Editingglossary

UNITED NATIONS STATISTICAL COMMISSION
and
ECONOMIC COMMISSION FOR EUROPE

CONFERENCE OF EUROPEAN STATISTICIANS
METHODOLOGICAL MATERIAL

GLOSSARY OF TERMS ON
STATISTICAL DATA EDITING

UNITED NATIONS
Geneva, 2000

_______________________________________________________________________________________ iii

CONTENTS

PREFACE.........................................................................................................................................iv
EXECUTIVE SUMMARY ...............................................................................................................v

GLOSSARY OF TERMS ON STATISTICAL DATA EDITING......................................................................1

iv ________________________________________________________________________________________

PREFACE

The methodological material “Glossary of Terms on Statistical Data Editing” was prepared
based on the request of countries participating in the activities on Statistical Data Editing
organised by UN/ECE Statistical Division within the framework of the programme of work
of the Conference of European Statisticians. It represents an extensive voluntary effort on
the part of the participants of the UN/ECE Work Sessions on Statistical Data Editing.

The document was reviewed at the Work Session on Statistical Data Editing in June 1999.
National Statistical Offices of the UN/ECE member countries and the Food and Agriculture
Organisation (FAO) participated in this meeting. The material reflects the outcome of the
discussion on the document.

At its 2000 plenary session, the Conference of European Statisticians agreed to have this
document reproduced and distributed to interested statistical offices as methodological
material.

_______________________________________________________________________________________ v

EXECUTIVE SUMMARY

By its very nature, a glossary of terms will never be complete. This is particularly true when
it concerns a topic that is relatively new, as editing and imputation certainly is. This
notwithstanding, the Work Sessions on Statistical Data Editing under the Conference of
European Statisticians programme of work and its predecessor, the Joint Group on Data
Editing under the Statistical Computing Project of the United Nations Development
Programme, took it upon itself to come up with such a glossary as part of the programme of
work of the Conference of European Statisticians. The present document is the fruit of that
labour over the past decade. Many of the group’s members contributed, but particular thanks
are owed to Dania Ferguson (U.S. National Agricultural Statistics Service) and William
Winkler (U.S. Bureau of the Census) for their efforts in, respectively, initiating and
completing the work.

The reasons for starting this endeavour were varied. Some no longer hold, but most do. The
primary concern was that many different editing and imputation activities were often referred
to by the same name. By contrast, other activities carried many different names, even within
a single organization, let alone between different countries. In fact even the term ‘editing’
was, and still is, understood by many to mean many different things. Standardization of the
editing terminology is clearly needed. It is only when we have a common understanding of
the terminology that we can pursue meaningful and efficient discussion of the topic. We
hope this document will help experts working on data editing in national statistical offices.

This glossary is a repository of terms related to statistical data editing and imputation. It
contains definitions of concepts, principles, techniques as well as methods. The current
version defines over 180 terms. By making the glossary available to novices and experts
alike, it at the same time instructs as well as invites critique. It also promotes a common
understanding of terminology and thus initiates a framework for sharing expertise and
experience.

What this glossary does not contain are practical examples and specific solutions. In
particular, a list of computer systems that put into practice the defined techniques is notably
absent. Related terms need to be linked, missing terms added, outdated definitions modified.
As noted above, this is clearly not a complete document, but its usefulness will be derived
only if it is made public, and if the public is invited to make it grow. Proposed additional
terms, commentaries and examples are invited from readers in order to enable this ‘living
document’ to evolve in a way that would make it remain as current and up to date as possible.
The comments and suggestions should be sent to: info.stat@unece.org.

Glossary of Terms on Statistical Data Editing ________________________________________ 1

ACCEPTANCE REGION data adjustment results in a file of corrected
The set of acceptable values defined by the (edited/imputed) records with accompanying
edits for each record. For categorical data the messages to report on the computer actions taken to
acceptance region can be represented as a set of make the adjustments.
lattice points in N-Space. For numerical data it is a
set of convex regions in N-Space (N-dimensions of AUTOMATED DATA REVIEW may occur in a
real numbers). Also called a feasible region. batch or interactive fashion. It is important to note
that data entered in a heads-down fashion may later
ACCEPTANCE RULE be corrected in either a batch or an interactive data
Logical or arithmetic condition applied to a data review process.
item or data group that must be met if the data are to
be considered correct. AUTOMATED IMPUTATIONS generally fall
into one of six categories:
ACTIVE FIELD
A data item (field) for which some values of a. Deterministic imputation- where only one
this data item create a conflict in combination with correct value exists, as in the missing sum at the
values of other data items. bottom of a column of numbers. A value is thus
determined from other values on the same
ANALYSIS OF CORRECTION RULE questionnaire.
SPECIFICATIONS
Verifying consistency of correction rule b. Model based imputation - use of averages,
specifications, mainly in an extensive set of check medians, regression equations, etc. to impute a
and correction rules. value.

