Decision Analyst is a global marketing research and analytical consulting firm, serving a number of Fortune 500 corporations, advertising agencies, and management consulting firms. Custom research and analytical consulting are Decision Analyst’s foundational competencies. Here we have videos about marketing research, brand loyalty, advanced analytics. Many of the videos come from our educational Insider Series Webinars. We host these webinars once a month: www.decisionanalyst.com/webinar/
Thanks for sharing, just a quick question regarding price elasticity, to calculate it we need to choose two time periods and see how things are changing right? if we choose period A and B, there is a value, if we change to period A and C or B and C, the elasticity result can be quite diff. So over time this is quite dynamic, how do we approach this variance problem and get a reliable elasticity estimate ?
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
Hello! I liked this video. I ran an OLS regression with 10 variables. Some of the unstandardized beta coefficients are negative. This makes it difficult for me to decide what threshold I should use for the y-axis. Normally people just take the mean of all betas, but in this case when they are negative it doesn't seem right. What suggestions do you have in this case?
Hi I have a data analyst degree and can code but I do not have a maths background such as calculus, linear algebra, statistics etc. Is this something that’s heavily part of the job or can I apply?
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.
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.
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.
Very interesting video! Thank you for sharing. I have a question and would appreciate it if you can answer. The process that you present here is a very solid clinical process which of course I do like, but its hard to imagine that it can be applicable for smaller businesses or startups as it is costly and time-taking with heavy focus on a marketing approach. Do you use this approach for big corporates that might need rebranding or merge of brands? or is it merely for consumer products and not the brand itself? Most of the brand studios and agencies take a much faster approach and the whole process of branding or rebranding can take something around 2-4 months (with exceptions). Of course research is always a part of brand strategy and brand identity design, but it is usually done in a month or so by very few people, and the focus is most on working with personas, brand personality archetypes, and iterative concept development.
"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.
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.
Hey John, Thank you for your video - I was trying to explain Econometric modelling to one of my previous colleagues to help them in their new role, and this has proven very useful - Kind regards, Kat
Great Content Guys! I am in the middle of installing an attrbution model at the company I work for. This def helped me understand the models & how attribution truly works!
If I recall correctly, the accuracy (usability) of the model results is limited to scenarios that fall within the general range of the input data. For example, if your model uses monthly data and your historical marketing spend has varied from $10k-30k/month and total revenue has varied from $50k-100k/month, then the expected return on investment per marketing dollar spent is only valid when the scenario marketing spend is limited to $10k-30k/month and the forecast revenue is within $50k-100k/month. Isn't that correct? Any scenarios going outside the range are extrapolations of the model, and the reliability can drop (perhaps with unknown probability).
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.
Great video! If two attributes are collinear, how would random forrest model interpret their %reduction in MSE? Will it be non-sensical? will the two collinear attributes have similar % reduction in MSE?
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.
This video is absolutely amazing. I am taking on my first MMM for Microsoft and am having to cram study so I can meet my project deadline. This has made my life much more simple. Much appreciated!
