Challenges in using SAE for official data production

Here are some of the reasons that hinder the use of SAE for official data production, as suggested by selected National Statistical Offices:

  • Lack of interest and support from the top management, hence lack of resource;
  • Lack of dedicated resources for SAE research and implementation. Compared to other output of a National Statistical Office, SAE is usually a relatively minor one.  While one household survey can produce a large number of indicators, great efforts are necessary for deriving just one indicator for small domains;  
  • Lack of in-house technical capacity;
  • Lack of proper input data;
  • Reluctance about the use of model-based estimates;   
  • Difficulties in communicating the technical aspects to users


Challenges for national statistical offices

"Overall, the main challenges for NSIs when producing small area estimates is the ability to master the complexities of the required statistical theory (e.g. the assessment of the estimation error is recognised as a complex problem in the small area estimation context), the availability of relevant administrative data and the capacity to overcome internal and external barriers for the acceptance of model based estimates as trustworthy official statistics outputs."

Source: Silva, D.B.N. and Clarke, P. (2008)

Challenges in using SAE for official data production

From National Statistical Offices

  • "We did an experiment using small area estimation method for poverty but the results were not consistent with our own estimates so we did not pursue it again."
  • "We do not have good input data source for SAE - census data are outdated, and administrative data sources do not have good coverage and lack proper auxiliary variables."
  • "SAE method is complicated and we are not comfortable with independently developing the method"
  • "It is very difficult to convince the managers to use model-based estimates."
  • "Producing SAE requires a lengthy period of looking for input data, finding the right auxiliary variables, testing different models and their assumptions and validating the estimates."
  • "The policymakers do not want to see the confidence intervals for the SAE estimates - only 1 number please!"  


  • No labels