For many decades the face to face interview (FTF) was the norm. In the 70 years, the FTF interview gave way to the telephone interview, which at the time provided adequate coverage and a substantial reduction in data collection costs. In the 90 years, however, new technologies such as mobile phones have made traditional telephone interviews less costly due to high rates of rejection and non-response, creating new concerns about coverage.
These innovations have led to a methodological change from a survey environment dominated by a personal interaction between the interviewer and the respondent to a method that is highly impersonal and guided by the interviewee (Dillman 2000).
The polls mixed-mode use multiple methods to contact or receive information from respondents.
One of the main reasons for the growing interest in this type of survey is the concern that coverage in any detection mode may be problematic and therefore the use of multiple modalities can compensate for the problems affecting the use of an individual.
The basic question concerning the use of survey methods mixed-mode is whether the data is reliable and consistent in all the survey methods.
Our research attempts to answer this question by examining the differences between non-response bias and modality effects between a self-administered questionnaire (SAQ) and a questionnaire administered by an interviewer (IAQ).
Methods and Data:
The substantial interest in our surveys was in assessing voter electoral experience, which focused on various aspects.
We used the voters registered on the November 2008 New Mexico electoral lists as our sampling list for independently conducted surveys and by an interviewer.
SAQ: we have championship 6000 registered voters to the state voters' archive provided by the Secretary of State of New Mexico, offering the possibility to respondents to reply via mail instead of through the internet. The 80% of the respondents chose the internet option and the remaining 20% the mail option. The response rate was 13,9%.
IAQ: a random sample of voters was interviewed by telephone. A total of 800 interviews were completed, with a global response rate of 17,4%.
Using the voters register allows us to compare the characteristics of the respondents of the sample with the characteristics of the population to determine the quality of the data collected.
We compare the population averages with those of respondents to the different types of investigations to determine the representativeness of the data. We also compare socio-demographic and religious characteristics between the two samples.
To remove from the surveys any sampling effect due to the lack of response, we use matching methods. These methods use the propensity scores that are calculated for each observation by estimating the conditional probability that a respondent is in the interview or self-administered survey, and that they are then used by the matching algorithms to organize the samples so that the observed covariates are balanced between the two groups.
Any remaining differences between the detection modes for our variables of interest can not be caused by differences in demographic characteristics between the samples and therefore may only be due to the effects of the detection mode.
According to our expectations, both methods of investigation should represent well the population of interest, despite the relatively low response rates.
In fact, we find that both surveys represent substantially the actual results of the election.
We find strong evidence for the effects in ways due to social desirability and the presence of an interviewer, so these aspects need to be taken into account in surveys that combine SAQ and IAQ surveys.
The simple combination of surveys with identical formulations of questions and answers is problematic, could in fact create unreliable measures and influence the relationships between variables.
Lonna Rae Atkeson, Alex N. Adams, R. Michael Alvarez