WesVar Features
General Features
Usability
- Easy-to-use Windows® interface
- No coding of procedure statements needed
- Simple workbook structure for organizing analysis requests
- Workbook output can be shared without having the source data sets
- Multiple files with replicate weights can be imported in one run
- Preferences control default settings for output, tables, regression models, and file locations
Data Import/Export
- SAS V7/V8/V9
- SAS transport file
- S-Plus
- Microsoft Access®
- Paradox
- Quattro Pro
- ASCII
- SAS for PC/DOS
- SPSS
- STATA
- Microsoft Excel®
- Lotus
- Minitab
- ODBC
- Many more
Working with WesVar Data Sets
- Add variable labels
- Perform recodes
- Subset populations
- Log file for tracking the history of changes made to a data file
Weighting
Create Weights
Import existing replicate weights or create them in WesVar.
Adjust input base weights using:
- Nonresponse adjustments
- Post-stratification
- Raking
Make adjustments for the full sample and separately for each replicate.
Replication Methods of Variance Estimation Supported
Stratified two primary sampling units (PSUs)/stratum
- Balanced repeated replication (BRR)
- Fay’s BRR method (Fay)
- Jackknife 2 (JK2)
Stratified >2 PSUs/stratum
- Jackknife n (JKn)
Unstratified
- Jackknife 1 (JK1)
Maximum Replicates
- BRR, Fay—up to 512 with bundled orthogonal matrices, up to 9,984 with user supplied matrices
- Jackknife (JK1, JK2, JKn)—up to 9,999
Tables
General Features
- Specify up to 8-way tables
- Easily generate all two-way and three-way tables
- Analyze data sets with multiple imputations
Generated Statistics
- Estimates
- Standard errors
- Coefficients of variation
- Confidence intervals
- Wilson method for confidence intervals on extreme proportions
- Unweighted sample size in each table cell
- Design effect for estimate
- Effective sample size
- t statistic for test that estimate is zero, associated p values
- Standardized rates
Computed Statistics
- A statistic that is a function of one or more survey variables or of statistics derived from a survey (e.g., RATIO=INCOME/ASSET). For example, the ratio of two means, the log-odds in a two-way table, and the difference in two ratios can be calculated using computed statistics.
- Other functions include standard operators (+, -, x, /), Log, Exp, Sqrt, Mean, Median, Quantile, and Geometric Mean
- Plausible values can be computed for education assessment tests or for multiple imputation analysis
Tests of Independence
- Pearson’s chi-square
- RS2—Rao-Scott statistic using a design effect adjustment to the Pearson chi-square
- RS3—Rao-Scott statistic based on Satterthwaite approximation
Options for Tables
- Set alpha for confidence intervals
- Set finite population correction factor for variance estimates
- Set degrees of freedom for estimated variances
- Choose method for handling missing values
- Listwise deletion (complete case analysis)
- Separate treatment of each analysis variable and computed statistic (available case analysis)
- Control method for computing medians or quantiles
- Group method
- No group method (default)
- Compute direct replicate estimates of standard errors
- Print diagnostic to measure whether replicate contributes heavily to variance estimate
Function Statistics
- A statistic computed as a function of the cell entries in a table, such as the difference between males and females (up to 4-way tables)
Output for Tables
- Requested statistics for all cells and marginals of table
- Statistic type
- Sum of weights
- Percentage of total
- Row percentages
- Column percentages
- Labels for variables and values
- Fixed decimal or scientific notation, with ability to control number of decimal places
- Replicate estimates saved in auxiliary output file
Table Viewer
- Allows users to view the output of a Table Request in a tabular format and copy the data into Microsoft Excel or other programs
Regression
General Features
- Point-and-click selection of dependent and independent variables
- Independent variables—continuous and categorical variables are allowed (class variables can have up to 256 categories)
- Automatic creation of dummy variables
- By variable for subgroup modeling
- Models with or without intercept
- Interactions can be specified among predictors
- Significance of linear combinations of parameters can be tested
- Degrees of freedom can be set for estimated variances
Types of Models
- Linear regression
- ANOVA—up to three-way interactions
- Dichotomous logistic regression
- Multinomial logistic regression (unordered categories)
Linear Regression
- Specify a transformation of a source variable as the dependent variable
- Find parameter estimates using weighted least squares algorithm
- Analyze data sets with multiple imputations
Logistic Regression
- Use modified Newton-Raphson iteration procedure to find parameter estimates
- Specify maximum number of iterations
- Set convergence criterion—relative difference of log-likelihood
- Control starting values for parameter estimates
- Define success value for the dependent variable in dichotomous logistic regression
- Define reference category for the dependent variable in multinomial logistic regression
- Compute odds ratios—estimates and confidence intervals
Output
- Model, error, and total sums of squares
- Parameter estimates
- Standardized linear regression coefficients
- Standard error of parameter estimates
- Confidence intervals for parameters
- t tests on individual parameters
- F tests on main effects and interactions
- User-defined tests on linear combinations of parameters
- Score tests for logistic parameters
- p values for hypothesis tests
- R2 values
- Linear—squared multiple correlation coefficient
- Logistic—Cox-Snell, entropy, Estrella
- Correlations of parameter estimates
- Covariances of parameter estimates
- Estimates and confidence intervals for odds ratios in dichotomous and multinomial logistic regression
- Fixed decimal or scientific notation, ability to control number of decimal places
- Replicate estimates saved in auxiliary output file
- Iteration history for logistic saved in auxiliary output file
Descriptive Statistics
Generated Statistics
- Extreme observations—five smallest and five largest unweighted observations
- Influential observations—five most influential weighted observation with respect to the mean and five most influential observations with respect to the total
- Univariate statistics:
- Weighted and unweighted count of observations, sum, mean, variance, skewness, kurtosis, percentiles, and geometric mean
- Standard errors for weighted statistics
- Both replication and Woodruff standard errors for percentiles
- Minimum, maximum, and unweighted count of missing observations
- Weighted and unweighted correlation coefficients
- Replication standard errors for weighted correlation coefficients