ANALYSIS OF EDIT RULE SPECIFICATIONS c. Deck imputation - A donor questionnaire is used
An activity by which the consistency of a set of to supply the missing value.
check rules is ascertained, implied (derived) check
rules are created, and an economical form Hot-deck imputation - a donor questionnaire is
(reduction) of specifications of the originally large found from the same survey as the
number of edit (check) rules is determined. questionnaire with the missing item. The
"nearest neighbour" search technique is often
ANALYTICAL EDITING used to expedite the search for a donor record.
Edit rule proceeding from a logical reasoning. In this search technique, the deck of donor
questionnaires comes from the same survey and
AUDIT TRAIL shows similarities to the receiving record,
A method of keeping track of changes to values where similarity is based on other data on the
in a field and the reason and source for each change. questionnaire that correlates to the data being
Audit trails are generally begun after the initial donated. For example: similar size and location
interview is completed. of farm might be used for donation of fuel
prices.
AUTOCORRECTION
Data correction performed by the computer Cold-deck imputation - same as hot deck
without human intervention. It makes particular use except that the data is found in a previously
of redundancy. Exclusion (elimination) of incorrect conducted similar survey.
records or substitution of a record or its part by data
from other records or the correction base. Auto- d. Mixed imputation - In most systems there
correction is generally done according to rules that usually is a mixture of categories used in some
assure the final (corrected) record fails no edits. fixed rank fashion for all items. For example,
first a deterministic approach is used. If it is not
AUTOMATED DATA ADJUSTMENTS occur as successful, then a hot deck approach is tried.
a result of computer actions. A desirable option in This is followed by a model based approach. If
-
any system allowing computer actions is to allow for all these approaches fail, then a manual
the overriding of those actions at some level. Batch

2 __________________________________________________ Glossary of Terms on Statistical Data Editing

imputation occurs through a human review neural networks allow to create an edit system
process. directly from the data and to develop the edits
over time through periodic retraining.
e. Expert Systems - An expert system is an
intelligent computer program that uses BALANCE EDIT
knowledge and inference procedures to solve An edit which checks that a total equals the sum
problems that are difficult enough to require of its parts. Also called an accounting edit.
significant human expertise for their solution. Example: Closing inventory = Opening Inventory +
Every expert system consists of two principal Purchases - Sales.
parts: the knowledge base and the inference
engine . The knowledge base contains both BATCH DATA REVIEW
factual and heuristic knowledge. Factual A review of many questionnaires in one batch
knowledge consists of items commonly agreed occurs after data entry. It generally results in a file
upon by spokesmen in a particular field. of error messages. This file may be printed for use
Heuris tic knowledge is the less rigorous, more in preparing corrections. The data records may be
experiential and more judgmental knowledge of split into two files. One file containing the "good"
performance or what commonly constitutes the records and one containing data records with errors.
rules of "good judgement" or the art of "good The other file can be corrected using an interactive
guessing" in a field. A wisely used process.
representation for the knowledge base is the rule
or if /then statement. The "if part" lists a set of CAI-COMPUTER-ASSISTED INTERVIEWING
conditions in some logical combination. Once uses the computer during interviewing. Any
the "if part" of the rule is satisfied, the "then part" contradictory data can be flagged by edit routines
can be concluded or problem solving action and the resultant data can be immediately adjusted
taken. Expert systems with knowledge by information from the respondent. An added
represented in rule form are called rule -based benefit is that data capture (key-entry) is occurring
systems . The inference engine makes inferences at interview time. CAI assists the interview in the
by determining which rules are satisfied by facts, wording of questions and tailors succeeding
ordering the satisfied rules, and executing the questions based on previous responses. CAI has
rule with the highest priority. been mainly used in Computer-Assisted Telephone
Interviews (CATI) or Computer-Assisted
Expert data editing systems make so-called Personal Interviewing (CAPI).
intelligent imputations based on a specified
hierarchy of methods to be used in imputing an CASIC
item. One item may use a deterministic approach The acronym “CASIC” stands for computer
followed by a hot-deck approach, while another assisted survey information collection. This
item might require a model based approach.
- encompasses computer assisted data collection and
Each item on the questionnaire would be resolved data capture. CASIC may be more broadly defined
according to its own hierarchy of approaches, the to include the use of computer assisted, automated,
next being automatically tried when the previous or advanced computing methods for data editing and
method has failed. imputation, data analysis and tabulation, data
dissemination, or other steps in the survey or census
f. Neural networks - information processing process.
paradigm based on the way the mammalian brain
processes information. The neural network is CHECKING RULE
composed of a large number of interconnected A logical condition or a restriction to the value
parallel processing elements tied together with of a data item or a data group which must be met if
weighted connections. These connection weights the data is to be considered correct. In various
store the knowledge necessary to solve specific connections other terms are used, e.g. edit rule.
problems. The network is prepared for solving a
problem by "training", i.e. the connection weights
are iteratively adjusted based on examples or a
verified set of input/output data. In data editing

