140 likes | 300 Views
http://www.ifs.org.uk/elsa. The English Longitudinal Study of Ageing. Rebecca Taylor National Centre for Social Research. Sample design Who was sampled? Why? Any implications from design? Response rate to Wave 1 Analysis accounting for non-response accounting for sample design.
E N D
http://www.ifs.org.uk/elsa The English Longitudinal Study of Ageing Rebecca Taylor National Centre for Social Research
Sample design Who was sampled? Why? Any implications from design? Response rate to Wave 1 Analysis accounting for non-response accounting for sample design Overview
Aim of survey - investigate ageing Capture individuals before and during ageing Sample needed to represent a population that ages Interested in individuals and interactions of individuals within households In practice - select a sample that is representative of population aged 50 and over Sample Requirements
Two choices randomly select addresses generating a general population sample and screen to remove youngest draw from an existing survey sample Advantages and disadvantages with each method Drawn from Health Survey for England (HSE) Annual cross-section of approximately 16,000 adults Representative of private households in England Three years of HSE used - 1998, 1999 & 2001 Sample Design 1
Advantages Health measures provide ‘pre-baseline’ information Core content and varying supplementary modules No expensive eligibility screening in field Anticipate higher levels of response Disadvantages Does not include institutions (e.g. care homes) Individual adult response to HSE (67% to 70%) Ethical considerations for re-approaching respondents Sample Design 2
Eligibility at Wave 1 • Core sample members born on or before 29th Feb 1952 • Other sample members invited for interview • Younger partners cohabiting with core sample at HSE • New partners of core sample who joined household since HSE • Not contacted directly if refused re-approach at HSE • BUT had chance to be interviewed if others in household • No interviews in institutions or outside England at Wave 1
Response Rates • Issued sample of ~11,600 households; ~18,800 individuals • Achieved sample of ~7,900 households; ~12,100 individuals • ~11,400 interviews with core sample members • ~640 interviews with partners under 50 and ~70 with new partners • Response among core sample members • Overall individual response rate of 67% • Household response 70% • 96% responding within households
Achieved Sample Size - All Sample Types Numbers % Age Men Women Total Men Women Total Under 50 104 472 576 2 7 5 50-54 920 1156 2076 17 17 17 55-59 1030 1171 2201 19 17 18 60-64 813 883 1696 15 13 14 65-69 806 912 1718 15 13 14 70-74 680 797 1477 13 12 12 75-79 498 596 1094 9 9 9 80+ 485 777 1262 9 11 10 Total 5336 6764 12100 100 100 100
Accounting for response • Most significant level of non-response from households • Created weight to take account of household non-response and post-stratified to the Census 2001 age-sex profile of individuals • HSE data used to model non-response • Method = calibration weighting • Produced identical household and individual weights • Weight only produced for core sample members
Elements of non-response • Some interviews were not completed - ‘partial’ • Some interviews were completed on behalf of the respondent by a ‘proxy’ • High response to elements within survey • Housing 99.7% • Income and assets 99.0% • Self-completion 92.0% • Consents c.80% • Imputation for missing values in economic variables
Analysing the data • Select analysis sample • Core sample members (discard younger / new partners) • In-person interviews (discard partial/proxy interviews) • Weight the data • Account for the effect of the sample design
Accounting for sample design • Complex sample design affects standard errors • HSE design used stratification and clustering • Standard errors too small if do not take into account • Can imply significant effects when not significant • Affects some variables more than others • Stata has facility to account for design (svy commands) • SPSS ‘Complex samples’ module has limitations
Later waves • Interview every two years • Follow all who responded at Wave 1 • Split households • Follow all core sample members • Younger/new partners followed for one interview • Refreshment sample at Wave 3 • New cohort based on age to replenish the youngest • 50-53 years old at Wave 3