Evaluation and Implementation of EDR in School-Based Research

Evaluation and Implementation of EDR in School-Based Research Lesli Scott Ashley Bowers Sue Ellen Hansen Robin Tepper Jacob Survey Research Center, University of Michigan Third International Conference on Establishment Surveys Montreal, Quebec, Canada June 18 -21, 2007

Introduction Survey Research Center conducts national school-based studies of elementary/secondary education and of children and adolescents. Typically, SRC school-based studies include a data collection activity to obtain information about each school from the principal. Focus of this presentation is on assessing the potential effectiveness of using Web surveys to obtain school information from principals in the U. S.

Introduction Currently, most researchers choose the traditional mail mode • Use multiple contact – tailored approach (Dillman), scannable instruments, and incentive. • Attain 60 -85+% response rates. One SRC research effort over past couple years used Web mode • Principals in a large district completed daily activity logs during 3 one-week periods each of three years (70 -90% response). • The same principals also completed annual Web surveys for three years (70 -90% response). • These studies provided some evidence that Web surveys can attain acceptable response rates in regional school studies.

Key Research Questions This presentation reports on two very recent studies that help evaluate the ability to transition some national school studies to Web mode. Key research questions considered include: • What’s the impact of Web mode on coverage error? • What’s the impact of Web mode on sampling error? • What’s the impact of Web mode on nonresponse error? • What future research is needed?

Two Recent Studies (1) Study 1: Principals’ Use of Internet Study (WEB) • In Spring 2007, we administered a brief Web survey. • Main purpose was to assess properties of the sample through phases of the study. • We developed a national sample of 500 schools randomly drawn from U. S. Department of Education database. • We carried out Internet searches and schools calls to obtain principal names and email addresses. • We sent principals UPS delivered pre-notification letter with $5 incentive then followed with up to 3 email contacts.

Two Recent Studies (2) Study 2: Principals’ Mode Preference Study • Our 2 nd study analyzed similar questions in the WEB study and a School Health Policy (MAIL) survey. • Main purpose was to learn about principals’ access to Internet and preference for mail or Web mode. • Surveys used nationally representative samples of 500 schools covering K-8 th grade (WEB) & 600 schools covering 8 th-12 th grade (MAIL).

Web Mode & Coverage Error Coverage error considers whether the list used to create the study sample includes all members of the population and probability of selection is known. Hypothesis: Web mode provides nearly universal access to principals. Principals we can and can’t access are not different.

Coverage – All Principals Listed? U. S. Department of Education provides a census of schools. Recent reports indicate that virtually all schools have high speed Internet access. What about principals? Our studies helped us answer: • Do principals use high speed Internet? • Can we obtain email addresses for principals? • How similar are schools where we can deliver email to principals and schools where we cannot?

Coverage – Internet Access National Center for Education Statistics report (2006) – 100% of schools have Internet. – 97% of schools have high speed connections. Answers to questions on our Web and mail surveys showed that principals use high speed connections. Hi-Speed Internet Access At School WEB Study 97% MAIL Study 98% Preliminary Results

Coverage – Email Available? U. S. Department of Education databases do not include principal name or email addresses. Our Web study showed that we could use procedures to obtain names and emails. Obtaining Email Addresses Obtained During 1 st Internet Search 55% Obtained by Adding School Calls 90% Successful Email Delivery to Principal 84% Preliminary Results

Coverage–Similar to Sample? Our study showed no difference in key school characteristics for principals with and without deliverable emails. Preliminary Results

Coverage Error - Conclusions • Nearly all principals have access to high speed Internet. • It is possible to obtain email addresses for most principals using a procedure that involves Internet searches and telephone calls to schools. • There are no statistically significant differences between schools where we can deliver email to principals and those we cannot on key school measures – that is, there is no evidence of coverage bias.

Web Mode & Sampling Error Sampling error considers the precision of estimates that can be made based in part on the number of units included in the random sample. Hypothesis: Web gains more precision for same cost as mail mode. Our study helped us identify the cost drivers for the two modes and we estimated the magnitude of difference between them.

