Phase III grading complete.

Class,

The project grading is complete.

Descriptive summaries:
Out of the 14% course grade and across 43 project groups, the mean is 7.61, the std dev is 2.04 and the range is (3.5, 11.25).

The overriding considerations in grading were more on the logic employed to get to actionable recommendations than on analysis output that did not directly support the recommendations. Thoughtful execution showed itself quite clearly here, I must add.

The mandatory requirements were there for a reason - they essentially specified data descriptions, input and output summaries, matrix sizes etc be mentioned - to help gauge what the quality of logic and implementation. Sadly, few groups did well on this count.

The level and detail of analysis in many instances was quite impressive, I must add. How much they supported the research objectives was not always clear, though. Some of the wackier recommendations actually seemed to make sense - implying both creative, out-of-box thinking as well as the crucial element of having fun with the project.

Anyway, this was intended to be a major learning exercise. From applying a variety of software tools to project planning, resource allocation and prioritization - the project was designed to bring together all the major elements of design and implementation that I went on and on about in the lectures. Sure, a team situation, time pressures, uncertainty and lack of clarity on some aspect or the other etc all added to the realism and the challenge factor.

Hopefully, inquiring and interested folks did find learning, meaning and application in the course of executing the project.

Chalo, enough said.

Goodbye and Good luck.

Sudhir

Evaluating Phase III - prelim thoughts

OK, just started this vast exercise. Slow going so far. Some quick prelim observations -

1. Some groups haven't met even the minimum thresholds. Have seen at least 1 that did no factor and cluster analysis. Inexcusable when the thing was set in very much a clear cut and structured manner.

2. The minimum thresholds weren't met by some groups who did factor and cluster analysis - in that they factor-analyzed psychographics but left out Q21 - car attrib preferences. Some groups do factor and cluster analyses but in parallel with no connection between the 2 procs.

3. HUGE confusion and is application of fairly well-defined technical terms. Some examples: Referring to factor analysis results as 'segments' (They're not!), some group saying "respondents loaded onto the factors well...", some segmenting prior to factor analysis - makes it hard to understand what is being conveyed, or indeed if there is a point at all being conveyed.

4. Some groups have used basic descriptive analyses of primary data well indeed. Lots to see in basic analysis, basically. But the more involved analyses, such as cross-tabbing demographics, psychogr, reach info, preferences etc should have happened at the chosen target segment level.

Don't get me wrong, the sometimes extensive basic analysis I saw has merit, but contributes less to the recomms than does the same exercise carried out at the chosen recommended customer segment level.

5. Few groups so far have bothered to estimate target popn size --> hence have said nothing about relative segment sizes --> recomms are general stuff w/o focused analysis backing.

6. Some solutions so neat as to be contrived - e.g. a perfect clustering along 6 clusters with zero overlap when an additional demogr variable is introduced as a clustering variable.

7. A project objectives slide was asked for, marked mandatory. More often than not, this has been missed out.

8. Q1 went off OK more often than not. Seems, structured guidance paid off. Demand estimates have come out well. log(sales) against log(price) and other Xs easily trumps the Rsquare tables. Most groups have also bothered to show how the Xs were selected, alternative specs considered etc. Good!

From what I gather, this may well be your last brush in your MBA with model building and demand estimation.This learning should serve most of you well into your work lives.

9. Q1 aslo saw issues though - lots of confusion could also be seen, particularly in the interpretation of Q1 results. Partial regression coeffs were interpreted more as total regression coeffs. Relative importance of the variables (standardized coeffs magnitude) was often ignored.

OK. These are essentially prelim observations based on only the first few ones I saw. More later.

Ciao

Sudhir

Interesting WSJ article

Article titled:The Henry Ford of Heart Surgery shows how assembely line mass production technqiues can substantially lower healthcare costs in ops even as complex as heart surgeries.

Of course, the process innovations that enabled this miracle to come about - by leveraging massive scale economies - happens yet again in Yindia.

And while the cost declines are nothing to sneeze about, the quality uptick coming in is also compelling indeed. Chew on this a while:

- The typical US cost is $20-100k whereas the very same Op in Banglore costs $2k.
- At the same time, the avg hospital profitability is higher in India than in comparable heart surgery hosps in the states!
- And all this while Quality figs - measured in terms of mortality rates for heart patients a month after surgery is lower in India (at 1.4%) than in the States (at 1.9%).

Munch on that a while and consider the implications.

The entrepreneur doctor who embarked on this brave route has a lot to look forward to in terms of expansion and new mkts. He's setting up a hosp in the Caribbean to service the US mkt from its near abroad. On more in Iceland/Ireland or central Europe to invade the EU mkt should be on the cards. All funded by pvt equity play, by the way.

Anyway, read the article in full. Some choice excerpts:

Dr. Shetty, who entered the limelight in the early 1990s as Mother Teresa's cardiac surgeon, offers cutting-edge medical care in India at a fraction of what it costs elsewhere in the world. His flagship heart hospital charges $2,000, on average, for open-heart surgery, compared with hospitals in the U.S. that are paid between $20,000 and $100,000, depending on the complexity of the surgery.

The approach has transformed health care in India through a simple premise that works in other industries: economies of scale. By driving huge volumes, even of procedures as sophisticated, delicate and dangerous as heart surgery, Dr. Shetty has managed to drive down the cost of health care in his nation of one billion.

His model offers insights for countries worldwide that are struggling with soaring medical costs, including the U.S. as it debates major health-care overhaul.

"Japanese companies reinvented the process of making cars. That's what we're doing in health care," Dr. Shetty says. "What health care needs is process innovation, not product innovation."

At his flagship, 1,000-bed Narayana Hrudayalaya Hospital, surgeons operate at a capacity virtually unheard of in the U.S., where the average hospital has 160 beds, according to the American Hospital Association.

Narayana's 42 cardiac surgeons performed 3,174 cardiac bypass surgeries in 2008, more than double the 1,367 the Cleveland Clinic, a U.S. leader, did in the same year. His surgeons operated on 2,777 pediatric patients, more than double the 1,026 surgeries performed at Children's Hospital Boston.

Next door to Narayana, Dr. Shetty built a 1,400-bed cancer hospital and a 300-bed eye hospital, which share the same laboratories and blood bank as the heart institute. His family-owned business group, Narayana Hrudayalaya Private Ltd., reports a 7.7% profit after taxes, or slightly above the 6.9% average for a U.S. hospital, according to American Hospital Association data.

And...

The group is fueling its expansion plans through private equity, having raised $90 million last year. The money is funding four more "health cities" under construction around India. Over the next five years, Dr. Shetty's company plans to take the number of total hospital beds to 30,000 from about 3,000, which would make it by far the largest private-hospital group in India.

At that volume, he says, he would be able to cut costs significantly more by bypassing medical equipment sellers and buying directly from suppliers.

Then there are the Cayman Islands, where he plans to build and run a 2,000-bed general hospital an hour's plane ride from Miami. Procedures, both elective and necessary, will be priced at least 50% lower than what they cost in the U.S., says Dr. Shetty, who hopes to draw Americans who are uninsured or need surgery their plans don't cover.

By next year, six million Americans are expected to travel to other countries in search of affordable medical care, up from the 750,000 who did so in 2007, according to a report by Deloitte LLP. A handful of U.S. insurance plans now give people the choice to be treated in other countries.

The growth in Americans seeking treatment abroad has skyrocketed and can safely be expected to grow even further. If somebody is going to take advantage of this burgeoning trend, why not a desi do so?

Quality issues tackled:

But Jack Lewin, chief executive of the American College of Cardiology, who visited Dr. Shetty's hospital earlier this year as a guest lecturer, says Dr. Shetty has done just the opposite -- used high volumes to improve quality. For one thing, some studies show quality rises at hospitals that perform more surgeries for the simple reason that doctors are getting more experience. And at Narayana, says Dr. Lewin, the large number of patients allows individual doctors to focus on one or two specific types of cardiac surgeries.
...
Dr. Shetty's success rates appear to be as good as those of many hospitals abroad. Narayana Hrudayalaya reports a 1.4% mortality rate within 30 days of coronary artery bypass graft surgery, one of the most common procedures, compared with an average of 1.9% in the U.S. in 2008, according to data gathered by the Chicago-based Society of Thoracic Surgeons.

It isn't possible truly to compare the mortality rates, says Dr. Shetty, because he doesn't adjust his mortality rate to reflect patients' ages and other illnesses, in what is known as a risk-adjusted mortality rate. India's National Accreditation Board for Hospitals & Healthcare Providers asks hospitals to provide their mortality rates for surgery, without risk adjustment.

Dr. Lewin believes Dr. Shetty's success rates would look even better if he adjusted for risk, because his patients often lack access to even basic health care and suffer from more advanced cardiac disease when they finally come in for surgery.

On the bottom-of-the-pyramid opprtunity in India and how the high value consumer subsidizing the low income ones all leveraged using scale is an excellent example of "it happens only in Yindia". TN based Arvind Eye Hospital also famously follows the same model.

Would be great is some Northern states like HP or Uttaranchal get their act together and offer these kind of hospitals land on favorable terms. Such health cities should be there in all 4 zones of the country, at least. Right now they are pursuing manufacturing units giving tax breaks and concessional land rates without much success.

