[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: Re: no room to add more variables...
I am inclined to think you don't need more memory, but like Michael, I
cannot guess why your model takes so long. I can only hope you are not
using -xi- to construct dummies. On my 3GHz P4 a similarly large model
takes 20 minutes to run while I am working the CPU hard elsewhere:
xtreg y x* dy*, i(id) fe
Fixed-effects (within) regression Number of obs = 45625
Group variable (i): id Number of groups = 3803
R-sq: within = 0.8557 Obs per group: min = 1
between = 0.7146 avg = 12.0
overall = 0.8494 max = 12
F(779,41043) = 312.55
corr(u_i, Xb) = 0.0009 Prob > F = 0.0000
On 5/8/07, Michael Blasnik <email@example.com> wrote:
John Worrall <firstname.lastname@example.org> wrote:
Sent: Monday, May 07, 2007 3:45 PM
Subject: st: no room to add more variables...
> I have a panel dataset of about 3800 cases over 12 years (45625
> observations). I'm estimating fixed effects models, so I assume that
> adds unit dummies. I enter year dummies manually. I also have some
> "growth cells" that are dummies (768 in total). Altogether, my
> variables, unit dummies, year dummies, and growth cells = just shy of
> 4600 variables. My system has .99GB Ram. The data file is 39873KB. I
> have set maxvar, memory, and matsize to various sizes. I either get
> the "no room to add more variables due to width" or it takes an
> eternity to run the model--longer than I want to wait. Any ideas? Do
> I need more memory?
Short answer: Yes.
Better answer: Who knows?
You need to be much more specific and follow Statalist advice to tell us exactly
what you typed. What memory settings did you try? What fixed effects estimator
are you using? How big is the dataset after you add the "growth cells"? Are
those added variables bytes? How long is an "eternity"? Have you closed all
other running programs to maximize the physical memory available to Stata? By
the way, most (all?) of Stata's fixed effects estimators do not add unit
dummies -- the unit effects are conditioned out in the estimation process. So I
would guess that you actually have about (4600-3800) 800 variables. No way to
really tell what the best settings would be for your machine without more
* For searches and help try: