]. TIQM is a
professional methodology. A significant variety of real-life examples and case studies
is discussed in English [1999], ranging from customer relationship management to
telemarketing and healthcare. A large number of DQ tools are also compared. The
managerial principles discussed in the book and the numerous success stories reported
in the related Internet site, provide examples of relevant profes-
sional experiences, especially in the area of leveraging data quality for cost reduction,
improvement of information value, and business effectiveness.
CIHI has been conceived and applied for many years in organizing, maintaining, and
improving over 20 health administrative databases, many of which are person oriented
and population based [Long and Seko 2005]. Evidence is mounting across Canada that
reflects CIHI’s impact in supporting effective health care management and developing
public policy [Chapman et al. 2006]. CIHI is also used as a reference model in other
countries; in fact CIHI together with TDQM have also been used to establish a data
quality strategy for the Ministry of Health in New Zealand [Kerr and Norris 2004].
After long-term application, the authors provide success stories that show that the
evaluation process when using CIHI appears to have been successful in meeting the
primary objective of identifying and ranking the most critical issues of data quality im-
provement. Not only does it highlight adequate areas or conversely, target problematic
areas within a database, it also appears to facilitate the seemingly overwhelming task
of understanding the state of data quality for numerous data holdings. Furthermore,
the evaluation scores can be used to describe data quality for a generic institution.
It appears that strategic, corporate-wide planning might be facilitated when evalua-
tion results are summarized across the holdings. It is also noted that, even though the
number of evaluations is still low, numerous database improvements have already been
implemented and many of these improvements might not have been detected otherwise.
The main limitations experienced with CIHI are that the measurement properties of
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:31
the evaluation process are not yet known and only preliminary data were available at
the time of the study.
The authors of QAFD claim that their work is the result of many years of experience
in the financial data quality area and that data quality analysis techniques in QADF
are easy to develop and do not require the use of expensive solution packages. They
also provide a real-case scenario, although, for security reasons, they do not disclose all
available evidence.
In Batini and Scannapieco [2006], a large-scale experience of the application of CDQ
is reported, referring to the reorganization of Government to Business (G2B) relation-
ships in Italy. The interactions with government are needed for several business events,
such as starting a new business and evolving a business, which includes variations
in legal status, board composition, senior management, and number of employees. In
their interactions with businesses, agencies manage information common to all busi-
nesses, typically official name, headquarters, branch addresses, main economic activity.
Since every business independently reports to each agency, the copies have different
levels of accuracy and currency. CDQ has been applied in this context leading to, (1)
the execution of record-linkage activities with the goal of linking the diverse business
identifiers in the registries of different administrations; and to (2) the reengineering
of the G2B relationship. In the reengineered scenario, businesses interact with only
one administration, that is in charge of sending the same piece of information to other
administrations. Considering a three-year period limited to costs and savings related
to data quality, benefits from the application of CDQ have been estimated around 600
million Euros. This provides evidence of the validity of the methodology, in which both
data- and process-driven activities are considered, and the data quality program is
chosen by optimizing the cost/quality ratio. Other applications of CDQ are reported in
Basile et al. [2007], where a tool adopting CDQ as reference methodology has been used
for the evaluation, measurement, and mitigation of the operational risks related to the
Basel II process, leading, in some cases, to significant yearly savings.
In this article we have addressed the issue of methodologies for data quality, describing
and comparing thirteen of them. The whole DQ research field is currently evolving,
and cannot be considered mature. Methodologies for data quality measurement and
improvement are evolving in several directions: (1) considering a wider number of data
types, moving from data quality to information quality, (2) relating data quality issues
more closely to business process issues; and (3) considering new types of information
systems, specifically Web and P2P information systems. We discuss the three areas
separately and discuss open problems.
4.1. From Data Quality to Information Quality
Our previous analyses highlight the fact that information quality issues are only
marginally addressed in the area of semistructured data, unstructured data, and mul-
timedia. More precisely, methods and techniques have long been developed for such
types of data in separate research areas, such as natural language understanding for
documents, with scarce or no cross-fertilization with the area of data quality.
The limited number of research contributions focusing on semistructured and un-
structured data in the DQ domain are a consequence of the tight historical relation-
ship between DQ and database design. Even complete DQ methodologies are biased
by an underlying focus on large collections of structured data, which still represent
the most mature information resource in most organizations [Batini et al. 2008]. The
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:32 C. Batini et al.
interest in semistructured and unstructured data as organizational resources is more
recent. Knowledge management, Web preservation, and geographical information sys-
tems represent important research fields where DQ techniques for unstructured and
semistructured data are currently investigated [Batini and Scannapieco 2006]. Devel-
oping DQ techniques for semistructured and unstructured data in these fields requires
a higher degree of interdisciplinarity, which, in turn, may involve an additional delay.
4.2. Data Quality and Process Quality
The relationship between data quality and process quality is a wide area of investi-
gation, due to the relevance and diversity of characteristics of business processes in
organizations. The different impacts of data quality at the three typical organizational
levels, namely the operational, the tactical, and the strategic levels, are analyzed in
Redman [1998], reporting interviews and outcomes of several proprietary studies. Data
quality and its relationship with the quality of services, products, business operations,
and consumer behavior is investigated in very general terms in Sheng and Mykytyn
[2002] and Sheng [2003], where generic propositions such as “the information quality of
a firm is positively related to the firm’s performance” are substantiated with empirical
evidence. The problem of how improving information production processes positively
influences data and information quality is also analyzed in English [2002].
A few papers address more specific issues, and, consequently, present more con-
crete results. Vermeer [2000] examines the issue of electronic data interchange (EDI),
which concerns the exchange of data among organizations using standard formats, and
its influence on the efficiency and effectiveness of business processes. EDI and, more
generally, markup languages, are seen as a DQ technology enabler, since they poten-
tially reduce paper handling activities, data-entry errors, and data-entry functions.
The impact of data quality on the effectiveness of decision-making activities is inves-
tigated in Raghunathan [1999], with a focus on the accuracy dimension. The analysis
shows that the effectiveness of decisions improves with a higher data quality only if the
decision maker has knowledge about the relationship among problem variables, while
it may degrade in the opposite case.
The influence of data quality in extreme process conditions, such as disasters and
emergencies, is discussed in Fisher and Kingma [2001]. Flaws in accuracy, complete-
ness, consistency, and timeliness are considered with reference to critical situations,
such as the US Navy cruiser Vincennes firing at an Iranian Airbus, that brought 290
people to their deaths.
The role of information in the supply chain is considered in Dedeke [2005], where
the quality robustness of an information chain is proposed to measure the ability of
the information production process to also build the final information product in case
of threats that cause information distortion, transformation variabilities and informa-
tion failures. A methodological framework called process quality robustness design is
proposed as a framework for diagnosing, prescribing, and building quality into infor-
mation chains.
4.3. Data Quality and New Types of Information Systems
With the evolution of technology, the nature of information systems is changing and
new types of information systems are emerging. We analyze issues related to Web and
P2P information systems.
Concerning Web information systems, methodologies address several problems: (1)
the quality of unstructured data and, in particular, documents; (2) new types of quality
dimensions, such as accessibility.
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:33
Pernici and Scannapieco [2003] propose a model that associates quality information
with Web data, namely with each item in a Web page, with pages, and with groups of
pages. This model is applied within a methodology for data quality design and manage-
ment in Web information systems. The authors discuss how to enrich methodologies for
web information system design (such as Mecca et al. [1998] and Isakowitz et al. [1995])
with additional steps specifically devoted to data quality design. Several dimensions
are considered, such as volatility, completability, and semantic and syntactic accuracy.
The quality of Web documents is of increasing relevance, since the number of doc-
uments that are managed in Web format is constantly growing. Several studies (e.g.,
Rao [2003]) have shown that 40% of the material on the net disappears within one
year, while a further 40% is modified, leaving only 20% in its original form. Other stud-
ies [Lyman and Varian 2003] indicate that the average lifetime of a Web page is 44
days and the Web changes completely about four times in a year. As a consequence,
the preservation of Web data becomes more and more crucial. The term Web preserva-
tion indicates the ability to prevent the loss of information in the Web, by storing all
significant versions of Web documents.
