The Composition metric, expressed as a percentage, tells us the number, or share, of people in the audience who have an attribute or trait.
The Composition numbers for all attribute values under an attribute may not always add up to 100%.
There are three primary reasons why you can encounter these metrics adding up to less than 100%:
- Multi-select survey question
- Follow-on survey question
Probabilistic Data Projection
1. Multi-select survey question
Our survey contains questions that allow survey-takers to pick multiple responses. Because of this, these responses are not going to be mutually exclusive and responses will overlap with one another. And when answers overlap, the sum of the composition numbers for all attribute values in an attribute is likely to be over 100.
For example, the survey question asking about Retailer Selection Traits allows participants to identify three traits.
If all survey-takers picked three Retailer Selection Traits, the percentages will add up to over 100% since survey-takers are asked to pick three answers.
2. Follow-on survey question
Some questions are conditional follow-ons and are only asked if a survey-taker meets that condition, based on previously provided answers. In this case, the sum of composition numbers will represent the percentage who qualified to answer the follow-on question.
For example, we ask survey-takers whether they have any pets. If someone responds with yes, that participant will get a follow-on question about the types of pets she owns. In this case, "Which of the following pets do you have in your household?" is a follow-on question asked only of those people who said yes to being a pet owner. Because not everyone in the online adult population owns a pet, the composition numbers for an audience owning a dog and owning a cat will not add up to 100%. Instead, their sum will refer to the percentage of people who said yes to parent question of owning a pet.
3. Probabilistic Data Projection
You're analyzing a single-select attribute (like Gender or Household Income - where survey respondents can only have one answer) against a Connected Profile audience (Tag, 1st party, contextual, etc.) and results aren’t summing exactly to 100% as you believe they should for a single-select attribute. Is this to be expected? Yes, this is expected and okay.
At times, Resonate’s Ignite platform may reflect compositions of single-select demographic attributes that sum to slightly over or under 100% based on Resonate’s data optimization methodology which is designed to provide rich insights in seconds. Insights are calculated by linking the activity from Tags, 1st Party Data, or Contextual Studies to the projections from our National Consumer Survey across 13K insights. As the activity from these data sources can be based on a combination of up to two to three trillion total data points, we compute the projections based on a probabilistic methodology that is highly accurate and biases towards speed to insights.
To provide greater speed to insights for our customers, Resonate leverages a probabilistic data projection methodology for these high cardinality data sets. This probabilistic methodology, leveraged by best-of-breed companies that work with large data sets including Adobe and Google in the advertising ecosystem and Nielsen and comScore in the research ecosystem, is highly accurate with a standard error of <2%. However, in rare circumstances, it is possible that these probabilistic projections may result in distributions with decimal points that round up or down in the Resonate Ignite platform. With this rounding, the total of all options across a set of mutually exclusive attributes (like Gender, or Household Income) may sum to be near or slightly above 100%. This is expected and enables us to benefit our customers by providing insights in seconds rather than minutes or hours.