Mailbag

Hi all,

Received this in the mail today and responded to it. Am putting up the exchange here coz I think it merits further dissemination.

The email I got:

Hi Sudhir,

I am a CBA Batch -3 student from Section-A

I am facing issues relating to very fundamental meanings of terminology introduced in DCBA. It may be due to the reason that I am not a business guy who is well versed with business terminology.

eg I am not comfortable with the following keywords: construct, dichotomy, Costs of Capital, trickier proposition, business meta-process and so many keywords introduced on Slide 16 Problem Formulations Examples of R.O.s, and then in psychometric scaling

Considering the example of Baskin Robbins. I am not able to get how psychology is coming into the picture here?

I may sound as asking stupid questions but I know I need to do something about it so that I can get comfortable with this subject.

Please provide me some directions..

Thanks,

P

My response:

Hi P,

Let me try to systematically answer what I can.

1. Regarding what a 'construct' means in our context, pls refer to this blogpost (from the PGP class):

http://marketing-yogi.blogspot.in/2014/09/session-2-exposition-what-are.html

2. The definitions of 'dichotomy', 'business processes' and meta-processes, 'cost of capital' can be had from a google search. A dichotomy means a branching into two separate streams. Thus, Data types exhibit a dichotomy - primary versus secondary data etc.

3. 'trickier proposition' is an expression in speech that means "is more problematic" or "is more challenging".

4. Not every construct need have profound psychological drivers. Many are fairly routine and habit driven.

5. The Baskins Robbins example has nothing to do with psychology. Its merely meant to illustrate the primary-secondary dichotomy.

I hope that helps clarify things somewhat, at least. Thanks for reaching out. I might put up this entire exchange on the blog, in case other students are also facing the same problem.

Thanks,

Sudhir

***************************

Updates. Received two more email queries. My responses are also putup below.

Hi Professor,

I am a student of CBA batch 3. I just had a query around the R code for text analysis (filename: textanalysis R code.R), I have gone through the entire code and wanted to understand the last part i.e. Bayes Factor Model selection and thereon. Can you kindly guide me on this?

I am not able to conceptually grasp the concept of Factor Model and the output from that point onwards.

Look forward to hearing from you.

RT

My response:

Hi RT,

>> I have gone through the entire code and wanted to understand the last part i.e. Bayes Factor Model selection and thereon. Can you kindly guide me on this?

Your query concerns what we call 'model selection' in statistics. A model is a set of relations which we fit upon data to explain them and/or make predictions about them. However, there maybe multiple models that fit the same data.

One way to sort through this multiplicity of models and select the *best* one is to first find how well each model 'fits' the data (i.e. has the least squared error). Accordingly, various 'goodness of fit' criteria have been developed and deployed. The Bayes Factor is one such, very important fit metric in Bayesian statistics.

For our purposes, just take the model results and use them to select the model with the optimal number of components (optimal, as decided by the log bayes factor). Going beyond that would be beyond the scope of the DC course. Wikipedia and other web resources are available however, in case you want to do a deep-dive into fit statistics in general and Bayes Factors in particular.

>> I am not able to conceptually grasp the concept of Factor Model and the output from that point onwards.

When we 'factorize' something (say, a), we break it down into pieces (say, b,c and d) such that the product of b*c*d will yield a back.

In general, any number can be 'factorized' into a product of primes. Similarly, when we factorize a matrix, we break it down into 'factors' whose product yields the original matrix again.

We took the TDM and 'factorized' it (conceptually only the LDA is more complex in its assumptions and its estimation) into 'factors' - terms that together can be interpreted as topics.

For our purposes, all we need to know is that using the latent topic factor model, we 'broke down' the corpus into distinct 'topics' or themes that can be interpreted and used for further analysis.

I hope that helps clarify.

Sudhir

Another one below:

Respected Professor,

I'm a CBA student from technology background. I need your help regarding data collection:

1. Is there any book that I can refer? I feel I'm lost with so much of info/topics. Also with no audio for first class, it seems I don't have way to revisit the fundamentals discussed.

2. It will be extremely helpful if you can please provide some practice papers and solutions. (hope that's possible)

3. Could you please also clarify whether Facbook assignment is group or individual H.W.?

SP

My response:

Hi SP,

>> 1. Is there any book that I can refer? I feel I'm lost with so much of info/topics. Also with no audio for first class, it seems I don't have way to revisit the fundamentals discussed.

I don't use any one text book for DC. The material is collected and collated from multiple sources. However, wikipedia is your friend in case you need more detail on particular topics. Also pls check the early blogposts for yourbatch on analytics-yogi.blogspot.in where some additional links and material was putup.

>> 2. It will be extremely helpful if you can please provide some practice papers and solutions. (hope that's possible)

The exam is open book-open notes. The questions are all short answer quetions (no essay length stuff) for more grade-ability and objectivity. I can;t make any promises regarding the practice exam at this point as I plan to modify the exams I have from previously for this batch as well.

>> 3. Could you please also clarify whether Facbook assignment is group or individual H.W.?

Individual. Because each of you has to pullup your own FB data.

Hope that clarifies.

Sudhir

Ciao.