_______________________________________________________________________________________ 3

CLASS ATTRIBUTE CHECK should be coded to specify whether the respondent
Verifying whether the value of the common was inaccessible or has refused, this information can
attributes (class attributes) of a logical unit or its be used in verification procedures.
components are identical.
COMPOSITION CHECK
CODE LIST Verifying whether the structure of a logical unit
LIST OF CODE WORDS (e.g. - household) is consistent with the definition
List of all allowed (admissible) values of a data (e.g., at least one adult).
item.
CONDITIONAL EDIT
CODE REDUNDANCY An edit where the value of one field determines
When a character or group of characters in a the editing relationship between other fields and
code word can be partially or completely deduced possibly itself. For example, suppose there are three
from the remaining characters of the code word. fields A, B, and C. A conditional edit would exist if
the relationship between fields B and C as expressed
CODE SPACE through the edits depended on the value in A.
A set of all combinations of admissible values
of a particular record of data. Cartesian product of CONFLICT RULE
the code word lists of individual data in a record. REJECTION RULE
A logical condition or a restriction to the value
CODE STRUCTURE VALIDATION of a data item or a data group which must not be met
CODE STRUCTURE CHECK if the data is to be considered correct.
Verifying whether the characters of the correct
type (e.g., digits, letters) are at the correct positions CONSISTENCY CHECK
of the code word. Detecting whether the value of two or more
data items are not in contradiction.
COLD-DECK
A correction base for which the elements are CONSISTENCY EDIT
given before correction starts and do not change A check for determinant relationships, such as
during correction. An example would be using prior parts adding to a total or harvested acres being less
year's data. A modified cold-deck may adjust cold- than or equal to planted acres.
deck values according to (possibly aggregate)
current information (see also HOT-DECK). CONSISTENCY ERROR
Occurrence of the values of two or more data
COMPLETE SET OF CONFLICT RULES items which do not satisfy some predefined
A set of explicitly given conflict rules and of all relationship between those data items.
implied conflict rules.
CONSISTENT EDITS
COMPLETE SET OF EDITS A set of edits which do not contradict each
The union of explicit edits and implied edits. other is considered to be consistent. If edits are not
Sufficient for the generation of feasible (acceptance) consistent, then no record can pass the edits.
regions for imputation (that is if the imputations are
to satisfy the edits). CORRECTION BASE
A set of correct data or records from which date
COMPLETENESS CHECKING or records are retrieved for imputation in the
... at survey level ensures that all survey data (probably) erroneous data or records.
have been collected. A minimal completeness check
compares the sample count to the questionnaire CORRECTION CYCLE
count to insure that all samples are accounted for, A cycle in which corrections (changes) are
even if no data were collected. made to a record according to an existing
edit/imputation strategy. If the record fails at the end
... at questionnaire level insures that routing of a cycle, it is sent through the edit/imputation
instructions have been followed. Questionnaires


software subroutines until it passes or until a pre- DATA ITEM
determined number of cycles has been exceeded. DATA FIELD
The specific sub-components of a data record.
CORRECTION RULE For instance, in a population census, specific data
A rule for correcting certain types of errors. Its items might be last name, first name, sex, and age.
general form is as follows: "If errors are detected by
the checks e 1 , ......, ek , make a correction in this way: DATA IMPUTATION
". Substitution of estimated values for missing or
inconsistent data items (fields). The substituted
CREATIVE EDITING values are intended to create a data record that does
A process whereby manual editors (i.e. those not fail edits.
doing the manual review) invent editing procedures
to avoid reviewing another error message from DATA REDUNDANCY
subsequent machine editing. When the value of data items (fields) can be
partially or completely deduced from the values of
DATA CAPTURE other data items (fields).
The process by which collected data are put in a
machine-readable form. Elementary edit checks are DATA REVIEW / DATA CHECKING
often performed in sub-modules of the software that Activity through which the correctness
does data capture. conditions of the data are verif ied. It also includes
the specification of the type of the error or condition
DATA CHECKING not met, and the qualification of the data and its
Activity through which the correctness division into the "error-free" and "erroneous" data.
conditions of the data are verified. It also includes Data checking may be aimed at detecting error-free
the specification of the type of the error or condition data or at detecting erroneous data. Data review
not met, and the qualification of the data and its consists of both error detection and data analysis,
division into the "error free" and "erroneous data". and can be carried out in manual or automated
Data checking may be aimed at detecting error-free mode.
data or at detecting erroneous data.
Data review/error detection may occur at many
DATA COLLECTION levels:
The process of gathering data. Data maybe
observed, measured, or collected by means of a) within a questionnaire
questioning, as in a survey or census response.
Item level / editing of individual data - the
DATA CORRECTION lowest logical level of checking and correction
CORRECTION OF ERRORS IN DATA during which the relationships among data
Activity of checking data which was declared items are not considered. Validations at this
(is possibly) erroneous. level are generally named "range checking".
Example: age must be between 0 and 120. In
DATA EDITING more complex range checks, the range may
The activity aimed at detecting and correcting vary by strata or some other identifier.
errors (logical inconsistencies) in data. Example: if strata = "large farm operation",
then the number of acres must be greater than
DATA IMPUTATION 500.
Substitution of estimated values for missing or
inconsistent data items (fields). The substituted Questionnaire level / editing of individual
values are intended to create a data record that does records - a logical level of checking and
not fail edits. correction during which the relationships
among data items in one record/questionnaire
are considered. Example 1. If married = 'Yes'
then age must be greater than 14. Example 2.