On returning to India in 1989, Dr. Shetty performed the first neonatal heart surgery in the country on a 9-day-old baby. He also confronted the reality that almost none of the patients who came to him could pay the $2,400 cost of open-heart surgery.

"When I told patients the cost, they would disappear. They literally didn't even ask about lowering the price," he says.
...
Four years ago, Dr. Shetty scrutinized his annual bill for sutures -- then $100,000 and rising by about 5% each year. He made the switch to cheaper sutures by Centennial, cutting his expenditures in half to $50,000.

"In health care you can't do one big thing and reduce the price," Dr. Shetty says. "We have to do 1,000 small things."

He says he would also like to find lower-cost versions of his priciest medical equipment. But the Chinese makers that have brought good quality, cheaper machines to market don't yet have enough local service centers to ensure regular maintenance.

And its not as if surgeons here are paid poorly at all.

Cardiac surgeons at Dr. Shetty's hospitals are paid the going rate in India, between $110,000 and $240,000 annually, depending on experience, says Viren Shetty, a director of the hospital group and one of Dr. Shetty's sons.

Dr. Shetty was paid almost $500,000 last year, according to the group's audited financial statements.

Here, too, Dr. Shetty finds additional savings on the per-patient cost. His surgeons perform two or three procedures a day, six days a week. They typically work 60 to 70 hours a week, they say. Residents work the same number of hours.

In comparison, surgeons in the U.S. typically perform one or two surgeries a day, five days a week, operating fewer than 60 hours.

Dr. Shetty says doctor fatigue isn't an issue at his hospital, and in general, his surgeons take breaks after three or four hours in surgery.

Anyway, read it all. The lower-end of the market is a HUGE opprtunity. And addressing it will drive India's innovative edge. The solutions required will have to be Yindian and not imported. The localized market gyan and mkt access will also have to be India-heavy.

Read it all before I end up copy-pasting the entire article.

Sudhir

Project submission mailbag - latest updates on top

Hi all,

Got these emails today:

Hi Sir,

Just a small doubt .. hope the title and mid-title slides are not a part of 30 slide limit.


regards
R

My reply:

OK. The title and section title slides if any shall not be part of the 30 slide limit.

Pls number the slides, in that case.

Suhdir

One more:

Hi,
I have one more query. We got an input that recommendations should be in 10 slides. But concise recommendations wont take more than 2 slides. Are you expecting us to give recommendations in 10 slides?

My reply:

No. I don’t recall anywhere suggesting recomm slide limits.

Kindly use your judgement on how long the recomm slides should be.

Sudhir

Hi Professor,

How do you want us to submit the assignment.
Should we mail the PPT to the TA or upload it via blackboard ?

Thanks,

My response:

Hi all,

Got a few queries on project submission modalities.

Kindly submit your final project deliverable PPTs only in the blackboard dropbox. Do not email as these files are large in size.

Pls name the file as yourgroup_project.pptx.

The deadline expires at noon today (Sunday).

Peer evals should be emailed to the AAs with an email with subject heading “peer evaluation” before Monday noon.

Hope that clarifies.

Another one:

Hi,
What is the maximum number of slides for the project? Will the name and presentation outline slide also be counted towards that number.

Regards,

My reply:

Yes, max number of slides is 30. A further annexure section is optional with at most 10 slides – containing supporting tables only. However, pls ensure nothing important goes into the annexures because there is no guarantee that these slides will be considered or even looked at.

Hope that clarifies.

Sudhir

A third one:

Dear Prof,

We are going ahead with our recommendation based on the internet survey sample.
...

Our take is that as our survey is primarily based on internet and “snowball” sampling, the taste and economic indicators would be definitely a lot higher than the general sample.
...
To do justice to our analysis and the data collect, we are basing our recommendation on the data collected and painstakingly cleaned and analyzed. I hope that this would be fine and the bias would be considered during the grading part of the assignment!

Thanks,

My response:

As long as your logic is sound, well-written and clearly expressed, you are fine.

Sudhir

More mailbag...Saturday

Got this email:

Dear Professor,
I have read on your blog about a clarification on cluster analysis for Q.7 and Q.21. These questions are about Car Attributes and their importance.

We have approached the project from a different direction and have done factor and cluster analysis on psychographic and demographic variables. We have found clusters using these variables and are trying to characterize these clusters in terms of the importance they attach to car attributes and other variables.

Does our approach make sense ? And is there any significant advantage of starting with the factor and cluster analysis of car attributes ?

Sincere thanks for constantly willing to help the students in this project.

Thanks and Best Regards,
S

My response:

Hi S,

IMO, its OK to take either approach, provided you have good reason to do so.

In your approach, you are segmenting the target popn along psychographics and demographics whereas in what was proposed by me, you would segment the target popn primarily along car attribute preferences.

While certainly there would be some overlap between the 2 sets of segments/clusters resulting from the 2 approaches, the logic underlying the approaches is quite different, IMHO.

Hope that clarifies.

Sudhir

2 Project related announcements

This is from an email I sent just now to the entire class:

Class,

Two separate announcements to make.

1. Office Hours - Saturday: Am in office today 11.15 am – 2.30 pm and after that over email. Am not on campus on Sunday but will be reachable over email. This is just FYI in case project related last minute queries come up. Project deadline is Sunday noon, late submissions are not acceptable.

2. Phase III Grading: Partly because grading in phases I and II may have been overly loose, that in phase III promises to be relatively tougher. Phase III carries 14% of course grade.

Pls read the blog posts relating to project to ensure that you are aware of what is being asked for. The minimum thresholds if met will yield half the grade for phase III i.e. 7%. The quality of your arguments, how well-supported they are, how well they lead to recommendations etc will decide the other 7%.

Once again, pls be aware that grading here will be tough. Every 1% above the 7% for minimum thresholds will have to be hard earned. There is a small but significant chance that the class average in phase III may not reach double digits.

I will be heavily involved in phase III grading, (unlike in Phases I and II). Admittedly there is quite some subjectivity and hence, discretion involved in phase III; and should you later choose to dispute phase III marks, pls be aware that in return, I shall be equally prepared to stand my ground and push back if I have to.

This is FYI.


Sudhir Voleti

More mailbag.... Friday

I got this email:

Hello Prof .
I came across one of you blog where you mentioned that the no of X columns for regression should be as few as possible.

While creating the X variables. I have taken 11 brands as mentioned in the DataSet construction example. This will end up in 11 dummy variables already.
So essentially will it make sense to reduce the no. of brands to a few.

Also as per the secondary data set . a lot Brand/Model dont have a corresponding sales/PRice figure. would it make sense to remove those from the regression data.


Thanks
D

My reply:

Yes, D.
Take only the top 4-5 brands as dummy variables and club all the others into an 'others' variable. Be careful to not include the others as a dummy else the regression will fail due to multicollinearity.

Some brand-models are not available in some years. Simply drop such observations from the dataset.

Sudhir

Measuring Competition

Hi all,

Someone asked me if there is any easy way to measure/quantify competition for each of the segments we find.

See, it is not necessary that the largest segment be the most attractive. The largest segment also attracts all manner of rival firms and brands. An average of the number of brand-models competing directly for customers in a particular segment (weighted perhaps by their market size) is one easy measure of how 'crowded/competitive' that segment is.

You may ultimately choose to pitch the second largest segment to the client perhaps, because it is the less crowded, perhaps. Whatever you do, ensure you have good arguments and data based evidence to back you up.

Another competition/concentration measure that comes to mind is the HH Index detailed below:

http://en.wikipedia.org/wiki/Herfindahl_index

It is the sum of the squares of marketshares and yields a measure of market power (i.e. inverse degree of competition). It is easy to compute and can be used in support of particular arguments you make.

Hope that helped.

Sudhir

A Project Checklist

Hi all,

Am putting together a structured checklist of things that should be there in your deliverable - at a minimum. Have already said all this in various posts but a summary will do good, IMO.

The deliverables should contain:

1. Title page (Mandatory) - Group Name, member names and UIDs & MKTR section.

2. Project objective (Mandatory) - Paraphrased in your words (don't life this off the scope document!), in at most a slide or 2, explain what you are trying to achieve through the project.

3. Presentation Outline (Optional) - a Contents page that outlines your presentation along sections (e.g. methodology, Q1, Q2, recommendations, etc).

4. A Methodology Section (Mandatory) - In preferably a graph or flowchart form, lay down what analysis procedures you used in what order to answer Q2 to reach the recomms based on Q2.

5. A Data section (mandatory) - explains that nature and structure of the data that were used. Be very brief but very informative - write (ii) the dimensions of the data matrices used as input to in different procedures, and (ii) the sources of data - cited sources if secondary data are used, and Question numbers in the survey questionnaire if primary data are used.

I strongly suggest using a tabular format here. Packs a lot of info into copact space. Easy to read and compare too.

6. Model Expression (Mandatory)- for Q1. Write the expression of the model used and a descriptives table of the input data, a brief explanation of the X variables used (apart from units and price), and of course, output tables along with interpretation.

7. Appendix Section (Optional): Some of the less important tables can be plugged into a separate appendix section (outside the 30 slide limit) in case you are running out of slide space. Have only the most important results tables in the main portion.

