Iowa Caucuses: Tracking and Privacy
Dstillery was recently featured in USA Today for our work exploring the correlations between victorious candidates in the Iowa caucuses and the distinctive behaviors and brand affinities of the corresponding populations. The insights we developed have provided a new perspective on the sources of political strength for each of the candidates, representing a potential new frontier in exit polling. The response to this work has been generally positive.
There were, however, dozens of comments on social media regarding the privacy practices of Dstillery, as well as the legality of our data collection practices. This blog post is an attempt to outline exactly what we did and how we did it. By being transparent about our methodologies, we hope to emphasize our ongoing commitment to consumer privacy, notice and choice.
There are 198 caucus sites in Iowa, one for each each party in each of its 99 counties. During the caucuses, Dstillery collected anonymous device IDs (IDFA or Android ID) from any devices we identified at each of these sites. We collected data via two means:
- License deals with companies that collect anonymous device information through partnerships with publishers and marketers. They have the right to resell data to parties such as Dstillery. These companies collect location data when the user has opted in to share their device location information.
- Advertising exchanges with which Dstillery has an integration. We use data from exchanges only when we are explicitly permitted to do so.
All of our data partners have warranted that they are committed to disclosing their practices of collection and use of consumer data in their privacy policies. Disclosure includes the sale of collected data to companies like Dstillery. We conduct audits to check compliance as part of our commitment to follow industry best practices as outlined by the Network Advertising Initiative.
For the device IDs at the caucus sites, the only information collected was the device’s ID showing up at a latitude/longitude pair that matched the location of one of the caucus sites. We then matched these devices via probabilistic means to other devices in order to understand the general browsing behaviors of the populations at each caucus location. It’s important to note that we did not match device IDs collected at the caucuses to any individual people, their voting behavior, or any piece of personal data that identifies who owns the devices or how they voted.
Once the caucus results were tallied and reported on February 2nd, we identified all counties where one of the candidates had won by at least 5 percentage points. We then pooled all of the device IDs we had seen at all of the counties won (by >5%) by a given candidate. We had no information as to whether that device ID was associated with a vote for the winning candidate. In fact, we have no information whatsoever about the individual owner of the device. We looked at patterns of behavior among the pools of devices aggregated across caucus sites won by each candidate in order to understand the distinctive characteristics of Iowans in counties won by each candidate.
The sample sizes we collected were judged by our data science team to be statistically significant enough to use as proxies for the general population of counties favoring each candidate.
Dstillery takes consumer privacy seriously. We were built on the foundation of privacy by design, which is an approach to systems engineering and data collection that takes privacy into account throughout the whole engineering process. We continue to believe that transparency in our processes and methodology for data analytics and advertising targeting is the right approach and benefits everyone.