Have a question?

Find the answer on our list of frequently asked questions:

What is PII

PII, or Personally Identifiable Information, is any data that can identify a specific individual. In our data, this includes full names, phone numbers, email addresses, state ids, etc.

If I don’t have enough targets, how should I expand my universe?

Sometimes we find that there are not enough available records to fulfill the requested amount of targets. In these cases, you can receive the available records or expand your universes. If you have a specific county or district, expand to surrounding areas or statewide. If you are using a flag you can instead use the score and use a lower score cutoff value.

What is a DV or Dependent Variable?

This is the thing you want to predict.  It might be whether or not someone voted in a past election, who they say they will support if surveyed, or the amount they will give if solicited for a campaign contribution.  A Dependent Variable (often abbreviated DV) can be binary, categorical or continuous.  

A binary dependent variable is true or false, yes or no, 1 or zero.  It can have only two known categories (for some individuals the DV might be unknown).   

Categorical dependent variables are those that fit into distinct categories.  A binary dependent variable is a subset of categorical DVs where there are only two different categories, but there can be categorical DVs with more than two categories.  For example, you might want to predict a voter’s top issue from a menu of five different issues.   The categorical DV for that question would have five possible categories plus unknown.   Often, this type of prediction will be made by building separate models for each possible category, so the five-category DV would be translated into five separate binary dependent variables each one giving a true or false (one or zero) value indicating whether or not the voter has picked a particular issue as their top concern.  But, there are some types of modeling algorithms that directly predict multi-category categorical dependent variables.

What are the flags?

There is an art and science to the modeling process.  The flag process is more of an art.  Years ago, when we only had scores, there were digital vendors that wanted a quick way to find the best targets on each score, thus we created the flags.  In short, it is a combination of the distribution of the score (where there is a drop-off in score quality) and what the polling numbers on the issue show. Haystaq goes through and looks at the distribution of each score and comes up with the best target audience based on where the distribution happens.  In the user guide you can see where the flag was set with each of the models that we built.  These flags range from 10% to 35%.  You can see the first two models in the user guide, abortion, we set the flag of pro-choice at 30%  and the flag for pro-life at 20%.  We often tell folks that are looking for the best audience to start with flags to get a sense of where you can go and expand/shrink from there.  Finally, as noted polling also plays a factor in finding the flags – for example if only 10% of people are in favor of cutting Social Security via a poll, but there is a natural drop at 30%, we still limit to 10%. 

How do you determine flag cutoffs?

Haystaq goes through and looks at the distribution of each score and comes up with the best target audience based on where the distribution happens.  In the user guide you can see where the flag was set with each of the models that we built.  These flags range from 10% to 35%.  You can see the first two models in the user guide, abortion, we set the flag of pro-choice at 30%  and the flag for pro-life at 20%.  We often tell folks that are looking for the best audience to start with flags to get a sense of where you can go and expand/shrink from there.  Finally, as noted polling also plays a factor in finding the flags – for example if only 10% of people are in favor of cutting Social Security via a poll, but there is a natural drop at 30%, we still limit to 10%.

What is an Ntiled Score?

An ntiled score is a model score that’s been converted from its raw value into a rank-based bucket, typically 1–100 (percentiles) or 1–10 (deciles). 

How do you conduct surveys?

We randomly select contacts from our database using a set of generic demographic buckets, often four age buckets and two to three race/ethnicity buckets (though this varies by the region of interest). We send texts to these contacts with a link to our survey. We aim to receive 5k complete responses, with approximately an equal number per demographic bucket – often this requires following up with phone calls to contacts in harder-to-reach demographics.

What does AUC mean? What is a good ROC/decile graph?

AUC, or Area Under the Curve, is derived from the Receiver Operating Characteristic (ROC) curve. The area under this curve quantifies the model’s ability to separate positive and negative classes, providing a robust, threshold-independent measure of a model’s quality. An AUC = 1.0 is perfect classification, though often indicates an overfitting problem with the model and a poor ROC graph. An AUC <= 0.5 is essentially random, indicating that the model can’t distinguish between classes and a poor ROC graph. An AUC score approximately between 60 and 90 is the sweet spot.

What is the stable polling score?

The stable polling score was requested by some of the pollsters who use our scores.  It is a raw score (not changed into a percentile) based on the same question every year.  It uses the same DV definition and algorithm to create the score each year.  For existing voters, this year’s score is blended with last year’s score (New voters are assigned this year’s score).  This creates a partisanship score that moves very gradually over time, making new samples pulled from it very similar to past samples using the same criteria.

Can I use your scores to buy digital ads?

Yes, our scores are available on a variety of digital platforms through L2 and will shortly be available through TargetSmart.  Please call for more details.

What’s the difference between the L2 Voterfile and Commercial file?

The L2 voterfile comes with PII of voters along with basic demographic fields released by the state. L2 buys its voterfile from each state and updates new registrants and vote history as each state releases files. This file is then enhanced with either publicly sourced or commercial data to provide a more complete view of each voter. The majority of states do NOT allow the voterfile to be used for commercial purposes and limit its use to political outreach and advocacy. 

On the other hand, the Commercial file, though having a similar appearance to the voterfile (PII and demographic fields followed by augmented commercial fields), CAN be used for commercial purposes because it is not sourced from the state’s election office. The Commercial file includes both voters and non voters and is often used to find citizens who are eligible to vote but have not yet registered.

Can I use your scores in VAN/Targetsmart/Catalist?

Yes! We can transfer scores using statefileid (the id assigned to each voter by a State’s secretary of state), countyvoterid, county FIPs code and PII. These fields can then easily be matched on import into other CRM systems such as VAN or Catalist. Our scores will be in TargetSmart soon, however you can always get them at www.l2datamapping.com


Still have questions? Reach us directly!

Contact Us