Cappiello et al. [2003a] propose a methodology to support the preservation pro-
cess over the entire life cycle of information, from creation, to acquisition, catalogu-
ing, storage, and access. The main phases of the methodology are summarized in the
(1) Each time a new page is published, data are associated with metadata, describing
their quality, in terms of accuracy, completeness, consistency, currency, and volatil-
ity, as defined in Section 2.3.
(2) In the acquisition phase, the user specifies acceptable values for all quality di-
mensions. If new data satisfy quality requirements, they are incorporated into an
archive. Otherwise, data are returned to their owner and are not catalogued until
their quality is satisfactory.
(3) In the publishing stage, when a new page replaces an old Web page, the volatility
of old data is evaluated. If old data are still valid, data are not deleted and are
associated with a new URL.
Assessment methodologies for evaluating specific qualities of Web sites are proposed
in Atzeni et al. [2001], Mecca et al. [1999], and Fraternali et al. [2004]. Atzeni et al.
[2001] is specifically focused on accessibility, as defined in Section 2.3, evaluated on
the basis of a mixed quantitative/qualitative assessment. The quantitative assessment
activity checks the guidelines provided by the World Wide Web Consortium in [World
Wide Web Consortium]. The qualitative assessment is based on ex-
periments performed with disabled users. Fraternali et al. [2004] focus on the usability
of the site and propose an approach based on the adoption of conceptual logs, which
are Web usage logs enriched with metadata inferred from the conceptual schema of the
Web site. While more traditional measures of the quality of a Web site are based on the
hypertext representation of information, in this approach new indicators are proposed
based on a conceptual representation of the site.
P2P systems are completely open and raise a need to assess and filter data. Trust-
related quality dimensions in peer-to-peer systems are still an open issue. A possible
solution is to rely on the reputation (or trustworthiness) of each peer. As an example,
the rating service of DaQuinCIS (see Figure 4) associates trust values with each data
source in the information system. The rating service is centralized and is supposed
to be performed by a third-party organization. Trust values are used to determine
the reliability of the quality evaluations made by organizations. Metrics are based
[De Santis et al. 2003] on complaints the users of the data sets produced by each
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:34 C. Batini et al.
peer. The metrics proposed in Gackowski [2006] consider as likely accurate, the data
sources that exchange data with accurate or high-quality data sources. These metrics
also include a temporal component that takes into account data source past accuracy.
Other proposals suggest investigating techniques that can improve the credibility of
data values [Gackowski 2006].
4.4. Further Open Issues
Further open problems in DQ methodologies concern:
(1) the identification of more precise statistical, probabilistic, and functional correla-
tions among data quality and process quality, with a focus on (1) the empirical
validation of the models; and (2) the extension of the analysis to a wider set of
dimensions and to specific types of business processes;
(2) the validation of methodologies; Often, a methodology is proposed without any large-
scale specific experimentation and with none or only a few, supporting tools. There
is a lack of research on experiments to validate different methodological approaches
and on the development of tools to make them feasible;
(3) the extension of methodological guidelines to a wider set of dimensions, such
as performance, availability, security, accessibility, and to dependencies among
dimensions. An example of a dependency among currency and accuracy is the follow-
ing: if an item is not current it is also inaccurate in 70% of the cases. Knowledge on
dependencies can be acquired with data mining techniques. Further, dependencies
can be analyzed with statistical techniques, providing new knowledge for improving
the efficiency and effectiveness of the improvement process [De Amicis et al. 2006].
(4) In Web information systems and in data warehouses, data are managed at different
aggregation levels. Quality composition should be investigated to obtain aggreate
quality information from the quality metrics associated with elementary data.
In this appendix we provide a description card of each methodology. The description
cards have a common structure composed of three parts summarizing: (1) the phases
of each methodology and their mutual dependencies and critical decisions, expressed
with a diagram and with a textual description, including inputs, outputs, and steps;
(2) a general description highlighting the focus of each methodology and original con-
tributions to the data quality assessment and improvement process; and (3) detailed
comments discussing the applicability of each methodology.
In the diagrammatic and in the textual descriptions of methodologies, we report the
terms used in the methodology, together with the standard terms we have used in
Section 2.1 (reported in parentheses).
A.1. The TDQM (Total Data Quality Management) Methodology
General description. The TDQM methodology was the first general methodology pub-
lished in the data quality literature [Wang 1998]. TDQM is the outcome of academic
research, but has been extensively used as a guide to organizational data reengineering
initiatives. The fundamental objective of TDQM is to extend to data quality, the princi-
ples of Total Quality Management (TQM) [Oakland 1989]. In operations management,
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:35
Fig. 6. Phases of TDQM.
TQM has shifted the focus of reengineering activities from efficiency to effectiveness, by
offering methodological guidelines aimed at eliminating discrepancies between the out-
put of operating processes and customers’ requirements. Given requirements, reengi-
neering must start from modeling operating processes. Consistent with these tenets,
TDQM proposes a language for the description of information production (IP) processes,
called IP-MAP [Shankaranarayan et al. 2000]. IP-MAP has been variously extended,
towards UML and also to support organizational design. IP-MAP is the only language
for information process modeling and represents a de facto standard. Practical experi-
ences with TDQM are reported, for example, in Kovac and Weickert [2002].
Detailed comments. TDQM’s goal is to support the entire end-to-end quality improve-
ment process, from requirements analysis to implementation. As shown in Figure 6(a)
TDQM Cycle consists of four phases that implement a continuous quality improvement
process: definition, measurement, analysis, and improvement.
The roles responsible for the different phases of the quality improvement process are
also defined in TDQM. Four roles are distinguished: information suppliers, which create
or collect data for the IP, information manufacturers, which design, develop, or maintain
data and related system infrastructure, information consumers, which use data in their
work, and information process managers, which are responsible for managing the entire
information production process throughout the information life cycle.
TDQM is comprehensive also from an implementation perspective, as it provides
guidelines as to how to apply the methodology. In applying TDQM, an organization
must: (a) clearly understand the IPs; (b) establish an IP team consisting of a senior
executive as the TDQM champion, an IP engineer who is familiar with the TDQM
methodology, and members who are information suppliers, manufacturers, consumers,
and IP managers; (c) teach IQ assessment and IQ management to all the IP constituen-
cies; and (d) institutionalize continuous IP improvement.
TDQM relies on the information quality literature for IQ Criteria and IQ improve-
ment techniques. In particular, it explicitly refers to Wang and Strong [1996] for the
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:36 C. Batini et al.
data quality dimensions specification. TDQM relates quality issues to corresponding
improvement techniques. However, in the recent literature no industry-specific tech-
nique is referred and no support is offered to specialize general quality improvement
A.2. The DWQ (Data Warehouse Quality) Methodology
General description. The DWQ methodology has been developed within the European
Data Warehouse Quality project [Jeusfeld et al. 1998]. This methodology studies the
relationship between quality objectives and design options in data warehousing. The
methodology considers the subjectivity of the quality concept and provides a classifica-
tion of quality goals according to the stakeholder group that pursues these goals. On
the other hand, they consider the diversity of quality goals and define corresponding
Detailed comments. The DWQ methodology states that data warehouse metadata
should account for three perspectives: a conceptual business perspective focusing on
the enterprise model, a logical perspective focusing on the data warehouse schema, and
a physical perspective representing the physical data transport layer. These perspec-
tives correspond to the three traditional layers of data warehousing, namely sources,
data warehouse, and clients. The methodology associates with each perspective, a cor-
responding metadata view called Quality Measurement.
From a data quality perspective, four main phases characterize the methodology:
definition, assessment, analysis, and improvement (see Figure 7). One of the main
contributions provided by this methodology is the classification of data and software
quality dimensions in the data warehouse context. Three categories of data and meta-
data are defined:
Design and administration quality: the former refers to the ability of a model to
represent information adequately and efficiently, while the latter refers to the way
the model evolves during the data warehouse operation.
Software implementation quality: the quality dimensions of the ISO 9126 standard
are considered, since software implementation is not a task with specific data ware-
house characteristics.
Data usage quality: it refers to the dimensions that characterize the usage and query-
ing of data contained in the data warehouse.
