International Statistical Review (2005),73, 2, 253–254, Printed in Wales by Cambrian Printersc© International Statistical Institute

Rejoinder: Frederick W.H. Ho

Census and Statistics Department, Hong Kong, China. E-mail: fwhho@censtatd.gov.hk

The two discussants of the paper, Biggeri and Kalton, offer useful views that serve to stimulatefurther deliberation on the subject of survey quality.

Biggeri talks about integrating survey data with other information sources such as administrativedata or other statistical surveys. The benefit of establishing such a linkage is obvious as it wouldhelp reduce respondent burden. Indeed the use of data from administrative records provides anattractive option to official statisticians as a supplement to, or even replacement of, the currentsurvey programme. This is particularly so in places where cooperation of survey respondents isdeclining. However the use of administrative records for statistical purposes is often loaded withformidable technical issues, the solution of which may lie beyond the ambit of the Statistics Office asit involves the operations of other government departments. Commonly met issues include differencesin definitions of data items, classification schemes and reference periods for reporting, which wouldrender the administrative information generated not amenable to the intended statistical purposes.Very often the Statistics Office has to secure the joint efforts of the administrative agencies in evolvingand adopting standardised practices for data capture, recording and management. Compounding tothese issues is the presence of legal provisions for protection of confidentiality of data pertaining toindividual persons, households and establishments. This is a standing feature of modern society, but isone that poses significant obstacles to the Statistics Office in accessing the administrative databases.A Statistics Office would have to engage strenuously in processes to bring about legislative changesto enable its access to relevant administrative information.

On the trade-off between accuracy and resources, Biggeri makes a point that timeliness shouldalso be considered as an important aspect in survey planning. This point should not attract anydissenting view. Indeed, resources interpreted in the broad sense should cover not just manpowerand finance, but also the time frame expected by data users. In coping with exceptionally tightschedules, the release of provisional survey estimates may be a tenable strategy. But the strategyneeds to be underpinned by a well-designed revision plan and publicity policy as revisions of evenseemingly small magnitude may cause undue concern to less statistically sophisticated users. As aresult, unwarranted adverse impact may impinge on the credibility of the Statistics Office.

On the assessment of survey quality, Kalton draws our attention to the issue of measurementerrors as a major source of error in survey estimates. He mentions that these errors are difficultto measure. Few would disagree with the statement. The difficulty may be explained by the manydifferent sources contributing to this category of errors (such as unsatisfactory questionnaire design,ineffective data collection method, insufficient interviewer training and unreliable responses). Thereis therefore a need to engage in intensive and elaborate research or experimental studies to quantifythe error arising from each source. Nevertheless, the lack of quantification of measurement errorsmay be partially, though not fully, compensated by qualitative analysis of the error sources and anassessment of the impact of these errors on the usefulness of survey estimates. Also, informationabout the survey design and procedures should be furnished to data users as a basis for judging thequality of the survey.

Finally, use of the Total Quality Management (TQM) Model mentioned by Biggeri as a usefulapproach to guarantee the integrity of decisions for the whole survey process is certainly agreed.However, since TQM is basically a management philosophy with focus on the overall quality of

254 F.W.H. HO

products and services, and does not offer any guidance to its practical implementation, StatisticalOffices would need to develop appropriate tools for ensuring their own performance with respectto the values advocated in TQM,viz. leadership, strategic planning, customer orientation, staffdevelopment, process management and business results. These are long-term commitments and mustinvolve leadership and staff support at all levels.

[Received March 2005, accepted May 2005]