代做Marketing Analytics、代写data analysis、R代写代写、代做R程序语言

日期：2019-05-14 08:19

Take-Home Assignment

Marketing Analytics, 2019

Instructions

1. Answer all questions. The total number of points is 150.

2. Every question requires you to run data analysis and interpret the results. While the former is obviously crucial, it is really important that you do explain your results and conclusions. If you modify the data in any way, e.g., drop rows, create new variable – indicate clearly what you’ve done and why you’ve done it.

3. For each question, please attach the relevant code as an Appendix to the Word file (very similar to the documents I used to give you in class). This avoids loading too many R files and keeps most of your answer in one document. Note that it is important you supply the relevant code - I have to see evidence that you actually ran the analysis on which your conclusions are based. I should be able to copy your code in R and run it on whatever data file you used to come up with your results.

4. To avoid too much clutter, feel free to put tables, etc. in an Appendix. I should, however, be able to follow what you’ve done, what your conclusions are, and how you arrived at those conclusions, from reading the body of your answer.

3. It is hard to set a word limit on this exam. For the most part, your work lies in running the analysis. You can explain the results of practically any analysis in a paragraph. Nevertheless, to prevent an arms race in the number of pages, I am going to set a limit of 12 pages for the entire exam. This does not include the Appendix. If you have run a great deal of supplementary analysis before coming up with the results you present, please put details of the supplementary analysis in the Appendix, and just refer to it in the body of the paper. To reiterate – the 12 pages are what I will be reading really carefully, so make sure you give a great deal of thought to what goes in there.

4. You are not allowed to consult anyone in the course of doing the assignment. You are free to use any other material – your notes, material available online, etc. As always, the hope is that you’ll challenge yourself and learn something while doing the assignment. To that end (not that I need to tell you), it is fine to email me with questions/concerns. And, of course, if you find errors while running the data, do let me know (I’ve checked, but as you all know, data is frequently messy (this is especially true for the conjoint data, which is a real-world survey)).

1. Regression

Data for sales of Chromebooks across 400 regions is given in the file Chromebooks.xlsx. The tab labelled Code Book describes the variables used.

a) Fit a regression model with only main effects (i.e., no interactions) to explain what affects sales of Chromebooks across different regions. Explain in words what each of your coefficient estimates mean. [Note: You don’t have to run diagnostics to check for multicollinearity, heteroskedasticity, outliers, etc. Just run the model.]

b) Now add any interaction effects you think might be meaningful. It is important that you have a justification for why you are including what you are including (the fact that you ran it and found significance is not the right justification). It is fine if you think something should have an effect and it does not. [Note: Please be parsimonious here – there is no need to show 15 models that you ran for all possible combinations of interactions.] Again, explain in words what the coefficients in your new model mean.

c) The company wants to predict median sales, i.e., sales with all the predictor variables set to their median level (for any categorical variables, pick a suitable level). Do the calculation and predict median sales (for prediction, retain only variables that were significant in the models you ran above).

(40 points)

2. Machine Learning (Classification) Models

A retail store chain is interested in marketing products to pregnant women (more precisely, a number of manufacturers of the relevant products are interested in marketing to these women). To do this, the retailer needs to develop a machine learning algorithm that gives it insight into which women are likely to be pregnant, so they can be sent promotional messages for relevant products. As a first step, the retailer has gathered data on the purchase history of a set of its customers; the set has a mix of pregnant and non-pregnant women. (To clarify, these purchases are at the household level; the household could have a pregnant woman in it, or not.) The data is given in the file labelled Pregnancy Data.xlsx.

Your task is to come up with an appropriate prediction model for the firm. To make the task more specific, please do the following.

i) Start by splitting up your data into training and test samples.

ii) You have a range of models you can consider – logit, bagging, random forests, boosting. Essentially, you are faced with a classification task, so start with logit and move on from there.

iii) Interpret each of the models you run. This means, explaining what you found and explaining how well the model does (make sure to show the confusion matrix and explain it).

iv) Pick a final model and give recommendations to the retailer based on this model. You might pick a model because it has the highest AUC, or you might pick it because it has a slightly lower AUC but is easier to understand managerially, or … whatever criterion you use, justify it.

(50 points)

3. Choice-based Conjoint

An entrepreneur has what he thinks is a wonderful idea for a dating app. He wants his service to fill the gap between rigid speed dating events and regular socialising, in order to provide a new concept of “thematic dating.” He has a configuration for the service in mind, but is unsure of market preferences, and hires a group of MSc Marketing students at LSE to help him out. The students, being extremely well-trained and bright, figure that they need to conduct a survey to address the client’s needs. In particular, they decide to perform a conjoint analysis. Data for a set of respondents is given in the file labelled Conjoint Vee Data.xlsx. The tab labelled Code Book gives details on the choice task and describes the attributes and levels.

a) Run a choice-based conjoint on the data provided. You’ll have to make some choices (e.g., there may be missing observations), but the data is generally fairly clean.

b) Explain in words what you infer from the coefficient estimates, calculate importance weights, and calculate the exchange rate.

c) Pick a set of product profiles and run a market share simulator. (You obviously cannot include competitors like Tinder, because the attributes are different.) Explain the output.  

d) Figure out appropriate segmentation for the service. To do this, you’ll probably have to run a model with heterogeneity and then do a cluster analysis on the output. Given the small number of respondents, don’t go overboard on the number of clusters. Again, make sure to explain what you’ve got.

e) Come up with a final recommendation for the entrepreneur. What should he be doing?

(60 points)