For each dimension contained in the listed classes, suitable measurement methods are
identified. The list of these methods together with the relevance degree associated with
each dimension by stakeholders are the input for the effective measurement step. In
the quality assessment phase, there is the storage of the following information about
each data quality dimension: (1) quality requirements—an interval of expected values;
(2) the achieved quality measurement; (3) the metric used to compute a measurement;
(iv) causal dependencies to other quality dimensions. Information about dependencies
among quality dimensions is used to trace and analyze quality problems. The identi-
fication of critical areas is the last step analyzed in the methodology. Indeed, it only
mentions the improvement phase but does not contain constructive knowledge about
how to improve the quality of a data warehouse.
A.3. The TIQM (Total Information Quality Management) Methodology
General description. The TIQM methodology [English 1999] has been designed to
support data warehouse projects. The methodology assumes the consolidation of
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:37
Fig. 7. Phases of DWQ.
Fig. 8. Phases of TIQM.
operational data sources into a unique, integrated database, used in all types of aggre-
gations performed to build the data warehouse. This consolidation eliminates errors
and heterogeneities of source databases. TIQM focuses on the management activities
that are responsible for the integration of operational data sources, by discussing the
strategy that has to be followed by the organization in order to make effective tech-
nical choices. Cost-benefit analyses are supported from a managerial perspective. The
methodology provides a detailed classification of costs and benefits (see Section 3.4).
Detailed comments. Figure 8 shows the phases of the TIQM methodology. From the
TIQM’s managerial perspective, there are three main phases: assessment, improve-
ment, and improvement management and monitoring. One of the valuable contribu-
tions of the methodology is the definition of this last phase, which provides guidelines to
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:38 C. Batini et al.
Fig. 9. The PSP/IQ model.
Fig. 10. Phases of AIMQ.
manage changes in the organization’s structure according to data quality management
requirements. Furthermore, the economics approach introduces cost benefit evaluation
to justify data quality interventions. The goal is not only the achievement of higher data
quality level, but to undertake improvement actions only if they are feasible; thus only
if benefits are greater than costs.
A.4. The AIMQ (A Methodology for Information Quality Assessment) Methodology
General description. The AIMQ methodology is the only information quality method-
ology focusing on benchmarking [Lee et al. 2002], that is an objective and domain-
independent technique for quality evaluation.
The foundation of the AIMQ methodology is a 2x2 table, called the PSP/IQ model (see
Figure 9), classifying quality dimensions according to their importance from the user’s
and manager’s perspectives. The axes of the table are conformity to specifications and
conformity to users’ expectations. Accordingly, four classes of dimensions are distin-
guished (sound, dependable, useful, and usable) and quality dimensions identified in
Wang and Strong [1996] are classified along these classes. Benchmarking should rank
information within each class.
The PSP/IQ model is an input to the AIMQ methodology whose phases are summa-
rized in Figure 10. The publications describing AIMQ mainly focus on the assessment
activities, while guidelines, techniques, and tools for improvement activities are not
Detailed comments. IQ is mainly assessed by means of questionnaires. A first pilot
questionnaire is used to identify relevant quality dimensions and attributes to be bench-
marked. Then, a second questionnaire addresses the dimensions and attributes previ-
ously identified in order to obtain IQ measures. Finally, these measures are compared
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:39
Fig. 11. Phases of CIHI.
with benchmarks. Lee et al. [2002] provide a list of standard quality dimensions and
attributes helping the definition of questionnaires.
The literature on AIMQ does not provide any description of the benchmarking
database that is required for the application of the methodology. Gap Analysis tech-
niques are advocated as a standard approach to conduct benchmarking and interpret
results. In particular, two Gap Analysis techniques are suggested: Information Quality
Benchmark Gaps and Information Quality Role Gaps. The former compares the quality
values of an organization with those of best-practice organizations. The latter compares
the information quality assessments provided by different organizational roles, that is
the IS professional and the information user. IQ Role Gaps searches for discrepancies
between the evaluations provided by different roles as an indication of potential qual-
ity issues. Discrepancies are associated with a direction. The direction of the gap is
positive if the assessment of IS professionals is higher than the assessment of users. A
large positive gap is considered dangerous, since it indicates that IS professionals are
not aware of quality issues that information users have detected. If the size of the gap
is small, the location of the gap should be analyzed. If the location is high, indicating
high IQ, incremental improvements are most appropriate, whereas if the location is
low, major improvement efforts are likely to be required.
A.5. The CIHI (Canadian Institute for Health Information) Methodology
General description. The CIHI methodology has implemented a method to evaluate
and improve the quality of Canadian Institute for Health Information data [Long and
Seko 2005]. In the CIHI scenario, the main issue is the size of databases and their
heterogeneity. The CIHI methodology supports the selection of a subset of data to focus
the quality assessment phase. It also proposes a large set of quality criteria to evaluate
Detailed Comments. The CIHI Data Quality strategy proposes a two-phase approach
(see Figure 11). The first phase is the definition of a Data Quality Framework, and the
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:40 C. Batini et al.
second is an in-depth analysis of the most frequently accessed data. The Data Quality
Framework is defined in three steps: (1) standardization of data quality information;
(2) development of a common strategy for data quality assessment; (3) definition of a
work process for CIHI’s data management that identifies data quality priorities and
implements continuous data improvement procedures.
The implementation of the CIHI framework is cyclical, according to the continuous
improvement approach. The following is required for a successful implementation:
—definition of the time period for the cycle;
—definition of time objectives for different quality targets;
—allocation of ad hoc resources for data quality analysis, evaluation, and documenta-
—allocation of ad hoc resources for data quality improvement.
The analysis of the most frequently used data is performed in three steps: data qual-
ity analysis, evaluation, and documentation. Documents report the quality problems
detected by the data quality analysis and evaluation.
Data quality evaluation is based on a four-level hierarchical model. At the first level,
86 basic quality criteria are defined. These criteria are aggregated by means of com-
position algorithms into 24 quality characteristics at the second hierarchical level, and
further aggregated into five quality dimensions at the third level; namely, accuracy,
timeliness, comparability, usability, and relevance. Finally, the five dimensions are ag-
gregated into one overall database evaluation at the fourth level.
The basic evaluation of the 86 data quality criteria is performed by means of ques-
tionnaires reporting criteria as items to be scored on a four-point ordinal scale as “not
applicable,” “unknown,” “not met,” or “met.” Then, at each aggregation level, evalua-
tions are validated. The validation process ensures that the interpretation and scoring
of each criterion is as standard as possible.
A.6. The DQA (Data Quality Assessment) Methodology
General description. The DQA methodology [Pipino et al. 2002] has been designed to
provide the general principles guiding the definition of data quality metrics. In the
literature, data quality metrics are mostly defined ad hoc to solve specific problems
and thus, are dependent on the considered scenario. The DQA methodology is aimed at
identifying the general quality measurement principles common to previous research.
Detailed comments. The classification of metrics of the DQA methodology is sum-
marized in Figure 12. The methodology makes a distinction between subjective and
objective quality metrics. Subjective metrics measure the perceptions, needs, and expe-
riences of the stakeholders. Objective metrics are then classified into task-independent
and task-dependent. The first assess the quality of data without contextual knowledge
of the application, while the second are defined for specific application contexts and in-
clude business rules, company and government regulations, and constraints provided
by the database administration. Both metrics are divided into three classes: simple
ratio, min or max value, and weighed average.
A.7. The IQM (Information Quality Measurement) Methodology
General description. The fundamental objective of the IQM methodology [Eppler and
unzenmaier 2002] is to provide an information quality framework tailored to Web
data. In particular, IQM helps the quality-based selection and personalization of the
tools that support Webmasters in creating, managing, and maintaining Web sites.
Detailed comments. The IQM methodology provides guidelines to ensure that
software tools evaluate all the fundamental information quality dimensions. The
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:41
Fig. 12. Phases of DQA.
Fig. 13. Phases of IQM.
methodology provides two sets of guidelines: the information quality framework
defining quality criteria, and the action plan explaining how to perform quality
The main phases of IQM methodology are reported in Figure 13. The first phase de-
fines the measurement plan. The information quality framework is defined as a list of
relevant information quality criteria identified by interviewing the information stake-
holders. The framework is the input for an information quality audit that associates
the information quality criteria with the methods and tools that will be used in the
measurement process. Some criteria require multiple measurement methods. The IQM
methodology coordinates the application of multiple measurement methods.
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:42 C. Batini et al.
Fig. 14. Phases of ISTAT.
