Interpretability

Characteristic Name:	Interpretability
Dimension:	Usability and Interpretability
Description:	Data should be interpretable
Granularity:	Information object
Implementation Type:	Process-based approach
Characteristic Type:	Usage

Verification Metric:

The number of tasks failed or under performed due to the lack of interpretability of data

The number of complaints received due to the lack of interpretability of data

GuidelinesExamplesDefinitons

The implementation guidelines are guidelines to follow in regard to the characteristic. The scenarios are examples of the implementation

Guidelines:	Scenario:
Standardise the interpretation process by clearly stating the criteria for interpreting results so that an interpretation on one dataset is reproducible	(1) 10% drop in production efficiency is a severe decline which needs quick remedial actions
Facilitate the interaction process based on users' task at hand	(1) A traffic light system to indicate the efficiency of a production line to the workers, a detail efficiency report to the production manage, a concise efficiency report for production line supervisors
Design the structure of information in such a way that further format conversions are not necessary for interpretations.	(1) A rating scale of (poor good excellent ) is better than (1,2,3) for rate a service level
Ensure that information is consistent between units of analysis (organisations, geographical areas, populations in concern etc.) and over time, allowing comparisons to be made.	(1) Number of doctors per person is used to compare the health facilities between regions. (2) Same populations are used over the time to analyse the epidemic growths over the tim
Use appropriate visualisation tools to facilitate interpretation of data through comparisons and contrasts	(1) Usage of tree maps , Usage of bar charts, Usage of line graphs

Validation Metric:

How mature is the process to maintain the interpretability of data

These are examples of how the characteristic might occur in a database.

Example:	Source:
when an analyst has data with freshness metric equals to 0, does it mean to have fresh data at hand? What about freshness equals to 10 (suppose, we do not stick to the notion proposed in [23])? Is it even fresher? Similar issues may arise with the notion of age: e.g., with age A(e) = 0, we cannot undoubtedly speak about positive or negative data characteristic because of a semantic meaning of “age” that mostly corresponds to a neutral notion of “period of time”	O. Chayka, T. Palpanas, and P. Bouquet, “Defining and Measuring Data-Driven Quality Dimension of Staleness”, Trento: University of Trento, Technical Report # DISI-12-016, 2012.
Consider a database containing orders from customers. A practice for handling complaints and returns is to create an “adjustment” order for backing out the original order and then writing a new order for the corrected information if applicable. This procedure assigns new order numbers to the adjustment and replacement orders. For the accounting department, this is a high-quality database. All of the numbers come out in the wash. For a business analyst trying to determine trends in growth of orders by region, this is a poor-quality database. If the business analyst assumes that each order number represents a distinct order, his analysis will be all wrong. Someone needs to explain the practice and the methods necessary to unravel the data to get to the real numbers (if that is even possible after the fact).	J. E. Olson, “Data Quality: The Accuracy Dimension”, Morgan Kaufmann Publishers, 9 January 2003.

Example:

Source:

when an analyst has data with freshness metric equals to 0, does it mean to have fresh data at hand? What about freshness equals to 10 (suppose, we do not stick to the notion proposed in [23])? Is it even fresher? Similar issues may arise with the notion of age: e.g., with age A(e) = 0, we cannot undoubtedly speak about positive or negative data characteristic because of a semantic meaning of “age” that mostly corresponds to a neutral notion of “period of time”

O. Chayka, T. Palpanas, and P. Bouquet, “Defining and Measuring Data-Driven Quality Dimension of Staleness”, Trento: University of Trento, Technical Report # DISI-12-016, 2012.

Consider a database containing orders from customers. A practice for handling complaints and returns is to create an “adjustment” order for backing out the original order and then writing a new order for the corrected information if applicable. This procedure assigns new order numbers to the adjustment and replacement orders. For the accounting department, this is a high-quality database. All of the numbers come out in the wash. For a business analyst trying to determine trends in growth of orders by region, this is a poor-quality database. If the business analyst assumes that each order number represents a distinct order, his analysis will be all wrong. Someone needs to explain the practice and the methods necessary to unravel the data to get to the real numbers (if that is even possible after the fact).

J. E. Olson, “Data Quality: The Accuracy Dimension”, Morgan Kaufmann Publishers, 9 January 2003.

The Definitions are examples of the characteristic that appear in the sources provided.

