We've spoken about a number of survey modes already, I've referred to some of them as the classic survey modes. It's increasingly common to combine modes into a single survey or single study. And these are known as mixed-mode design. So, in the current segment we'll introduce a number of different mixed-mode designs, and then elaborate on the implications of these designs for survey error. So, the goal of mixed-mode designs, or the goals, are to want optimize the kinds of trade offs we've been talking about so far between different costs and errors. To increase response rates, the goal being to reduce non-response error, and as we'll discuss, lowering response rate doesn't necessarily reduce non-response error. But the two can be related, and it's certainly desirable to reduce non-response and increase response rates. So those are goals, but the concern is creating what are known as mode effects, which really are different estimates coming out of data collected from the same sample generally, in different modes. This is certainly a source of error, and something one would want to minimize. Different modes might not only be used for data collection, but for communicating with sample members in different phases of the study. We call this pre-contact in which the sample members are alerted to the fact that an invitation is forthcoming. Recruitment, in which they are actually invited to take part in the survey. And then reminders after they've been recruited to make a decision about participating. So let's discuss mixed-mode designs that mix one mode for pre-notification and another for the data collection. Pre-notification, as I said, alerts the sample member to an upcoming survey invitation. It's not actually the invitation, the invitation will follow. But an example might be in a telephone interview or a survey in which the data collection is done by telephone interviews. The recruitment and pre-notification are done face to face in some areas. Or another example might be mail recruitment, including pre-notification for a web survey, a web only survey. Which is actually a very common combination of recruitment mode and data collection. So again the goals are to increase response rate by increasing the study's salience and if the pre-notification, recruitment etc. happened to be in the same mode as the data collection then we don't have to worry about mode effects. As an example of how pre-notification can be conducted in a mode that differs from data collection Bosnjak and his colleagues tested the impact on response rate in web surveys of three pre-notification modes: SMS, email, or none; and two invitation modes: SMS or email. They found that overall, SMS prenotification was most effective, that it's response rates were higher when the prenotification was texted to respondents. They found that the invitation, which followed the prenotification was most effective when it was sent via e-mail. And they found that the combination of an SMS pre-notification and then email invitation was most effective overall. So, we can see that in the graph here that the line, or curve, that's highest is that combination of SMS pre-notification and an email invitation. When it comes to mixing modes in data collection, as opposed to between pre notification or recruitment and data collection, there are a number of different designs. And these have been well documented by de Leeuw in 2005 in an article I encourage you to read. First is what is known as concurrent mix-mode design in which there's one sample, one time point, a single questionnaire but the sample persons are interviewed or they data are collected in different modes. A key aspect of this type of design is mode choice, giving the sample members, the respondents, an option of how their data will be collected. Sequential mixed-mode is far more common. This refers to the situation where, again, there's a single sample, a single time point, But the sample members are recruited in increasingly effective but increasingly expensive modes if previous contact attempts are not successful. So this would be a case in which the recruitment or prenotification is intended to be in the same mode as data collection. But if the sample members don't respond or can't be contacted in, for example, a paper mailed questionnaire. Or for those who did not respond, there's a followup in a telephone interview, assuming the sampling frame contains both addresses and phone numbers. And then for those who don't respond in telephone contact, a personal visit is made by an interviewer. So you can see that the idea is to Maximise response rate while minimising expense. So only using the most expensive mode face to face interviews for those sample members who hadn't responded to either telephone interview or a paper mailed up questionnaire. Other designs include switching modes within the questionnaire. So the most well known example of this would be a face-to-face interview in which The interviewer turns the computer over to the respondent for the most sensitive, so Cassie or audio Cassie, as we've discussed. So there are actually two modes in a single interview. Or for a single questionnaire. Longitudinal mixed mode studies in which there is one sample but their behaviours and attitudes are measured at multiple time points. And the idea is that at one time more often the initial data collection is done in one mode and then the subsequent interviews or data collection sessions are done in another mode. So you can imagine a face-to-face interview to really engage and extract a committment from households to take part in the longitudinal study and then with the follow-ups being conducted by telephone. And then parallel or separate mixed mode designs refer to situations in which there are actually different samples. Data are collected in different modes but these are all combined into a single study. So multinational surveys are really the primary example of this approach. It makes sense to use different modes Predominately in different countries, and then integrate the data in the analysis. And so, in the next segment we will focus in more detail on the first of those mixed modes, mixed data collection modes, namely, concurrent mixed modes data collection.