A.8. The ISTAT (Italian National Bureau of Census) Methodology
General description. The ISTAT methodology [Istat 2004; Falorsi et al. 2003] has been
designed within the Italian National Bureau of Census to collect and maintain high
quality statistical data on Italian citizens and businesses. The fundamental issue faced
by the methodology is how to guarantee the quality of data integrated from multi-
ple databases of local Public Administrations. This issue is particularly challenging
within the Italian context, where the Public Administration is organized in three ge-
ographical levels, Central, Regional and Peripheral, each managing its own data au-
tonomously. The ISTAT methodology focuses on the most common types of data ex-
changed among different levels of the Public Administration, namely private data. The
methodology is strongly focused on formal norms, since it is aimed at regulating data
management activities in such a way that their integration can satisfy basic quality
Detailed comments. The fundamental phases of the ISTAT methodology are (see
Figure 14):
—the assessment phase, that is initially performed on the central databases
owned and managed by ISTAT, to detect quality issues from a data integration
—the global improvement phase, which is in charge of performing record linkage among
national databases and designing the improvement solution on processes including
the decision to make, buy, or adapt existing solutions;
—Improvement activities on databases owned and managed by local Administrations.
These activities should be performed by local Administrations themselves, with the
aid of tools and courses provided by ISTAT.
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:43
—Improvement activities that require the cooperation of multiple Administrations.
These activities are typically process oriented, since they address flows of data ex-
changed during the execution of specific operating activities. Central databases may
be required for coordination purposes. These activities are centrally planned and
The ISTAT methodology provides a variety of simple but effective statistical tech-
niques for quality measurement. It also provides tools for the most relevant data clean-
ing activities. Local Administrations are helped in tailoring tools to specific geographical
or process issues. In the ISTAT methodology, data owners are defined at a high level of
detail, corresponding to individual attributes, such as MunicipalityCode. The method-
ology supports the standardization of data formats and their expression in a common
XML schema allowing the integration of the databases of local Administrations. Data
exchanged among different Administrations are redesigned using an event-driven soft-
ware architecture, based on publish and subscribe mechanisms.
A.9. The AMEQ (Activity-based Measuring and Evaluating of Product
information Quality) Methodology
General description. The main goal of the AMEQ methodology [Su and Jin 2004] is
to provide a rigorous basis for Product Information Quality (PIQ) assessment and
improvement in compliance with organizational goals. The methodology is specific for
the evaluation of data quality in manufacturing companies, where product information
represents the main component of operational databases. In manufacturing companies,
the association between product information and production processes is straightfor-
ward and relatively standard across companies. The schema of product databases is
also similar across different organizations. The methodology provides an approach
and methodological guidelines to model both information and related production
Detailed comments. The AMEQ methodology consists of five phases for measuring
and improving PIQ (see Figure 15). The first phase assesses the cultural readiness of
an organization, using the Information Quality Management Maturity Grid, a template
to conduct interviews for key managerial roles. In this phase, the dimensions of PIQ are
also defined and classified according to their relevance for different business activities.
The second phase specifies the information product. Each information product is asso-
ciated with a corresponding business process, modelled by means of an object-oriented
approach (OOA). In the AMEQ methodology, eight types of objects are modelled: human
resources, information resources, enterprise activities, resource inputs, resource pro-
cesses, resource outputs, performance measures, and enterprise goals. In this phase, a
model of measurement methods is also produced. The third phase focuses on the mea-
surement activity. In the fourth phase, the IQ team should investigate the root causes
for potential PIQ problems by analyzing the quality dimensions that have received a
low score. Finally, the PIQ improvement phase can start. Note that for the fourth and
fifth phases, AMEQ does not provide operating methods and tools, but only general
A.10. The COLDQ (Cost-Effect Of Low Data Quality) Methodology
General description. The fundamental objective of the COLDQ methodology [Loshin
2004] is to provide a data quality scorecard supporting the evaluation of the cost-effect
of low data quality. Similarly to TIQM (Section A.3), the methodology provides a detailed
classification of costs and benefits (see also Section 3.4). Direct benefits are obtainable
from the avoidance of poor quality costs due to the adoption of improvement techniques.
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:44 C. Batini et al.
Fig. 15. Phases of AMEQ.
The goal is to obtain a quantitative assessment of the extent to which business processes
are affected by bad information.
Detailed comments. Six interrelated phases are proposed to evaluate the cost-effect
of low data quality (see Figure 16). In the first phase of the methodology, the business
context is modelled by identifying two data flow models: the strategic data flow, used for
decision-making, and the operational data flow, used for data processing. Both models
represent a set of processing stages that describe the information flow from data supply
to data consumption. Based on these models, the objective and subjective analyses
of the business context are conducted. Internal and external users, employees and
customers, are interviewed in order to identify flawed data. Then, errors are attributed
to faulty activities in the strategic and operational models of the business context. This
association between errors and activities provides the basis for cost evaluations. The
COLDQ methodology provides a thorough and valuable classification of operational,
tactical, and strategic economic impacts that have to be considered. Each class of costs
is assigned an economic value based on contextual knowledge. Costs represent the input
to the final improvement phase.
Finally, the COLDQ methodology supports cost-benefit analyses by evaluating and
aggregating the cost of quality improvement projects. Methodological guidelines sup-
port the calculation of the return on investment (ROI) and break-even points of im-
provement initiatives.
A.11. The DaQuinCIS (Data QUality IN Cooperative Information Systems) Methodology
General description. The DaQuinCIS methodology [Scannapieco et al. 2004] addresses
data quality issues in Cooperative Information Systems. Cooperation raises two
context-specific data quality issues. First of all, data quality is predicated upon in-
terorganizational trust. Second, poor data quality can hinder cooperation and, thus,
has far-reaching consequences. To address the first issue, the DaQuinCIS methodology
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:45
Fig. 16. Phases of COLDQ.
introduces the concept of data quality certification, which associates data with cor-
responding quality measures that are exchanged among organizations along with
data. The second issue is addressed by providing quality-based data selection mecha-
nisms. These selection mechanisms identify the highest-quality data among overlap-
ping databases owned by different cooperating organizations. In this way, cooperation
is leveraged to improve quality.
Detailed comments. The DaQuinCIS methodology provides an innovative model to
represent data quality called data and data quality (D2Q) that includes: (1) constructs
to represent data, (2) a set of data quality properties, (3) constructs to represent data
quality properties and (4) the associations between data and quality metadata. Source
trustworthiness is included among quality properties. The value associated with this
dimension is assigned by a third-party organization on the basis of several parameters,
including the number of complaints made by other organizations and the number of
requests issued to each source.
Figure 17 reports the fundamental phases of the DaQuinCIS methodology: qual-
ity analysis,quality assessment,quality certification, and quality improvement. The
methodology is supported by an architecture composed of an internal infrastructure
and an external infrastructure (see Figure 18). Methodological phases are implemented
by corresponding modules of the Quality Factory that are implemented in each organi-
zation belonging to CIS. The communication among the organizations involved in the
CIS is enabled by a quality notification service that is a publish/subscribe engine used
as general message bus between the different architectural components.
In the Quality Factory, the requests from external users define data to be retrieved
and corresponding quality requirements. An apposite module evaluates the quality of
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:46 C. Batini et al.
Fig. 17. Phases of DaQuinCIS.
Fig. 18. The DaQuinCIS architecture.
data and compares the quality values with the quality requirements expressed by the
users. If data do not satisfy the quality requirements, an alert is sent to the user. On
the contrary, if values of quality are satisfactory, a quality certificate is associated with
the data and sent to the user. Quality improvement is carried out by the data quality
broker. This module, by collaborating with the Data Quality Repository, translates
queries according to a global schema and selects data that maximize quality. A query
submitted by a specific organization is issued to all organizations, specifying a set of
quality requirements on requested data. Different copies of the same data received as a
response to the request are reconciled and best-quality values are selected and returned
to the requesting organizations.
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:47
Fig. 19. Phases of QADF.
As regards the techniques used in the improvement phase, DaQuinCIS methodology
proposes a new algorithm for record matching. The Record Matcher is a component of
the Data Quality Broker. The Record Matcher implements a method for record matching
based on the quality data exported by cooperating organizations.