Sampling – Main Cost Factors Web and mail survey costs include: • • • Sample Development Pre-Notification with Incentive Questionnaire Development Application Programming Survey Administration Post Collection Processing

Sampling – Costs (1) (Hypothetical 2 -page Qstr) Web Mode (n=500) Sample Development Pre-Notification with Incentive Paper Questionnaire and Materials Development Mail Mode (n=500) NCES Data Internet Search School Calls Preload ~$2, 700 NCES Data Purchase Names School Calls Mail Merge ~$1, 700 Materials Postage Incentive Labor ~$5, 900 Questions Provided Format Paper Qstrs Print Questionnaires Print Letters Postage ~$9, 500

Sampling – Costs (2) Web Mode (n=500) Programming Web Survey Administration Data Processing Data Handling TOTAL Mail Mode (n=500) ~$500 1 st Email 2 nd Email 3 rd Email ~$10, 000 (Hypothetical 2 -page Qstr) ~$550 ~$350 DDE ~$500 1 st Mail Logging 2 nd Mail Logging 3 rd Mail Logging DDE Data Handling ~$20, 000 ~$1, 800 ~$600

Sampling Error - Conclusions • The cost to conduct a Web survey of principals should almost always be less than the cost to conduct the survey using the traditional mail mode. • Cost savings can be used to increase sample size – thus, reducing sampling error. • In addition, the marginal cost of adding a case in a Web survey is generally smaller than the cost of adding a case in a mail survey.

Web Mode & Nonresponse Error Nonresponse error considers whether sample members complete the survey and whether the responders have different characteristics than nonresponders. Hypothesis: Participation in Web surveys is at least as good as mail. Principals who do participate and those who do not look similar on key school characteristics.

Nonresponse – How Willing? Our studies helped us answer the following: • Are principals receiving requests and choosing to respond to Web surveys? • Do principals have a preference for Web or mail surveys? • Can we obtain response rates for a Web survey of principals that are comparable to mail? • Do those principals who participate in a Web survey look similar to those who do not on key school characteristics?

Nonresponse – • On average, MAIL Study principals received more than one Web survey request during past year. – Most completed at least one. • WEB study principals have participated in Web surveys at even higher rates. Web Survey Experience WEB Study Principals (%) MAIL Study Principals (%) 0 requests 21 21 1 -3 requests 54 29 4 -6 requests 17 27 7+ requests 8 23 Preliminary Results

Nonresponse – Preference Web vs. Mail • Most MAIL study and WEB study principals indicated that they prefer Web surveys over mail surveys. • Few differences in preference by school characteristics or available principal demographics (e. g. , gender).

Nonresponse – WEB Response Rates • The response rate achieved in the WEB study was comparable to typical mail survey response rates. • The length of time needed to conduct a survey is greatly reduced using the Web compared to traditional mail data collection.

Nonresponse – Similar to Sample? Our study showed no difference in key school characteristics for principals who participated in the WEB Study and those who did not. Preliminary Results

Nonresponse Error - Conclusions • Principals are receiving Web survey requests and responding to them. • Many principals express a preference for Web surveys over mail surveys. • Our experience conducting a Web survey of principals in a random sample of U. S. public schools provides evidence that Web surveys can achieve response rates comparable to mail surveys. • Principals who participated in the Web study generally looked similar to those who did not on key school characteristics.

Future Research (1) Measurement Error Research • Are Web surveys measuring differently than other modes? • What aspects of Web design influence quality of response?

Future Research (2) Complexity Found in Establishment Studies • What are impacts of introducing long and complex instruments? • Will principals input information collected from school files into a Web instrument? • Will principals forward portions of Web instruments to other school staff?

Future Research (3) Some Coverage Error Research Questions: • What might we learn if we make contact with principals in schools where we could not obtain deliverable email addresses? • Are there some specific groups of principals that don’t want to be contacted and if so, what can we learn about them? • Are email services in schools provided by school, district, state or other entities? • How much do principals use home computers for work related activities? How often would they choose to complete work-related surveys at their homes? • In addition to principals, can we obtain email addresses for multiple informants at the school who might assist with school surveys?

Future Research (4) Experimental Design • Ideally, our next step will be to set up research using random assignment to Web and mail modes.

Final Conclusions Special Thanks to: Andrew Hupp Timothy Wright Jackie Mc. Bride