80 likes | 193 Views
Clustering and Design Effects: Moldova Example. Kathleen Beegle DECRG January 2008 Module 2: Sampling for Surveys. Moldova Household Budget Survey. 1,620 households in each quarter (6,480 households per year) 45 PSUs (cluster/EA) 36 households in each PSU in each quarter
E N D
Clustering and Design Effects: Moldova Example Kathleen Beegle DECRG January 2008 Module 2: Sampling for Surveys
Moldova Household Budget Survey • 1,620 households in each quarter (6,480 households per year) • 45 PSUs (cluster/EA) • 36 households in each PSU in each quarter • Issue: is the design sufficient to ensure precision due to the cluster effect? • In other words, does the relatively large number of households offer much benefit with so few PSUs?
Moldova Labor Force Survey • 8,208 households in each quarter • 114 PSUs (not same PSUs as HBS) • 72 households in each PSU in each quarter • Issue: is the design insufficient to ensure precision due to the cluster effect?
LFS 2002 quarter 2Rate of unemployment • Design effect is 6.92.
Proposal: Integration for Improved Data • Conduct the two surveys over a common sample of 160 PSUs • In each PSU in each quarter: • 12 HHs do both HBS and LFS • 36 HHs do only LFS
Proposal: Integration for Improved Data • Costs: • Approximately same manpower • Approximately same number of questionnaires • Benefits: • Precision of estimates • Value of data for analysis: analysts can link detailed labor force data to HBS information
Expected SE • HBS with 12 households in 160 PSUs: • SE: 9.33 Lei (compared 15.52) [for HBS 2002 quarter 2] • CI: 315 - 351 (compared to 302 - 364) • LFS with 28 households in 160 PSUs: • SE: 0.58 percent points (compared to 0.68) • CI: 5.2 – 7.5 (compared to 5.0 - 7.7)