Skip to end of metadata
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

Version 1 Next »

Opportunity-to-see

Once an ad unit is plotted on the map, Geopath determines its viewshed. The viewshed is the physical area where the ad unit’s message can be seen by people traveling toward it. It is typically teardrop-shaped, and its perimeter is defined by the Maximum Noticing Distance and Right and Left Viewing Angles.

Maximum Noticing Distance

The maximum noting distance is the furthest someone can be from the display and still have a meaningful viewing experience. The average person first notices an object when it is close enough to take up 1.5 degrees of their field of view, this is known as the minimum angular size (α min).

The formula for calculating the maximum noting distance is:

For example, the maximum noting distance for a 14 ft x 48 ft frame is 1,909 ft, meaning if someone is traveling directly towards the ad unit and the ad unit is oriented directly at them, the message would start being clearly viewable and able to be noted when the person is 1,909 ft away.

Maximum Viewing Angle

Media is optimally viewed when it is head on, but people’s paths vary as they pass by and have the opportunity to view media. When the media is oriented more than 60 degrees from a viewer, it is considered unviewable. This maximum viewing angle modifies the extent of the viewshed on each side of the media unit.

Further Refining the Viewshed

Only people traveling towards the ad unit in the viewshed have an opportunity-to-see the ad unit.

 

 

If obstacles, such as a tree or bridge, blocks the view of the ad unit from a specific vantage point, people traveling in that blind spot are not counted towards the audience.

Digital Ad Exposure

For digital units, with ads running in a rotation, the opportunity to see any individual ad is a function of the amount of time a person spends in the viewshed (dwell time), how long the ad is displayed each time in runs (spot length) and the number of ads running in the loop (share of voice). Time spent in the viewshed is based on vehicle speed estimates from HERE Speed.

First, we calculate the number of opportunities a person has to view an ad each time they pass through the viewshed. Let’s call this the opportunity per person (OPP). For example, if a person spends 30 seconds in the viewshed on average and the average spot length is 10 seconds, then the person has 3.9 chances to see the ad each time they pass through the viewshed.

Next, we calculate how likely it is for a person to be exposed to the individual ad based on the number of ads running in the loop. Let’s call this the opportunity per spot (OPS). To continue with our example, if the spot is one of eight running in rotation on the display, then the spot has a 1/8 share of voice, and an individual person passing through the viewshed would have a roughly 49% chance of being exposed to that ad.

So finally, if 100 people passed through the viewshed, 49 people would have had an opportunity to see the individual ad running in rotation on the digital unit.

Likely-to-see Audience

Geopath applies a Visibility Adjustment Index (VAI) to the OTS audience to estimate the number of people who likely viewed the message on the ad unit. This is a measure of engagement.

To create the VAI, Geopath equipped 300 vehicles with Go Pro Cameras and provided the front seat passenger with Eye Tracking Glasses. These test subjects traveled a fixed route containing 50 roadside ad units of varying size and at differing distances and angles from the roadway, producing over 15,000 data points (300 x 50). A detailed description of this research is available in the next section.

This data is used to extrapolate a unique VAI for each of the 6 million ad units measured by Geopath. The key factors that impact an ad unit’s VAI are:

Size of the Ad Unit

  • The width and height (i.e., area) of the ad unit face.

Right-hand versus left-hand read

  • Whether the ad unit is on the right or left side of the person.

Perpendicular read

  • Whether or not the ad unit’s surface is perpendicular or parallel to the person.

Traffic angle

  • The angle between the direction of travel and the ad unit orientation. This is a more low-level feature than a perpendicular read.

Roadway classification

  • A categorical variable associated with a roadway’s physical and operational characteristics, e.g., interstate, highway, arterial, residential.

Illumination

  • The impact of artificial light on the ad unit. Digital ad unit may be illuminated all day, while non-digital ad units may have spotlights at night.

 

Visibility Adjustment Index (VAI)

Here is a detailed description of Geopath’s original field research into outdoor billboard viewership.

Data Collection Method

Geopath conducted a study in 2013 where a total of approximately 300 respondents were driven by about 50 billboards in two markets: Peachtree Street in Atlanta, Georgia and I-35E/I-30 in Dallas, Texas. The respondents and the billboards are uniformly distributed across both markets.

