Martina Mysíková, Štěpán Tourek, Martin Zelený Attrition in EU-SILC in the Czech Republic Presentation for ESRA Conference June 28th 2007 Martina Mysíková, Štěpán Tourek, Martin Zelený
Design of EU-SILC Four year rotational panel households are visited continuously for four years each year ¼ dropped and ¼ newly selected sample persons moved to another private households are traced and visited First wave: 4351 households responded (response rate 64,6%) Second wave: 87,7% Response, 11,2% Non-response, 1,1% Out of survey population
Design of EU-SILC 2005 2006 2007 2008 2009 1 2 3 4
Households
Household distribution NUMBER OF HOUSEHOLD MEMBERS The distribution by number of household members is the same at „response“ and „non-response“ Among „out“ prevail households with one member
Household distribution NUMBER OF CHILDREN The distribution by number of children is similar at „response“ and „non-response“ Among „out“ prevail households without children
Household characteristics SIZE OF MUNICIPALITY The „non-response“ is highest in big cities (18%) Small – less than 49999 inhabitants Middle – 50000-99999 Big – more than 100000 inhabitants
Household characteristics INTERVIEWER The change of the interviewer has some impact on „non-response“ (15%) Domácnosti jsou ochotnější odpovídat stejnému tazateli. Noví tazatelé jsou méně zkušení
Individuals
Individual distribution NET INCOME DECILES The greatest share of „non-response“ falls in higher deciles
Individual characteristics GENDER No differences by gender
Individual characteristics AGE Higher „non-response“ at middle-age groups Highest „out“ at oldest group
Individual characteristics MARITAL STATUS Higher „non-response“ at separated (18%), divorced (13%) and single (13%) Highest „out“ at widowed
Individual characteristics ECONOMIC ACTIVITY Higher „non-response“ at self-employed (17%) Highest „response“ at retired (not working) and those receiving parental allowance Highest „out“ at retired (not working)
Individual characteristics EDUCATION „non-response“ rises with higher educational level „out“ increases moderately with lower educational level
Summary Household level „Non-response“ „Out“ Higher in households with working members Lower in households with retired (not working) members Higher in bigger cities „Out“ Mainly one member households (85%) – retired (institutions or decease)
Summary Individual level Tight connection with time and economic activity „Non-response“ No impact of gender Higher at middle-age groups, self-employed, higher educated, high-income groups Lower at old groups, widowed, retired, parental leave - probably spend more time at home „Out“ Mainly old, widowed and retired (institutions or decease) and partly young (move out of country) Mainly low-income groups
Impact on weights Focus on impact of attrition on weights and on selected variables Take the values from the 1st wave Simply eliminate those who didn't response in the 2nd wave (the same weights) Recalculate the weights excluding those who didn't respond in the 2nd wave
Impact on weights 1. Original sample Original weights 2. Exclusion “Non-response” and “Out” households eliminated Weights recalculated Optimally: the 3rd step should bring the values closer to the original values 2. Exclusion Original weights 3. Exclusion New weights
Impact on weights Dwelling 3rd step lowers the difference by one half
Impact on weights Problems with the dwelling the 3rd step brings the values closer in all cases
Impact on weights Can not afford... the 3rd step can also increase the difference
Impact on weights Regular monthly inter-household transfer great decrease at received and only slight increase at paid transfer
Impact on weights Annual household income 3rd step brings the values closer to the original ones
Conclusion The 3rd step brings the values back in most cases (e.g. dwelling) increases the difference in some cases (e.g. affordability) The influence of the attrition on weights and results in next wave is relatively negligible but can be significant for certain variables (diagnostics)