Survey data quality: Does verbal ability and text readability matter?

Similar documents
May 15, Dear Parents and Community Members:

Confirmatory Factor Analysis of the BCSSE Scales

DISTRICT LETTERHEAD. REVISED TEMPLATE (Letter Sent on District s Letterhead) School Annual Education Report (AER) Cover Letter

School Annual Education Report (AER) Cover Letter

Body Weight Behavior

Assistant Superintendent of Business &

School Annual Education Report (AER) Cover Letter

ANN ARBOR PUBLIC SCHOOLS 2555 S. State Street Ann Arbor, MI www. a2schools.org Pioneer High School Annual Education Report (AER)!

Estimating Behavior Frequencies: Do Vague and Enumerated Estimation Strategies Yield Similar Results? James S. Cole. Ali Korkmaz

MACOMB MONTESSORI ACADEMY School Annual Education Report (AER) Cover Letter - REVISED

GENENTECH S INNOVATIVE PILOTS TO ALTER CERVICAL CANCER S TRAJECTORY

Patient Instructions for Health Literacy and Language Barriers: Values & Challenges. Charles Lee, MD President & Founder Polyglot Systems, Inc.

22932 Woodward Ave., Ferndale, MI School Annual Education Report (AER) Cover Letter

Guidelines & Examples for Preparing Plain Language Summaries for. Authors of ACSM Journal Articles

Common Measure: Nutrition, High School

16 WEEKS TO A HEALTHIER YOU

Attitude Towards Tobacco Use

Providing Highly-Valued Service Through Leadership, Innovation, and Collaboration

Ruffalo Noel-Levitz Student Satisfaction Inventory Results: All Students Gallaudet University Spring 2018 Report

Dietary Behavior High School Version

Billing Code: P DEPARTMENT OF HEALTH AND HUMAN SERVICES. Centers for Disease Control and Prevention. [30Day-18-17AUZ]

PSYC3010 Advanced Statistics for Psychology

Warren Consolidated Schools

Where We Are: NCI s Informed Consent Template and Updates to Address Changing Needs

Accepted Manuscript. The readability of ipledge program patient education materials. Rachel Howard, BS, Gideon Smith, MD, PhD

This SOP applies to all IRB administrative staff, members and investigators. IRB members, administrative staff and investigators.

Evaluation of a Computer-Based Toolkit for Hematological Cancer Patients, Caregivers, and Family/Friends

2009 JEMF Project. Survey to Inform Development of the Genetic Counseling Cultural Competence Toolkit (GCCCT)

Georgia Cancer Quality Information Exchange

Reimagining Parent and Family Orientation. Ashlea Coulter Alison Welch Rutgers University

ACDI. An Inventory of Scientific Findings. (ACDI, ACDI-Corrections Version and ACDI-Corrections Version II) Provided by:

THE EMERGE SURVEY ON TAKING PART IN BIOBANK RESEARCH: VERSION A

FOOD FOR THOUGHT Topical Insights from our Subject Matter Experts THE NFL S SENSORY PANEL PHILOSOPHY

EFSA s Reputation Barometer

Jeanne M. Moseley, MPH Senior Lecturer & Director Global Health Program Division of Nutritional Sciences Savage Hall, B15 Cornell University

Identifying Barriers to Cervical Cancer Screening Faced by South Asian Muslim Immigrant Women in Calgary

Lessons Learned from HPV Vaccine Study Among Young Adult Women and Men

Division of Research Policy

Competency Rubric Bank for the Sciences (CRBS)

Validation of an Analytic Rating Scale for Writing: A Rasch Modeling Approach

STRESS LEVELS AND ALCOHOL USE AMONG UNDERGRADUATE STUDENTS: A QUANTITATIVE STUDY. Noemi Alsup California State University, Long Beach May 20, 2014

Comparing Multiple-Choice, Essay and Over-Claiming Formats as Efficient Measures of Knowledge

These materials are Copyright NCHAM (National Center for Hearing Assessment and Management). All rights reserved. They may be reproduced

