Page 222 - Contributed Paper Session (CPS) - Volume 7

CPS2068 Jan-Philipp Kolb et al.
Furthermore, a survey evaluation is carried out at the end of each wave
questionnaire. The panelists are for example asked if they had difficulties with
questionnaire comprehension or with finding an adequate answer on a
question. Also, further survey-related questions are asked. Survey
participation, for example, gathers the information whether the person
participates in other surveys in addition to the GESIS Panel.
Also, it is possible to access process-based para-data because the GESIS
panel is self-administered. We have information on, e.g. how long it took the
online panelists to answer one specific question. However, some of these
variables are only available for the online part. Nonetheless, this type of
information might be of particular importance, since the impact of para-data
for predicting nonresponse was highlighted recently (Lugtig and Blom 2018).
In addition, we have information about panel management. The GESIS
panel maintains all contacts with panelists in a panel protocol database. This
includes, for example, whether there was a complaint or concern from a
panelist. For online users, it records whether they have identified a technical
problem with the online portal. It is then also recorded whether the technical
problem could be solved. Some panelists also have problems with the content
of surveys and express these. All in all, this information provides a very
interesting picture. In some cases it is quite easy to explain why there is a case
of unit nonresponse. However, it must also be noted that the number of cases
is unfortunately relatively small and that these variables therefore do not play
a major role in the statistical learning models.
Another example of an administrative variable is the mode of invitation.
Two ways of participation are available, online and offline. We also generated
the variable survey participation, to control the frequency and regularity a
person responds to the survey. We divided the number of waves a panelist
responded by the number of waves he could have participated. This takes into
account that panelists who have been in the panel since 2016 were able to
participate in fewer waves than panelists who have been in the GESIS panel
since 2013. We use wave fa for the analysis in this paper. Another interesting
variable in this context is the latency. This variable is used to record how long
it takes the panelists to respond to the invitation to participate in the
questionnaire. This can be recorded relatively precisely for online users. For
offline panelists, the time span between the invitation and the day on which
the questionnaire is filled out and returned is recorded.

2. Methodology
We applied two types of techniques. On the one hand, we have parametric
methods like logit or lasso regression. That means that, in this case, we have a
predefined additive and linear functional form. In the model, we use a
logarithmic function for the link between probability and logits. Thus, we have

209 | I S I W S C 2 0 1 9