THE MAIN INNOVATIONS OF DATA EDITING AND IMPUTATION FOR THE 2010 ITALIAN
AGRICULTURAL CENSUS
G. Bianchi, R. M. Lipsi, P. Francescangeli, G. Ruocco, A. M. Salvatore, F. Scalfati
Work Session on Statistical Data EditingLjubljana, Slovenia, 9-11 May 2011
UNECE - CONFERENCE OF EUROPEAN STATISTICIANS
Outline
• Introduction
• E&I strategy guidelines
strategy and census stages
during data collection
stages after data capturing
• E&IS tools and innovations
• Conclusions
• References
IntroductionFor the 6th Italian Agricultural Census, a new Editing and Imputation System (E&IS) has been implemented in order to reduce the total census error
The main purpose of the E&IS is to identify and treat the non sampling errors, in order to provide a complete and consistent set of data
E&I strategy guidelines (1) Quality oriented approach by performing the
E&I process from data collection to the final figures
Data editing and detection of outliers and influential errors (selective editing) during data collection
After data capturing, scheduling of two main
correction stages, centrally managed by Istat
E&I strategy guidelines (2)
Use of techniques that minimize the number of changes especially for the treatment of not influential random errors
Quality indicators to monitor the main steps of E&I
Ad-hoc documentation to evaluate the outcome of the procedures, paying particular attention to changes due to the E&I process
E&I strategy and census stages (1) According to the E&I strategy, all variables
are separated into different related subsets to identify the most appropriate treatment for each of them
The E&I process will feature three main stages:
1. E&I during data collection
2. Provisional figures dissemination (primary variables)
3. Final results dissemination
E&I strategy and census stages (2)
E&I during data collection (1)In order to prevent and correct fatal errors and missing values during data capturing
Questionnaire editingHoldings/enumerators
A subset of 220 checking rules (fatal and query) hasbeen implemented in the web based data entry System
Automatic checkData collection staff
Before the final release of data to the census DB, to localize potential errors slipped during data gathering
Census Data Collection System
E&I during data collection (2)Before the end of field enumeration operations, and while data collection network is still in force, two distinct procedures have been implemented and launched by Istat to detect influential errors and outlier values
Outliers detection
-Forward Search Technique-manual review of anomalous values by data collection staff
Micro-editing check
Underlines inconsistent data by analyzing at unit level the coherence between the answers referring to related topics
E&I SYSTEM
E&I during data collection (3)Forward Search Technique: outliers detection among strata, defined according to the crop type and the farm size
Census
Administrative Register
-Regression lineY=aX+b
-Parameters estimation a and b, with and without outliers
-Statistical significance and goodness of fit of the regression model (R2)
E&I stages after data capturing (1) In order to achieve maximum coherence
between provisional and final data at regional level, the strategy adopted is firstly to correct all the primary variables and then the secondary ones
After data collection, two main correction stages are scheduled. In the first stage, all the variables for the dissemination of provisional figures (primary variables) are corrected
In each E&I stage, the following steps are repeated: automatic error detection and treatment of errors
E&I stages after data capturing (2)
First step of each E&I stage
Automatic error detection
Macro level editing
- Uses all (or large part) of data to identify errors- Enables to evaluate the accuracy of preliminary estimates such as totals (or subgroups main figures)- Outliers detection
Micro level editing
Erroneous values in individual records are automatically identified by means of edit rules
E&I stages after data capturing (3)
Second step of each E&I stage
Treatment of errors
Selective editing
Treatment of the outliers and influential errors, having substantial impact on data dissemination is based on manual review
Random errors
Treatment of not influential random errors is based on minimum change approaches
Imputation
Model based techniques or nearest neighbour donor will be used for the imputation of not influential random errors
E&IS tools and innovations (1) Inclusion of a subset of edit rules in the data
capture stage Use of Forward Search methods for the outliers
detection Use of administrative sources for micro and
macro data checks Use of score functions to prioritize records to
be manually reviewed Use of minimum change based model or
nearest neighbour approach for localizing residual random errors
Mix of different imputation methods as nearest neighbour approach or model based imputation
E&IS tools and innovations (2) The core of E&IS is the software DIESIS (Data
Imputation Editing System – Italian Software), used for dealing with non influential errors in quantitative variables
DIESIS was developed in 2001 by ISTAT and academic researchers of the University of Rome “ Sapienza”
In DIESIS, optimization techniques were implemented for the simultaneous treatment of qualitative and quantitative variables
Joint use of data driven and minimum change approaches
DIESIS localization performance has been compared with the performance of the Canadian software BANFF
E&IS tools and innovations (3) The scheduling and the monitoring of all
procedures and the interactive corrections will be managed by CONCERT, a Java web application
To test the E&IS while implementing the scheduled procedures, an Oracle database was implemented
The whole process of E&I will be documented by a set of quality indicators both, on the data collected and on the results of the different editing steps
E&IS tools and innovations (4)
E&IS tools and innovations (5)Some simulation studies have been carried out for:
identifying for each section of the questionnaire, the most appropriate correction approach
evaluating the imputation of missing non linearly dependent data through conditional Copula functions (developed by ISTAT and the University of Bologna)
assessing the use of Forward Search techniques (robust statistical methods) for outliers detection (developed by ISTAT and the University of Parma)
ConclusionsThe innovative E&I strategy will reduce the efforts of coping with timeliness constraints and will increase data consistency and accuracy
The results of the procedures implemented in the E&IS are very encouraging and allow to trust in a good improvement of census data quality
Thank you!!! Thank you!!!
Thank you!!!
References • Bianchi G., Di Lascio F. M. L., Giannerini S., Manzari A., Reale A., Ruocco
G. (2009-a) Exploring copulas for the imputation of missing nonlinearly dependent data, Seventh Scientific Meeting of the CLAssification and Data Analysis Group of the Italian Statistical Society Università di Catania (Italy). September 9-11, 2009.
• Bianchi G., Francescangeli P., Manzari A., Reale A., Ruocco G., Salvi S. (2009-b) An overview of Editing and Imputation System of 2010 Italian Agriculture Census. Round. Roundtable Meeting on Programme for the 2010 Round Census of Agriculture . Budapest 23-27 november 2009.
• Bianchi G., Manzari A., Reale A., Salvi S. (2009-c) Valutazione dell’idoneità del software DIESIS all’individuazione dei valori errati in variabili quantitative. Istat - Collana Contributi Istat – n. 1 – 2009.
• Cotton C. (1991) Functional description of the generalized edit and imputation system. Business Survey Methods Division - July 25 Statistics Canada.
• Kovar J.G., MacMillian J.H., and Whitridge P. (1988) Overview and strategy for the generalized edit and imputation system. Report, Methodology Branch - April 1988 (updated February 1991) Statistics Canada.
• Luzi et al. (2007). EDIMBUS. Recommended Practices for Editing and Imputation in Cross-Sectional Business Surveys, August 2007.
• Riani M., Atkinson A. C. (2000). Robust Diagnostic Data Analysis: Trasformations in Regression. TECHNOMETRICS. vol. 42, pp. 384-394 ISSN: 0040-1706. With discussion.