In this post, we explore the impact of screen selection on tracking variations and provide insights on how to optimize your campaign success.
To this end, we analyzed variations in measured tracking per day of the week between consecutive weeks on different sensor subsets within a given network.

For each experiment, a random subset of all available screens (with sensors) in a given network was selected and analyzed, e.g. 20 random screens out of 100 available.
By calculating the average number of trackings per day of the week and their standard deviation, we obtain a metric for variation (standard deviation divided by mean).
The experiments were repeated many (10000) times, each time selecting a different large random subset of the network.
Finally, the average of the variation in the trackings was plotted across all experiments with the same subset (see Figures 1-3).
In other words, if 16 different experiments were performed, each selecting 20 random screens, then the average of the variation of the trackings was plotted for the 16 experiments.

The experiments were also repeated for 3 different time periods: the first figure (Fig. 1) includes data between April 2022 and April 2023,
the second figure (Fig. 2) includes data between January 2023 and April 2023, and the last figure (Fig. 3) includes data from the last 2 weeks (April 17 – April 30).

As expected, the variations of trackings (and contacts) decreases as the number of screens used increases:
The more screens used, the lower the spread of trackings and thus contacts achieved for your campaign.

In summary, using a larger number of screens (>10) can help minimize variation in predicted contacts and allow you to plan the campaign with more reliable and accurate estimates.
Therefore, we recommend to plan a campaign with at least 5 screens.

Figure 1: Variations in contacts tracking per day of the week considering data between April 2022 and April 2023.

For a small number of selected screens (fewer than five), the variations in contacts for all days of the week range from 20-70%, except for Sunday, which shows fluctuations of about 100+/-60%.
Clearly, a small number of screens can have large variations in contacts between consecutive weeks and days of the week.
With a medium number of screens (5-20), tracking fluctuations drop to 10-50% for all days of the week, with Sunday still showing the largest fluctuations.
This shows that the variations can be reduced significantly with a moderate number of screens.
With more than 40 screens, the variations drop below 5-20% for all days of the week.

Figure 2: Variations in trackings per day of the week considering data between January 2023 and April 2023.

Figure 3: Variation of trackings per weekday considering data between April 17, 2023 and April 30, 2023.

We observe that the overall variation in trackings becomes smaller as the time span becomes shorter. We explain this behavior as follows:
The variation between several consecutive weeks is larger (than between a few consecutive weeks) because the number of trackings also follows seasonal effects.
But considering that the standard deviation is relatively large at small screen numbers, there may still be considerably large variations between subsequent weeks and even within a single week at one location.