8. Recommendations (Mandatory) - crisp, clear, in simple words directed towards the client. Emphasize the usability and actionability of the recomms.

Hope that clarifies.

Sudhir

Project related mailbag - Friday

Got the following emails and also postiong my responses:

Dear Professor,

I tried to do cluster analysis for Q 21 and 7. I could not find cluster centroid data (as discussed in the lec 8) under any of the three classification methods. Can you please guide me towards proper menu option?

Regards,

My reply:

>> I tried to do cluster analysis for Q 21 and 7.

What does that mean exactly? Hope you filtered out respondents that don’t belong to the target popn before commencing on a factor analysis of Q21 and 7 before commencing cluster analysis using the factor scores.

>> I could not find cluster centroid data (as discussed in the lec 8) under any of the three classification methods.
Once you have Cluster membership, merely sort the segments by cluster number and find variable means for each cluster. That is the set of cluster centroids.

>> Can you please guide me towards proper menu option?
I would if I knew. Fact is, SPSS is new to me. I would suggest exploring the menu option yourself. Else, compute cluster centroids as mentioned above.

Sudhir

Then this one:

Dear Prof. Voleti,

Needed a clarification from you on factor and cluster analysis.

Can we use nominal or ordinal variable for factor analysis? From lecture slides, I understand that we cannot use such variable for factor analysis.

If we want to use nominal or ordinal variables for cluster analysis, but want to use factor analysis to reduce the number of variables first, how can we do this? If we cannot do this, then can we use such variables directly for cluster analysis?

Regards,

And my reply:

Nonmetric variables cannot be used for factor analysis. Nominal variables will have to be first transformed into dummy variables before they can be cluster analyzed.

My suggestion: Use only metric variables for factor and cluster analyses. Later, after clusters have been formed, use the nominal variables to characterize the segments.

Sudhir

Hope that helped.

MKTR deadline extension

Just now sent out this email:

People,

Am receiving quite a few requests for project deadline extension.

The argument against an extension is that having extended the deadline once already, it is liable to become a slippery slope.

The arguments for it are persuasive as well. Given that this was exam week with different folks having different exams, people's availability was one issue. Also, in view of the peer review, am hoping people who had slacked off till now can make up by stepping up to the plate.

But primarily, from my POV, the project *is* the single most important applied learning tool and take-away from the entire course. Also, am aware that timelines have been rather constricted throughout the project. Hence, I am inclined to take a favorable view of such as extension.

Am calling for a 48 hour extension to Sunday noon as the new deadline. Am sorry if this upsets some folks (hopefully a small minority only).

To the AAs:
This pushes our grading schedule back by about 48 hours. We should be done by Tuesday evening latest.

Regards,

Sudhir

Groups that are on track to complete early, kindly do so and submit early.

MKTR project peer evaluation clarifications

Hi all,

Yesterday, I sent in the following email alongwith a form attachment:

Class,

The group project allocates equal group marks to everybody. However, this also creates a classic moral hazard problem – a few folk might choose to shirk because they figure they might get the same grade with less work. That would be unfair to those who did put in effort.

To ensure such is not the case, I am asking you to answer a simple constant-sum question and rate the contribution of your group peers.

Pls find attached a peer evaluation form modeled on what some other courses have done.

Fill it up and email it to your respective section AAs after your group has submitted its project deliverables, latest by Sunday noon. Write “peer evaluation” in your email subject line.

Of course, I totally expect the vast majority of the class groups to have found equivalent contribution from all their members. However, in case there are any seriously dysfunctional groups out there, I want to know and ensure the folks who put in the most effort are not subsidizing those who never showed up at all.

I have received a few responses. I quote a diligent, thoughtful one below:

Dear Sudhir,

This is a very good exercise, and I believe should be done for all group assignments. But I feel, the expectation for the same needs to be set in the beginning of the project.

This is very late in the game to announce this for the reasons below -

• We have not done such evaluation for any courses so far, so we’re unfamiliar with this.
• Since the groups are really independent for each course (therefore moral hazard can easily kick in.)
• Some of our team-mates are already travelling and shall be back on Monday. Many others are tied up in the exam and not able to contribute genuinely.
• There are people who show up but can do nothing without hand-holding and it is difficult to account for it in this process.
• Lastly, it is best to announce such course evaluation mechanisms upfront.

In the current light, if you plan to use this exercise only to identify negative outliers and penalize them, then it may be ok, otherwise it may become highly controversial.

E.g. in our group even though everyone mayn’t have contributed equally intellectually/ effort wise, but everyone has performed above the minimum threshold, and I would not want them to get different marks based on peer evaluation and hurt relations, because we never set the performance criteria up-front. If we had set such a criteria up-front, then we could have done that.

Pls. see if the above logic makes sense, and accordingly let us know your thoughts.

Regards

Excellente. The mail above brings up many of the points that merit clarification.

Clarifications:
1. Because different people have had different course and exam workloads, their availability might have differed. So, kindly do *not* compare contributions in literal/ratio terms, for example in hours put in or something like that. Contributions also differ qualitatively, not just quantitatively.

2. Use more of a 3-bin ordinal or categorical scale - "Met or exceeded fair share/minimum threshold", "Did not meet minimum threshold", "never showed up at all".

3. The attempt is to ID the negative outliers rather than sow discord anywhere. I fully expect the vast majority (like 90%+) groups to assess "did fair share keeping in mind different constraints" and rate peers accordingly. Ideally, the entire class rates its group peers well and saves us (the graders and the instructor) a lot of headache.

But I wouldn't want free-riding to prevail on 30% of the course grade. Free-riders if clearly identified would tend to lose upto 5% of course grade maybe, not all 30%, precisely because, as the mail mentions above, its late in the day to announce this.

The deadline for emailing the peer reviews is extended to Monday midnight. If someone does not email their peer review, I will assume all is well and they would have allocated equal points to their peers.

Hope that clarifies.

Sudhir

Q1 analysis dataset construction

Hi class,

Just now sat with a student trying to build an analysis input dataset based on the secondary data I sent you. Pls find below an very preliminary and incomplete dataset.

Google spreadsheet showing analysis dataset construction

Point here is, this is how the Y and X columns are constructed in excel, and when complete, from here we can directly read into SPSS and run regression models.

If this example dataset is too confusing or anything, light lo and proceed as you were. If you were totally lost on how to input Q1 data for analysis, this might help to look at.

Sudhir

Improving MKTR for next year

Hi all,

I met some MKTR janta over the past week who shared their opinion on what can be done to make MKTR better for the class of 2011. Of course, it being my first prep, teaching your class was both a unique and a difficult experience. Would like to know what can be done to smooth things over for next year.

Some of the broader points that I do intend to incorporate:

1. Reduce course scope - reduce the number of topics covered and spend more time on each method. Something I do intend to do. Shall get rid of lec 10 altogether and spend that time on modeling and estimation.

2. Do away with the quizzes, have HW assignments instead based on excel datasets. For example, extracting broad descriptive summaries and cross-tabs for the beer and the ice cream datasets.

3. More cases/applied focus for each and every method/tool covered.
Now, am not a great guru on the case method. Personally, I prefer the data method - hands on immersion into data works best IMHO, especially for a course like MKTR.

Am also considering going easy on HBS style long cases and instead focussing on caselets - 1-2 page real business situations. Am sure I can come up with a dozen odd India based examples over the next 1 year.

4. Have project on much earlier on. Get phase I complete by week 2, phase II done by week 3 and then the other 2 weeks devoted to phase III.

Provide as much structure and guidance as feasible early on, not just for the project but also for the assignments and the exam.

5. Make blog more interactive (but how?). Would be great if your class occasionally drops in to give advice to the class of 2011.

6.Get software issues settled. Some things that I would like to see happen and shall push for at ASA is getting SPSS licenses from week 1 itself. Dunno how it will work, but will push for it for sure. Also, will make Stats I and II core courses pre-requisites for MKTR just so that folks coming in know revising stats would be helpful.

Well, that's about all I could think of and gather from people.

Pls feel free to add your own bit of advice about course content modification in the comments section. Thoughtful, actionable feedback would be much appreciated.

Sudhir

Q1 guidance

Hi all,

The project is pretty structured now, so I don't expect many issues at this stage. Kindly ensure your group has done the minimum steps required. The textbook examples and data are a good way to first practice implementation of any procedure on SPSS.

Spent some time a while back looking over one group's attempts with building a demand model using secondary data.

Some pointers:

1. Know the dimensions of your data matrices: How many observations/rows? How many columns in the X variables?

More # observations (or rows in your Y and X data matrices) is good. More # columns in your X matrices is not so good.

So use the secondary data sent to setup as many rows as you can get info for.

The group I looked at was planning to pick up the top 15 brand-models for one year and had collected some 20 X variables (brand dummies, size dummies and so on). Obviously the model gave nonsense results.

A good rule of thumb is that the # X rows should be at least 3x or 4x the # X columns! Else, inference is severely impaired.

2. Reduce the number of X variables. I suggested using a size proxy (like car length - available from brand/model/specs/dimesnsions at sites like carazoo.com) instead of using dummy variables for different size categories. This reduces the number of size related variables to 1 ratio-measured variable.

3. Be creative in the Xs you choose.
carazoo.com specs have plenty of info. Is mileage important? Is engine capacity (in litres) important enough to be included as an X? These are calls you have to take.

