1 / 32

Item Nonresponse in Mail, Web, and Mixed Mode Surveys: A Summary of the AAPOR Session

Item Nonresponse in Mail, Web, and Mixed Mode Surveys: A Summary of the AAPOR Session. Benjamin L. Messer Washington State University PAPOR Mini-Conference, Berkeley, CA June 24, 2011. 2011 AAPOR Session.

gin
Télécharger la présentation

Item Nonresponse in Mail, Web, and Mixed Mode Surveys: A Summary of the AAPOR Session

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. Item Nonresponse in Mail, Web, and Mixed Mode Surveys: A Summary of the AAPOR Session Benjamin L. Messer Washington State University PAPOR Mini-Conference, Berkeley, CA June 24, 2011

  2. 2011 AAPOR Session • Are Measurement and Item Nonresponse Differences a Problem in Web and Mail Mixed-Mode Surveys? • 1) Millar & Dillman, “Do mail and web produce different answers? Mode differences in question response and item nonresponse rates” • 2) Smyth & Olson, “Comparing numeric and text open-end response in mail and web surveys.” • 3) Lesser, Olstad, Yang, & Newton, “Item nonresponse in web and mail response to general public surveys.” • 4) Messer, Edwards, & Dillman, “Determinants of web and mail item nonresponse in address-based samples of the general public.” • 5) Israel & Lamm, “Item nonresponse in a client survey of the general public.”

  3. Background • Item nonresponse and measurement differences have not been thoroughly tested in web and mail surveys, particularly of the general public • Both types of error are thought to be similar between the modes since they are self-administered and visual, although some have found that mail obtains higher error rates (i.e. lower data quality)

  4. PAPER 1: Millar & Dillman • Tested item nonresponse and measurement error in web and mail surveys • Used data from the WSU Student Experience Survey, in which the population has near universal access to the web and both postal and email addresses, in the Spring and Fall of 2009 • Spring 2009 study: 100 items, 36 questions • Fall 2009 study: 76 items, 33 questions

  5. Response Rates

  6. Methods • Item nonresponse: Calculated percent of respondents missing a response and compared rates with z-tests • Measurement: Used chi-square tests to test for differences in the distribution of responses

  7. Item Nonresponse • No statistically significant differences in overall item nonresponse rates between web and mail modes in each experiment

  8. Item Nonresponse, cont. • Some individual items exhibited significant mode differences in item nonresponse rates • Multi-item questions • Branching questions • Open-end questions

  9. Measurement Differences • Very few significant differences exist • Multi-item questions are most likely to exhibit measurement differences

  10. PAPER 2: Smyth & Olson • Tested web vs. mail measurement differences in open-ended questions, a question type more prone to error in self-administered, visual modes • Used data from the 2009 Quality of Life in Changing Nebraska Survey (QLCN) • 45.6% total response rate

  11. Methods • Four experimental treatment groups • Mail-only, Mail+Web, Web+Mail, Web-only • Tested two different open-end question formats • 7 number box items (e.g. date of birth) • Measurement differences (distributions of answers to questions) • Item nonresponse rates (missing or not missing) • 2 text box items (e.g. descriptions or narratives) • Item nonresponse rates (missing or not missing) • Quality (e.g. length, number of themes, elaboration)

  12. Results • Number Box: • Regression analyses resulted in very few significant differences in either responses or measurement indicators between web and mail modes • What few differences occurred were largely due to differential participation in web and mail modes, as well as questionnaire design • Text Box: • Analyses show that item nonresponse rates and data quality were very similar across modes • However, the differences found to be significant were largely due to questionnaire design and mode

  13. PAPER 3: Lesser, Newton, & Yang • Compared item nonresponse rates across modes and question types and compared unit response between mail, web, and telephone modes • Used data from the 2008 & 2010 general public surveys conducted for the Oregon Department of Transportation • Address-based samples

  14. Methods • Four groups: • Telephone, Mail-only, Web+Mail, Web/Mail (choice) • Five question types: • Likert, Open-end, Filtered, Tabular, & Demographic

  15. Item Nonresponse Results Type Mail Web+Mail Web/Mail Likert 2.4 2.6 2.4 Filter 0.9 2.7 2.7 Table 4.9 5.6 5.2 Open 3.2 4.1 4.3 Demo 4.0 4.3 3.3 Overall 3.5 4.1 3.9

  16. Item Nonresponse Results: 2008 Type Mail Web/Mail Telephone Likert 1.8 1.7 0.65 Filter 1.3 1.4 0.44 Table 4.1 3.2 0.20 Open 5.5 4.3 2.55 Demo 3.7 3.6 4.05 Overall 3.0 2.6 0.87

  17. Unit Response Results

  18. PAPER 4: Messer, Edwards, & Dillman • Tested for item nonresponse differences controlling for survey mode and design, question types, and demographic characteristics • Used data from three general public surveys with address-based samples: • 2007 Lewiston & Clarkston Quality of Life Survey (LCS) • Web+Mail, 55%; Mail-only, 66% • 92 items, 51 questions • 2008 Washington Community Survey (WCS) • Web+Mail, 40%; Mail-only, 50% • 110 items, 52 questions • 2009 Washington Economic Survey (WES) • Web+Mail, 50%; Mail-only, 62% • 96 items, 57 questions

