Brands can’t get enough of MRP — and we don’t blame them! Understandably, they’re overjoyed to finally have access to reliable and accurate brand insights. However, many still find the concept a bit tricky to fully grasp.
A frequently asked question we deal with often is "What is MRP?". A top-level answer is that MRP is advanced data science. However — to be fair — we don’t know too many marketers who double as data scientists.
Therefore, we’ve written a comprehensive guide to MRP-powered brand tracking. However, we thought it would also be useful to provide you with a shorter, explanatory article focused on answering the most popular questions we receive.
We hope you find the answers you’re looking for here. If not, please don’t hesitate to send your question to us. We’d be happy to provide you with an answer!
1. What is MRP?
MRP, or Multilevel Regression and Poststratification, is a form of advanced statistical modeling that was made popular by Professor Andrew Gelman. Based upon the Bayes’ theorem — which was formulated by English statistician and philosopher Thomas Bayes in the 18th century — MRP takes into account prior data to determine probability.
First used as a means of forecasting election results, at Latana we use MRP for brand tracking purposes.
In essence, MRP uses data to create a model and then uses said model to generate estimates for responses in a survey. So, when given a set of respondent characteristics, the model can produce an estimate for how a certain respondent would answer a particular survey question.
MRP then organizes respondents’ characteristics into groups — which allows it to better understand and capture how variables interact in real life. Essentially, using MRP enables Latana to make accurate predictions by assuming that certain audience segmentations have similar preferences to other segmentations in different areas.
In the last step, this method takes weighted averages of all the predictions to ensure that the model has a fair sample of respondents.
2. How is MRP different from traditional quota sampling?
One of the biggest differences between MRP and the quota sampling used by traditional brand trackers is that traditional methods are unable to accurately measure opinion within small target audiences.
Why? Well, because traditional quota sampling narrows in on specific respondents within a target audience — which, often results in very low sample sizes that drive up a much higher margin or error on the data. Furthermore, traditional quota sampling delivers raw data with very limited processing. This means the accuracy is limited and brands won’t be able to identify as reliable insights.
MRP, on the other hand, does the opposite by not restricting itself to the small number of respondents and taking into consideration surrounding and past data patterns. This, in turn, allows MRP to achieve much higher precision — even for niche audiences.
Essentially, MRP uses information from the entire sample to create a model that can predict a KPI output such as brand awareness based on respondents’ characteristics. MRP also has the ability to recognize real-world changes and deal more effectively with outliers.
The outcome? MRP can provide more reliable insights with higher precision and a smaller margin of error.
3. What is the role of sample size in MRP?
This is another nice thing about MRP — it provides direct information regarding the number of respondents needed in a sample to achieve good, representative estimates.
Sample size, among other factors, is directly linked to the accuracy of a result. This accuracy is represented by margins of error.
The margins of error produced during MRP imputations can give direct feedback as to the sample size requirements of an audience, based on the scope of brand data we require. Therefore, at Latana we adjust sample sizes accordingly to ensure we always provide reliable insights, no matter what the audience or brand size may be.
So, where traditional quota sampling may need 3000 respondents or more to achieve something close to representative data for a singular niche audience, MRP works dynamically to reduce this requirement to 1000 or fewer.
4. How is significance measured in MRP?
MRP uses a Bayesian model to predict brand awareness, which is based on respondents’ characteristics. This Bayesian framework provides us with an advantage, as we’re able to figure out the measure of the uncertainty within our estimate for free.
Called “error bounds”, with MRP, the margin of error gets smaller the more information we feed to our model — e.g. by including prior information from the past or larger sample sizes. Thus, MRP can achieve statistical significance with less data input when compared to traditional quota sampling.
5. Why are some KPIs (e.g. the brand associations) the same for a variety of audiences for any given brand?
To accurately identify the effects of characteristics in a population, an MRP model needs a certain amount of information.
For hard-to-reach audiences — e.g. consumers who are aware of a very new brand or 75+ females with a pet — it can be very difficult to access enough information during the first sampling waves.
However, our MRP model will accumulate information over time and, after a couple of months, it should have enough information to tell us the difference in brand associations for different audiences — even for small brands.
6. Why is MRP particularly good at detecting changes over time?
As previously mentioned, MRP is a Bayesian framework that always comes with an estimate of uncertainty in its prediction. Why is that? It’s because MRP uses current and past data to correct for fundamental skews within a sample — thus allowing it to better detect real-world changes over time.
This estimate of uncertainty also helps with detecting changes over time since it allows us to make statements such as: “With a probability of 83%, there was a change in brand awareness between March and April within X audience”.
Essentially, knowing the estimate of uncertainty provides our clients with a more accurate picture of the reliability of their data.
7. Do my brand tracking insights get better over time?
They most certainly do! Since an MRP model learns over time, brand KPIs which only have a little information at the beginning will improve significantly over time as more data is collected.
This means that every wave of data you receive from Latana is better than the last — making it easier and easier to roll out successful brand marketing campaigns using your brand tracking data.
To be fair, there’s much more we could discuss concerning MRP. It’s a complicated, fascinating process that creates amazing results. As the first brand tracking software to use MRP, we’re excited to be at the forefront of the brand tracking world.
If you’re interested in learning more about MRP, feel free to check out our in-depth whitepapers, which provide more detailed information and examples. And if you think it’s time your brand had access to reliable, highly-accurate consumer insights, set up a demo with our Sales team to see if Latana is the right solution for you.
Updated by: Cory Schröder on 04.02.22