• Design effect (cluster sampling)
  
• Interviewer effect (measurement error, design effect)
  
• These usually confounded
  
• Interpenetrated design (but costly and rare)
  
• Cross-classified model
  
• Controls for the area (census data) and respondent
  
• Most research focus on items
  
• Focus on interviewer characteristics
  

   
   

Business case for Natcen

1. Nature of research
   
   1. Interviewer error - variance and bias
      
   2. Problems with estimating interviewer variance
      
   3. Different models to estimate interviewers variance
      
      1. HHB
         
      2. Kish
         
      3. Multilevel (Colm, Hox, Schnell)
         
   4. Interpenetrated design
      
   5. Cross-classified model
      
   6. Complex variance 1 model
      
2. Dataset - NTS because
   
   1. Rolling design
      
   2. Natcen information on Interviewers
      
   3. Potential variables: different formats of questions
      
   4. Census data (?) SOAs
      
3. Research questions
   
   1. There is no work that looks at different interviewer's variability depending on interviewer's characteristics e.g. Do female interviewers get more variability than male interviewers? This sort of questions could potentially be answered by complex variance component models.
      
4. Requested variables
   
   1. NTS questionnaire (including respondent characteristics)
      
   2. NatCen Interviewers
      
   3. Geographical reference