Читать книгу Handbook of Web Surveys - Jelke Bethlehem - Страница 19

Version 2.0 of HTML made it possible to implement questions on a web page. The <input> tag can be used to define different types of questions. With type=radio this tag becomes a radio button. A closed question is defined by introducing a radio button of each possible answer. See Figure 1.2 for an example. Not more than one radio button can be selected. This corresponds to a closed question for which only one answer must be selected.

Figure 1.2 A closed question in HTML

Sometimes respondents must be offered the possibility to select more than one answer, like in Figure 1.3. Respondents are asked for their means of transport to work. Some people may use several transport means. For example, a person may first take a bicycle to the railway station and then continues by train. Such a closed question is sometimes also called a check‐all‐that‐apply question. It can be implemented in HTML by means of a series of checkboxes. A checkbox is obtained by stetting the type of the <input> tag to checkbox.

Figure 1.3 A check‐all‐that‐apply question in HTML

Figure 1.4 shows the implementation of an open question. Any text can be entered in the input field. A limit may be set to the length of the text. An open question is defined with type=text for the <input> tag.

Figure 1.4 An open question in HTML

If an input field is preferred that allows for more lines of text to be answered, the <textarea> tag can be used for this.

There are no specific types of the <input> tag for other types of questions. However, most of these question types can be implemented with the input field of an open question. For example, Figure 1.5 shows a numeric question. The question is basically an open question, but extra checks on the answer only allow numbers to be entered within certain bounds.

Figure 1.5 A numeric question in HTML

Date question can be specified as a set of three input fields: one for the day, one for the month, and one for the year.

In the first years of the World Wide Web, use of web surveys was limited by the low penetration of the Internet. Internet penetration was higher among establishments than among households. Therefore, it is not surprising that first experiments tested the use of web business surveys. Clayton and Werking (1998) describe a pilot carried out in 1996 for Current Employment Statistics (CES) program of the U.S. Bureau of Labor Statistics. They expected the web to offer a low‐cost survey environment. Because it was a form of true online data collection, an immediate response to the answers of the respondents was possible. This could improve data quality. They also saw the great flexibility of web survey questionnaires. They could be offered in a form layout or in a question‐by‐question approach. The drawback was the limited number of respondents having access to the Internet. Only 11% of CES respondents had access to Internet and a compatible browser.

Roos and Wings (2000) conducted a test with Internet data collection at Statistics Netherlands for the construction industry. Respondents could choose between three modes:

 Completing a form offline. The form was sent as an HTML file that was attached to an e‐mail. The form is downloaded, completed offline, and returned by e‐mail.

 Completing a form online. The Internet address of an online web form was sent by e‐mail. The form was completed online.

 Completing an e‐mail form. An e‐mail is sent containing the questionnaire in plain text. Respondents clicked the reply button, answered the questions, and sent the e‐mail back.

A sample of 1,500 companies was invited to participate in the experiment. 188 companies were willing and able to participate. Of those, 149 could surf the Internet, and 39 only had e‐mail. Questionnaire completion times of all three modes were similar to that of a paper form. Respondents preferred the form‐based layout over the question‐by‐question layout. The conclusion of the experiment was that web surveys worked well.

General population web surveys were rare in the first period of existence of the Internet. This was due to the low Internet penetration among households. This prevented conducting representative surveys. However, there were polls on the Internet. Recruitment of respondents was based on self‐selection and not on probability sampling. Users could even create their own polls on websites like Survey Central, Open Debate, and Internet Voice (see O'Connell, 1998).

Also in 1998, the Survey2000 project was carried out. This was a large self‐selection web survey on the website of the National Geographic Society. This was a survey on mobility, community, and cultural identity. In a period of two months, over 80,000 respondents completed the questionnaire. See Witte, Amoroso, and Howard (2000) for more details about this project.

It seems to be typical for this type of self‐selection web surveys that they make it possible to collect data about a large number of respondents in a relatively short time. Other examples are given by Bethlehem and Stoop (2007). The survey 21minuten.nl has been conducted a number of times in the Netherlands. This survey supposed to supply answers to questions about important problems in Dutch society. Within a period of six weeks in 2006, about 170,000 people completed the online questionnaires. A similar survey was conducted in Germany. It is called Perspektive Deutschland. More than 600,000 participated in this survey in 2005/2006.

It should be noted that these large sample sizes are no guarantee for proper statistical inference. Due to under‐coverage (not everyone has access to the Internet) and self‐selection (no proper random sampling), estimates can be biased. This bias is independent of the sample size.

Internet penetration is still low in many countries, making it almost impossible to conduct a general population web survey. Since data collection costs can be reduced if the Internet is used, other approaches are sought. One such approach is mixed‐mode data collection. A web survey is combined with one or more other modes of data collection, like a mail survey, a telephone survey, or a face‐to‐face survey. Researchers first attempt to collect as much data as possible with the cheapest mode of data collection (web). Then, the nonrespondents are re‐approached in a different (next cheapest) mode. Example 1.6 describes a survey run using a mixed‐mode approach.