Going DEEP into Oklahoma with the Diabetes Empowerment Education Program

D Scerner: Critical Appraisal of a Review Article on the Role of Dopamine in Addiction

School Consultation Project Application

Advantages of Within Group Analysis of Race/Ethnicity

Acute Withdrawal Symptoms of Alcohol and Other Drugs EvidEncE BasEd PracticEs (EBP): Consistency in the Group Setting

Sexual Agreements and HIV Risk Among Gay Male Couples

Capacity to Consent to Research in Patients with Acute Pain: A Pilot Study

Leveraging Social Networks to Promote Cancer Prevention Health Behaviors

Items pertaining to drunk driving

Preliminary Thoughts

2016 Pharmacist Re-Licensure Survey Instrument

CONSIDERATIONS WHEN ENGAGING THE COMMUNITY IN BRAIN HEALTH: Lessons Learned from Memory Screening and Memory Training

AMERICAN SIGN LANGUAGE GREEN BOOKS, A TEACHER'S RESOURCE TEXT ON GRAMMAR AND CULTURE (GREEN BOOK SERIES) BY CHARLOTTE BAKER-SHENK

D Scerner: Critical Appraisal of a Review Article on the Role of Dopamine in Addiction

Early Intervention in Psychosis Program (EIP)

Differences in Alcohol Use among Ohio State Students. Center for the Study of Student Life

www. russellresearch.com

EVALUATIONWEB 2014 DIRECTLY FUNDED CBO CLIENT-LEVEL DATA COLLECTION TEMPLATE

FUNDAMENTALS OF PEDIATRIC DENTISTRY BY RICHARD J. MATHEWSON DOWNLOAD EBOOK : FUNDAMENTALS OF PEDIATRIC DENTISTRY BY RICHARD J.

Sensitivity and specificity of depression screening tools among adults with intellectual and developmental disabilities (I/DD)

Examining Temperament in Exercise Dependence and Eating Disorders

ALABAMA SELF-ASSESSMENT INDEX PILOT PROGRAM SUMMARY REPORT

ESSA Accountability Alignment

Internal Consistency and Reliability of the Networked Minds Measure of Social Presence

N Utilization of Nursing Research in Advanced Practice, Summer 2008

Moms Help Organization Helping Moms to be the best Moms they can be! West Sample Road, #24 Coral Springs, FL

INTRODUCTION TO THE AEROBIC BUILDING PHASE

Optimizing Communication of Emergency Response Adaptive Randomization Clinical Trials to Potential Participants

MAKING YOUR GROUP A SUCCESSFUL EXPERIENCE THE GROUP PROCESS

Meaningful Inclusion of People on the Autism Spectrum

Review of Individual Budgets Advice to Social workers Draft 1 Introduction

HVTN 505 Study HIV Vaccine Candidate Not Effective

START & RUN A MARIJUANA DISPENSARY OR POT SHOP: WHEREVER IT IS LEGAL! (START & RUN A BUSINESS SERIES) BY JAY CURRIE

Elements of Nonfiction

Creating a Positive Professional Image

Wellness Assessment: Intellectual Wellness. Center for the Study of Student Life

Breaking News English.com Ready-to-Use English Lessons by Sean Banville

Bibliography. General Requirements for Informed Consent. 46 C.F.R. Sec Office of NIH History.

Reliability and Validity checks S-005

Wellness Assessment: Spiritual Wellness. Center for the Study of Student Life

Publishing while female

Wellness Assessment: Financial Wellness. Center for the Study of Student Life

Ch. 1 Collecting and Displaying Data

SNAP Access: Improving Seniors' Health Outcomes

Accessible Print Materials

Wellness Assessment: Creative Wellness. Center for the Study of Student Life

SUBSTANCE USE AND ABUSE ON CAMPUS: RESULTS FROM THE UNIVERSITY OF MICHIGAN STUDENT LIFE SURVEY (2011)

Limited English Proficiency Plan

ADMS Sampling Technique and Survey Studies

Cultural Beliefs About Cancer Screening: Are They Deadly?