A.12. The QAFD (Quality Assessment of Financial Data) Methodology
General description. The QAFD methodology [De Amicis and Batini 2004] has been de-
signed to define standard quality measures for financial operational data and thus min-
imize the costs of quality measurement tools. The methodology combines quantitative
objective, and qualitative subjective assessments to identify quality issues and select
the appropriate quality improvement actions. Context-dependent indices, data qual-
ity rules, measurements, and strategies for quantitative and qualitative assessments
are defined. Overall, it represents the only methodology for the quality assessment of
financial data.
Detailed comments. The main phases of this methodology are reported in Figure 19.
First, the methodology selects the most relevant financial variables. Selection is usually
based on knowledge from previous assessments, according to their practical effective-
ness. Variables are grouped in categories of “related issues” that similarly affect the
behavior of investors and consumers and are characterized by the same risk, business,
and descriptive factors.
The second phase aims at discovering the main causes of errors. The most relevant
data quality dimensions are identified in this phase and data quality rules are produced.
Data quality rules represent the dynamic semantic properties of variables that cannot
be measured along quality dimensions.
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:48 C. Batini et al.
Fig. 20. Phases of CDQ.
In the third phase, the objective assessment is performed based on quantitative in-
dexes. Authors propose a mathematical model for the objective assessment resulting,
in an overall ranking of data along each quality dimension.
The subjective assessment is performed in the fourth phase from three different
perspectives; business experts, customers, and data quality experts. Each interviewee
has to assess the quality level along each quality dimension. An overall assessment is
obtained as the mean value of the subjective assessment of each class of experts.
Finally, objective and subjective assessments are compared in the fifth phase. For
each dimension, the difference between the objective and subjective assessments is
calculated. If the difference is positive, the objective assessment must be reconsidered
to point out quality issues relevant from the experts’ point of view.
A.13. The CDQ (Complete Data Quality) Methodology
General description. The CDQ methodology [Batini and Scannapieco 2006; Batini et al.
2008] is conceived to be at the same time complete, flexible, and simple to apply. Com-
pleteness is achieved by considering existing techniques and tools and integrating them
in a framework that can work in both intra- and inter-organizational contexts, and
can be applied to all types of data, structured, semistructured and unstructured. The
methodology is flexible since it supports the user in the selection of the most suitable
techniques and tools within each phase and in any context. Finally, CDQ is simple since
it is organized in phases and each phase is characterized by a specific goal and set of
techniques to apply.
The CDQ methodology is innovative since it provides support to select the opti-
mal quality improvement process that maximizes benefits within given budget lim-
its. Second, it emphasizes the initial requirements elicitation phase. In fact, the other
methodologies implicitly assume that contextual knowledge has been previously gath-
ered and modelled. The focus is on how to reach total data quality without providing
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:49
indications as to how to use contextual knowledge. A goal of CDQ is instead to obtain a
quantitative assessment of the extent to which business processes are affected by bad
Detailed comments. Three main phases characterize the methodology: state recon-
struction, assessment, and choice of the optimal improvement process (see Figure 20).
In the first phase of the methodology, the relationships among organizational units,
processes, services, and data are reconstructed. These relationships are modelled by
using matrixes that describe which organizational units use data and their roles in
the different business processes. Furthermore, in this phase, processes are described
along with their contribution in the production of goods/services and the legal and or-
ganizational rules that discipline workflows. The second phase sets new target quality
levels that are needed to improve process qualities, and evaluates corresponding costs
and benefits. This phase locates the critical variables affected by poor quality. Since im-
provement activities are complex and costly, it is advisable to focus on the parts of the
databases and data flows that raise major problems. Finally, the third phase consists
of five steps and is aimed at the identification of the optimal improvement process: the
sequence of activities that has the highest cost/effectiveness ratio. New target quality
levels are set by considering costs and benefits. Different improvement activities can be
performed to reach new quality targets. The methodology recommends the identifica-
tion of all the data-driven and process-driven improvement techniques for the different
databases affected by poor quality. A set of mutually consistent improvement tech-
niques constitutes an improvement process. Finally, the most suitable improvement
process is selected by performing a cost-benefit analysis.
ABITEBOUL, S., BUNEMAN,P.,AND SUCIU, D. 2000. Data on the Web: From Relations to Semistructured Data
and XML. Morgan Kaufmann Publishers.
AIKEN, P. 1996. Data Reverse Engineering. McGraw Hill.
ARENAS, M., BERTOSSI, L., AND CHOMICKI, J. 1999. Consistent query answers in inconsistent databases. In
Proceedings of the 18th ACM SIGMOD-SIGACT-SIGART Symposium on Principles of Database Systems
(PODS). ACM, New York, 68–79.
ATZENI,P.AND ANTONELLIS, V. D. 1993. Relational Database Theory. Benjamin/Cummings.
ATZENI, P., MERIALDO,P.,AND SINDONI, G. 2001. Web site evaluation: Methodology and case study. In Pro-
ceedings of International Workshop on data Semantics in Web Information Systems (DASWIS).
BALLOU,D.AND PAZER, H. 1985. Modeling data and process quality in multi-input, multi-output information
systems. Manag. Sci. 31,2.
, D., WANG, R., PAZER, H., AND TAYI , G. 1998. Modeling information manufacturing systems to deter-
mine information product quality. Manage. Sci. 44,4.
BASILE, A., BATINI, C., GREGA, S., MASTRELLA, M., AND MAURINO, A. 2007. Orme: A new methodology for
information quality and basel II operational risk. In Proceeedings of the 12th International Conference
of Information Quality, Industrial Track.
BASILI, V., CALDIERA, C., ROMBACH, H. 1994. Goal question metric paradigm.
BASKARADA, S., KORONIOS, A., AND GAO, J. 2006. Towards a capability maturity model for information quality
management: a tdqm approach. In Proceedings of the 11th International Conference on Information
BATINI, C., CABITZA, F., CAPPIELLO,C.,AND FRANCALANCI, C. 2008. A comprehensive data quality methodology
for Web and structured data. Int. J. Innov. Comput. Appl. 1, 3, 205–218.
BATINI,C.AND SCANNAPIECO, M. 2006. Data Quality: Concepts, Methodologies and Techniques. Springer
BERTOLAZZI, P., SANTIS,L.D.,AND SCANNAPIECO, M. 2003. Automatic record matching in cooperative infor-
mation systems. In Proceedings of the ICDT International Workshop on Data Quality in Cooperative
Information Systems (DQCIS).
BETTSCHEN, P. 2005. Master data management (MDM) enables IQ at Tetra Pak. In Proceedings of the 10th
International Conference on Information Quality.
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:50 C. Batini et al.
OHM, C., DRABA, K., NAUMANN,F.,AND WEIS, M. September 2005. Automatic data
fusion with HumMer. In Proceedings of the VLDB Demonstration Program.
BOVEE, M., SRIVASTAVA, R., AND MAK, B. September 2001. A conceptual framework and belief-function ap-
proach to assessing overall information quality. In Proceedings of the 6th International Conference on
Information Quality.
BUNEMAN, P. 1997. Semi-structured data. In Proceedings of the 16th ACM Symposium on Principles of
Database Systems (PODS).
ı, A., CALVANESE, D., DEGIACOMO,G.,AND LENZERINI, M. 2004. Data integration under integrity con-
straints. Inform. Syst. 29, 2, 147–163.
CALVANESE, D., DEGIACOMO,D.,AND LENZERINI, M. 1999. Modeling and querying semi-structured data. Net-
work. Inform. Syst. J. 2, 2, 253–273.
CAPPIELLO, C., FRANCALANCI,C.,AND PERNICI, B. 2003. Preserving Web sites: A data quality approach. In
Proceedings of the 7th International Conference on Information Quality (ICIQ).
in cooperative information systems: a multi-dimension certificate. In Proceedings of the ICDT Interna-
tional Workshop on Data Quality in Cooperative Information Systems (DQCIS).
CATARCI,T.,AND SCANNAPIECO, M. 2002. Data quality under the computer science perspective. Archivi Com-
puter 2.
CHAPMAN, A., RICHARDS, H., AND HAWKEN, S. 2006. Data and information quality at the Canadian in-
stitute for health information. In Proceedings of the 11th International Conference on Information
CHENGALUR-SMITH,I.N.,BALLOU,D.P.,AND PAZER, H. L. 1999. The impact of data quality information on
decision making: An exploratory analysis. IEEE Trans. Knowl. Data Eng. 11, 6, 853–864.