In general, its better to collect more data than less. One can always exclude variables that are there but not important. But how do you include variables that are not there?

End of the day, you want to have identified the major variables that account for a clear majority of Y variation. Look at R-square to judge fit and all.

4. Try out different specs - try using a quadratic term for some variable you believe follows a diminishing return or ideal point pattern. Try using the log-log specification. Compare model fit using adjusted R-square. And select the best fitting and most intuitive demand model.

5. See if the demand model fits into the recommendations: For a given size of car type, for a given price range that you have in mind, you can compute predicted sales. Now this may or may not help in the recommendation you make to the client. But do check and see if it helps.

Hope that helped.

Sudhir

The next big thing?

Today's TOI carries this story:

India targets 1,000mw solar power in 2013

Very interesting, IMO.

The Union Cabinet is going to consider the mission document, which requires India to generate 1,000 mw of solar power every year by 2013. A complete package has been proposed to propel the power sector into `solar reforms' that could lead to annual production of 20,000 mw by 2020 if phase I of the solar mission goes well. The country currently produces less than 5 mw every year.

In the first phase, between 2010 and 2013, the government is also proposing to generate 200 mw of off-grid solar power and cover 7 million sq metres with solar collectors.

The mission, if approved by the Cabinet, will entail three phases with the ambitious targets and financial mechanisms for the latter two phases being reviewed on the basis of performance in the first three-year phase.

By the end of the final phase in 2022, the government hopes to produce 20,000 mw of grid-based solar power, 2,000 mw of off-grid solar power and cover 20 million sq metres with collectors.

Wow.

1GW/yr? That's mighty ambitious and even a significant fraction actually realized on the ground is great progress. To compare, our current total installed capacity is around 135GW (as compared to some 400GW in China and some 1250 GW in the US).

1GW/yr would be enough to power intra-city mass-transit (metros and overground high speed rail) in out top 20 cities. Having driven around in Hyd on a weekday in non-peak hours, I see no way other than a metro-like system for our top cities. NYC has a well planned and laid out metro. Why is Hyd or Banglore any less in need of one?

It would also dent our fuel bill and import dependence on oil since IMO our big cities account for a lot of our private petrol consumption.

And it would be green and non-polluting and all. Advantages raining all around. Wish the cost poart of the equation found more mention., What is the GoI planning to invest? How much does each MW of installed solar capacity cost? Etc.

BTW, did someone say entrepreneurial opportunity? You bet!

The government also plans to do away with customs and excise duty on import of capital equipment as well as ease the duty rates for raw material and inputs.

Roof-top solar power will be promoted by providing a generation based incentive for self-use as well as putting the power on to the grid.

To enhance human resources involved in the sector, it wants to train at least 1,000 engineers to specialise in the field besides providing scholarships to 100 engineers to study abroad in the first phase of the mission. IITs and other engineering institutes will be asked to set up specialised courses to meet the industry's demand.

To promote innovative uses of solar power, a venture capital fund will be set up to promote start-ups in collaboration with institutes like IIMs. A national centre of excellence for research on solar power is proposed as well to push research and development in the emerging field.

The government has proposed a Solar Energy Authority or a Mission with an additional secretary rank official to head the executive arm.

Project mailbag for Tuesday

Shall keep updating this one post - most recent updates on the top - rather than create separate posts for interesting emails received through the day.

Dear Professor,

We have identified three distinct factors in the car data and have divided the data into 4 clusters.

In order to characterize the clusters, analyzing the data manually and eyeballing for differences between clusters is proving to be difficult, even with pivot tables.
We were wondering if there are any statistical procedures that will help us identify the differences between the four clusters.

N

My response:

Hi N,

The point of cluster analysis is segmentation. The segments should have some coherence, some identity.

A natural way to characterize clusters is thus, by finding the means of the clustering variables for each segment and seeing how different they are from that of other segments. IMO, this is what the book calls ‘cluster centroids’.

Also, the means of variables that were not used to ID the clusters – say, demographics, can also be used to profile that segment. Find the means of the segments along these variables also and see if these could be used as demographic markers to reach that segment.

As for statistical tests, a 2-sample t-test can be used to test if clusters (say) 1 and 2 differ on emphasis on Price (say). The mean, stdev and number of respondents are available for both segments, after all.

Another option could be mapping the centroids of the 2 most important variables across all clusters on a 2 dimensional map to see which clusters diverge most on those variables.

Hey, just throwing ideas out there. You don't have to spend a lot of time on this. Just try to zero in on which few variables are most actionable from the client POV - how to target the segment, how to reach the segment, segment size etc and which also can adequately differ one cluster to the next. And you are good to go.

Hope that helped.

Sudhir

Project miscell

Hi all,

With some groups, I sat and tried factor/cluster analyses. Typically Q21 responses are used. Try to filter out respondents who are part of the target population, i.e., not prospective car buyers.

Some observations -

1. Use consistent variable transformations to numeric values.

2. Safe to remove missing valued rows from the analysis because we have enough rows left over at the end (>1500).

3. Not all variables may load well in factor analysis. The communalities table gives an idea of which variables have very little (<50%) of their variance explained by the factors. Kindly consider removing such variables from the factor analysis and consider them as separate factors in their own right.

4. Cluster analysis using the factors as bases doesn't always work great under the two-step method. In one case I saw, it was saying have 15 clusters based on the AIC method. In such a case, consider adding more clustering variables. If it still doesn't work, then use k-means clustering wherein you will have to specify the # clusters desired a priori. Go for a number between 4 and 6 for convenience and handling sake.

5. Define your target population well! Some janta continue to confuse the target population with the general population. The 2 are different. The target population IMHO in this project is the prospective car buying popn in the next 12 months. Estimate the size of the target population to the nearest lakh using secondary sources, sound reasoning and assumptions where appropriate. List these clearly for the project though.

6. You don't have to use a particular method just because you can or because it was covered in the course. Use only those methods that you need to arrive at recommendations. The logic used to arrive at the recomms will be considered for grading.

7. The recomms should be actionable from the client POV. Shall elaborate on this later.

P.S.
Am happy to hear the exam wasn't bad. I didn't ask for it to be placed next to the OFD exam though. Anyway, on has to go with the hand one is dealt, I guess. Jai Ho.

Rs 3000/- gift cheque winner found

Hi all,

Just now did a random draw using the Excel function RANDBETWEEN() and found the winner of the Rs 3000/- gift cheque.

This is one Mr Dinesh K Agarwal from Banglore, contacted by Team Noida.

Would the Noida team member who contacted him kindly send me an email?.

I will ask for the winner's address details over his email.

Thanks to tireless efforts of Shri Ankur Vaish, I finally got down to getting this random draw done. Thanks, Ankur.

Sudhir

Project phase III grading guidance

People,

Met with the AAs this afternoon to decide grading criteria for phase 3 of the project.

Of the 14% course grade for phase III, Q1 will have 4% and Q2 will have 10%.

The 4% for Q1 relates to how you have modeled demand (the functional form f(.)), whether you tried alternative specifications, what Xs you have used apart from price and why, and what results you have to report.

Of the 10% for Q2, upto 5% will be for the bare minimums asked of you in this blogpost:

http://marketing-yogi.blogspot.com/2009/11/some-more-project-guidance.html

The rest (5%) will be discretionary and will depend on the quality of recommendations and how they connect with the analysis done. Your logic and reasoning will become important here.

It is strongly advised to include a flowchart - some sort of overall process map - showing your methodology leading upto the recommendations.

Will be in touch over the next few days over the blog.

Kindly feel free to meet with me in case you need more guidance anytime by appointment over the next few days.

Sudhir

P.S. Pls also consult the following blogposts for more clarity on what you are doing:

http://marketing-yogi.blogspot.com/2009/11/project-related-mailbag-resurrected.html

http://marketing-yogi.blogspot.com/2009/11/project-research-question-way-ahead.html

http://marketing-yogi.blogspot.com/2009/11/more-project-related-guidance.html

Negative marking clarification

Class,

Sorry about the confusion. Let me specify more clearly.

There is no negative marking in the MCQs! Pls feel free to attempt them without any tension.

In the non-MCQs, out of the 15% course weightage, there is some negative marking for some 3% of course weightage only. The rest does not carry negative marks. Feel free to attempt sans any second guessing.

The 3% that do carry negative marks will have it written down specifically in the quiestion that there is negative marking.

The rationale for negative marking sounds silly, but it is that, thanks to the 15% restriction, I found myself making questions which had 3 subparts - that taken together would constitute, like, 1 mark.

Now, to allocate 0.33 across the subparts would be overkill. So the scheme I decided upon is that if any 2 wrong, all points lost for these 1 mark questions.

Sorry aboiut the confusion, Its just 3 marks out of 40 for negative marking. Don't worry much about it.

Sudhir

Errata for the solution set

Hi class,

There is an erratum to the prepared solution set. Pls find my response below to the email Swati originally sent.

Hi Swati,

1) >> While calculating the predicted sales, why has the component for “light” (-2.205*0.47) been included when specification (model) 1 does not state “light” as an independent variable.

You are right. The expression for specification (1) misses Light. I simply looked up the table results for coeffs and multiplied mean value with coeffs directly.

