Chronic Murmuring

Why TSP? Why not STATA?

You still have a legitimate gripe -- for such a trivial mistake, the software should have given you an understandable error message.

I am a recovering TSPer - I have been clean for about five days. I have switched to STATA and while I still type genr instead of gen, I think in the long run STATA might be the right move. Perhaps I will fall off the wagon the next time I need MLE -

I would gladly, with deep, satisfying enthusiasm, use Stata. But Prof. Atkinson, after briefly considering allowing us to use Stata for our course, in the end told us to stick with TSP instead. He said that Stata cannot yet handle, very well, the things we're doing on qualitative dependent variables - specifically on taking first derivatives. I think it may also be that Dr. Atkinson knows TSP very well, and does not know Stata yet, and so would just prefer to stick to what he knows. He's main complaint of TSP is that it is all "canned". He feels like we need to know how to do this ourselves, rather than just point-and-click.

I hope (for my sake) that Atkinson just has not had enough time with STATA. Upon entering the program (same on Scott's in for any of you readers at home) we were told that starting with us the entire program would be STATA centric in order to give us a certian economies of scale when it comes to the classes. You know less time learning yet another program and more time doing econometrics.

Scott is the work your doing in TSP actual estimation or is it writing some program that demonstrates some concept he wants you to have deeper understanding of. He had us do one of the latter last semester and one brave soul did it in STATA but supposedly it was not alot of fun. The rest of us jsut used the TSP code he had in an example and mod-ed it enough to get the job done.

Anyone know anything about SEM. A friend in Marketing is taking a sequence on it and thinks its the bees knees. She thinks that I could get some milage out of it at the outskirts of Econ Town near the Poli-Sci border where I will likely live.

No, it's estimation. The first problem set wasn't difficult. It was using MLE, and used bootstrapping and Monte Carlo sampling. It gave me some trouble, but now that it's over, I think I see it wasn't so bad. It was really just the programming that was giving me the most trouble, but I think that's just the learning curve.

Things may be better for you when you do reach this course, though. Atkinson projects that soon, Stata will take over the world and all his complaints about it will be fixed. I can't really explain the specific problem he had with Stata, though - it had to do with, like I said, its inability to take the first derivatives for what I'm assuming is probit or logit. You had to instead take the numerical first derivatives and insert them in. Or something like that - honestly, I didn't know what he was talking about. Anyway, TSP can do it well, and so after a few minutes of considering STATA, he told us he wanted us using TSP instead. I honestly think it's mostly because he is very comfortable with TSP and really doesn't want to have to reinvest and acquire more human capital on another programming language. Plus, I got the sense that he had something to do with TSP's evolution over the years. He may have a certain fondness for it, since he seems to be friends with the developer.

Anyway, it'll be a lot of estimation. We're using some data of his from the NLSY on smoking. By chance, have you started working on Cornwell's? I'm going to start on it now that I finished Atkinson's but as of yet, I hadn't.

What's SEM?

I take that back - I didn't mean he wasn't sincere that it had problems. But I also think he's deeply invested in TSP, knows it very well, can help us with it very well, and therefore doesn't want us using it yet (a) until he learns it more, (b) until they fix the things he says are bad about it. But he's optimistic. He seems to expect convergence towards a single language, and thinks it'll be STATA. It appears to have improved considerably over the last couple of years, and now everyone and their mom is using it.

I won't start Corwell's project until after my Macro exam on Thursday, but it does not look too bad (famous last words).

I can devote all of Friday-Monday (with breaks for a Micro Prelim study session, LTC and Church) and all of Wednesday on it if need be. I fired up STATA and messed around with the crime data a little. It looks like its really a sweet little program. I really hated using SAS on the linux servers. STATA feels to me more like I am doing research and less like I am programing a DOS program. But my experience is pretty limited so YMMV.

Oh SEM is Structural Equation Modeling. From what my friend tells me its all based on regression somehow but it deals with "constructs" like intelligence or attitude. I am wondering if it may be useful for looking at things like....(drumroll please)... ability. That age old problem for the Labor guys looking at returns to schooling.

Just wondering if anyone (specifically econ types) has any experience with it.

There is a FAQ for it here: http://www.gsu.edu/~mkteer/sem.html

"its inability to take the first derivatives for what I'm assuming is probit or logit. You had to instead take the numerical first derivatives and insert them in."

Huh?

It seems that you guys are grad students at UGA, like I once was. When I was in Scott Atkinson's class we also learned TSP (this was in 1992/1993) - TSP hasn't improved much since then. At that time we learned SAS for Cornwell, TSP for Atkinson, RATS for Lastrappes and I also threw in LIMPDEP for good measure for Cornwell's panel data course.

In the end I used TSP, mainly because of its flexibility and reasonably intuitive programming language, but looking back on things if I had STATA on my machine over the past few years my productivity (such as it is) would not have decreased at all and may have even improved a bit.

STATA has the canned routines that are not useful when teaching PHD econometrics, although I use SAS and canned routines in my Masters level Econometrics Course. However, when you graduate and are trying to write your way out of being fired, the canned routines are less of a concern than good data and ideas.

That's a little trick the faculty probably won't think to tell you.

Jim, you had to take the derivatives by hand, find numerical approximations, then insert those numerical approximations. I think this has to do with generating a Hessian matrix, possibly. Regardless, the gist of the complaint was that TSP was currently programmed to handle this, and Stata currently was not. It's still all new to me, so I don't quite understand what the problem it, but it's something along those lines, although I may have garbled it.

Craig, that's right. University of Georgia PhD student. I visited your website and noted the work you've been doing. Your research in economics of sports is interesting. That area seems like a ripe one.

Lastrapes still recommends RATS for time-series, and Atkinson still recommends TSP for his classes. Cornwell has made the full switch to Stata. TSP is, thankfully, as you said, intuitive, which I appreciate.

Just out of curiousity, how long did it take you to graduate? Precisely when did you settle on your dissertation topic? I ask because, I can't quite gauge how long to expect the process to take. I've an idea that I'm going to pursue a ways, but I'm always wondering how I'll know when the topic is the right topic, or if I should scrap it and keep searching.

Scott,

No, that's better.