COREY, D., COBLER, L., HAYNES, K., AND WALKER, R. 1996. Data quality assurance activities in the mili-
tary health services system. In Proceedings of the 1st International Conference on Information Quality.
,T.AND JOHNSON, T. 2003. Exploratory Data Mining and Data cleaning. Probability and Statistics
series, John Wiley.
DATA WAREHOUSING INSTITUTE. 2006. Data quality and the bottom line: Achieving business success through
a commitment to high quality data.
DEAMICIS, F., BARONE,D.,AND BATINI, C. 2006. An analytical framework to analyze dependencies among
data quality dimensions. In Proceedings of the 11th International Conference on Information Quality
(ICIQ). 369–383.
DEAMICIS,F.AND BATINI, C. 2004. A methodology for data quality assessment on financial data. Studies
Commun. Sci. SCKM.
WOO,C.,AND YU, E. 1997. Cooperative Information Systems: A Manifesto. In Cooperative Information
Systems: Trends & Directions, M. Papazoglou and G. Schlageter, Eds. Academic-Press.
DESANTIS, L., SCANNAPIECO, M., AND CATARCI, T. 2003. Trusting data quality in cooperative information sys-
tems. In Proceedings of the 11th International Conference on Cooperative Information Systems (CoopIS).
Catania, Italy.
DEDEKE, A. 2005. Building quality into the information supply chain. Advances in Management Informa-
tion Systems-Information Quality Monograph (AMIS-IQ) Monograph. R. Wang, E. Pierce, S. Madnick,
and Fisher C.W., Eds.
DQI. 2004. Data quality initiative framework. Project report. Framwork Update Letter 160604.pdf
ENGLISH, L. 1999. Improving Data Warehouse and Business Information Quality. Wiley & Sons.
ENGLISH, L. 2002. Process management and information quality: how improving information production
processes improved information (product) quality. In Proceedings of the 7th International Conference on
Information Quality (ICIQ). 206–211.
EPPLER,M.AND HELFERT, M. 2004. A classification and analysis of data quality costs. In Proceedings of the
9th International Conference on Information Systems (ICIQ).
UNZENMAIER, P. 2002. Measuring information quality in the Web context: A survey of
state-of-the-art instruments and an application methodology. In Proceedings of the 7th International
Conference on Information Systems (ICIQ).
the quality of toponymic data in the italian public administration. In Proceedings of the ICDT Workshop
on Data Quality in Cooperative Information Systems (DQCIS).
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
Methodologies for Data Quality Assessment and Improvement 16:51
FELLEGI,I.P.,AND HOLT, D. 1976. A systematic approach to automatic edit and imputation. J. Amer. Stat.
Assoc. 71, 353, 17–35.
FISHER,C.AND KINGMA, B. 2001. Criticality of data quality as exemplified in two disasters. Inform. Man-
age. 39, 109–116.
FRATERNALI, P., LANZI, P., MATERA,M.,AND MAURINO, A. 2004. Model-driven Web usage analysis for the
evaluation of Web application quality. J. Web Eng. 3, 2, 124–152.
GACKOWSKI, Z. 2006. Redefining information quality: the operations management approach. In Proceedings
of the 11th International Conference on Information Quality (ICIQ). 399–419.
HAMMER, M. 1990. Reengineering work: Don’t automate, obliterate. Harvard Bus. Rev. 104–112.
, J. 2001. Reengineering the Corporation: A Manifesto for Business Revolution,
Harper Collins.
HERNANDEZ,M.AND STOLFO, S. 1998. Real-world data is dirty: Data cleansing and the merge/purge problem.
J. Data Min. Knowl. Dis. 1,2.
ISAKOWITZ, T., BIEBER, M., AND VITALI, F. 1998. Web information systems - introduction. Commun. ACM 41,7,
ISAKOWITZ, T., STOHR, E., AND BALASUBRAMANIAN, P. 1995. RMM: A methodology for structured hypermedia
design. Comm. ACM 58,8.
ISTAT. 2004. Guidelines for the data quality improvement of localization data in public administration (in
JARKE, M., LENZERINI, M., VASSILIOU,Y.,AND VASSILIADIS, P., Eds. 1995. Fundamentals of Data Warehouses.
Springer Verlag.
JEUSFELD, M., QUIX,C.,AND JARKE, M. 1998. Design and analysis of quality information for data warehouses.
In Proceedings of the 17th International Conference on Conceptual Modeling.
KERR,K.AND NORRIS, T. 2004. The development of a healthcare data quality framework and strategy. In
Proceedings of the 9th International Conference on Information Quality.
KETTINGER,W.AND GROVER, V. 1995. Special section: Toward a theory of business process change manage-
ment. J. Manag. Inform. Syst. 12, 1, 9–30.
KOVAC,R.AND WEICKERT, C. 2002. Starting with quality: Using TDQM in a start-up organization. In Pro-
ceedings of the 7th International Conference on Information Quality (ICIQ). Boston, 69–78.
LEE, Y. W., STRONG, D. M., KAHN, B. K., AND WANG, R. Y. 2002. AIMQ: A methodology for information quality
assessment. Inform. Manage. 40, 2, 133–460.
LENZERINI, M. 2002. Data integration: A theoretical perspective. In Proceedings of the 21st ACM Sympo-
sium on Principles of Database Systems (PODS).
LIU,L.AND CHI, L. 2002. Evolutionary data quality. In Proceedings of the 7th International Conference on
Information Quality.
LONG,J.AND SEKO, C. April 2005. A cyclic-hierarchical method for database data-quality evaluation and im-
provement. In Advances in Management Information Systems-Information Quality Monograph (AMIS-
IQ) Monograph, R. Wang, E. Pierce, S. Madnick, and Fisher C.W.
LOSHIN, D. 2004. Enterprise Knowledge Management - The Data Quality Approach. Series in Data Man-
agement Systems, Morgan Kaufmann, chapter 4.
LYMAN,P.AND VARIAN, H. R. 2003. How much information.
MECCA, G., ATZENI, P., MASCI, M., MERIALDO,P.,AND SINDONI, G. 1998. The Araneus Web-based management
system. In Proceedings of the ACM SIGMOD International Conference on Management of Data,L.M.
Haas and A. Tiwary, Eds. ACM Press, 544–546.
MECCA, G., MERIALDO, P., ATZENI,P.,AND CRESCENZI, V. 1999. The (short) araeneus guide to Web site develop-
ment. In Proceedings of the 2nd International Workshop on the Web and Databases (WebDB) Conjunction
with Sigmod.
MOTRO,A.AND ANOKHIN, P. 2005. Fusionplex: Resolution of data inconsistencies in the data integration of
heterogeneous information sources. Inform. Fusion, 7, 2, 176–196.
, S., WITHMAN,L., AND CHERAGHI, S. H. 1999. Business process re-engineering : a consolidated method-
ology. In Proceedings of the 4th annual International Conference on Industrial Engineering Theory, Ap-
plications and Practice.
NADKARNI, P. 2006. Delivering data on time: The assurant health case. In Proceedings of the 11th Interna-
tional Conference on Information Quality.
NAUMAN N, F. 2002. Quality-driven query answering for integrated information systems. Lecture Notes in
Computer Science, vol. 2261.
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
16:52 C. Batini et al.
NELSON, J., POELS, G., GENERO, M., AND PIATTINI,EDS. 2003. Proceedings of the 2nd International Workshop
on Conceptual Modeling Quality (IWCMQ). Lecture Notes in Computer Science, vol. 2814, Springer.
OAKLAND, J. 1989. Total Quality Management. Springer.
OFFICE OF MANAGEMENT AND BUDGET. 2006. Information quality guidelines for ensuring and max-
imizing the quality, objectivity, utility, and integrity of information disseminated by agencies.
PERNICI,B.AND SCANNAPIECO, M. 2003. Data quality in Web information systems. J. Data Semant. 1, 48–68.
PIPINO, L., LEE,Y.,AND WANG, R. 2002. Data quality assessment. Commun. ACM 45,4.
RAGHUNATHAN, S. 1999. Impact of information quality and decision-maker quality on decision quality: a
theoretical model and simulation analysis. Decis. Supp. Syst. 26, 275–286.
