Some Admin Announcements

Update: Got this email with Qs from SK. My response is also appended in it.

Hi S,

My answers to your Qs are prefaced with '>>'

1. In Q1, you have ordered even insignificant X variables. Can they be ordered if they are insignificant?

>>You can't infer anything from insignificant variables. Still, an ordering by 'importance' was asked for, t-stats used as proxy (ideally standardized coeffieicnts should be used but in their absence t-stats will do too even though the correlation isn't perfect) and hence an ordering was given. Yes, I wouldn't read too much into ordering or impact of non-significant variables.

2. In Q4, we see that LnSize is insignificant. How then can we conclude that if average size could be raised by one ounce, average sales would go up by close to 2%.

>>Tricky one, this. LnSize is insignificant at the 90% level but signif at the 85% confidence level. True that we take an arbitrary, 'safe' cut-off of 95%, however when calculating fitted values or taking marginal effects or making predictions etc, the variables below the 95% threshold don't go away, they continue to count. However, that said, I see the point of your objection. Such a Q, if it were to occur in the exam would be redundant - as in both TRUE and FALSE would get marks on this one.

3. In Q5, I know that adjusted R-square is not a measure of variation explained in the sample. On the net, some people interpret adjusted R-square as the percentage of variation explained in the population and not the sample. If you look at this question in that light, the question seems ambiguous. Please clarify.

>>That would not be the right interpretation of Adj-R-square. The simple or multiple R-square is the answer to that Q, period.

4. In Q6, again “Promo” and “Light” are both insignificant. How can we conclude anything?

>>Agreed. Again, a redundant Q - both TRUE and FALSE will count for this one. In the exam there's a third category called 'CAN'T SAY' and it would be best suited to Qs dealing with insignificant variables.

Sudhir

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

Update: I have sent an email to you all with the attachments as LMS is playing games and both the AAs are on leave today.

Received this email from DB:

Dear Sir,
Pl clarify one doubt, in the following table, one asked about “Importance” we have to just refer to “t”values and not Betas?

My response:

Hi D,

You're right in that ideally, we would look at 'standardized beta values'. This means we have standardized the X variables (so no scaling effects are there) and the betas we get after that are comparable with one another.

However, in the absence of standardized beta values (which may need to be computed separately), we usually look at t-stats which are often very highly correlated with the standardized betas. Its not perfect but it does the job. In this case, with SPSS outpout, since you have standardized betas given to you, you can go with that, no problem.

So yes, you could put LnPrice before Miller in the example given and that would be correct. IN the exam however, if standardized betas are not given, pls rely on t-stats.

Hope that clarifies.

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

Hi all,

Based on what the AAs have told me, a 'solution' to the sample end-term is in order. Hence, I have uploaded a "solution to HW 8" document into the session 8 folder. If it is not visible for any reason, pls email me and let me know.

Some folks have emailed saying they might need hard copies of lecture slides for sessions they may have missed. If you email me the sessions you have missed, I will keep printouts ready with my secretary Ms Sireesha. You can pick the same up during working hours anytime.

I'm teaching the PhD class (FPM students) between 21 and 23-Nov and am off to Mohali for term 6 on 24-Nov. However, if you want to meet for any reason, pls email fro an appointment and I'll ensure I'm around at the appointed time.

Was a pleasure teaching you. Class size does make a difference - smaller classes are inherently nicer to interact with. I've come to know more students by name in this year than I have in all the 3 previous years combined. Thanks!

Sudhir