1 / 21

Alternative Methods of Unit Nonresponse Weighting Adjustments: An Application from the 2003 Survey of Small Business Fin

Alternative Methods of Unit Nonresponse Weighting Adjustments: An Application from the 2003 Survey of Small Business Finances *. Lieu N. Hazelwood, Traci L. Mach, and John D. Wolken ICES 2007 June 20, 2007

morrisa
Download Presentation

Alternative Methods of Unit Nonresponse Weighting Adjustments: An Application from the 2003 Survey of Small Business Fin

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Alternative Methods of Unit Nonresponse Weighting Adjustments: An Application from the 2003 Survey of Small Business Finances * Lieu N. Hazelwood, Traci L. Mach, and John D. Wolken ICES 2007 June 20, 2007 * Published as a Federal Reserve Board Working Paper 2007-10. The views expressed herein are those of the authors. They do not necessarily reflect the opinions of the Federal Reserve Board or its staff.

  2. Reasons for applying a Nonresponse Adjustment (NRADJ) • Biased estimates (both relative and absolute) • Inappropriate variances in the weights • Invalid confidence statements

  3. Nonresponse Adjustment Methods • Traditional Weighting Class • Direct Propensity Scoring • Propensity Stratification • 2003 SSBF Hybrid Method

  4. Talk Overview • Survey overview • NRADJ methods overview • Application using the 2003 SSBF • Results

  5. Survey of Small Business Finances (SSBF) • Survey conducted every five years • Data from the 2003 Survey released in October 2006 • Three previous surveys – 1987, 1993, and 1998

  6. 2003 Overview • Short telephone screening interview to determine main eligibility Screener response rate: 62% • 4,240 completed main interviews Main response rate: 52% • Overall response rate: 32%

  7. Current Analysis • Used only the screener interview • Focuses solely on unit nonresponse • Future study may deal with the main interview

  8. Traditional Weighting Class Adjustment • Widely used • Partitioned into homogenous cells • NRADJ=inverse of cell response rate • Cell counts maintained • Population count maintained

  9. Propensity Weighting • Estimate a propensity score using a probit or a logit model • Propensity scores are used to form the NRADJ cell

  10. Traditional versus Propensity • Traditional Cell Weighting Preserves cell counts Preserves population counts Limited number of variables • Propensity Weighting May not preserve cell counts May not preserve population counts Allows large number of variables

  11. Direct Propensity versus Propensity Stratification • Direct Propensity NRADJ=mean propensity score NRADJ=inverse of the propensity score • Propensity Stratification Stratify sample using propensity scores NRADJ=inverse of cell response rate

  12. 2003 SSBF Hybrid Method • Size class (“super strata”) • Propensity strata: propensity scores • NRADJ=inverse of cell response rate • Cell and population counts maintained within super strata

  13. Implementation • Traditional cell weighting used 98 strata variables for stratification • Propensity Methods estimated a logit model that incorporated most frame variables • Direct Propensity divided the sample into 5 cells and used the inverse of the mean propensity score within each cell • Propensity Stratification divided the sample into 5 cells and used the inverse of the cell response rate • Hybrid method used the inverse of the response rate w/in propensity strata w/in super strata

  14. Super Strata Selection • Size class important to maintain • Size Classes: 0-19 employees 20-49 employees 50-99 employees 100-499 employees • Smallest size class had 25 NRADJ cells and the rest had 5

  15. Results: Hybrid versus Traditional Cell Weighting • Similar to Traditional method • Mixed results within subcategories Different weight variances Different point estimates • But differences are small

  16. Results: Hybrid versus Direct Propensity Method • Small differences in point estimates • Smaller weight variances • Difference in population counts by 7.5%

  17. Results: Hybrid versus Propensity Stratification • Slightly higher weight variances • Slightly different point estimates • Preserves population count but not size class counts

  18. Total Firms and Variance of Screener NRADJ Weight

  19. Variance of the NRADJ Screener weight by Business Type

  20. Conclusion • Similar weight variances and point estimates • Easier implementation • More homogenous NRADJ cells

  21. Contact Information • Lieu.N.Hazelwood@frb.gov • Traci.L.Mach@frb.gov • John.D.Wolken@frb.gov • www.federalreserve.gov/pubs/oss/ oss3/nssbftoc.htm

More Related