That also explains why average size has been missed. The mean value for size should be taken is 144. I pasted that expression directly into excel, hence my answer is wrong.

The correct answers for part (j) should be:

(i) Predicted sales = = 66.773 + 4.419*0 + 1.792*1 -2.205*0.47 +0.015*144 -3.174*7.8757 +2.148*0.5 = 45.765

(ii) Predicted sales = = 66.773 + 4.419*1 + 1.792*0 -2.205*0.47 +0.015*144 -3.174*7.8757 +2.148*0.5 = 48.3921


Am surprised nobody – not a single person – caught this error so far. Thanks for pointing this out! Appreciate it. Shall fwd this email to the entire class, just to make sure, everyone gets it.

>>When calculating the predicted sales, can we take the mean value for ‘size’ given as 144 in descriptive statistics or do we need to find the hidden component of size-square also for it. The mean value of size that has been plugged in the formula has somehow got deleted from the solution key. Reverse calculation show the mean for size taken in this case to be 1641.

No, spec (1) has no quadratic term. So no need to try to find size square. Besides, mean of size square needn’t have the same value as square of mean size.

Sudhir

Am crashing now, its 10 pm

OK. No new emails in over an hour. I won't be able to answer any emails till say about 4.30-5 am tomorrow.

G'nite all. Don't forget to get some solid hours of shut-eye. Critical before things like exams, IMHO.

And you'll do pretty fine in the MKTR exam, IMHO. It is designed to test and not to harass.

Ciao.

Sudhir

Latest Exam mailbag

Added later:

This one on z-stats:

Dear Professor,

In one question from the quiz,
Q) The z value for a point is the number of standard deviations a point is away from the mean.
Indicate True or False.
True ( ) False ( )

The answer mentioned is False.
However, as far as I know, Z=(x-mu)/sigma Which means that x=mu+sigma*Z.

So shouldn't the correct answer be True?

My response:

I agree. I also believe it is TRUE.

I remember a discussion about this with the AAs. The question bank, IMO had got the answer wrong. In any case, to avoid confusion, we ended up giving everybody the benefit of doubt on this one.

Sudhir

This one on MDS

When two or more brands are close to each other on a dissimilarity map, does that mean they are more dissimilar or they are more similar? Is a similarity map the opposite of a similarity map?

My reply:

No, spatial maps *always* depict similarities only. SO 2 brands close to each other on a perceptual map are more similar to each other than to any far off brand.

Dissimilarity scores can be transformed into similarity scores and then used as a distance measure in the MDS program.

Sudhir

Got these mails:

Dear Sudhir,



It is very confusing, in terms of which Lectures to study. It would have been great, if apart from the book chapters the Lectures were specified.



Most of us study for exam based on the lecture notes, and now I am not sure which lectures to really focus on.



Ex: One response on blog says MCQ questions based on Lec 3, 6 and 7 alone. So shall I skip Lectures 1, 2, 4 and 5 and be safe for the exam?



Regards,

R

My reply:

The MCQs are based on chapters only. Within the chapters, if anything has not been covered/mentioned in the slides, ignore it.

BTW lec 5 is the modeling primer. The non-MCQ part borrows from there quite a bit, as you can see from the practice exam.

I would say, read all the slides. And only the book chapters mentioned for MCQs. And within those chapters, only that portion which was covered/mentioned in the class.

Hope that helps.

Another common question:

Dear Professor,



The exam instruction mentions that 2 A4 papers, 2-sided cheat sheets allowed.

Just to clarify, does this mean 4 different sides of the cheat sheets?

My reply:

Yes Priyom.

The instructions mention this, IMO. 4 different faces of A4 page are allowed.

Another one:

Prof Voleti,

Sorry for bothering you with this mail-I haven't been able to cull out the relevant information from your blog. The chapters for MCQs have been specified in the TA's earlier mail so there's no ambiguity on that. However, the chapters for the rest of the exam are still not clear. Could you tell me specifically which chapters are coming and which portions to leave out.

Tks and Rgds

My response:

For the non-MCQ part, feel free to ignore the chapters and go by the lecture slides alone.

Use the textbook only for reference to clarify anything in the slides.

The practice exam is very much like what the real exam will be like. YOu can see that its based more on the modeling primer and regression than on anything from the textbook directly.

Project Q1 Announcement

Class,

Please find here a google spreadsheet with data for Q1 - a common dataset that can then be used by the class.

If you already have collected data and estimated a model for Q1, then ignore this post and continue as you had planned.

The issue with secondary data for Q1 as I see it, is that there was little, if any, variation in the monthly price of cars. In fact, nothing on the right hand side of the equation (i.e., in the Xs) varies by month at all. That puts estimation in jeopardy.

So I next ventured to find where I could dig up price data from.

If such data were not available, the logical next step would be to try to find proxies for car price (for example, if I have monthly revenue data, then coupled with units data I can impute the average model price per month).

That is similar to what I ended up doing, but for annual sales data and not the monthly one originally sent.

I found annual unit sales data for cars by brand-model for the last 5 years, found annual revenue increase for the industry from LRC databases and open source data (SIAM website)and imputed the average prices by model in the dataset.

Now you have both units and prices as well as brand information and can easily get basic attribute information by model. IOW, you have all the ingredients for a simple demand estimation exercise. Q1 is now quite straightforward, IMHO. You got the dataset, you next design a simple demand model, and then all you have to do is run regression in SPSS for Q1. Take your call on which observations to keep and which to discard.

Am sure you are busy with exams and stuff. After the exams are done, kindly go through these communications again before doing anything with Q1 in the project.

Shall send another communication after the exam is done. Keep an eye on the blog now and then for more info on the project and the exam.

Sorry that the whole project thing is not as clear-cut and straightforward as it should have been. Now that both the datasets are in place - primary and secondary - data uncertainty is gone. Also, now that you have explicit guidance on what to do with each dataset, the methodological uncertainty is gone too. In case you are not clear about it, read up the project related guidance for the past few days on the blog. I expect things will settle down much more nicely now, IMHO.

Tks.

Sudhir

Teacher's POV: Why practise sets are not a good idea

Janta,

Have prepped and emailed solutions to all but 1 of the practice questions created yesterday.

Admittedly I was (and am) not keen on making either the practice questions or their solutions.

From a teacher's POV - the concerns are twofold - learning and grading. While the learning part isn't harmed by the practice sets, the grading part is now certainly more difficult.

IMHO, there is precious little leeway to make good questions that can test your mettle in MKTR_105 simply because while the 5 week course offers good breadth and scope, it doesn't always offer sufficient depth to permit the kind of testing I have in mind.

Further, it didn't help that we'd already bound ourselves into the "25% MCQ from these 6 chapters only" straitjacket. More flexibility lost.

So when I first heard that practice question sets are expected, I was like "What?! No Way!" because I didn't (and don't) think I would have much left to work with. Then the AAs prevailed that "no, practice exams are SOP @ ISB, they are totally done only" etc. I figured maybe I can tweak the questions I have sufficiently to make them quite disguised etc.

And then I was clear, "no solution sets!" Now I'd really have nothing left to work with and everybody will ace the exam!

Then this morning, the mailbox was full of student emails with this query or that. I figured it's less of a hassle to make indicative solutions that don't have to give much away. But when I actually got down to making solutions, I figured "Hey I'm already writing solutions, might as well do a good job".

So there you have it, the whole story.

My main concern now is that the class may be outside the optimal zone in terms of overall performance. I fear the class average on the project may be too low and on the exam may be too high making grading difficult in either case. Ideal situ a good variation in these critical course components - the project and the exam - that permits a natural ordering to arise.

Anyway, you have the solution sets now. They're fairly precise and representative of the exam non-MCQs.

Should anyone find any mistakes, wrong calculations, wrong logic or other such issue with either the questions or the solutions, do please let me know! Would appreciate it.

P.S.

Am away 11.30-8 pm today. Will be in office 8.30-10 pm in case people want to meet with any exam queries. I was in office quite a while yesterday but fewer groups showed up than had expressed interest in meeting over email. Anyway, over the next few days, I intend to be around as much as possible. Pls send meeting requests because they go directly to my calendar and don't rely on my admittedly shaky memory.

Sudhir

More Exam mailbag

Got this email today:

Dear Sir,

First of all thank you for applying labels to blog posts. That surely helps.

Question I have is whether we need to know the concepts which are in book but are neither in slides nor covered in any lectures ?
For ex in chapter 7 there are few concepts which were not touched upon in class like the following .

1) Controlled test market
2) Simulated test market
3) Test marketing
4) Latin square design
5) Randomized block design.


Thanks and Regards,

My Response:

No, if it hasn’t been covered in the lectures and/or the slides, simply drop it from consideration .

For terms that do show up on slides though but haven’t found much mention or discussion in the lecture, basic definitions are quite enough – at least a bird/plane level understanding of what it is would be safe to have though.

But things like non-parametric tests and all are simply way beyond the scope pf the course. I mean they would be nice and relevant to do if we were to seriously undertake a study of basic economic or business modeling but not in this course.

Hope that clarifies.

Sudhir

Project related mailbag - resurrected questions happening

Got this email just now:

Dear Professor

I do not understand why you are asking us to construct a demand function totally based on secondary data? Why did we do the survey and ask questions like will you buy in the next 12 months, what type of car will you buy etc?

