Methodology FAQs

Survey FAQs  

How is the survey administered online?  

Surveys are administered through our survey collection partners who work through an online platform which houses our entire survey experience. Respondents are invited to take the survey and come from a double opt in vendors. The survey is not Resonate branded and is not a pop up. Additional information in our Methodology overview.  

How do you choose to which devices to send survey requests?   

We don’t directly target devices. In order to be eligible to take the survey, a respondent has to be in our survey vendor’s panel, over the age of 18, live in America, fit into a non-full quota (age/gender, income, etc), and meet our strict screening requirements on device and behavior data.  

Once someone has taken the survey, when could they take another survey from Resonate?  

Survey respondents are not re-contacted for 90 days after they’ve completed a survey with us.  

How frequently are vertical specific categories asked in your survey?  

Vertical surveys are rotated throughout the year and scheduled based on business requirements. For example, if a certain vertical is more important to our clients, we may ask it more frequently.  

What characteristics are survey respondents weighted on?  

We weight on the following parameters:  

1. Age/Gender composite
2. Ethnicity
3. Income
4. Region
5. Education
6. Presence of Children
7. Registered to Vote
8. Time Spent Online

Do you capture mobile browser/mobile app data in your behavioral database?  

We capture desktop and mobile behavior in our behavioral database.  

Do you capture search activity from our behavioral database?  

No.  

Modeling FAQs  

How do your models work?  

Our models estimate the probability that a device on the internet is operated by someone with a given attribute value. 

Under the hood, our predictive models assess whether the presence of some feature (website visits, NLP category visits, etc) increases the odds of the person behind the device has the attribute value. We retrain our predictive models as new research data becomes available. 

How do you build an accurate model?  

There are several steps to creating accurate models:  

1. We start with psychometrically guided survey question structures  
2. Exhaustively examine all possible survey response patterns to identify quality survey respondents  
3. Execute advanced models on a 5-fold cross validation with 20% hold outs, leveraging an L2 regularization to correct for complexity  

For updating models, we randomly hold out 20% of the data for each question. This means that none of the calculus that tunes model parameters even knows these responses exist; they are purely for internal validation. The hold out set was already surveyed, we know their responses, and they passed our survey data quality QA.  

Note that this hold out set has nothing to do with the survey design and audiences that are created from survey data have no random hold out set.  

Audience Building FAQs  

Why does the projected audience size change depending on what attributes are used in an audience definition?  

We are predicting the presence of an attribute and every attribute across our cookie jar. When OR statements are created when building an audience, the Resonate platform will first try and accomplish that by looking at waves where all “or”s are present. If the audience that creates is above the existing threshold then it will present that audience, if it isn’t then it will look at imputed data across all waves and present that audience. This doesn’t directly relate to cookie modeling.