Definition:	Source:
Comparability of data refers to the extent to which data is consistent between organisations and over time allowing comparisons to be made. This includes using equivalent reporting periods.	HIQA 2011. International Review of Data Quality Health Information and Quality Authority (HIQA), Ireland. http://www.hiqa.ie/press-release/2011-04-28-international-review-data-quality.
Data is not ambiguous if it allows only one interpretation – anti-example: Song.composer = ‘Johann Strauss’ (father or son?).	KIMBALL, R. & CASERTA, J. 2004. The data warehouse ETL toolkit: practical techniques for extracting. Cleaning, Conforming, and Delivering, Digitized Format, originally published.
Comparability aims at measuring the impact of differences in applied statistical concepts and measurement tools/procedures when statistics are compared between geographical areas, non-geographical domains, or over time.	LYON, M. 2008. Assessing Data Quality , Monetary and Financial Statistics. Bank of England. http://www.bankofengland.co.uk/ statistics/Documents/ms/articles/art1mar08.pdf.
The most important quality characteristic of a format is its appropriateness. One format is more appropriate than another if it is better suited to users’ needs. The appropriateness of the format depends upon two factors: user and medium used. Both are of crucial importance. The abilities of human users and computers to understand data in different formats are vastly different. For example, the human eye is not very good at interpreting some positional formats, such as bar codes, although optical scanning devices are. On the other hand, humans can assimilate much data from a graph, a format that is relatively hard for a computer to interpret. Appropriateness is related to the second quality dimension, interpretability.	REDMAN, T. C. 1997. Data quality for the information age, Artech House, Inc.

Completeness of mandatory attributes

Characteristic Name:	Completeness of mandatory attributes
Dimension:	Completeness
Description:	The attributes which are mandatory for a complete representation of a real world entity must contain values and cannot be null .
Granularity:	Element
Implementation Type:	Rule-based approach
Characteristic Type:	Declarative

Verification Metric:

The number of null values reported in a mandatory attribute per thousand records

GuidelinesExamplesDefinitons

The implementation guidelines are guidelines to follow in regard to the characteristic. The scenarios are examples of the implementation

Guidelines:	Scenario:
Specify which attributes are required to maintain a meaningful representation of an entity.	1) A sales order should at least have values for order number, Quantity, Price and Total (Sales order is the record)
Specify the states of an entity where the above identified attributes become mandatory values	(1)Order number quantity and total should be available as mandatory by the time order is created whereas price will become mandatory when the order is approved. (States :"Order created" "Order approved") (2) Product is retired and now has a product-last-available-date
Specify the dependencies of entities in operational context to identify the mandatory values	(1)Invoice number should exist to create a gate pass
Specify default values where possible	(1) Default country is Australia for those who fill the application from Australian IP addresses

Validation Metric:

How mature is the creation and implementation of the DQ rules to handle mandatory values

These are examples of how the characteristic might occur in a database.

Example:	Source:
1) Let us consider a Person relation with the attributes Name, Surname, BirthDate,and Email. The relation is shown in Figure 2.2. For the tuples with Id equalto2,3,and 4, the Email value is NULL. Let us suppose that the person represented by tuple 2 has no e-mail: no incompleteness case occurs. If the person represented by tuple 3 has an e-mail, but its value is not known then tuple 3 presents an incompleteness. Finally, if it is not known whether the person represented by tuple 4 has an e-mail or not, incompleteness may not be the case. ID 1 2 3 4 Name John Edward Anthony Marianne Surname Smith Monroe White Collins BirthDate 03/17/1974 02/03/1967 01/01/1936 11/20/1955 Email smith@abc.it NULL NULL NULL not existing existing but unknown not known if existing Fig. 2.2. The Person relation, with diﬀerent null value meanings for the e-mail attribute 2) if Dept is a relation representing the employees of a given department, and one speciﬁc employee of the department is not represented as a tuple of Dept, then the tuple corresponding to the missing employee is in ref(Dept),and ref(Dept) diﬀers from Dept in exactly that tuple.	C. Batini and M, Scannapieco, “Data Quality: Concepts, Methodologies, and Techniques”, Springer, 2006.
if a column should contain at least one occurrence of all 50 states, but the column contains only 43 states, then the population is incomplete.	Y. Lee, et al., “Journey to Data Quality”, Massachusetts Institute of Technology, 2006.
1) A database contains information on repairs done to capital equipment. How- ever, it is a known fact that sometimes the repairs are done and the information about the repair is just not entered into the database. This is the result of lack of concern on the part of the repair people and a lack of enforcement on the part of their supervisors. It is estimated that the amount of missing information is about 5%. This database is probably a good-quality database for assessing the general health of capital equipment. Equipment that required a great deal of expense to maintain can be identiﬁed from the data. Unless the missing data is disproportionately skewed, the records are usable for all ordinary decisions. However, trying to use it as a base for evaluating information makes it a low-quality database. The missing transactions could easily tag an important piece of equipment as satisfying a warranty when in fact it does not. 2) a BIRTH_DATE value left blank would not be accurate because all of us have birth dates.	J. E. Olson, “Data Quality: The Accuracy Dimension”, Morgan Kaufmann Publishers, 9 January 2003.

