Download presentation
Presentation is loading. Please wait.
1
Data Quality Class 4
2
Goals Discuss Project Midterm Statistical Process Control Data Quality Rules
3
Project Informtion is now on web site Final version is due on July 26 Data will be available by end of the week We will spend some time discussing goals today
4
Midterm Written exam on July 5 th Will cover: –Cost of low data quality –Dimensions of data quality –domains and mappings –SPC –Data Quality Rules
5
Statistical Process Control Developed by Shewhart at Bell Labs in the 1920’s through 1950’s Notions of Variation vs. Control Important in original context of both equpiment manufacture and service quality
6
Variation Natural variations Defects Errors Mistakes Some variations are meaningful, some are not
7
Causes of Variation Common, or Chance causes –minor fluctuations or differences –not necessarily important to correct –observed to form a normal distribution Assignable, or Special causes –(self explanatory) We expect to see the normal variations, but assignable cause variations are interesting
8
Example Measure railroad on-time performance –Trains are typically on time or a few minutes late –One night, the trains are all 1 hour late due to electrical problems – a special cause
9
Statistical Control State in which variations observed can be attributed to common causes that do not change with time
10
Pareto Principle In a population that contributes to a common effect, relaively few of the contributors account for the bulk of the effect Example: code performance analysis Can be used to direct analysis
11
Control Chart
12
Control Chart 2 Used to look for distinct variations from the mean Goal: predictable behavior Plot series of data over time Variations are represented as distance from the mean
13
Control Chart 3 Center Line: can be computed as mean of variable points Upper Contril Limit: three standard deviations above center line Lower Control Limit: three standard deviations below center line
14
Control Chart 4 As long as all points are between UCL and LCL, the variations are due to common causes, and the process is said to be in control, or stable Points above UCL or below LCL are indicative of abnormal variation, and are due to special causes – the process is not in control
15
Control Chart 5 Select variables chart or attributes chart Use data quality dimensions as guideline Select meaningful variables to measure (i.e., stuff that will point at a diagnosible problem)
16
Interpreting the Control Chart Lack of stability indicates potential problem Look for: –points utside of control limits –zone testing (clusters of points within certain standard deviation limits) –potential to split out data points into different logical data sets Look for cycles
17
SPC and Data Quality “The Information Factory” Use data quality dimensions as guideline for investigation Analyze the state of data as it passes through the information chain Probing can be automated with data quality rules
18
Inserting the Probes FInd a location in information chain that is: –nondisruptive –easy to access –easy to retool
19
Data Quality Rules Definitions Proscriptive Assertions Prescriptive Assertions Conditional Assertions Operational Assertions
20
Definitions Nulls Domains Mappings
21
Proscriptive Assertions Describe what is not allowed Used to figure out what is wrong with data Used for validation
22
Prescriptive Assertions Describe what is supposed to happen with data Can be used for data population, extraction, transformation Can also be used for validation
23
Conditional Assertions Define an assertion that must be true if a condition is true
24
Operational Assertions Define an action that must be taken if a condition is true
25
9 Classes of Rules 1) Null value rules 2) Value rules 3) Domain membership rules 4) Domain Mappings 5) Relation rules 6) Table, Cross-table, and Cross-message assertions 7) In-Process directives 8) Operational Directives 9) Other rules
26
Null Value Rules Null value specification –Define GETDATE for unavailable as “fill in date” Null values allowed –Attribute A allowed nulls {GETDATE, U, X} Null values not allowed –Attribute B nulls not allowed
27
Value Rules Value restriction rule Restrict GRADE: value >= ‘A’ AND value <= ‘F’ AND value != ‘E’
28
Domain Rules Domain Definition Domain Membership Domain Nonmembership Domain Assignment
29
Mapping Rules Mapping definition Mapping membership Mapping nonmembership
30
Relation Rules Completeness Exemption Consistency Derivation
31
Completeness Defines when a record is complete (I.e., what fields must be present) IF (Orders.Total > 0.0), Complete With {Orders.Billing_Street, Orders.Billing_City, Orders.Billing_State, Orders.Billing_ZIP}
32
Exemption Defines which fields may be missing IF (Orders.Item_Class != “CLOTHING”) Exempt {Orders.Color, Orders.Size }
33
Consistency Define a relationship between attributes based on field content –IF (Employees.title == “Staff Member”) Then (Employees.Salary >= 20000 AND Employees.Salary < 30000)
34
Derivation Prescriptive form of consistency rule Details how one attribute’s value is determined based on other attributes IF (Orders.NumberOrdered > 0) Then { Orders.Total = (Orders.NumberOrdered * Orders.Price) * 1.05 }
35
Table and Cross-Table Rules Functional Dependence Primary Key Assertion Foreign Key Assertion (=referential integrity)
36
Functional Dependence Functional Dependence between columns X and Y: –For any two records R1 and R2 in a table, if field X of record R1 contains value x and field X of record R2 contains the same value x, then if field Y of record R1 contains the value y, then field Y of record R2 must contain the value y. In other words, attribute Y is said to be determined by attribute X.
37
Primary Key Assertion A set of attributes defined as a primary key must uniquely identify a record Enforcement = testing for duplicates across defined key set
38
Foreign Key Assertion When the values in field f in table T is chosen from the key values in field g in table S, field S.g is said to be a foreign key for field T.f If f is a foreign key, the key must exist in table S, column g (=referential integrity)
39
In-process Directives Definition directives (labeling information chain members) Measurement directives Trigger directives
40
Operational Directives Transformation Update
41
Other Rules Approximate Searching rules Approximate Matching rules
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.