International Choice Modelling Conference, International Choice Modelling Conference 2015

Font Size: 
An analysis of the impact of screen size on stated choice behavior for credit card applications
Jeffrey Dumont

Last modified: 11 May 2015


With the increased use of mobile technology by survey respondents to complete online surveys, potential bias from the reduced screen sizes of mobile devices will be important for researchers to account for when designing and modeling stated choice (SC) experiments. In particular, traditional discrete choice models make specific assumptions that choice-makers are well-informed and fully take into account all of the information presented in an experiment. With a reduction in screen size, there is the potential for respondents to miss or ignore information presented in an experiment. It also places additional burden on the mobile survey respondent to potentially scroll and zoom to fully understand the choice scenario presented to them. This additional burden may encourage the use of simplifying choice heuristics like lexicography and attribute non-attendance.

It also strikes us that simply excluding respondents based on the type of device used to complete survey is not a valid solution to this potential bias. There is a potential of excluding key demographics of respondents who can only be reached via mobile devices.

In this paper, we take a first look at the potential bias that screen size has on choice model results using data collected from a stated choice survey for credit card applications. This work is on-going and not yet fully completed but initial results suggest that the screen size of the potential respondent is important for researchers to consider.

There is a recent body of literature on understanding attribute non-attendance and its impact on choice models developed from stated choice data. There is also literature on incorporating decision rule heterogeneity in choice models. We feel that this work extends both sets of literature by looking at the impact that the overall survey experience (in particular screen size) has on the both of these.

Data description
We use data collected as part of stated choice survey about preferences for travel credit card offers fielded in September of 2014. This sample comprises of 4,526 US respondents who were targeted based on their likelihood of being interested in this specific type of credit card. Each respondent was presented with a total of 8 stated choice experiments where they were asked to select from a set of up to 4 credit card offers. Respondents were also allowed to select that they would not be interested in any card. Four multi-level (including annual fee) and 21 binary attributes were tested as part of the stated choice experiment. Due to the confidential nature of this work, we can only provide a high-level overview of the attributes tested; however, this limitation should not affect our ability to demonstrate our paper’s objectives.

In addition to the choice experiments and socio-demographics, data were collected about each respondent’s survey experience. Data were collected around the browser, operating system, device type (mobile, table, vs desktop and laptop) and screen size. Screen pixel width and height were collected for respondents who completed the survey via a mobile device (smart-phone or tablet). For non-mobile survey respondents, the dimensions were collected for the browser window in which the survey was being completed. This data was not self-reported but collected by the survey fielding software.


Analysis methods
For this work, we are planning on developing latent class models that will account for the impact that screen width and screen height have both on respondent taste and decision rule heterogeneity. These models will allow us to identify if there is a bias due to screen size and how it impacts respondent preference and decision rule heterogeneity.

Preliminary conclusions

As stated above, this is a work-in-progress but preliminary results suggest that the size of the screen in which the respondent takes the survey can be a significant determinant of respondent’s preferences. For example, respondents who took this survey on smaller screens selected the left most alternative twice as frequently. Another initial finding is that decreases in screen size (both height and width) have a strong, positive impact on the use of the opt-out alternative. Finally, there is also evidence that willingness-to-pay for certain attributes also varies by screen size.

Conference registration is required in order to view papers.