1 / 26

Assessing Quality for Integration Based Data

Assessing Quality for Integration Based Data. M. Denk, W. Grossmann Institute for Scientific Computing. Contents. Introduction Data Generating Processes Data Quality for Integration Based Production Assessing Quality for Integration Based Data Conclusions.

amadis
Download Presentation

Assessing Quality for Integration Based Data

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Assessing Quality for Integration Based Data M. Denk, W. Grossmann Institute for Scientific Computing

  2. Contents • Introduction • Data Generating Processes • Data Quality for Integration Based Production • Assessing Quality for Integration Based Data • Conclusions

  3. Introduction – Aspects of Quality • Quality is discussed from two different points of view • The Processing View • What methods can be used in production of statistics ? • Specific statistical techniques for specific statistics • Development of models of best practice or standards • The Reporting View • How should Quality reports look like?

  4. Introduction – Reporting View • Numerous formats for Quality Reports • SDSS, DQAF, Fed Stats, StaCan,…. • Logic of the proposals according to so called hyperdimensions • For example ESS: • Institutional Arrangements • Core Statistical Processes • Dimensions for Statistical Output • Inside the hyperdimensions so called quality dimensions • Relevance, Accuracy, Timeliness, Accessibility,……

  5. Introduction – Reporting View • Not so much agreement about the dimensions • Possible Reason: Different methods / levels of Conceptualization • Concepts of mental entities • e.g. quality dimensions in DQAF • Concepts as meaning of general terms • e.g. quality elements in DQAF • Concepts as units of knowledge • e.g. quality indicators of DQAF • Concepts as abstracts of kinds, attributes or properties • measureable quantities like sampling error, …

  6. Introduction – Reporting View • Stronger matching of the processing and the reporting view seems necessary • Starting point can be attributes and properties of statistical processes necessary for assessing quality • From basic quality concepts we build higher level elements by aggregation • Prerequisite for definition of necessary basic quality concepts: • Empirical analysis of different production processes • Final result is a User Oriented Quality Certificate

  7. Data Generating Processes • We can distinguish two broad classes of data generating processes • The survey based data generating process • The integration based data generating process

  8. Data Generating Processes – Survey based • Most considerations about reporting quality start from the traditional survey process • Characteristics of the traditional survey process • One well defined target population (e.g. persons) • A rather homogeneous method for data collection (e.g. questionnaire) • A more or less linear sequence of processing steps (e.g. data cleaning, data editing, data imputation, output) • Final Output is one Output File

  9. Data Generating Processes – Integration based • Many Statistics do not follow such a linear production scheme • Examples: Indices, numerous balance sheets, National Accounts, …. • Common characteristic: Data are produced from many different sources • Let us call such processes as integration based processes • Data produced in such way are called integration based data

  10. Data Generating Processes – Integration based • Characteristics of integration based data processing • Population: • The underlying population may be split into segments • Example: Expenditures for education: government, private enterprises, households • Many times more than one population is involved, possibly also one population at different times • Example: calculation of indices

  11. Data Generating Processes – Integration based • Characteristics of integration based data processing • Data collection: • Data collection is different for different segments and populations • Many times the collected data are the output of already existing data products • Main processing activities are alignment procedures making the different sources comparable • Output may be a set of organized Data Files

  12. Data Generating Processes – Workflow View • Workflow for Survey Process

  13. Data Generating Processes – Workflow View • Workflow for Integration Based Process

  14. Data Quality for Integration Based Production • Two important aspects of data quality • Content quality • Are the measured “concepts” really the target “concepts” ? • Production quality • Are the used methods sound?

  15. Data Quality for Integration Based Production – Content Quality • Main reasons for lack of content quality • Slight difference in the measurements of the variables (“concepts” ) in case of reuse of already existing data • Example: • Transport of goods on Austrian rails • Transport of goods according to data from railway authorities (taking not into account that transport may use partly German rails) • Slight differences in the definition of the segments in the underlying population

  16. Data Quality for Integration Based Production – Content Quality • Conclusion: Using data already collected for other purposes gives often only proxy variables for the intended variables • Question: Is this in coincidence with your mental concept of the term “Non-Sampling Error”? • Manuals of international organizations are many times rather vague with respect to such problems

  17. Data Quality for Integration Based Production – Content Quality • Possible Strategies for Solution • Statistical Models for aligning the concepts • More detailed description of the concepts by using additional variables characterizing the differences as formal properties of the data • More detailed description of the underlying populations by using additional variables characterizing the differences

  18. Data Quality for Integration Based Production – Processing Quality • Elements of processing quality • Quality of methods used for the different components of the integration based statistic • This implies that we do not have one method of collection, one editing, one imputation,… but many activities of that kind • Quality of methods used in the integration process • Alignment of variables in order to overcome differences in concepts • Standard activities like plausibility, editing, imputation necessary for the integration activities

  19. Assessing Quality for Integration Based Data • If we know the quality of all the components used in the integration process we have to think about transmission of quality in the integration steps • Starting point should be an “Authentic Data System” • All data used in the integration process • Quality information about the different data sets of the system

  20. Assessing Quality for Integration Based Data • Distinguish two types of quality transmission • Quality compilation • Methods for representing quality of the overall product • Quality calculations • Algorithms for assessing quality • In both cases we need • Methods for assessing quality • Models of best practice / standards

  21. Assessing Quality for Integration Based Data – Quality Compilations • In some cases the best we can do is better representation of the quality dimensions of the used components • Distribution of quality indicators • Concentration of quality indicators

  22. Assessing Quality for Integration Based Data – Quality Compilations • Example: Coverage for integration based data • Structure of integrated sources together with coverage information

  23. Assessing Quality for Integration Based Data – Quality Compilations • Coverage distribution

  24. Assessing Quality for Integration Based Data – Quality Compilations • Coverage concentration with respect to target concept

  25. Assessing Quality for Integration Based Data – Quality Calculations • Methods will be in most cases not formulas but advanced statistical procedures for different quality dimensions • Examples: • Measurement of accuracy using variances, standard errors or coefficient of variation • Could be done by using bootstrap (e.g. applied for indices by NSO-GB) • Simulation techniques • Sensitivity analysis (“robustness”)

  26. Conclusions • Assessing quality of integration based statistics needs • Clear separation of content based quality and processing based quality • Better documentation / representation of complex production processes, Usage of Workflow Models • Documentation of the authentic data file • Definition of best practice / standards for integration processes • Algorithms for calculation quality dimensions • Methods for representation of quality indicators

More Related