Why bother with MRP?

Traditional Quota Sampling

Quota sampling can often give us a rough idea of our KPI of interest but may suffer from noise. Noisy data can often be identified by large fluctuations up and down over time. The problem with such noise is that it gets difficult to say exactly how high or low our KPI actually is, and which increases or decreases are legitimate.

One consequence of this is larger margins of error (MoEs) in quota sampling. These difficulties exist at the general population level, and get worse as we look into more niche audiences, especially those with smaller sample sizes.

MRP

Latana uses MRP models to address these problems in a couple of ways. The first comes from not treating data as isolated in time, like quota sampling does. Opinions and awareness take time to form and change, meaning that brand and consumer behaviour KPIs tend to evolve smoothly. Latana’s MRP models consider data from all waves up to a point in time simultaneously to construct the story which makes the most sense across time. Waves with more data give us more information and help to anchor the narrative. Those waves with less data are more informed by the surrounding waves so that noise does not result in a misleading conclusion.

This is not just a trend line drawn across the quota sampling estimates. This method of connecting data through time is applied at the most fine-grained level to estimate the KPIs for every audience simultaneously. Due to the coherence of all these estimates, we still get a plausible story after aggregating to our target audience.

By working with the data in its entirety, any KPI fluctuations are greatly reduced, so real changes are easier to identify. The margins of error are significantly narrower, which means higher confidence in the data, and the ability to look at segments with confidence.