_______________________________________________________________________________________ 5

Sum of field acres must equal total acres in DEDUCTIVE IMPUTATION
farm. An imputation rule defined by a logical
reasoning, as opposed to a statistical rule.
Hierarchical - This level involves checking
items in sub-questionnaires. Data relationships DETECTION OF ERRORS IN DATA
of this type are known as "hierarchical data" (ERROR DETECTION)
and include situations such as questions about An activity aimed at detecting erroneous data.
an individual within a household. In this Usually predefined correctness criteria are used.
example, the common household information is
on one questionnaire and each individual's DETERMINISTIC CHECKING RULE
information is on a separate questionnaire. A checking rule which determines whether data
Checks are made to ensure that the sum of the items are incorrect with a probability of 1.
individual's data for an item does not exceed the
total reported for the household. DETERMINISTIC EDIT
An edit, which if violated, points to an error in
b) across questionnaires / editing of logical the data with a probability of one. Example: Age 5
units and Status = mother. Contrast with stochastic edit.
A logical level of checking and correction
during which the relationships among data in DETERMINISTIC IMPUTATION
two or more records are considered, namely in a The situation, given specific values of other
group of records that are logically coupled fields, when only one value of a field will cause the
together. The across questionnaire edits involve record to satisfy all of the edits. For instance, it
calculating valid ranges for each item from the might occur when the items that are supposed to add
survey data distributions or from historic data to a tota l do not add to the total. If only one item in
for use in outlier detection. Data analysis the sum is imputed, then its value is uniquely
routines that are usually run at summary time determined by the values of the other items. This
may easily be incorporated into data review at may be the first situation that is considered in the
this level. In this way, summary level errors are automated editing and imputation of survey data.
detected early enough to be corrected during the
usual error correction procedures. The "across DONOR (imputation)
questionnaire" checks should identify the In hot-deck edit/imputation, a donor is chosen
specific questionnaire that contains the from the set of edit-passing records based on its
questionable data. similarity to the fields in the record being donated to
(being imputed within). Values of fields (variables)
Across questionnaire level edits are generally in the donor are used to replace the corresponding
grouped into two types: statistical edits and macro contradictory or missing values in the edit-failing
edits. record that is receiving information. This type of
replacement may or may not assure that the imputed
DATA VALIDATION record satisfies edits.
An activity aimed at verifying whether the
value of a data item comes from the given (finite or EDIT RULE SPECIFICATION
infinite) set of acceptable values. For instance, a CHECK RULE SPECIFICATION
geographic code (field), say for a Canadian A set of check rules that should be applied in
Province, may be checked against a table of the given editing task.
acceptable values for the field.
EDITING BOUNDS
DATA VALIDATION ACCORDING TO A Bounds on the distribution of a measure used
LIST on survey data so that when raw, unedited data are
Verifying whether the data value is in the list of outside the bounds the data is subject to review and
acceptable values of this data item. possible correction (change).


EDITING EFFICIENCY EDITING OF INDIVIDUAL DATA
An efficient edit is good at identifying The lowest logical level of checking and
suspicious data. Edits that incorrectly flag large correction during which the relationships among
amounts of valid data are not very efficient. If such data items are not considered.
edits require an analyst to re-contact the respondent
to verify the data, the analyst will have less time to EDITING OF INDIVIDUAL RECORDS
perform other tasks, such as converting non- Logical level of checking and correction during
respondents. The hit rate may be used as an which the relationships among data items in one
indicator of editing efficiency. record are considered.

