Using Web Surveys to Collect Data from Candidates. Experience from the 2019 Greek Candidate Survey

Web-surveys are being used more and more often in social sciences, as a fast and low-cost mode of data collection. However, there are some serious drawbacks which are mainly related to the absence of an interviewer such as low response rates and low response quality (i.e. satisficing). Furthermore, the length of the survey instrument affects considerably the data quality. There is evidence that lengthy online questionnaires lead to lower response rates and lower quality responses. The aim of this paper is to study whether different web survey structures affect the data quality of a candidates survey. Specifically, we conducted four web experiments to test the impact of: i) questionnaire length ii) the duration of the breaks between the sub-questionnaires of the survey, iii) the web survey layout and iv) candidates recruitment on data quality. In order to test the impact of each factor we have manipulated the conditions of the survey in several ways. First, we divided the questionnaires into two or three parts in order to test whether shorter questionnaires increase response rate. When respondents completed the first part of the questionnaire, a follow up question appeared on the screen asking them if they wanted to answer the rest of the questions. We sent a second invitation to those who did not answer the second (or third) part of the questionnaire within a month after the completion of the first part. Second, we conducted an experiment on whether the layout of the survey affects the response rate. Specifically, there were two different designs of the questionnaire namely single page question and grid. Furthermore, some surveys were optimised for mobile devices to test if this affects the response rate. The final experiment is about the candidates recruitment. Most of the Greek candidate MPs have e-mail addresses which are available online especially during the period of electoral campaign. We collected their e-mail addresses with two methods: i) using search engines and visiting websites related to the candidates or to the Greek elections in general, and ii) asking political parties to provide us with a list of their candidates along with their email addresses. We also got in contact with the candidates whose e-mail could not be collected by the aforementioned methods through their facebook pages. To examine data quality, we measure the participation rate, in terms of response rate and drop-outs, as well as response behavior. We also applied data cleaning methods relying on response quality indicators such as the length of response to the open-ended questions, item-nonresponse (skipping), the time spent on questionnaire items etc.