SOCIAL DETERMINANTS OF HEALTH: A COMPARATIVE APPROACH BY ALAN DAVIDSON

Arthritis Physical Therapy Marketing Project. March 1, 2016 Erin Loomis, Westat

Syllabus Psy 371 Abnormal Psychology Spring :30 2:45 p.m. Tuesdays and Thursdays Physical Science Building 217

Survey Instrument. Appendix B

A Thesis. entitled. Laura Benge

Burden of Cancer in California

San José State University CASA/Kinesiology KIN 292B, Seminar in Sports Medicine, Spring, 2017

Transcription:

AIR Annual Forum Long Beach, CA May, 2013 Survey data quality: Does verbal ability and text readability matter? James Cole, Ph.D. Associate Scientist

Introduction Satisficing Krosnick, Narayan, and Smith (1996) identify three regulators of satisficing (a source of survey data error). One of them is task difficulty. Task difficulty has to do with how familiar the language is to the respondent. Survey creators try to decrease task difficulty and accommodate low verbal ability by writing items that use plain, simple language that is easily understood by the target audience (Fowler, 2008). Use of focus groups, cognitive interviews, and pilot testing are important components for determining task difficulty for the survey respondent. However, the verbal ability of survey respondents is one factor that can determine task difficulty and is overlooked (Krosnick, 1999).

Introduction Satisficing One tool for aligning text difficulty with appropriate reading skills has been the use of grade-level calculators. Though not widely used in survey research, they are used in fields such as health in the development of their promotional, educational materials, as well as surveys (e.g., NCI, NIH, CDC, various state depts of health). In addition, some health science IRB s are now starting to require this information for informed consent documents. There is some research regarding use of grade level estimations in survey research (e.g., Calderon, Morales, Liu, & Hays, 2006), but much is still not known.

Introduction Research Questions Research questions include: 1. Is low verbal ability, as compared to very high verbal ability respondents, associated with suspicious data quality? 2. Are commonly used readability programs that provide grade-level reading estimates associated with indicators of suspicious data quality?

Method Data Source Data for this study comes from the 2011 web administration of the National Survey of Student Engagement (NSSE) and only included full-time, first-year students with institution-reported SAT verbal scores. More than 81,000 respondents enrolled at 441 US institutions across the United States were included in the study. Average 2011 institutional response rate: 33% Respondent Characteristics Male 36.2% Female 63.8% African American 9.0% Asian American 7.8% Caucasian 61.8% Hispanic 10.5% Other 10.9% Non-First Generation 37.4% First Generation 62.6% TOTAL Count 81,164

Method Data Source NSSE annually collects information at hundreds of four-year colleges and universities about student participation in programs and activities that institutions provide for their learning and personal development. The results provide an estimate of how undergraduates spend their time and what they gain from attending college.

Method Method Key variables used in this study to identify suspicious data include: verbal ability survey duration straight-lining (indicated by the respondent checking the same response for the entire set of items on that screen) item skipping break-offs (non-completers) skip through (respondents that submit a screen in a matter of seconds) hanging (respondents that take an exceedingly long time to submit screen) grade-level calculations of survey text.

Method Verbal ability For this study, the SAT verbal test score measured verbal ability. The lowest 20% (scores between 200 and 460) were used as the low ability group. The top 20% (scores between 640 and 800) were considered the high verbal ability group. Survey duration By screen and per item (average duration divided by number of items on screen)

Method Straight-lining Indicated by the respondent checking the same response for the entire set of items on that screen Seven screens used where straight-lining was possible. These 7 screens used for all analysis.