ULLER, D., HONG-HAI, D., GOLOVIN,N.,AND KIRSTEN, T. June 2005. iFuice information
fusion utilizing instance correspondences and peer mappings. In Proceedings of the 8th International
Workshop on the Web and Databases (WebDB). located with SIGMOD.
RAO, R. 2003. From unstructured data to actionable intelligence. IT Professional 535, 6, 29–35.
REDMAN, T. 1996. Data Quality for the Information Age. Artech House.
REDMAN, T. 1998. The impact of poor data quality on the typical enteprise. Comm. ACM 41, 2, 79–82.
tecture: a platform for exchanging and improving data quality in Cooperative Information Systems.
Inform. Syst. 29, 7, 551–582.
SCANNAPIECO, M., PERNICI,B.,AND PIERCE, E. 2002. IP-UML: Towards a Methodology for Quality Improve-
ment based on the IP-MAP Framework. In Proceedings of the 7th International Conference on Information
Quality (ICIQ). Boston.
SCANNAPIECO, M., PERNICI,B.,AND PIERCE, E. 2005. IP-UML: A methodology for quality improvement-based
on IP-MAP and UML. In Information Quality, Advances in Management Information Systems, Informa-
tion Quality Monograph (AMIS-IQ), R. Wang, E. Pierce, S. Madnik, and C. Fisher, Eds.
SESSIONS, V. 2007. Employing the TDQM methodology: An assessment of the SC SOR. In Proceedings of
the 12th International Conference on Information Quality. 519–537.
SHANKARANARAYAN, G., WANG,R.Y.,AND ZIAD, M. 2000. Modeling the manufacture of an information product
with IP-MAP. In Proceedings of the 6th International Conference on Information Quality (ICIQ 2000).
SHANKARANARAYANAN,G.AND WANG, R. 2007. IPMAP: Current state and perspectives. In Proceedings of the
12th International Conference on Information Quality.
SHENG, Y. 2003. Exploring the mediating and moderating effects of information quality on firm’s endeavour
on information systems. In Proceedings of the 8th International Conference on Information Quality 2003
(ICIQ). 344–352.
SHENG,Y.AND MYKYTYN, P. 2002. Information technology investment and firm performance: A perspective
of data quality. In Proceedings of the 7th International Conference on Information Quality (ICIQ).DC,
STOICA, M., CHAWAT,N.,AND SHIN, N. 2003. An investigation of the methodologies of business process reengi-
neering. In Proceedings of Information Systems Education Conference.
,Y.AND JIN, Z. 2004. A methodology for information quality assessment in the designing and manufactur-
ing processes of mechanical products. In Proceedings of the 9th International Conference on Information
Quality (ICIQ). 447–465.
US DEPARTMENT OF DEFENSE. 1994. Data administration procedures. DoD rep. 8320.1-M.
modeling, design, control and execution of ETL processes. Inform. Syst. 26, 537–561.
VERMEER, B. 2000. How important is data quality for evaluating the impact of edi on global supply chains.
In Proceedings of the 33rd Haway Conference on Systems Sciences.
WAND,Y.AND WANG, R. 1996. Anchoring data quality dimensions in ontological foundations. Comm.
ACM 39, 11.
WANG, R. 1998. A product perspective on total data quality management. Comm. ACM 41,2.
WANG,R.AND STRONG, D. 1996. Beyond accuracy: What data quality means to data consumers. J. Manage.
Inform. Syst. 12,4.
WORLD WIDE WEB CONSORTIUM. Web accessibility initiative.
ZACHMAN, J. 2006. Zachman institute for framework advancement (ZIFA).
Received December 2006; revised December 2007; accepted May 2008
ACM Computing Surveys, Vol. 41, No. 3, Article 16, Publication date: July 2009.
... Processdriven strategies improve quality by redesigning the processes that create or modify data. As an example, a process can be redesigned by including an activity that controls the format of data before storage [13]. ...
... In this chapter, DQ assessments methodologies are presented (see Table 16). These methodologies consist of methods described in the last two chapters, that is, each one uses a combination of data quality dimensions to describe data quality along with different strategies and techniques to improve data quality [13]. ...
... Some may be able to apply to modern implementations as theoretical guidelines. The broad differences in focus across methodologies can be recognized by classifying methodologies into four categories, as shown in Figure 8 [13]. ...
... • Extensional conciseness (data level): meaning the absence of duplicated entities and relations in the KG. Batini et al. [2] refer to conciseness as "uniqueness." They define it as the degree to which data is devoid of duplicates in areas of breadth, depth, and scope. ...
... The report aims at answering the following questions: (1) What is the current maturity level? (2) What are the properties, that the reviewers have recommended? (3) How many reviews in total are needed for a comparison? ...
... A variety of studies were conducted to identify di↵erent quality issues within the data integration process. Most of these studies agree that data quality faces numerous challenges [Batini et al., 2009, Warth et al., 2011. Indeed, ETL is a crucial part of the data warehousing process where most data cleansing and curation are carried out. ...
... Data analytics has been used for many years to support business decision-making. Several studies uphold that poor data quality has direct and indirect impacts on the underlying business decisions [Batini et al., 2009, Warth et al., 2011. According to Redman [Redman, 1998], at least three proprietary studies have provided estimates of poor data quality costs between 8 and 12% of the revenue range. ...
L'émergence des données issues du web et de ses corollaires entraîne une révolution digitale qui touche tous les aspects d'une entreprise. L'évolution de ces données en terme de volume, variété et vélocité explique la nécessité d'intégrer dans les systèmes existants des services axés sur le Big Data. Dans les systèmes Business Intelligence & Analytics (BI&A), les décideurs utilisent les bases de données (BD) pour extraire des informations pertinentes afin d'améliorer la prise de décision. Ces systèmes sont aussi impactés par cette révolution digitale. En effet, à l'ère du Big Data, les BD NoSQL sont devenues omniprésentes en tant que systèmes hautement extensibles et sans schéma pour stocker des données de volume, variété et vélocité importants. Quoique NoSQL est largement utilisé aujourd'hui, son exploitation est limitée dans le cadre de la BI&A qui reste, pendant de longues années, liée essentiellement aux systèmes de gestion de BD conventionnels telles que les BD relationnelles.En revanche, dès les premiers jours de l'entreposage des données, le modèle relationnel a été fondamental dans la quête de la cohérence et de la qualité des données analytiques dans les systèmes BI&A. Pour obtenir de meilleures performances, le modèle NoSQL offre une grande variété structurelle et une flexibilité accrue de schémas et abandonne certaines règles comme les contraintes d'intégrité. Par conséquent, l'exploitation des données sans schéma, et souvent sans contraintes d'intégrité pour la prise de décision nécessite de revoir toutes les phases de l'architecture BI&A, notamment le processus Extract-Transform-Load (ETL) afin de les adapter au volume, variété et vélocité des données, telles que les BD orientées documents.À titre d'exemple, dans le processus ETL, joindre plusieurs collections en l'absence d'un schéma défini au préalable est un défi important. Détecter ces candidats manuellement s'avère laborieux, très coûteux en temps et infaisable dans les ensembles de données à grande échelle.L'objectif principal de cette thèse est d'explorer comment extraire, transformer et charger des BD NoSQL, particulièrement des BD orientées documents, pour des fins décisionnelles ? et comment préparer ces données variées et volumineuses pour répondre aux besoins des décideurs ?Dans un premier temps, nous avons mené une étude approfondie de la littérature sur ces problématiques. Cette revue nous a conduits à introduire une nouvelle approche de BI&A permettant d'extraire, transformer et de charger à la demande les données requises pour l'analyse OLAP à partir de BD orientées documents.Nous nous concentrons sur l'ETL à la demande où, contrairement aux travaux existants, nous considérons la dispersion des données sur deux ou plusieurs dans les BD orientées documents. Dans un second temps, nous étudions le problème de la découverte automatique des attributs clés de jointure. Nous proposons un algorithme qui vise à détecter automatiquement les identifiants et les références composés et non composés à partir de plusieurs sources de données orientées documents.L'approche est basée sur des caractéristiques et des règles d'élagage pour trouver les identifiants candidats. Pour trouver les pairs (identifiant, référence) entre chaque deux collections, nous avons mis en pratique node2vec, technique de plongement de réseau qui offre des avantages significatifs en utilisant les similarités syntaxiques et sémantiques. Pour illustrer notre étude, nous avons développé des prototypes comportant les deux niveaux : (i) détection des identifiants candidats ; et (ii) identification des paires candidates des attributs clés. L'étude expérimentale est basée sur deux benchmarks TPC-H et TPC-E et deux sources de données réelles Twitter et Musicians. Nous présentons les résultats qui montrent la faisabilité et la pertinence de notre approche et nous discutons les défis à relever dans nos travaux futurs.