EDITING INDICATORS (editing flag/editing EDITING OF LOGICAL UNITS
code) A logical level of checking and correction
A flag or code that is added during the edit during which the relationships among data in two or
process. The flag might indicate that, for example: a more records are considered, namely in a group of
field in a record was targeted for change because it records that are logically coupled together.
failed an edit, the field in the record was changed, or
an override code was entered so that the edit system EDITING PROCEDURE
would no longer fail the record or the field in the The process of detecting and handling errors in
record. data. It usually includes three phases:
• the definition of a consistent system of
EDITING MATCH (matching fields/statistical requirements,
match) • their verification on given data, and
For hot-deck imputation, the fields in an edit- • elimination or substitution of data which is in
failing record that do not fail edits are matched contradiction with the defined requirements.
against edit-passing records. Often the edit-passing
record that is closest to the edit-failing record in EDITING RATIONALITY
terms of some metric is chosen as the donor record. An editing process that focuses on improving
If the edit-passing records are in random order, then the incoming data quality and hence the overall
the (possibly erroneous) assumption is that the quality of the survey data. This may include moving
donation is at random from a valid set of donors. most of the editing as close as possible to the
This type of matching is sometimes referred to as collection of data, limiting manual follow-up to
statistical matching. those flagged records with the heaviest potential
impact on estimates, and applying a total quality
EDIT(ING) MATRIX management approach to data editing.
A matrix used in editing. In hot-deck
imputation, the matrix contains information from ELECTRONIC QUESTIONNAIRE
edit-passing records that is used to donate Questions are in a software system that can be
information to (impute values in) the edit-failing answered by an individual. Examples might be a
record. Typically, there will be a variety of matrices software system that is on a laptop computer where
that correspond to the different sets of variables that respondents can answer questions directly into the
are matched on. The matrices are updated laptop (possibly without knowledge on the part of
continuously as a file of records is processed and interviewers of the details of the answers) or through
additional edit-passing records become available for queries on an Internet page.
updating the matrices.
ERROR DETECTING CHARACTER
EDITING MEASURE Character added to the basic characters of the
A formal measure such as the Hidiroglu- code word. Its relationship to the basic characters is
Berthelot statistic that allows delineation of a specified previously. The relationship is specified
targeted subset of records for manual follow-up. so that typical errors in the code word transmitted
The statistic might be a measure such as size that break this relationship.
allows the most important respondents to be
manually reviewed.

_______________________________________________________________________________________ 7

ERROR DETECTING CHARACTERS procedures, similar to a human with specialized
CHECK-DIGIT knowledge in that field.
Code or given set of code words whose
relationship to the set of valid codes is known, such EXPLICIT EDIT
that when a transcription error occurs, the An edit explicitly written by a subject matter
relationship is violated and the error is detected with specialist. (Contrast explicit edits with implied or
certainty or very high probability. implicit edits.)

ERROR DETECTION EXPLICITLY DEFINED CONFLICT RULE
An activity aimed at detecting erroneous data, A conflict rule which in defined by the people
using predefined correctness criteria. The responsible for the correctness of the data.
correctness criteria can be defined through various
checking rules. FATAL ERRORS
Errors identified by fatal edits.
ERROR LOCALIZATION
The (automatic) identification of the fields to FAILED EDIT GRAPH
impute in an edit-failing record. In most cases, an As used by the U.S. Bureau of the Census, a
optimization algorithm is used to determine the graph containing nodes (corresponding to fields)
minimal set of fields to impute so that the final which are connected by arcs (an arc between two
(corrected) record will not fail edits. nodes indicates that the two fields are involved in an
edit failure.) Deleting a node is equivalent to
ERROR STATISTICS choosing that field to be imputed. A minimal set of
A statistical report of errors found. It usually deleted nodes is equivalent to a minimal set as
provides the error rate in individual data items and defined by Fellegi and Holt.
the occurrence of particular kinds of errors.
FATAL EDIT
EVALUATION OF EDITING Identifies data errors with certainty. Examples
There are several methods available to evaluate are a geographic code for a Canadian province that
an edit system. One method is to compare the raw does not exist in a table of acceptable geographic
(unedited) data file with the edited file for each codes.
question. By ordering the changes by descending
absolute magnitude, the cumulative impact of the FELLEGI- FACTOR CHECK
editing changes to the total editing change or to the Check of measuring units verifying whether the
estimated item total can be displayed in graphs and data has been given in correct units.
tables. This technique can be used to identify
questionnaire problems and respondent errors. FELLEGI-HOLT SYSTEMS, TENETS,
When forms are reviewed before data capture or PRINCIPLES
when general editing changes are noted on the In reference to assumptions and editing and
questionnaires, evaluation studies can be carried out imputation goals put forth by Fellegi and Holt in
by selecting a sample of forms and analysing the their 1976 Journal of the American Statistical
effect of the editing procedures on individual data Association paper. A key feature of the Fellegi-Holt
items. A widely used technique to evaluate new model is that it shows that implied edits are needed
editing methods is to simulate the new process using to assure that a set of values in data fields that are
a raw data file from the survey. By replacing values not imputed always lead to final (imputed) records
flagged according to the new method with the values that satisfy all edits.
from the tabulation file containing the data edited by
the alternative editing process, it becomes possible FIXED CONSTRAINT CHECK
to compare estimates from this newly edited file RANGE CHECK
with estimates from the tabulation file. Verifying whether the data item value is in the
previously specified interval.
EXPERT SYSTEM
Computer system that solves complex problems
in a given field using knowledge and inference


FUNCTIONAL CHECK correction. Typically edit-passing records from the
ARITHMETIC EDIT current database are used in the correction database
Verifying whether the given functions of two or (see COLD-DECK).
more data items meet the given condition.
HOT-DECK IMPUTATION
GENERATED (versions of) QUESTIONAIRES A method of imputation whereby values of
There are software systems that assist in variables for good records in the current (hot) survey
designing and electronically generating paper file are used to impute for blank values of
questionnaires. incomplete records (see COLD-DECK).

