Welcome back. This is our second of four lessons, and it's devoted to automated self-administration. Including web or online data collection. In this lesson, we'll talk first about the modes that are automated and self-administered but part of an interview, where a human interviewer asks the non-sensitive questions and turns over the laptop or other computer to the respondents to self-administer or directly enter their answers to the sensitive questions. And then the remainder of the lesson will be devoted to web questionnaires, online data collection and really in two components. One is kind of the error properties of web surveys, so the impact of coverage error and non-response and measurement error, and we'll look at different sampling procedures that are commonly in use and how these can affect the different types of error in web surveys. And then in the final part of the lesson, we'll drill down into measurement error and look at a particular kind of feature of web surveys, interactivity, which is something that's not available in other types of self-administration, in particular, in paper self-administration. And how interactive features of web surveys can enhance or not enhance the quality of the data that are collected in that mode. Now we'll turn to CASI and audio CASI and how they can help increase disclosure of sensitive information. So just as a reminder, the most frequently used modes really are both self-administered, mailed out paper questionnaires, and web. And telephone interviews and face-to-face interviews involve a human interviewer. There are of course hybrids used for sensitive questions, which is really what we're going to focus on now, in which the interviewer administers non-sensitive questions and the respondent self-administers the sensitive questions. And there are mobile web surveys increasingly being used. And again, just by way of reminder, when we say CASI, we're referring to Computer Assisted Self Interviewing. ACASI is Audio Computer Assisted Self Interviewing. So turning first to a CASI, computer assisted self interviewing, and how it affects a respondent's answers to sensitive questions. [COUGH] The initial comparison or the main comparison that evaluations of CASI have been concerned with is answers under self-administration that is with CASI, compared to answers to questions asked by interviewers. So CATI, for example, for personal or for face to face interviews. There are a number of studies which demonstrate there seems to be more disclosure of sensitive information under CASI than when interviewers ask the question. One of the early demonstrations was by Tourangeau & Smith, who observed more disclosure, where disclosure here means more reports of a socially undesirable behavior, more disclosure with CASI than both paper. We'll return to other comparisons to other types of self-administration in a moment, but also compared to CAPI where interviewers are asking the questions. So you'll see here that asking about number of sex partners at one year, five years and lifetime, generally we see higher bars for CASI, the light blue bars, then CAPI, the dark blue bars. At lifetime, there's really no difference between CAPI and CASI. It has been proposed that questions become less sensitive over the course of one's life or if one can answer them about one's lifetime. The assumption here is that it's undesirable, at least for females, to have lots of sex partners, so that fewer sex partners is more desirable. But it seems that lifetime the mode doesn't matter as much. And this could be because it is a less sensitive question over the course of a lifetime. But for one year and five year intervals, the last year and the last five years, there's a clear advantage in terms of disclosure, more reporting of undesirable behaviors with CASI, that is self-administration, than CAPI, interviewer administration. Tourangeau and Smith also demonstrated that self-administration, in particular audio CASI, increased the reporting of using illegal drugs [COUGH] at different intervals relative to interviewer administered questions, CAPI. So this table contains ratios of reported prevalence of cocaine and marijuana use at these three time periods, past month, past year, lifetime. Ratio of ACASI reports to reports under interviewer administration. And the point is if they're bigger than one, there are more reports with ACASI than with CAPI when the interviewers asked the question. These interpreted to mean more disclosure. So, you can see all these ratios are larger than one, and in some cases quite a bit larger than one. Suggesting that, particularly, in the past year, that respondents are more willing, able to disclose use of marijuana, and particularly use of cocaine when they self-administer the responses, the questions and answer by directly entering their answers, than when interviewers asked the questions. Returning to the domain of sex partners, opposite sex sex partners, Tourangeau and Smith showed that males and females seem to both become more candid, provide more undesirable answers under self-administration, both Audio-CASI and Text-CASI, or conventional CASI, than when an interviewer asks the question, but in opposite directions. And their suggestion, and others have made the suggestion, is that what's undesirable for males is not the same as what's undesirable for females. So under CAPI, males tend to overreport the number of opposite sex sex partners, compared to what they report under self-administration. And for females with CAPI, when interviewers are asking the questions, the tendency is to underreport the number of opposite sex sex partners. And of course this can't be the case because in some sense they are each other's partners, if you subscribe to the idea that the sample really represents the population. So you can see that audio-CASI and text-CASI narrows the gap between males and females. It doesn't eliminate the gap, but it does narrow the gap, suggesting that even different types of social desirability are reduced when under conditions of self-administration. Self-administration, clearly in these examples increases disclosure compared to interviewer administration. One can also ask whether computerized, automated self-administration is more effective, leads to more disclosure than paper self-administration, and whether there's any difference in disclosure between CASI and audio-CASI. So there actually is evidence of more disclosure of sensitive information in the automated modes than the paper modes. There's, like the automated modes the paper, SAQ, as we described it early, self administered questionnaires, do lead to an increase in reporting undesirable behaviors relative to interviews. But in a meta-analysis by Gnambs and Kaspar comparing computerized self-administration to paper-based self-administration, there were reliably more reports of sensitive information with automated self-administration than paper-based self-administration. And the effect was larger for the most sensitive items. So there is something about automation which seems to increase privacy relative to just self-administration. Now, ACASI is presumed to be, when it was introduced, its proponents argued that by hearing the questions, and they're not being presented on the screen, the sense of privacy given to respondents was increased relative to CASI, in which the questions were displayed on the screen. Couper and his colleagues, and really the only direct comparison, one of the few direct comparisons of CASI and ACASI found no advantage. Part of this was attributed to the fact that respondents in audio-CASI were actually not using the headphones. So they were just reading questions from the screen. In this implementation, questions were presented both on the screen textually and to headphones. But they found that even when respondents were using their headphones and the questions were presented auditorialy, there was no advantage over and above CASI. This is not to say in any way that these modes don't increase disclosure relative to interviewer administrate questions, but they're not different from each other, at least in the study by Couper and colleagues. One thing that's certainly the case about ACASI, even if it doesn't increase disclosure, is that it reduces the impact of low literacy compared to either CASI or paper self-administered questionnaires, because respondents who are not high ability readers are able to hear the questions, and so this eliminates that concern. A comparison by Charles Turner, one of the early comparisons of Audio-CASI to paper, found a large advantage in disclosure of some quite sensitive same-sex sex behaviors, suggesting again that there is some advantage to automation over and above the advantages of self-administration. When we return, we'll look in a little more detail at why self-administration, in particular with audio-CASI, may lead to more disclosure than interviewer administration.