We spent two whole days trying to make sense of the primary data – which seems quite random. We were able to get R2 of no more than .3 (mostly .1) with all kinds of combinations of parameters from the primary data.

I am not sure what the takeaway from the project is – if we are going to primarily rely on secondary data.
K

My response:

"Sigh."

and then

Hi K,

I’ve talked about this at length already. Let me repeat the story so far.

There are 2 questions to be answered.

Q1 relies on secondary data based demand estimation. Q2 relies on primary data based segmentation of the target population using car attribute preferences as segmenting (or clustering) variables.

Why do Q1?
Because demand modeling estimation is fundamental to your role as managers of firm resources. As students of economics and mgmt, you can’t but know how to build and estimate demand models. Regardless of what industry you go to, your understanding of the broad characteristics of the demand for that industry’s offerings is fundamental to your success.

OK, but why do Q1 in MKTR?
Because it is a natural corollary to and easily the most powerful, practical and useful application of your newly burnished model building skills. Capisch?

OK, why collect primary data and all then?
Because it’s complementary to (and *not*, as you imply, a substitute to) secondary data. There are actionable, practical things the client wants to know which simply won’t be available from secondary data analysis at an aggregate level, period. And these ‘things’ relate primarily to the distribution of attribute level preferences over the population, and an actionable segmentation based on these preferences. (Try getting that from secondary data!)

In Q2, you tease out answers to questions like “What are the attractive customer segments out there (based on attribute preferences)? What is the size of these target segments? Which is the most attractive? Which others could be similarly considered for targeting? How can these segments be reached? What should be stressed in the communications to these segments? Etc etc.”

Hope that helped.
But really, I have spoken and written about the above points multiple times. How come there is still so much confusion and the same questions resurrecting every time?

>> We spent two whole days trying to make sense of the primary data – which seems quite random. We were able to get R2 of no more than .3 (mostly .1) with all kinds of combinations of parameters from the primary data.

Welcome to the real world. These things happen. Empirical research can be frustrating indeed. If the data are totally random and senseless, then kindly build your case and say so. That would be a valid result if you can show sound evidence for the same.

P.S. to janta in general,
I am available to answer your questions. Make no mistake about that.
But I'd like to see some new questions too, once in a while.

Sudhir

Exam related mailbag

More email jussnow:

Hi Professor,

There are some portions in the prescribed chapters which you hadn’t mentioned during the class;
for e.g., most of the content under Non-Parametric Tests (K-S one-sample test, runs test, binomial test, etc.).
Some portions of these are also heavily quant-intensive.

Can we assume that we will not be tested on these concepts, either in the MCQ or non-MCQ sections?

P.S. – There are similar chunks in other chapters too, which I’m not listin here for brevity.
So is it safe to assume that if something was absolutely not mentioned in your class presentations,
then we can ignore those from the exam perspective?

Please clarify.

My reply:

Kindly safely assume that nothing that was not covered in class/mentioned on lecture slides will show up anywhere in the exam.

Tks.

Sudhir

Another email:

Dear Professor,

Just wanted to clarify a couple of things reg the exam. Do we need to know Phi Coefficient, Contingency Coeff., Cramers V, Lambda etc. Also, to what extent do we need to know about Chi Square, F-stat etc. Would we need to know the computation and calculation or just the concept

Thank you.

Regards,

My reply:

Hi S,

No. *None* of the stuff not referred to in the class or the slides is relevant for the exam - either in the MCQs section or the non-MCQ one.

Hope that clarifies.

Sudhir

Sudhir

MKTR secondary dataset emailed

Class,

This is just to ensure the email didn't get missed. The attachment contains secondary data on monthly car sales. Yippie!

Here's my merry email message:

Class,

Pls find attached the secondary data on monthly car sales units by brand-model for the months Dec-08 to Jul-09 and from Dec-07 to Jul-08.

Admittedly wasn’t easy getting the dataset, was buried somewhere in the data annexures of the Crisil industry research pages. Tks to the group that brought this problem about the inaccessibility of data secondary to my attn this morning.

The secondary data are a goldmine of info (but you already knew that). Marketshares, YoY trends, growth rates, segment-wise and brand-wise breakups – so much info at your fingertips. Kindly use this as supporting evidence for statements you do endup making in the project.

Again, given time constraints – don’t go overboard. My advice: select some top 15 or 20 brand-models and go with those.

Price info (ex-showroom) is available at carazoo.com. For price info in 07-08, one could conceivably deflate current prices by the average CPI but am not sure that’s a good idea. I would suggest you take just the Dec-08 to Jul 09 sales and proceed.

So your demand function will be this sales units (Y) = f(Price, size, brand and maybe other Xs). Take your call on what the Xs should be, if any. Would be nice to have at least 60% of variance in Y explained. Consider using previous year’s sales as an X, maybe?

Integrating primary data here will be darned tough because the integration has to happen at the brand-model level, which the primary data aren’t formatted according to. You needn’t go there unless you have to.

As for f(.), take your call. Try some basic specifications – simple, quadratic (in size? Price?),interactions, log-log and so on. Compare simple model fits and pick the best one. Try stepwise if you want to.

Chalo, I know it’s a tad late in the day, but, have fun with the data folks!

Sudhir

Update:
OK, Go easy on including lagged sales (i.e. last year's sales figure for the same month) as an X. Reason is it will suck up most of the explanatory power and render the other Xs insignificant. And its not really an independent variable, is it?

One more exam related practise question

OK, I forgot to add this question part to the Practice exam questions I had uploaded.

Add the following to the end of Q1) in the practice questions PDF.

(j) Keeping Bud=0, Miller =1 and (Price, size and Promo) at their mean values in the sample, can you predict the level of sales according to Specification (1)?

Can you predict sales for the above if Miller=0 and Budweiser=1?

That should be all. The point of the question was to have you see the predictive uses of basic demand modeling - which is just what we have done now.

Thank you.

Sudhir

Practise questions for exam put up

Hi all,

Pls find uploaded a PDF on blackboard.

It was harder than I had first estimated to create practice questions for the exam. It was tricky making questions for the final too, btw. There are only 4 questions in non-MCQ. I have put up 2 for practice that within them cover more or less all the question types you will see in the non-MCQ portion.

P.S.
Was banking on the exam along with the project creating sufficient variation in total marks so that grading becomes easier and more natural. And now, after putting up the practice questions, am not too sure. I get the feeling everybody's gonna do great and it will make the grading process tricky.

Sudhir

Another exam related email exchange

Got this email today:

Please confirm that we do not need to refer to slides for Lectures 1,2,4,5,8,9,10. Also, please confirm that for the quant portion of the paper – do we need to refer to the corresponding chapters in the book, or is knowing how to interpret results of the various tests wholly sufficient. Do we need to know the various classifications of the different tests etc. The quant part is very unnerving and overwhelming. Please provide some guidance on how to prepare for that part.

Also sir, we do appreciate your being available in your office today, but should I have doubts (as I plan to start MKTR today evening) could I meet you tomorrow late afternoon in your office. Would that be possible?

Thanks & regards
S

My reply:

Hi S,

All I can say is “Pls read up all the slides. MCQs will come from those chapters only.”

The sample questions I intend to putup today will give a broad idea of what kind of questions and difficulty level can be expected in the non-MCQ section. It needn't be overwhelming at all, even to the non-technical folks. I quite realize that no pre-requisites were asked for in the syllabus document.

Beyond that, I don’t want to promise anything.

Already, we (I and the AAs) find that a public announcement we made – that 25% will be MCQ - has come back to haunt us. I wanted more space for non-MCQs – upto 20% but was stuck with only 15%.

Henceforth, shall be more careful in my course related pronouncements. They have a knack of becoming constraints down the line.

As for availability tomorrow – tomorrow morning at most I will be available. Am off-campus in the afternoon right till dinner time. Email me if tomorrow morning works fine. Else, shall check and answer email queries every other hour or so.

Thanks.

Sudhir

Some more Project Guidance

Class,

Am better off setting expectations straight and streamlined so that everybody is well served.

Am told there are groups that are quite lost because they haven't figured out how to start also and are wasting time cleaning data.

Here is how I look at the project (and you DON"T have to agree with me. But if you have zero clue as to what to do, this may help):

The project says "client wants to know car attribute preferences in the Indian mkt".

So as part of Q2, find at least (i) if there are some underlying dimensions (factors?) beneath the attribute preference ratings that are generating the data we see.

(ii) see if these attribute preferences alongside other relevant variables can be used to segment the prospective (in the next 12 months) car buying population.

(iii) Estimate segment sizes for the estimated segments - project sample share onto target population size. For this you will have to come up with some rationale and a string of secondary sources for what the target population size would be - so many middle class households times how many intend to buy times etc. This is how, BTW, the McKinseys, BCGs, Gartners and forresters do their estimates from secondary sources, assumptions and extrapolations.

The above points (i)-(iii) should be done, at a minimum, if you are at a loss as to what is being asked of you.

(iv) Do a simple demand estimation exercise from secondary data for Q1 (at a minimum). Integrating primary data into the analysis can be tricky. You are welcome to try, if you want to. But just that you don't *have* to do it.

