We apply the most advanced techniques in software engineering, data mining, artificial intelligence, and human-computer interaction.
Rooted in technology and talent emanating from Carnegie Mellon University, our world-class team of technologists and social scientists operate at the forefront of the digital world. We apply this technology to solve major challenges in consumer marketing and public opinion research.
Respondent Identity and Validation
Every CivicScience respondent is internally identified by a directed acyclic graph. The vertices of the graph are respondent aliases of varying reliability, such as a social network user ID or a UUID stored in a browser cookie. An edge is created when two or more aliases are available to the server at the same time, and each edge is directed from the more reliable alias to the less reliable alias.
Attribute values are attached to the most reliable alias available at the time the attribute is observed. A respondent's complete attribute profile is the union of the profiles attached to each vertex in that respondent's graph. Matching respondent profiles with respect to certain high entropy attributes is used to recognize respondent identities which likely represent the same individual, for example one person using multiple devices, and the respondent identities are merged. Aliases with improbable or contradictory profiles suggest fraudulent voting or two individuals sharing one device, and those aliases are pruned from the respondent graph.
Attribute Value History
A respondent's answer to a survey question is stored as an attribute/value pair, along with a timestamp to enable tracking a respondent's changing opinions over time. The attribute value schema is also used to store response meta-information, derived attributes, and imported third-party attributes. Adding probability to each attribute value allows data mining processes to contribute predicted values for unobserved attributes. The database engine used for storing the attribute value history is designed to rapidly match respondent profiles against requested attribute values arranged in complicated, ad hoc boolean expressions.
The iQ Engine
The iQ Engine is responsible for asking a respondent the optimal question during a particular engagement. First the engine collects questions for which the respondent matches the required sample population. The engine checks each question against the locale and media constraints imposed by the current user interface. Questions are weighted by information gain and semantic relevance to the survey context. Lastly, an auction is conducted among the weighted questions, awarding the engagement to the researcher who most highly values the respondent's participation. The heuristics used by the iQ Engine to choose a single question can also be applied to a series of questions, allowing long, complicated questionnaires to be split apart and distributed to respondents over a number of short engagements.
Reporting
Researchers assemble custom reports by first designing a sample population and then choosing the attributes to include in the report for members of the sample group. As a researcher builds the custom report, the reporting tool monitors the size of the smallest group of respondents with reported profiles in common. To protect respondents from deductive disclosure, the report is not prepared if the smallest group sharing a reported profile is statistically significant, but fails to meet the privacy threshold. An iQ Engine simulation with respect to current response trends and the report description estimates the time remaining until the report will satisfy the privacy criteria and suggests to the researcher possible question bids to improve the estimate.
In addition to custom reports designed by researchers, data mining processes investigate correlations in the attribute value history. The resulting auto-insight reports are provided to subscribed researchers as suggestions for additional study. As researchers provide feedback to these insights using a simple thumbs up/down mechanism, the auto-insight algorithm learns to identify and report more interesting insights.
We are pioneering a new method of attitudinal research that captures opinions from broad segments of the consumer population everyday.
The field of quantitative attitudinal research is facing a profound period of change as two primary forces converge on the industry:
First, rapid migration away from land phones, rejection of bulk emailing, and the deterioration of online panels have made it nearly impossible to engage high-value consumers in reliable research. The overall decline in survey participation and growing levels of respondent fatigue seem insurmountable under traditional research methods. The emerging field of sentiment analysis, namely through blog- and Twitter-monitoring, shows some level of promise from a qualitative standpoint, however, the shortcomings in Natural Language Recognition technology severely limit the ability to produce scalable, representative data.
Secondly, consumer attitudes are changing more rapidly than ever before. Pervasive social influences, 24/7 media, and endless product choices have created opinion volatility that leave marketers and advertisers guessing from one day to the next. So, consumers are not only proving increasingly difficult to reach. Monitoring their changing attitudes is also more critical than ever.
CivicScience is pioneering a new method of quantitative attitudinal research that captures the opinions of large, broad segments of the consumer population every minute of every day. By engaging respondents in strict, one to three-questions intervals where they travel online, we maximize participation and eliminate respondent fatigue. Then, by following these consumers across multiple sites and engagements, we can append responses to a unique anonymous profile in our dataset. The result is a dynamic panel of millions of consumers who answer dozens if not hundreds of questions over time. Analysis of this massive dataset allows researchers to model for perceived biases and apply marginal, if any, weighting methodologies to achieve representative samples.
CivicScience is on pace to gather, organize, and analyze more than 400,000,000 opinions in the next calendar year, with that number rising dramatically. Our goal is to provide our data to professional researchers, brand managers, government agencies, and other decision-makers to enable more reliable and affordable research.
To request a methodology White Paper from one of our esteemed Scientific Advisors, please contact us here: contact@civicscience.com
