Відео

Thanks for your question. One would expect that the impact of a price change manifests over time since buyers would not, typically, be immediately aware of the price change. When price elasticity is based on stated choices in a survey, one approach is to include an awareness-build over time to the simulation equations to simulate short- or long-term impacts of price changes. The survey results, unadjusted, are more representative of a long-term price impact since survey respondents are 100% aware of prices. When price elasticity is estimated based on econometric modeling of historical data, distributed lag models are often used to account for lagged impacts of a price change.

@@DecisionAnalystArlington thanks, i suppose survey based result is more accurate but hard to obtain.... result based on historical data, elasticity between 2021/2022 can be very diff from that of 2022/2023, having time series data can produce many diff results.. will take a look at the distributed lag models you referred to

Glad you liked it!

We do not have a PowerPoint presentation. We have a short article on the same topic: www.decisionanalyst.com/blog/distributionstrategy/

Sample size for a MaxDiff exercise should be much larger than 8 respondents. As a general guideline, if the MaxDiff exercise contains a total of 20 items, the total sample should be 200 or more.

Thank you!

Market Segmentation helps a company or a brand focus its marketing energies to maximize the return on investments. Market Segmentation answers four basic strategic questions: What market segment or segments should be targeted? What is a brand’s optimal positioning to reach its targets? What advertising themes and messages should be explored? What new product opportunities exist within a given marketplace? A segmentation should be done, when you need answers to the questions above.

You're very welcome!

Thanks

You're welcome!

Advanced regression models, such as Random Forest models and Relative Weights Regression, tend to perform well. Categorical drivers should be dummy coded prior to entering the regression. You could also consider transforming your data (there are a number of transformations you can use) before conducting the analysis.

"The approach I propose is too expensive for most smaller companies and for many startups. In a world of perfect worlds, however, all of the proposed steps are important and, taken together, lead to much better branding solutions. True, many branding agencies and branding consultants work on a short time cycle, but most of these branding agencies and consultants rely very little on research-based or evidence-based results to drive and guide the branding work. Most pay lip-service to research, but the reality is research plays a small role in the final outcome. Or, at least, that has been our experience and observations over a few decades.

@@DecisionAnalystArlington thank you for answering my question. 🙏🏼

Although Shapley Values are not regression coefficients and cannot be used in making predictions, these values can be used as measures of derived importance. Other measures of derived importance include correlation values and regression coefficients.

Glad it helped!

Good point. It is always true that predicting an output (e.g. revenue) based on values of independent variables that are outside the range of the historical data used to build (or train) the model must be done with great caution and is generally not recommended. Additional data is required to predict certain spend levels. For example, one could design an A/B market test to explore whether the relationship between digital marketing spend (of a particular type) and unit sales still holds, if the digital spend were to increase to a level that is 50% higher than experienced within the historical data. John Colias, SVP, Advanced Analytics

The Excel Simulators are custom designed for our clients as part of a marketing mix modeling project. Each Simulator is unique based on the project details, client specifications and goals, and the type of data collected and also provided. If you are interested in having us conduct a marketing mix modeling project, please contact jcolias@decisionanalyst.com.

RF is just as subject to collinearity problems as any other multiple regression technique. However, because RF tests many different combinations of predictors as well as number of predictors, the potential for unintended impact of collinearity is reduced. It is best practice to examine the correlations among the potential predictors and remove any predictors that are very highly correlated with others. Still, though, if collinearity exists among the predictors submitted to the RF procedure, the % RMSEs that result from these highly related predictors are generally lower than they would be otherwise.