As for time, there's 6 of you in each group. Q1 and Q2 can proceed in parallel. Divide the workload among yourself and am sure a few hours contribution from each member should see a reasonably sound output emerging.

Hope that helps.

Sudhir

Added later:

Now before somebody emails saying "if you detail everything that is required, then there will be no variation in deliverables across groups" etc, let me preempt that and say that there is little if any chance that situation will come to pass.

There is *plenty* of flexibility and leeway within the guidelines set. What interpretation is given the factors? How are the clusters characterized? What is the rationale for arriving at target population size? What recommendations emerge about which segment to target, what communication to stress, what attributes to have in the car and why to each of these questions is quite broad in scope and will produce sufficient variation, IMHO.

The quality of analysis, of interpretation, of thinking and logic used will vary across groups, IMHO.

Sudhir

About Early Rising - on Nostalgia Lane

OT to Mktg.

Ran into some janta in the Goel Hall remarking I was an early riser. I am and not necessarily by choice. Sleep to me is a primal force of nature that I gave up trying to fight long ago. It literally overpowers the will. Something the Mrs never quite got as she's always found it somewhat difficult to fall asleep.

Even in my MBA days in Joka, Kolkata I would crash by 10 pm and rise by 4.30-5 am. And not because I wanted to. I mean, I missed a lot of the good stuff in the student hostels that only really starts at like 11 pm. The birthday bumpings, the mishti fights, the antaksharis and movie screenings, the card games and footer volley in the hostel quad, the smoke sessions and daru parties.... I slept through it all most of the time.

It also meant that I wasn't particularly popular as a project team mate - I wouldn't/couldn't show up for most project meets only. Even otherwise was more of a loner then, relatively speaking.

Anyway, the nice part was that I could get to wake up reeeally early, when most other janta would be crashing off reluctantly. Serene mornings by the Joka lakes..... Used to practise pranayam those days, have lost the good habits along with the bad ones I guess. My hostel name in those days was 'Monk' and has partly to do with my meditative attempts pre-dawn. Of course, there is also the 'Old Monk' connection.... And then when I troop down to mousees - the 24/7 roadside dhaba just outside the IIMC gate for puri-bhaji and chai, most janta would be trooping back from mousees to crash for the half day.

Almost never missed a class but almost always missed the party - admittedly not a braggable attribute. BTW, IIMC those days had no 'compulsory attendance' policy. Dunno about now, all tyhe IIMs implicitly accepted and copied the 'superiority' of the IIMA strict-regimen format. And the 'no compulsory attendance' suited everybody - separated the interesteds from the disinteresteds from the uninteresteds.

Another issue was that IIMC class profile those days (even now perhaps) was different - skewed towards younger people. Avg work-ex would be like <1 year. Close to half the batch had 0 work-ex. And what IIMC's lack of regimented structure did was it allowed people who should have known better to slack off. I have seen otherwise good people waste away because there was no structure to guide them, no pressure to meet some minimum in performance thresholds. Which is why I'm wiser now. And thankful sleep overpowers me the way it does. It helped me back at a time when I wouldn't a have known better.

Anyway, long ramble. Time to logoff. Shall be back in office today most of the time. Have lunch at Prof. Rishtee's place, so will be away in the afternoon 12.30-2 I think.

Sudhir

Emails sent recently

Class,

I will be around in my office 2118 (AC2, L1) most of today daytime - 10 am to 5 pm or so. Will break for lunch sometimes. So feel free to drop in anytime I am free. Easier that way than having to manage multiple calendar meeting requests and times.

Many folks have their mailboxes full it seems like, judging by the # of undeliverables I am getting.

Anyway, here are some of the emails sent out recently: (with some value added commentary as appropriate)

Hi all,

Am copy pasting below the Instructions sheet at the cover of the exam paper.

"ISB
MKTR END TERM EXAM
2009-10
General Instructions
• This exam is a closed book and closed notes exam. You have two hours for the exam.
• There are Two parts to this exam: Sections – A, & B. There should be 19 different numbered pages in the answer booklets (excluding Instructions Sheet & Evaluator’s Mark Sheet). Check your copy and request for a new exam booklet if you do not have 19 pages.
• Please read the ENTIRE question before starting to answer.
• MCQs are all SINGLE answer only. Choosing more than one answer will not result in any point awarded.
• You will be allowed Two A4 papers as Cheat Sheet for the exam with notes allowed on both sides, which is to be Hand Written Only. (No photocopies). No rough sheets will be provided.
• Use pencil only to mark OMR (Section A) and Pen to answer descriptive questions (Section B). There are a total of 40 points for the End-term exam. All intermediate steps must be shown in the space provided with individual questions.
• The exam is completely self-explanatory. The academic associates or faculty will not answer any questions during the exam. If you think something is not right or feels something needs to be clarified, make an assumption and proceed. Please make sure that you clearly state any assumptions you make.
• Be brief and to the point. Writing irrelevant arguments may lead to negative points.
• Please write legibly.
• You must observe the honour code at all times."


Hope this helps.

Suhdir

Ok. Shri Sreenath wrote that instruction set and am counting on his experience to have covered all the bases. If not, make common-sensical assumptions as necessary. If there are still things to clarify, let me know. Use the blog comments page preferably so that the entire class can also be freed from these doubts.

Another email I got:

Hi Professor,

Could you please share solution to the 4 quick fires on 4 khans in last session, MDS and Conjoint?

Thanks,

And had to be honest here - the answers if any to these Quickfires turned out to more involved and complicated than I had first (hurriedly) thought (or so it seemed to me judging by the response in Section D).

Hi A,

There are no real *solutions* to those quickfires that I could think of. The variation in what is acceptable based on what is assumed is simply too large. Besides, my own familiarity with conjoint and MDS is not so high that I could unambiguously judge assumptions and answers for these ones.

Sudhir

Then questions like this:

Dear Prof,
Can you please also let us know which sessions we need to cover for the exam ?

Regards,
S

To which my email to the class says:

Hi all,

Pls refer to the email below.

The MCQs will be based on those chapters. The non-MCQs can be based on anything covered in the lectures. But yes, they will emphasize the quant models we have covered.

Shall putup some sample non-MCQ questions today (Saturday), the MCQ questions will be similar to what you have seen in the quizzes so far.

P.S. shall fwd to the entire class so that there is no repetition of the same Q&A.

Sudhir

And more on a similar slant:

Dear Prof,

You have mentioned that the MCQs will come only from chapters 7, 8, 9, 11, 12 and 15. So just wanted to confirm that we only need to study slides for only lectures 3,7,6 (lectures corresponding to the mentioned chapters).

Thanks and Regards,

V

My reply:

I guess so.

The AAs have prepared the MCQs have been based on publisher’s question banks for those chapters. I have vetoed/modified MCQs that were not covered/mentioned in class slides.

Hope that clarifies.

Sudhir

Chalo, hope that takes care of most of the questions on this front.

Sudhir

End term ready

"Whew!"

Your end term exam paper is finally ready. Some 19 pages long. Somewhat long for a non-case-based exam paper, I guess.

But but but but its long only in length (tables couldn't be fit well and each takes up 60% of a page), *not* in time required to attempt.

The AAs estimate about 90-105 minutes is sufficient. You'll have 120 minutes.

Good luck and Jai Ho.

P.S.
Got some requests for sample questions for the non MCQ part. Shall put them up sometime tomorrow on blackboard or by email.

Sudhir.

Another interesting Bizweek article

Interesting. Worth a read.

How Western Business Schools Should Play Asia

Asia is where the growth is, so B-schools need to prepare their MBAs for an Asia-centric future. Here are a few things smart B-schools can do to keep up

Aha. So western B-school grads will polish up their Asia expertise. Am reminded of Shri Kishore Biyani's response to an audience question:

"Are you worried that Wal-Mart and Tesco are coming to India? How will your group compete with them?"

He replied (If I recall correctly):

"So what if they are coming? They should be worried about us because this is our backyard."

Or something to that effect.

Similarly, Asian B-school grads needn't exactly be worried IMO. Its always better to be prepared though.

As the ice hockey legend Wayne Gretzky astutely noted, one should "skate to where the puck will be, not where it is now." In management education, the puck is moving rapidly toward Asia. Western business schools that do not figure out how to play the Asia game effectively run a serious risk of ending up as regional players.

Aha. So it was Sri Wayne Gretzky who said those golden words. Its kind of the cornerstone of the formal analysis of strategic options, in many ways.

Well, where's the evidence Asia is going to dominate and all, you ask? Good, hopefully the MKTR course has taught janta the value of healthy skepticism. Don't buy anything at face value. Ask for evidence, reasoning. Look for assumptions implicitly or explicitly made. Etc etc.

Anyway, here's the good stuff (evidence) that Asia's rise is far from ordinary.

Consider a few facts: Asia today accounts for over a quarter of the world's gross domestic product and is growing at more than twice the pace of other regions. Within two decades it will constitute almost half of the world's GDP and be larger than the U.S. and Europe combined. Asia is also becoming more Asian. Intra-Asia trade presently accounts for about 54% of all cross-border trade by Asian countries. By 2025 this figure will be closer to 75%.