  19. Methods • Survey modes: web, mail, & mail-follow-up • Survey designs: web+mail& mail-only • Question Types • Screened, Multi-item, Open-end, Close-end • Question Formats • Factual, Attitudinal, Behavioral • Demographic Characteristics • Gender, age, education, income • Item nonresponse rate: missing responses divided by total number of possible responses for each respondent

  20. Item Nonresponse Rates by Survey Mode

  21. Question Effects • The same trend persists across questions types and formats, with web obtaining the lowest rate, followed by mail, and then mail follow-up in all three surveys • In regression analyses controlling for both survey mode and question type & format(e.g. screened, multi-item, etc.) are significant predictors of item nonresponse

  22. Demographic Effects • The same trends in item nonreponse rates are also found across different demographic subgroups. • We also found that older respondents with less education and income have higher item nonresponse rates • Regression analyses controlling for mode and demographics indicate this could be due to differential participation

  23. Item Nonresponse Rates by Survey Design • Web+mail and mail-only designs obtained statistically similar item nonresponse rates ranging between 4-8% • Question Type and Format • Regression analyses indicate that question format is a significant predictor of item nonresponse, controlling for design • Demographic Characteristics • Older respondents with lower levels of education and income tend to exhibit higher rates, controlling for design • Survey design was not a significant predictor

  24. PAPER 5: Israel & Lamm • Tested for item nonresponse differences controlling for survey mode, question characteristics, and respondent demographics • Used data from the 2008, 2009, 2010 University of Florida Extension Customer Satisfaction Survey

  25. Methods • Mail-only and Web-only modes • Had postal and email addresses of clients • Used logistic regression and HLM statistical methods • Demographic Characteristics: • Gender, education, age, race, & client participation status • Question Types • Open-end, screened, demographic, grid, & yes/no

  26. Item Nonresponse Rates by Mode Mail Web 7.3% 5.1% Mail Web 7.1% 6.3% Mail Web 7.1% 5.5% 2008 2009 2010

  27. Completed Survey Rate by Mode Mail Web Mail Web Mail Web 28.2% 47.2% 28.4% 48.2% 30.0% 41.3% 2008 2009 2010

  28. Determinants of Item Nonresponse • Logistic regression analysis shows that demographic characteristics of respondents have little affect on item nonresponse • Open-end and screened question types yielded the highest item nonresponse rates • HLM analyses result in few mode or demographic effects  question characteristics

  29. Tying it all together • Millar & Dillman found few item nonresponse rate differences between web and mail in a population with easy access to both, however questions types were found to influence data quality net of mode • For the general public, Smyth & Olson discovered that one particular question type obtained variable data quality depending on the survey mode and the format of the question (numeric vs. text), with web obtaining slightly lower nonresponse rates and better data quality • In multiple general pubic web and mail surveys, Lesser et. al., Messer et. al., and Israel & Lamm found that web obtained lower item nonresponse rates. However, when combined with a mail follow-up, web+mail item nonresponse rates approximate those obtained by mail alone. In addition, question characteristics and respondent demographics (Messer et. al.) were found by to influence item nonresponse rates. Lesser et al. also showed that telephone obtained the lowest item nonresponse rates, as could be expected.

  30. Where we stand now… • Using similar questionnaires, web may obtain slightly better data quality than mail, at least in general public surveys, but also currently obtains lower unit response rates, indicating a trade-off. • In addition, web may also attract types of respondents that have a greater propensity to complete the questionnaire compared to mail, and complex question types/formats can produce lower data quality net of survey mode.

  31. In addition…. • I also saw Danna L. Moore present “Driving Respondents to the Web: Experimental Trial of Benefit Appeals and Impacts on Survey Completion” • She tested the use of three different letter types • One basic letter with no appeals • One letter with an appeal to save the government money by using the web (instead of paper) • One letter with an appeal to help out the local community • Results indicated that the appeal to save money had a positive, significant effect on web response rates, compared with the other two letters • About 3 percentage points higher

  32. Citations and Contact Info • Millar, Morgan M. & Don A. Dillman. 2011. “Improving Response Rates to Web and Mixed-Mode Surveys.” Public Opinion Quarterly, 75:2(249-69) • morgan_millar@wsu.edu • Smyth, Jolene & Kristen Olson. Unpublished Manuscript. “Comparing Numeric and Text Open-End Responses in Mail & Web Surveys.” • jsmyth2@unl.edu • Lesser, Virginia, Andy Olstad, Danny Yang & Lydia Newton. Unpublished Manuscript. “Item nonresponse in web and mail response to general public surveys.” • lesser@science.oregonstate.edu • Messer, Benjamin & Don A. Dillman. Forthcoming. “Surveying the General Public Over the Internet Using Address-based Sampling and Mail Contact Procedures.” Public Opinion Quarterly • bmesser@wsu.edu • Israel, Glenn D. & Alex J. Lamm. Unpublished Manuscript. “Item-Nonresponse in a Client Survey of the General Public.” • gdisrael@ufl.edu

More Related