GRAPHICAL EDITING IMPLIED CONFLICT RULE
Using graphs to identify anomalies in data. Conflict rule which can be deduced from the
While such graphical methods can employ paper, the explicitly given conflict rules.
more sophisticated use powerful interactive methods
that interconnect groups of graphs automatically and IMPLIED EDIT
retrieve detailed records for manual review and An unstated edit derived logically from explicit
editing. edits that were written by a subject matter specialist.

“HEADS-DOWN” DATA ENTRY IMPUTATION
Data entry with no error detection occurring at A procedure for entering a value for a specific
the time of entry. Data entered in a heads down data item where the response is missing or unusable.
mode is often verified by re-keying the questionnaire
and comparing the two-keyed copies of the same IMPUTATION VARIANCE
questionnaire. Data entered in a "heads-down" A component of the total variance of the survey
fashion may later be corrected in either a "batch" or estimate introduced by the imputation procedure.
an "interactive" data review process.
INLIER
"HEADS-UP" DATA ENTRY An inlier is a data value that lies in the interior
Data entry with a review at time of entry. of a statistical distribution and is in error. Because
Heads up data entry requires subject-matter inliers are difficult to distinguish from good data
knowledge by the individuals entering the data. values they are sometimes difficult to find and
Data entry is slower, but data review/adjustment is correct. A simple example of an inlier might be a
reduced since simple inconsistencies in responses value in a record reported in the wrong units, say
are found earlier in the survey process. This mode is degrees Fahrenheit instead of degrees Celsius.
very effective when the interviewer or respondent
enter data during the interview (CAI). INPUT-EDITING
Editing that is performed as data is input. The
HIT RATE editing may be part of a data entry system.
The "success" rate of an edit; the proportion of
error flags that the edit generates which point to true INTEGRATED SURVEY PROCESSING
errors. The concept that all parts of the survey process
be integrated in a coherent manner, the results of one
HOLT METHOD FOR AUTOCORRECTION part of the process automatically giving information
Automatic correction method in which the least to the next part of the process. The Blaise system is
possible number of data items is changed and the an example of integrated software in which the
Fellegi-Holt model is used to determine acceptable specification of the Blaise Questionnaire gives rise
sets of values or ranges for the items that are to a data entry module as well as CATI and CAPI
imputed. Sequential or simultaneous imputation via instruments. The goals of Integrated Survey
cold-deck or hot-deck method may be applied. Processing include the one-time specification of the
data, which in turn would reduce duplication of
HOT-DECK effort and reduce the numbers of errors introduced
A correction base for which the elements are into the system due to multiple specifications.
continuously updated during the data set check and

_______________________________________________________________________________________ 9

INTERACTIVE DATA REVIEW/ MACRO-EDIT
INTERACTIVE EDITING/ SELECTIVE EDIT
ONLINE CORRECTION Detection of individual errors by: 1) checks on
Checking and correcting data in dialogue mode aggregated data, or 2) checks applied to the whole
using video terminals. It can be applied during data body of records. The checks are typically based on
entry or on data that are already in machine-readable the models, either graphical or numerical formula
form. The questionnaire is immediately reviewed based, that determine the impact of specific fields in
after adjustments are made. The results are shown individual records on the aggregate estimates.
on a video terminal and the data editor is prompted
to adjust the data or override the error flag. This MANUAL CORRECTION
process continues until the questionnaire is A human activity aimed at changing the values
considered acceptable by the automated review of data items deemed erroneous. The correction
process. Then results of the next questionnaire's specified usually on the diagnostic list is entered into
review by the auto review processor are presented. the data set by means of a program specially written
A desirable feature of Interactive Data Editing for this purpose.
Software is to only present questionnaires requiring
adjustments. MANUAL DATA REVIEW
May occur prior to data entry. The data may be
LEAST SQUARES METHOD FOR reviewed and prepared/corrected prior to key-entry.
AUTOCORRECTION This procedure is more typically followed when
FEERD-HASTLY METHOD heads-down data entry is used.
An automatic correction method in which:
MICROEDITING
1. the least possible number of data items are Finding errors by inspection of individual
changed observations. Editing done at the record, or
2. the changed record is the closest one questionnaire level.
(measured by the weighted sum of squares of
deviations of the changed data) to the original MINIMAL SET OF CONFLICT RULES
(incorrect) record. A minimal subset of the complete set of conflict
rules expressing the same erroneous data
LINEAR EDITS combinations as the complete set of conflict rules.
Edits arising from linear constraints. For
example, if v1 , v2 , v3 are variables and a, b, and c are MINIMAL SET OF FIELDS TO IMPUTE
real constants, the linear inequality edits are given The smallest set of fields requiring imputation
by: that will guarantee that all edits are passed. See also
weighted minimal set.
1. a <= v1 / v2 <= b (This is two edits. Each can
be converted to linear inequality.). MONITORING OF EDITING
2. a v1 + b v2 # c. Analyzing the audit trail and evaluating the
3. v1 + v1 = v3. edits for efficiency.