Importantly, too, the composition of the world's 500 largest corporations is changing rapidly. In 1995. companies headquartered in China or India accounted for only five of the 500 largest corporations in the world. By 2009 the number had grown to 44. It could easily reach 150 by 2025. Add to this the large number of corporate giants headquartered in Japan, South Korea, Southeast Asia, and West Asia, and it is not unlikely that by 2025, half of the world's 500 largest corporations could be headquartered in Asia.

"OK", you yawn. "That's old hat. Change is constant only. People will still go west for the top B-school edu and return to work in Asia maybe. I still don't buy the alternate hypothesis. The Null still rules - all this growth and all is just random variation. The pendulum tomorrow could swing away from Asia just as easily." You rationalize.

Think again....

Note also that, in the case of many non-Asian multinationals, Asia accounts for a large and growing proportion of its middle and senior managers' responsibilities. In short, the global market for management education is rapidly becoming Asia-centric.

Aaha. This lady's good. She's written a crisp, pointed no-BS column. I liked her style.

Most business schools are analogous to diversified companies with multiple product lines—related yet distinct in terms of target customers, value propositions, and value-creation and delivery systems. As such, the rise of Asia has different implications for different product lines: undergraduate, MBA, EMBA, and executive education. Since faculty research and doctoral programs have always been extremely global, these require the least amount of tinkering.

Wow. creative. Its a fresh new way to look at and thnk about the B-school.

In the case of programs that involve large numbers of students over a multi-year time frame (such as the four-year undergraduate degree program, the two-year full-time MBA, and the three-year part-time MBA), we deem it unwise to consider building an Asian footprint, except in those rare cases where a generous host government is willing to underwrite all or most of the costs. Large numbers and long duration make these programs very resource-intensive. For such programs, creating an Asian footprint would subject the Western business school to heavy startup costs, tough competition from local players, lower fee structures, and extremely severe resource constraints in terms of top-tier faculty.

Wow. See? Recommendation #1. Very simple, neat, elegant, sensible.

The right approach for these programs is to stay confined to the home campus, attract a reasonable proportion of bright Asian students, and transform the learning content and format with the goal of making graduates "Asia smart." One of the best ways to do this is to require that every student undertake immersion field visits to at least two Asian countries—preceded by intensive classroom learning about the country to be visited and capped by systematic debriefing and analysis of the lessons learned. Field visits, however, cannot be the sole strategy to transform the content of learning. Business schools also need to invest in helping their faculties become more knowledgeable about Asia so the content of regular on-campus courses becomes more global and less Western-centric.

Ok. Recos 2 and 2a served up.

You also know now where the competition for plum Asia region posts and jobs will come from. Again, before you start to feelunsure, remember, Asia is *your* backyard.

OK, but you aren't in some regular MBA. Yours is an accelarated, almopst exec style program. No?

For a growing number of business schools, the executive MBA has moved from the periphery to the center and is rapidly becoming one of the core programs. The short duration (typically 18 months) and modular format (four to six consecutive days once per month) of the EMBA program makes such offering extremely amenable to globalization.

Now for the whopper....

We predict that by 2020 the most successful EMBA programs will be those that are taught across multiple continents, include at least two Asian locations, and are the product of a formal long-term collaboration among two or three business schools, each with its home base at one of the teaching locations.

See? Projecting/extrapolating anything that far (into 2020 and all) should be treated with caution only. But still, how interesting and exciting, eh? "Go to where the puck will be!".

Now, look at her reasoning and the clean logical flow from premises to conclusion. Oh, I liked her style very much indeed. She knows what's she's talking about, quite clearly.

Our logic is straightforward. First, in order to excel, managers need deep, on-the-ground understanding of the world's major economies. In most cases such understanding will be better provided by local experts than by a faculty member based in the U.S. or Europe on a short visit to the foreign location. Second, ambitious managers also need tight interpersonal linkages with colleagues who are native to, grew up in, and are embedded in these economies. These linkages (in terms of both number and strength) are most naturally developed when the EMBA student undertakes the entire 18-month program with a cohort of peers rooted in different economies and cultures. Third, by leveraging the geographically complementary resources, relationships, and reputations of partner schools, collaborative EMBA programs are likely be both more effective and more efficient on all key dimensions: marketing, the teaching-learning process, operations, and placement. Finally, students who graduate from dual-degree programs (such as the INSEAD-Tsinghua EMBA or the Kellogg-Hong Kong UST EMBA) will have the added advantage of lifelong memberships in different and non-overlapping alumni networks across two or more continents.

It all boils down to ROI. And I have little doubt that Asia prsents a much healthier, sustainable low-cost high-growth killer-combo that is unmatched anywhere in the emerged markets.

For many of the leading Western business schools, the toughest issue regarding nondegree executive-education programs pertains to pricing. Notwithstanding the fact that companies such as Tata Motors (TTM), Infosys (INFY), Huawei, and Haier are large global corporations by any measure, it is critical to remember that they are creatures of an extremely cost-conscious environment. More often than not, a lower cost structure is also one of their primary sources of global competitive advantage. As a result, they tend to be price-conscious buyers and tough negotiators.

And

Whether or not a Tata Motors or a Huawei will be willing to pay a premium price to a Western business school—relative to a top-tier local school—will depend entirely on two factors: one, whether or not the Western business school can make a persuasive case that it offers added value; two, whether or not it brings brand cachet as a "thought leader." Faculty from a top-tier local school will almost always bring a better understanding of the local environment. The Western business school must be able to prove that the expertise of its faculty on globally important subjects will more than compensate for any weaknesses in local knowledge.

Like I say, India alone is more diverse than the entire G7 put together, what to say of the whole of Asia. Also, increasingly market opportunities rely heavily on localized conditions and peculiarities. IMHO, it will be a tough case to make for the non-local B-schools.

Partnering with a local school on an as-needed basis can be another strategy to address this weakness. Aside from making the case for added value, it is also important to note that premium pricing is impossible without a top-tier brand image. Even schools such as Harvard, Wharton, or INSEAD that enjoy the highest rankings in global surveys need to keep investing in brand-building efforts. Several of the leading local players in countries such as China, India, South Korea, and Singapore are extremely ambitious. They are also rapidly building the financial muscle to recruit top-tier PhD graduates and faculty from Western business schools to try to leapfrog into the ranks of the leading global schools.

In sum, Asia is different, diverse, and dynamic. To Western business schools it offers vast new opportunities but also serious competitive challenges. In designing their Asia strategies, business school deans need to undertake the analysis for each product line separately. They should also never forget the central maxim of business strategy: If you don't have a competitive advantage, don't compete.

Oh, Ok. It is a dual authored article. SO credit goes to both authors.

Read it all, though I believe I've covered like 80% of the article here.

Sudhir

More project mailbag

Got these emails and have noted my responses below each.

Dear Prof. Voleti,

Could you please share the Likert scale mappings for questions 23 through 26 regarding the perceptions about the current car brands?
Unlike the other questions, the mappings are not listed in the Word doc itself.

My reply:

OK. I usually take the neutral point on a likert as 0 and increment each side by 1 or -1. Those are the weights by which number of responses can be multiplied for a attribute-based brand map of sorts to be formed.

However, let me mention that the scaling and zero point are arbitrary. ANy consistent scaling will do and will produce similar brand maps like that one in lec 10, sec D. IMHO, of course.

Another mail came:

Hi Prof,

We started work on the analysis, and found that quite a few entries are incomplete. Wanted your suggestion on whether we should remove these entries. Also would be great if you could possibly suggest which entries to ignore and the basis for ignoring a field…

Thanks and Regards,

My reply:

Its your call.

Typically, missing data presents few, if any, happy options. In practice, it is best to remove rows with missing data. If there are too few rows to start with, impute the missing values. Failing that, replace missing values with column means.

IMO, the car dataset presents sufficient rows or observations to remove missing data rows.

Hope that helps.

Sudhir

Deliverables guidance

Hi all,

This is in response to a few queries I got.

The deliverable slides should contain:
(i) recommendations (and suggestions) summary you make to the client
(ii) decision flowchart depicting the methodology you have used to arrive at the recommendations
(iii) the Y=f(X) type of model equations for the analyses used in different stages of your flowchart.
(iv) the main results tables from SPSS (or other software) of the model estimation

Should you run out of space for (iv), include them in an appendix [there is *no* guarantee the appendix will be considered, though].

Shall putup more thoughts on the project and the exam as they occur to me and as I make the exam paper.

Sudhir

kal ka Times of India

Well, well.... it gets better and better.

Unlike what I'd first thought, VW's campaign in ToI was a PR blitzkreieg - 6 full pages, 4 half pages and couple of qtr pages.

All their main brands successful abroad were introduced - the Jetta, the Passat and the venerable VW Beetle.

The last full page advert also lets on that VW is building a car plant (makes sense, with a high Euro, they would have gotten next to nothing in sales of imports) in Pune that represents the largest German investment to date in India - some Euro 580 million or about 2/3rds of a billion dollars. Wow. Imagine what they must have spent on MKTR.

And the effect of MKTR research shows.

Two full pages - 2 and 7 are dedicated to the innovation theme. And page 11 is all about positioning a 'lowest cost of ownership' mantra.

Hmmm. Who would they be targetting now, eh?

Chalo, am signing off for the day.

Its the last class done. Time to celebrate. Was good fun teaching you all, most of the time anyway.

Ciao and do well.

Sudhir