1. 1 Introduction to Survey Data Analysis Linda K. Owens, PhD Assistant Director for Sampling & Analysis Survey Research Laboratory University of Illinois at Chicago

2. 2 Survey Research Laboratory  Data cleaning/missing data  Sampling bias reduction Focus of the seminar

3. 3 Survey Research Laboratory When analyzing survey data... 1. Understand & evaluate survey design 2. Screen the data 3. Adjust for sampling design

4. 4 Survey Research Laboratory 1. Understand & evaluate survey  Conductor of survey  Sponsor of survey  Measured variables  Unit of analysis  Mode of data collection  Dates of data collection

5. 5 Survey Research Laboratory  Geographic coverage  Respondent eligibility criteria  Sample design  Sample size & response rate 1. Understand & evaluate survey

6. 6 Survey Research Laboratory  Nominal  Ordinal  Interval  Ratio Levels of measurement

7. 7 Survey Research Laboratory ALWAYS examine raw frequency distributions for… (a) out-of-range values (outliers) (b) missing values 2. Data screening

8. 8 Survey Research Laboratory Out-of-range values:  Delete data  Recode values 2. Data screening

9. 9 Survey Research Laboratory  can reduce effective sample size  may introduce bias Missing data:

10. 10 Survey Research Laboratory  Refusals (question sensitivity)  Don’t know responses (cognitive problems, memory problems)  Not applicable  Data processing errors  Questionnaire programming errors  Design factors  Attrition in panel studies Reasons for missing data

11. 11 Survey Research Laboratory  Reduced sample size – loss of statistical power  Data may no longer be representative– introduces bias  Difficult to identify effects Effects of ignoring missing data

12. 12 Survey Research Laboratory Assumptions on missing data  Missing completely at random (MCAR)  Missing at random (MAR)  Ignorable  Nonignorable

13. 13 Survey Research Laboratory Missing completely at random (MCAR)  Being missing is independent from any variables.  Cases with complete data are indistinguishable from cases with missing data.  Missing cases are a random sub- sample of original sample.

14. 14 Survey Research Laboratory Missing at random (MAR)  The probability of a variable being observed is independent of the true value of that variable controlling for one or more variables.  Example: Probability of missing income is unrelated to income within levels of education.

15. 15 Survey Research Laboratory Ignorable missing data  The data are MAR.  The missing data mechanism is unrelated to the parameters we want to estimate.

16. 16 Survey Research Laboratory Nonignorable missing data  The pattern of data missingness is non-MAR.

17. 17 Survey Research Laboratory Methods of handling missing data  Listwise (casewise) deletion: uses only complete cases  Pairwise deletion: uses all available cases  Dummy variable adjustment: Missing value indicator method  Mean substitution: substitute mean value computed from available cases (cf. unconditional or conditional)

18. 18 Survey Research Laboratory  Regression methods: predict value based on regression equation with other variables as predictors  Hot deck: identify the most similar case to the case with a missing and impute the value Methods of handling missing data

19. 19 Survey Research Laboratory  Maximum likelihood methods: use all available data to generate maximum likelihood-based statistics. Methods of handling missing data

20. 20 Survey Research Laboratory  Multiple imputation: combines the methods of ML to produce multiple data sets with imputed values for missing cases Methods of handling missing data

21. 21 Survey Research Laboratory Multiple Imputation Software SAS—user written IVEware, experimental MI & MIANALYZE STATA—user written ICE R—user written libraries and functions SOLAS S-PLUS

22. 22 Survey Research Laboratory Methods of handling missing data: summary Listwise deletion assumes MCAR or MAR Dummy variable adjustment is biased, don’t use Conditional mean substitution assumes MCAR within cells Hot Deck & Regression improvement on mean substitution but... Multiple Imputation is least biased but computationally intensive

23. 23 Survey Research Laboratory Missing Data Final Point All methods of imputation underestimate true variance  artificial increase in sample size  values treated as though obtained by data collection Rao (1996) On variance estimation with imputed survey data. Jrn. Am. Stat. Assoc. 91:499-506 Fay (1996) in bibliography

24. 24 Survey Research Laboratory Types of survey sample designs  Simple Random Sampling  Systematic Sampling  Complex sample designs  stratified designs  cluster designs  mixed mode designs

25. 25 Survey Research Laboratory Why complex sample designs?  Increased efficiency  Decreased costs

26. 26 Survey Research Laboratory  Statistical software packages with an assumption of SRS underestimate the sampling variance.  Not accounting for the impact of complex sample design can lead to a biased estimate of the sampling variance (Type I error). Why complex sample designs?

27. 27 Survey Research Laboratory Sample weights  Used to adjust for differing probabilities of selection.  In theory, simple random samples are self-weighted.  In practice, simple random samples are likely to also require adjustments for nonresponse.

28. 28 Survey Research Laboratory Types of sample weights  Poststratification weights: designed to bring the sample proportions in demographic subgroups into agreement with the population proportion in the subgroups.  Nonresponse weights: designed to inflate the weights of survey respondents to compensate for nonrespondents with similar characteristics.  “Blow-up” (expansion) weights: provide estimates for the total population of interest.

29. 29 Survey Research Laboratory Syntax examples of design-based analysis in STATA, SUDAAN, & SAS STATA svyset strata strata svyset psu psu svyset pweight finalwt svyreg fatitk age male black hispanic SUDAAN proc regress data=”c:\nhanes.sav” filetype=spss desgn=wr; nest strata psu; weight finalwt subpgroup sex race; levels 23; model fatintk = age sex race;

30. 30 Survey Research Laboratory SAS proc surveyreg data=nhanes; strata strata; cluster psu; class sex race; model fatintk = age sex race; weight finalwt Syntax examples of design-based analysis in STATA, SUDAAN, & SAS

31. 31 Survey Research Laboratory In summary, when analyzing survey data...  Understand & evaluate survey design  Screen the data – deal with missing data & outliers.  If necessary, adjust for study design using weights and appropriate computer software.

32. 32 Survey Research Laboratory Bibliography On website with slides Big names in area of missing data are: Roderick Little Donald B. Rubin Paul Allison See also McKnight et.al. Missing Data: A Gentle Introduction Large government datasets (NSFG, NHANES, NHIS) often include detailed methodological documentation on imputation and weight construction.

33. 33 Survey Research Laboratory Thank You! www.srl.uic.edu