Friday, March 14, 2014

Seminars by the Number - Redux

In my second post on this blog, just over three years ago, I took a shot at seminars - economics seminars in particular. There's nothing there that I want to retract. I still remain bemused by the duration of economics seminars; the time that's wasted on details rather than "the big picture"; and the proportion of the allotted time that's taken up with audience "participation".

This being the case, I thought I'd update my earlier suggestion for streamlining these seminars. The focus is on seminars of an econometric nature - very occasionally we actually do have such events in my department.

Here's what I suggested in that earlier post:

"......all members of the audience would use an agreed-upon numeric code when asking questions. There is a fairly standard menu of questions that arise in seminars - the hardy perennials that go with the territory – and these could be codified. For example, the inevitable question ‘Did you test your data for unit roots’ could be assigned the number 4. Of course, even more efficiency could be introduced by encouraging the speaker to use a standard numeric code as well. On this side of the table matters would be quite simple: 1 = ‘Yes’; 2 = ‘No’; 3 = ‘That’s a good question’; 4 = ‘I tried it, but it didn’t affect the results’; and 5 = ‘It seemed like a good idea at the time’. Keen-eyed readers will notice that there is no code for, ‘I don’t know’.
A suitable code for audience members requires a little more thought. When this idea catches on (as it surely will) we could consider having a specific coding system for time-series seminars; a slightly different one for seminars dealing with cross-section data issues; and so on. The generic speaker’s code suggested above would suffice in all cases. For the meantime, though, here is a suggested basic coding to get things going:
Code Number          Question/Comment (Disruption)
1                                  Why didn’t you use a Logit/Probit model instead of OLS?
2                                  Did you test for weak instruments?
3                                  Are those standard errors ‘heteroskedasticity-consistent’?
4                                  Did you test your data for unit roots?
5                                  Are your data seasonally adjusted, or not?
                                    (This one is tricky – see the dialogue below.)
6                                  Couldn’t you simply bootstrap the confidence intervals?
7                                  Do you think that the number of replications that you used
                                    in your Monte Carlo experiment is really enough?
8                                  What about structural breaks in your
                                    data?                      
9                                  This would be much more easily accomplished if you took
                                    a Bayesian approach.
10                                I don’t understand why you are ignoring the obvious
                                    pre-test issues that arise here.
11                                Does this test have good power in the present context?
12                                Why are you trying to estimate a model that is obviously
                                    under-identified?
                                    (One of my personal favourites.)
As a refinement, and to avoid disturbing those members of the audience who really do know about opportunity cost, and have fallen asleep mid-way through the seminar, a verbal exchange of numbers could be replaced by having participants silently raise little ‘paddles’ with the desired numbers on them. I have in mind the sort of thing that is used at public auctions to identify bidders. 
To illustrate the obvious merits of the proposed system, here is a real-life example of an exchange of views at an econometrics seminar:
Dr. A:                          1?  
Speaker:                      4.
Dr. A:                          3?
Speaker:                      1.
Dr. Z:                           8?
Speaker:                      3. Yes,......, 3.
Dr. McQ:                     5?
Speaker:                      2 (and therefore 1).
Drs. D, A, E:                11, 6, 6      
Speaker:                      1, 3, 4.
Dr. A:                           But 12?
Speaker:                       5.
etc.
Now there's a lively seminar!"
O.K., so now for a few updates to all of this. First, let's broaden the range of the speaker's responses just a little, by adding:

6.  'I'm making a note of that' (which should be interpreted as 'I have no idea what you mean and I really don't care, but I'm not going to admit it').
7.  'Why don't we discuss that later?' (which means 'yikes! I was hoping no one would notice that!').
8.  'I'll be getting to that shortly' (which means 'I don't like that question, and I'm hoping that you won;'t notice if I never get around to it later in the talk').
9.   'I have another paper in progress that addresses exactly that point' (which means 'good suggestion - I'm going to write a paper about that').
10. 'It all generalizes without any problems' (which means, 'I hope it generalizes - I couldn't figure out how to do it').

Next, let's extend the range of inevitable questions/comments:

Code Number          Question/Comment (Disruption)


13                              Actually, I solved that problem in my '95 paper.
14                              Why are you using a 5% significance level when you've got 1.5 million observations?
15                              How sensitive are your results to your choice of prior?
16                              Just for clarification - is that a lower case "a" or an upper case "A"?
17                              Just a clarifying question if I may - Now, it seems to me that if you... blah, blah, .....
18                              Just for clarification - shouldn't that second negative sign be a positive sign?
19                              Sorry - are you really stupid? could you tell us again why you wrote this paper?
20                              Sorry to interrupt again, but......... (insert any number from 1 to 19 here)

Feel free to adopt this system at the seminar series nearest to you!

© 2014, David E. Giles

2 comments:

  1. Better yet would be to start requiring appendices in advance documenting every econometric procedure the researchers ever ran. Then only one answer to questions of the type "Did you do X?" is required: "Check the appendix."

    To make the appendix easily searchable, they should be required to label common issues like unit root tests with a unique, standardized code.

    ReplyDelete