Method Item skipping Items skipped per screen Mutually exclusive of those that break-off Break-offs (non-completers) Stopped submitting screens and did not return Skip through Respondents that submit a screen in a matter of seconds Calculated by taking bottom 5% in duration for each screen. Each screen calculated separately Hanging Respondents that take an exceedingly long time to submit screen Calculated by taking top 5% in duration for each screen. Each screen calculated separately

Method Grade-Level Calculations Four procedures used to calculate grade level reading estimates 1. Flesch-Kincaid (MS Word) 2. Flesch-Kincaid (online calculator)

Method Grade-Level Calculations 3. Gunning Fog Index 4. Automated Reading Index (ARI)

Method Grade-Level Calculation Procedures 1. Text from each of the seven screens were copied into the grade-level calculator 2. Text included instructions at top and item wording 3. Response category text was not included in the text analysis

Method Groups Three groups of respondents were created based on verbal test scores Respondent SAT Verbal Characteristics =<460 461 to 639 =>640 TOTAL 20.6% 59.0% 20.4% Male 18.5% 58.3% 23.2% Female 21.7% 59.3% 18.9% African Am 46.8% 48.0% 5.2% Asian Am 23.6% 59.7% 16.7% Caucasian 27.8% 53.1% 19.1% Hispanic 37.5% 53.3% 9.2% Other 25.4% 56.5% 18.1% Non-First Gen 12.2% 59.4% 28.4% First Gen 31.9% 58.5% 9.6% SAT Verbal =<460 461 to 639 =>640 Baccalaureate 19.9% 52.5% 27.6% Masters 29.3% 60.0% 10.7% Doctoral 12.1% 61.2% 26.8% Private 22.4% 59.3% 18.3% Public 17.1% 58.3% 24.6%

Results Is low verbal ability, as compared to very high verbal ability respondents, associated with suspicious data quality? Duration (mid 90% only) Screen (%) 1 2 3 4 5 14 15 Survey Bottom 20% 60.6 60.6 52.8 37.2 52.2 34.2 33.0 900.6 Middle 60% 55.2 57.0 49.8 34.2 46.2 33.0 32.4 857.4 Upper 20% 52.8 53.4 48.0 33.6 45.6 31.8 31.8 823.2 Difference Bottom/Upper 7.8 7.2 4.8 3.6 6.6 2.4 1.2 77.4 Significance 1 p<.001 p<.001 p<.001 p<.001 p<.001 p<.001 p<.001 p<.001 1 Scheffés post hoc test

Results Is low verbal ability, as compared to very high verbal ability respondents, associated with suspicious data quality? Item Missing Screen 1 2 3 4 5 14 15 Bottom 20% 5.8% 10.7% 6.2% 5.0% 3.5% 7.9% 9.1% Middle 60% 4.5% 8.7% 4.8% 3.6% 2.5% 5.9% 6.0% Upper 20% 4.0% 7.1% 4.4% 2.8% 2.0% 4.1% 4.6% Difference Bottom/Upper 1.8% 3.6% 1.8% 2.2% 1.5% 3.8% 4.5% Significance 1 p<.05 p<.05 p<.05 p<.05 p<.05 p<.05 p<.05

Results Is low verbal ability, as compared to very high verbal ability respondents, associated with suspicious data quality? Straight Lining Screen 1 2 3 4 5 14 15 Bottom 20% 2.6% 3.3% 6.8% 24.1% 1.6% 22.6% 12.4% Middle 60% 1.5% 1.8% 4.2% 18.8%.8% 17.4% 8.6% Upper 20%.9% 1.1% 2.3% 11.1%.4% 10.3% 5.1% Difference Bottom/Upper 1.7% 2.2% 4.5% 13.0% 1.2% 12.3% 7.3% Significance 1 p<.05 p<.05 p<.05 p<.05 p<.05 p<.05 p<.05

Results Is low verbal ability, as compared to very high verbal ability respondents, associated with suspicious data quality? Break-Offs Screen 1 2 3 4 5 14 15 Bottom 20% - 5.3% 2.2% 1.0%.6% 1.4% 1.3% Middle 60% - 5.0% 1.7%.8%.5% 1.0%.8% Upper 20% - 3.9% 1.6%.6%.3%.5%.5% Difference Bottom/Upper - 1.4%.6%.4%.3%.9%.8% Significance 1 - p<.05 p<.05 p<.05 p<.05 p<.05 p<.05