LOGICAL CONDITION CHECK MONTE-CARLO METHOD FOR
Verifying whether the given logical condition is AUTOCORRECTION
met. It is usually employed to check qualitative An automatic correction method in which the
data. corrected data value is randomly chosen on the basis
of a previously supplied probability distribution for
LOGICAL LEVEL OF THE CHECKING RULE this data item. The method employs computer
The logical level of the data structure to which algorithms for generating pseudo-random variables
the given checking rule refers (individual data item, with the given probability distribution.
record, logical group of records, and the like).


MULTI-LEVEL MODELING IMPUTATION uncorrected (raw) data are more likely to be
An imputation rule defined by a sequence of incorrect. Examples are data values that lie in the
decisions each based on exclusive sets of tails of the distributions of ratios of two fields (ratio
observations. edits), weighted sums of fields (linear inequality
edits), and Mahalanobis distributions (multivariate
MULTIVARIATE EDIT normal) or outlying points to point clouds of graphs.
A type of statistical edit where multivariate
distributions are used to evaluate the data and to find OVEREDITING
outliers. Editing of data beyond a certain point after
which as many errors are introduced as are
NONLINEAR EDITS corrected.
Edits from non-linear constraints. For example,
if v1 and v2 are variables and b are real constants, PROBABILISTIC CHECKING RULE
then nonlinear edits are: A checking rule causing, with some small
probability, incorrect qualification of data (i.e., it
1. v1 v2 # a. may identify actually correct data as incorrect and
2. v1 # exp(v2 ). identify incorrect data as correct).
3. conditional edits.
4. Mahalanobis-distance edits with multivariate PROBABILISTIC IMPUTATION
normal data. An imputation rule that is in part a function of a
randomization process exogenous to the
The importance of nonlinear edits is that they occur experimental observations.
often but are not amendable to theory in the
determination of a minimal set. Some nonlinear QUALITATIVE DATA
edits, such as ratio edits, can be cast in a linear form. Data describing the attributes or properties that
an object possesses. The properties are categorized
NIM (new imputation methodology) into classes that may be assigned numeric values.
A generalization of the hot-deck that employs However, there is no significance to the data values
sophisticated matching methods to choose potential themselves, they simply represent attributes of the
donors from a pool of edit-passing records that most object concerned.
closely resemble the edit-failing record being
donated to. The method uses additional metrics for QUALITY CONTROL
comparing numeric data and specific logic to assure ... of the data collection process assures that
that records satisfy edits that are not available with the underlying statistical assumptions of a survey are
traditional hot-deck methods. not violated, i.e. the meaning of the principal
statistical measures and the assumptions which
NORMAL FORM OF CONFLICT RULE condition their use is maintained.
A conflict rule which is defined by the logical
product of conditions on the values of individual ... in data review process measures the impact
data items in a record. For example, the conflict of data adjustment on the data.
(branch - (101, 107,112); production 104, 180;
efficiency 0.8) is a conflict in the normal form QUANTITATIVE DATA
(CAN-EDIT). Data expressing a certain quantity, amount or
range. Usually, there are measurement units
ON-LINE CORRECTION associated with the data, e.g. meters, in the case of
Correcting the values in erroneous data items of the height of a person. It makes sense to set
a previously checked data set on a video terminal. boundary limits to such data, and it is also
meaningful to apply arithmetic operations to the
OUTLIER data.
An outlier is a data value that lies in the tail of
the statistical distribution of a set of data values. QUERY EDIT
The intuition is that outliers in the distribution of Points to suspicious data items that may be in
error. An example could be a value that, compared