Example:

Source:

1) Let us consider a Person relation with the attributes Name, Surname, BirthDate,and Email. The relation is shown in Figure 2.2. For the tuples with Id equalto2,3,and 4, the Email value is NULL. Let us suppose that the person represented by tuple 2 has no e-mail: no incompleteness case occurs. If the person represented by tuple 3 has an e-mail, but its value is not known then tuple 3 presents an incompleteness. Finally, if it is not known whether the person represented by tuple 4 has an e-mail or not, incompleteness may not be the case.

ID 1

2 3 4

Name John

Edward Anthony Marianne

Surname Smith

Monroe White Collins

BirthDate 03/17/1974 02/03/1967 01/01/1936 11/20/1955

smith@abc.it NULL NULL NULL

not existing existing but unknown not known if existing

Fig. 2.2. The Person relation, with diﬀerent null value meanings for the e-mail attribute

2) if Dept is a relation representing the employees of a given department, and one speciﬁc employee of the department is not represented as a tuple of Dept, then the tuple corresponding to the missing employee is in ref(Dept),and ref(Dept) diﬀers from Dept in exactly that tuple.

C. Batini and M, Scannapieco, “Data Quality: Concepts, Methodologies, and Techniques”, Springer, 2006.

if a column should contain at least one occurrence of all 50 states, but the column contains only 43 states, then the population is incomplete.

Y. Lee, et al., “Journey to Data Quality”, Massachusetts Institute of Technology, 2006.

1) A database contains information on repairs done to capital equipment. How- ever, it is a known fact that sometimes the repairs are done and the information about the repair is just not entered into the database. This is the result of lack of concern on the part of the repair people and a lack of enforcement on the part of their supervisors. It is estimated that the amount of missing information is about 5%. This database is probably a good-quality database for assessing the general health of capital equipment. Equipment that required a great deal of expense to maintain can be identiﬁed from the data. Unless the missing data is disproportionately skewed, the records are usable for all ordinary decisions. However, trying to use it as a base for evaluating information makes it a low-quality database. The missing transactions could easily tag an important piece of equipment as satisfying a warranty when in fact it does not.

2) a BIRTH_DATE value left blank would not be accurate because all of us have birth dates.

J. E. Olson, “Data Quality: The Accuracy Dimension”, Morgan Kaufmann Publishers, 9 January 2003.

The Definitions are examples of the characteristic that appear in the sources provided.

Definition:	Source:
Domain Level: Data element is 1. Always required be populating and not defaulting; or 2. Required based on the condition of another data element. Entity Level: The required domains that comprise an entity exist and are not defaulted in aggregate.	B. BYRNE, J. K., D. MCCARTY, G. SAUTER, H. SMITH, P WORCESTER 2008. The information perspective of SOA design Part 6:The value of applying the data quality analysis pattern in SOA. IBM corporation.
A given data element (fact) has a full value stored for all records that should have a value.	ENGLISH, L. P. 2009. Information quality applied: Best practices for improving business information, processes and systems, Wiley Publishing.
Determined the extent to which data is not missing. For example, an order is not complete without a price and quantity.	G. GATLING, C. B., R. CHAMPLIN, H. STEFANI, G. WEIGEL 2007. Enterprise Information Management with SAP, Boston, Galileo Press Inc.
Completeness refers to the expectation that certain attributes are expected to have assigned values in a data set. Completeness rules can be assigned to a data set in three levels of constraints: 1. Mandatory attributes that require a value 3. Inapplicable attributes (such as maiden name for a single male), which may not have a value.2. Optional attributes, which may have a value.	LOSHIN, D. 2001. Enterprise knowledge management: The data quality approach, Morgan Kaufmann Pub.
An expectation of completeness indicates that certain attributes should be assigned values in a data set. Completeness rules can be assigned to a data set in three levels of constraints:1. Mandatory attributes that require a value, 2. Optional attributes, which may have a value based on some set of conditions, and 3. Inapplicable attributes, (such as maiden name for a single male), which may not have a value.	LOSHIN, D. 2006. Monitoring Data quality Performance using Data Quality Metrics. Informatica Corporation.