Percent Results Item missing, straight-lining, and break-offs are mutually exclusive activities. What percentage of respondents are displaying one of these activities? Bottom 20% Middle 60% Upper 20% 35 30 30.1 31.9 25 20 15 10 5 8.4 4.9 19.3 12.1 15.2 8.3 14.5 5.7 2.7 14.9 22.8 10.2 0 1 2 3 4 5 14 15 Screen

Results Is low verbal ability, as compared to very high verbal ability respondents, associated with suspicious data quality? Skip Through Screen 1 2 3 4 5 14 15 Bottom 20% 4.3% 6.7% 7.1% 6.0% 4.8% 7.0% 7.3% Middle 60% 4.1% 4.5% 4.5% 4.6% 4.8% 4.5% 4.3% Upper 20% 4.8% 3.1% 2.5% 3.0% 4.9% 2.7% 2.6% Difference Bottom/Upper -.5% 3.6% 4.6% 3.0% -.1% 4.3% 4.7% Significance 1 n/s p<.05 p<.05 p<.05 n/s p<.05 p<.05

Results Is low verbal ability, as compared to very high verbal ability respondents, associated with suspicious data quality? Hanging Screen 1 2 3 4 5 14 15 Bottom 20% 5.7% 5.2% 5.2% 5.6% 5.2% 5.3% 5.4% Middle 60% 4.9% 5.0% 5.1% 4.8% 4.9% 5.2% 5.2% Upper 20% 4.7% 5.2% 4.8% 5.0% 5.5% 4.7% 4.8% Difference Bottom/Upper 1.0%.0%.4%.6% -.3%.6%.6% Significance 1 p<.05 n/s n/s n/s n/s p<.05 n/s

Results Are commonly used readability programs that provide grade-level reading estimates associated with indicators of suspicious data quality? Grade Level Calculations Screen MS Word FK Flesch-Kincaid ARI FOG 1 8.8 10.0 10.0 10.7 2 10.0 9.7 8.5 10.8 3 11.9 11.8 11.2 13.4 4 13.9 14.6 14.4 16.8 5 7.4 7.7 6.7 7.1 14 13.3 13.9 12.5 13.6 15 10.6 11.3 10.0 12.9

Results Are commonly used readability programs that provide grade-level reading estimates associated with indicators of suspicious data quality? Grade Level Calculations Screen Mean Grade Level Easiest 1 5 7.2 2 2 9.8 3 1 9.9 4 15 11.2 5 3 12.1 6 14 13.3 Hardest 7 4 14.9

Results Are commonly used readability programs that provide grade-level reading estimates associated with indicators of poor data quality? Grade Level Calculations Grade Level Rank D/item IM SL BO ST Ha 5 1 15 1 5 5 1 2 2 2 14 2 1 4 5 3 1 3 1 3 2 15 4 5 15 5 2 5 3 14 2 14 3 4 3 4 15 3 14 15 14 6 4 14 14 2 3 4 4 7 5 15 4-15 1

Summary Summary This study provides additional evidence that survey responders with lower verbal ability are more likely to provide suspicious data. In particular this study found that low verbal scores on the SAT were associated with: Significantly shorter duration time to complete the survey Significantly more likely to skip items and straight items sets on a screen. Significantly more likely to not complete the survey., or to skip through some survey screens quickly. In addition, this study found: Moderate agreement between grade level calculations. Rank ordering the screens based on the grade level calculations was associated with straightlining, but no other suspicious data. Overall, college students with low verbal ability are more e likely to submit suspicious data on surveys. Use of grade level calculators did not prove very useful in identifying screens that may be problematic.

Thank you! Copy of this and past presentations can be found at: nsse.iub.edu/html/pubs.cfm Additional NSSE information can be found at: nsse.iub.edu/ Feel free to contact me with any questions regarding NSSE. Jim Cole colejs@indiana.edu