_______________________________________________________________________________________ 11

to historical data, seems suspiciously high. Contrast (e.g., by using the record from a hot deck or cold
query edit to fatal edit where data item is known deck).
with certainty to be in error.
SPECIFICATIONS GENERATOR
QUERY ERRORS A module in an editing system from which files
Errors identified by query edits. for paper questionnaires, data entry modules, editing
software, CATI, CAPI, and summary software are
RATIO EDIT generated. The specifications generator is the
An edit in which the value of a ratio of two unifying feature in Integrated Survey Processing
fields lies between specified bounds. The bounds software. In the Blaise system, the Blaise
must be determined through a priori analyses Questionnaire can be considered to be a
(possibly involving data sets in which truth data are specifications generator. The specifications
available) or via exploratory data analysis methods. generator contains information relating to the data to
be collected as well as to the edits and routes to be
REJECTION RULE / CONFLICT RULE applied to the data.
A logical condition or a restriction to the value
of a data item or a data group which must not be met STATISTICAL EDIT
if the data is to be considered correct. In various A set of checks based on statistical analysis of
connections other terms are used, e.g. Y-rule. respondent data, e.g., the ratio of two fields lies
between limits determined by a statistical analysis of
REPEATABILITY that ratio for presumed valid reporters. A statistical
The concept that survey procedures should be edit may incorporate cross-record checks, e.g., the
repeatable from survey to survey and from location comparison of the value of an item in one record
to location; the same data processed twice should against a frequency distribution for that item for all
yield the same results. (Also called reproduc ibility.) records. A statistical edit may also use historic al
data on a firm-by-firm basis in a time series
REPORT ON ERRORS modeling procedure.
ERROR DIAGNOSTICS
A report which usually contains the record STOCHASTIC EDIT
identification, violated conditions, items that are An edit which if violated points to an error in
probably erroneous, etc. the data with probability less than one. Example:
80<yield<120. Contrast with deterministic edit.
SCORE FUNCTION Compare to query edit.
A numerical indicator used to prioritize micro
data review in selective editing. STOCHASTIC IMPUTATION
Probabilistic imputation.
SELECTIVE EDITING
A procedure which targets only some of the STRUCTURE ERROR
micro data items or records for review by The absence or presence of a data record not
prioritizing the manual work and establishing following the hierarchical order into which the data
appropriate and efficient process and edit file is organized.
boundaries.
SUBJECT-BASED EDIT
SEQUENTIAL CORRECTION Checks incorporating real-world structures
SEQUENTIAL IMPUTATION which are neither statistical nor structural.
A correction where the items intended for the Example: wages paid/hours worked $
correction are corrected sequentially. minimum wage.

SIMULTANEOUS CORRECTION SUBSTANTIAL EDIT
SIMULTANEOUS IMPUTATION Edit rule proceeding from knowledge of the
A correction in which all the data in a record substance of the subject matter.
intended for correction is corrected at the same time


SUM CHECK VALIDATION EDITS
Verifying whether the sum of the values of the Edit checks which are made between fields in a
given data group equals the value of the particular record. This includes the checking of
corresponding data item which should represent every field of every record to ascertain whether it
their total. contains a valid entry and the checking that entries
are consistent with each other.
SURVEY MANAGEMENT
The processes used to monitor, administer and VALIDITY ERROR
control the survey. Survey Management includes An occurrence of the value of a data item which
preparing mailing labels and calling lists, making is not an element of the set of permissible codes or
enumerator assignments and tracking the status of values assigned to that data item.
each questionnaire through the data editing
processes (data collection, data capture and WEIGHTED MINIMAL SET
review/correction). Survey Management also A minimal set n which fields are weighted
i
includes monitoring the status and quality of the data according to reliability in generating imputations. If
editing processes. fields (variables) are given weights, then it is the set
of fields that give the minimum weight. There may
SYSTEMATIC ERROR be two of more sets that give the same minimum.
Errors reported consistently over time and/or
between responding units (generally undetectable by WINSORISATION
editing). A phenomenon caused either by the An imputation rule limiting the influence of the
consistent misunderstanding of a question on the largest and smallest observations in the available
survey questionnaire during the collection of data or data.
by consistent misinterpretation of certain answers in
the course of coding. The systematic error does not
lead necessarily to validity or consistency errors but
always seriously compromises statistical results.

TOTAL QUALITY MANAGEM ENT (TQM) is
an optimisation and integration of all the functions
and processes of an organisation in order to satisfy
customer demands through a process of continuous
improvement. Total Quality revolves not only
producing a good product, but on improving the
competitiveness, effectiveness and flexibility of the
whole organisation in providing better products and
services to customers. In data editing, the TQM
approach aims to prevent errors rather than correct
them, and to learn from past (editing) experiences in
order to improve the data collection process. One
element of the approach is to use the information on
errors and error causes collected at the editing
process to evaluate the performance of the survey
questionnaire, and to feed back the findings for
improvement in staff training, system design and
form design. The other element is the principle of
getting the data right the first time.

TRANSPOSITION CHECK
Detecting whether the data has been recorded at
the correct position (i.e., whether the data in various
character positions have been transposed).

Editingglossary

More Related Content

Viewers also liked (8)

Similar to Editingglossary (20)

Recently uploaded (20)

Editingglossary