The data consists of:

  • GPS data The trajectory of each respondent as they passed by the billboards

    • These GPS data were processed after the fact to obtain the speed and the dwell time of the respondents within the viewshed of each piece of inventory.

  • GoPro recordings Front-facing recordings showing the field of view of the respondent

    • These recordings were processed after the fact to mark the exact points where a billboard enters and exits the respondent's field of view.

  • Eye tracking glass Respondents' eye movements and focus points

    • These recordings were later used to identify a noting event. Noting events are defined as eye fixation that occurs for more than 60ms over a billboard (or spot if digital).

Combining the three different components above, we can study the relationship between the probability of noting versus the spatial and temporal characteristics of the pass. For the purpose of this study and the for the sake of calculating the billboard noting probability, the individuality of spots on digital billboards was ignored.

The viewshed is intersected with the moving field of view of the passers-by, which is a function of

  • The orientation of the billboard

  • The heading of the nearby roadway segment

  • The distance of the billboard from the road

When there is an intersection of the above two components, there is an opportunity to see a billboard. The probability of noting is influenced by the following.

  • Distance from audience The perpendicular distance of the billboard from the roadway segment

  • Apparent size The weighted average of the angular size of the board from the perspective of the audience passing by as they travel through the viewshed and while the billboard is in their field of view. Angular size is in degrees and is a rough estimate of the number of degrees in the audience’s field of view that is occupied by the billboard. Apparent size is a function of (1) how far a person is from a billboard and (2) at what angle they see the billboard. The figure below schematically demonstrates how apparent size changes with distance from the billboard.

  • Degrees off-center Similar to apparent size, this is the weighted average of how off-center the billboard is with respect to the center of eye sight, as a person drives through the viewshed and the billboard is within their field of view. This variable is a function of (1) the relative heading of the travel path and (2) the distance between the road and the billboard. The figure below shows the changes in degrees off-center through travel trajectory.

Overall, apparent size and degrees off-center have a non-linear relationship with the lateral distance between the location of the viewer and the billboard. To sum up, the figure below shows the apparent size and degrees off-center for a person traveling on a road that is perpendicularly 100 feet away (shown on the x-axis) from a 14-foot by 28-foot billboard.

  • Trajectory distance The distance traveled while the billboard was in sight.

  • Dwell The amount of time a person spent in the viewshed and while the billboard was in their field of view.

  • Speed Average speed of travel through the viewshed.

All of the components above show some level of correlation with the probability of noting. The distributions below show the variation in each of these features, whether or not the billboard is seen by the respondents.

Model Calibration

With the understanding from the correlation analysis above, a logistic regression model is calibrated to predict the probability of noting, given the components discussed earlier.

Out of the feature listed above (size of the ad unit, right-hand versus left-hand read, perpendicular read, traffic angle, roadway classification and illumination), only illumination has shown a statistically significant impact on the noting probability and thus was included in the final model. Moreover, the features with strong correlation (for instance, travel speed, dwell, and trajectory distance) are consolidated to avoid multicollinearity in the regression model. The model shows that, as expected, the probability of noting increases with an:

  • Increase in apparent size,

  • Decrease in degrees off-center,

  • Increase in dwell, and

  • Illumination.

The figure below supports this.

The model parameters were estimated using a 10-fold cross-validation technique. In 10-fold cross-validation, the original sample data are randomly partitioned into 10 equal-size subsamples. Of the 10 subsamples, a single subsample is retained as the validation data for testing the model, and the remaining 9 subsamples are used as training data. The cross-validation process is then repeated 10 times (the folds), where each of the 10 subsamples is used exactly once as the validation data. The 10 results from the folds can then be averaged (or otherwise combined) to produce a single estimated model. One of the advantages of this method is that all observations are used for both training and validation, and each observation is used for validation exactly once.

Model Validation

Applying the final model on the collected dataset yields a prediction accuracy of 62% in predicting individual noting events. And aggregated over each billboard, the predicted and actual percentage of noted correlate with an R-squared value of 0.58 and a root mean square error of 15.7 (persons - ~10% in percent noted). These stats are calculated stochastically over 100 iterations. The figure below shows the predicted noting probability distribution (left) and the predicted versus actual noting percentages by frame (right).

  • No labels