Based on our experience in the mixed-mode (CATI-CAWI) field, we perceived a difference in the length of interview (LOI) between these two modes: CAWI interviews are always briefer than CATI's. Validating this work hypotesis we tried to find out which questions show the greater gap and why. Another aspect that we've investigated is the link between LOI and some socio-demographic variables. With CAWI interviews, we've tried to validate the existance of a link between LOI and the number of questionnaires completed by panelists in the past . We've also tried to estimate CAWI's LOI using CATI's LOI as input and viceversa.
Methods & Data:
To carry out these analysis we've used both metadata, such as LOI and sinlge page completion time, and socio-demographic respondent's characteristics. Those data come from two mixed-mode CATI/CAWI surveys. In both cases we've used an online panel (Opinione.net) as CAWI framework, whereas CATI interviews were collected through a non-probabilistic quota sample design (geographically and demographically representative of Italian population). The first dataset has 1600 CATI interviews and 1020 CAWI interviews, the second one is composed by 752 CATI and 294 CAWI questionnaires respectively.
Results:
As we expected LOI for the CAWI interviews is shorter than the LOI for the CATI interviews (t-test: p-value = 0). Both datasets confirm this work hypotesis. The difference increases when taking into account matrix question type (significative p-value).
Correlation between LOI and socio-demographic variables is stronger in the CATI interviews than in the CAWI. In the former case Pearson's correlation index between LOI and the birth year is statistically significative for both datasets (p-value=0). In the latter case the correlation index is lower and statistically significative only in the first dataset.
We didn't find any correlation between the LOI and the habit of completing questionnaires in the CAWI subsets (ranking+# invitations) (correlation index = -0.014, p-value=0.63).
Using the first dataset we've created a model for estimating the CATI LOI/CAWI LOI ratio. Applying this model to estimate second dataset's ratio, we've obtained a good result (estimated ratio: 1.46; real ratio: 1.5).