Algorithms and technologies are essential tools that pervade all aspects of our daily lives. In the last decades, health care research benefited from new computer-based recruiting methods, the use of federated architectures for data storage, the introduction of innovative analyses of datasets, and so on. Nevertheless, health care datasets can still be affected by data bias. Due to data bias, they provide a distorted view of reality, leading to wrong analysis results and, consequently, decisions. For example, in a clinical trial that studied the risk of cardiovascular diseases, predictions were wrong due to the lack of data on ethnic minorities. It is, therefore, of paramount importance for researchers to acknowledge data bias that may be present in the datasets they use, eventually adopt techniques to mitigate them and control if and how analyses results are impacted. This paper proposes a method to address bias in datasets that: (i) defines the types of data bias that may be present in the dataset, (ii) characterizes and quantifies data bias with adequate metrics, (iii) provides guidelines to identify, measure, and mitigate data bias for different data sources. The method we propose is applicable both for prospective and retrospective clinical trials. We evaluate our proposal both through theoretical considerations and through interviews with researchers in the health care environment.
The quality of the healthcare environment has become an essential factor for healthcare users to access quality services. Smart healthcare systems use the Internet of Medical Things (IoMT) devices to capture patients’ health data for treatment or diagnostic purposes. This sensitive collected patient data is shared between the different stakeholders across the network to provide quality services. Due to this, healthcare systems are vulnerable to confidentiality, integrity and privacy threats. In the COVID-19 scenario, when collaborative medical consultation is required, the quality assessment of the framework is essential to protect the privacy of doctors and patients. In this paper, a ring signature-based anonymous authentication and quality assessment scheme is designed for collaborative medical consultation environments for quality assessment and protection of the privacy of doctors and patients. This scheme also uses a new KMOV Cryptosystem to ensure the quality of the network and protect the system from different attacks that hamper data confidentiality.
The United Nations prescribed the Sustainable Development Goals (SDGs) to various nations to provide enduring answers to widespread problems and to give long-lasting solutions to common issues being faced across the globe. SDG 5 in particular was aimed at minimizing gender inequality by employing 9 targets and 14 indicators. The indicators serve as a yardstick to measure the progress of each of the 9 targets. This research takes an in-depth look at the perspectives of SDG 5 –Gender Inequalities, its targets, and indicators. Furthermore, explanatory data analysis and numerical association rule mining alongside QuantMiner are applied to the generated Indian datasets on SDG 5 to extract patterns and associations among the fourteen indicators of SDG 5. The association rule mining carried out on the indicators reveals the pattern of association among these indicators. Legal provision for women and the rate of crimes against women have a perfect association of 100% while the association between legal provision for women and women who have experienced physical violence stands at 80%. The full relationships of all the 14 indicators are discussed extensively in the result and discussion section. Overall, it is established that these indicators are interdependent. This will make it easier for academics, the general public, and governmental and non-governmental organizations to understand the trends and form informed opinions on issues relating to gender inequality and SDG 5.
Background: In Australia, aged care and disability service providers are legally required to maintain comprehensive and accurate clinical documentation to meet regulatory and funding requirements and to support safe and high quality care provision. However, evidence suggests poor quality clinical data and documentation is widespread across the sector and can significantly affect clinical decision-making and care delivery and increase business costs. Objective: The OPTIMISE study uses an Agile Lean Six Sigma framework to: 1) identify opportunities for optimisation of clinical documentation processes and clinical information systems, 2) implement and test optimisation solutions, and 3) evaluate outcomes post-optimisation, in a large post-acute community-based health service providing aged care and disability services in Western Australia. Methods: A three-stage prospective optimisation study will be undertaken. Stage 1 (Baseline) will measure existing clinical data quality, identify root causes of data quality issues across services, and generate optimisation solutions. Stage 2 (Optimisation) will implement and test changes to clinical documentation processes and information systems using incremental Agile sprints, and Stage 3 (Evaluation) will evaluate change in primary and secondary outcomes from baseline to 12 months post-optimisation. The primary outcome is data quality measured in terms of Defects Per Unit (DPU), Defects Per Million Opportunities (DPMO) and Sigma level. Secondary outcomes are care delivery (direct care time), clinical incidents, business outcomes (cost of quality, workforce productivity), and user satisfaction. Case studies will be analysed to understand impacts of optimisation on clinical outcomes and business processes. Results: As of 1 June 2022, Stage 1 commenced with baseline data quality audits conducted to measure current data quality. Baseline data quality audits will be followed by user consultations to identify root causes of data quality issues. Optimisation solutions will be developed by January 2023 to inform optimisation (Stage 2) and evaluation (Stage 3). Conclusions: Study findings will be of interest to individuals and organisations in the healthcare sector seeking novel solutions to improve the quality of clinical data and support high quality care delivery and reduce business costs. Clinicaltrial: N/A.
Full-text available
Many information quality initiatives and projects need to demonstrate the potential benefits of their IQ-related activities already in their planning stage. In doing so, practitioners rely on cost estimates based on current non-quality data effects (that are then compared to data quality improvement costs). In producing such estimates on costs caused by low quality data, it is difficult to identify all potential negative monetary effects that are the result of low quality data (as well as all possible costs associated with assuring high quality data and their progression). Consequently, this article reviews and categorizes the potential costs associated with low quality data and examines their progression. This analysis can help practitioners to identify cost saving potentials and argue a more convincing business case of their data quality imitative. For researchers, the proposed classification framework and the cost progression analyses can be helpful to develop quantifiable measures of data quality costs and to prepare – subsequently – benchmarking studies, comparing different cost levels in different organizations. Thus, the paper contributes elements of a future cost-benefit analysis method for data quality investments.
Full-text available
Companies continue to reexamine and fundamentally change the way they do business. Intense competitive pressures and a sluggish economy provide the motivation for continued efforts to "deliver more with less." Properly executed, reengineering can be an effective tool for organizations striving to operate as effectively and efficiently as possible. This study examines various methodologies of business process reengineering (BPR) and the reasons for failure of BPR efforts. Our examination of BPR research shows that companies need a BPR methodology that takes a holistic and systematic approach.
In this chapter, we are concerned with exploring large unfamiliar data sets inexpensively, to learn characteristics of the data set. First, we introduce an example that will be used throughout the book to informally motivate the concepts of uncertainty, random variables and probability distributions. Our focus is on relating these concepts to exploratory data mining. Next, we introduce the concept of Exploratory Data Mining (EDM) and list the characteristics of a good EDM technique. We then discuss summaries (estimates) such as means, variances, medians and quantiles. We present complex estimates like histograms and the empirical cumulative distribution function (ECDF) that capture the variation in attributes across the attribute space. We discuss the challenges of EDM in higher dimensions, and we also discuss multivariate histograms that have linear boundaries, parallel to the original variable axes, that is, axis-aligned. Next, we discuss data depth, its variations and using depth to order data points in higher dimensions. We conclude with a discussion of the role of data depth and multivariate depth which play an important role in multivariate binning.
Measuring and improving data quality in an organisation or in a group of interacting organisations is a complex task. Several methodologies have been developed in the past, providing a basis for the definition of a data quality programme that guarantees high data quality levels. Since the main limitation of existing approaches is their specialisation on specific issues or contexts, this paper presents a Comprehensive Data Quality (CDQ) methodology. The main aim of the CDQ methodology is the integration and enhancement of the phases, techniques and tools proposed by previous approaches. In particular, the CDQ methodology is conceived to be at the same time complete, flexible and simple to apply. Completeness is achieved by considering an existing techniques and tools and integrating them in a framework that can work in any organisation. The methodology is flexible, since it supports the user in the selection of the most suitable techniques and tools within each phase and in any context. Finally, CDQ is simple, since it is organised in phases and each phase is characterised by a specific goal and a set of techniques to apply. The methodology is explained by means of a running example and significant cases of its application are reported.
This article discusses the existing situation and some of the expected changes in National Health Services, to provide a background for the other articles in this issue of "The Statistician".