Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | "Ploutz-Snyder, Robert (JSC-SK)[USRA]" <robert.ploutz-snyder-1@nasa.gov> |
To | "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu> |
Subject | RE: st: AW: changes in -xi: xtmixed- command from Stata 10 to 11 |
Date | Thu, 22 Apr 2010 08:42:50 -0500 |
Same for me, running 10.1 on a 64bit win machine, versus 11.0 on the same machine.. both up to date with patches/upgrades. Output: STATA 10.1: webuse melanoma, clear (Skin cancer (melanoma) data) . gen uv2 = uv^2 . xtmepoisson deaths uv uv2, exposure(expected) || nation: || region: Refining starting values: Iteration 0: log likelihood = -1169.4088 (not concave) Iteration 1: log likelihood = -1156.8957 (not concave) Iteration 2: log likelihood = -1102.0609 Performing gradient-based optimization: Iteration 0: log likelihood = -1102.0609 Iteration 1: log likelihood = -1096.5001 Iteration 2: log likelihood = -1089.4293 Iteration 3: log likelihood = -1089.4112 Iteration 4: log likelihood = -1089.411 Iteration 5: log likelihood = -1089.411 Mixed-effects Poisson regression Number of obs = 354 -------------------------------------------------------------------------- | No. of Observations per Group Integration Group Variable | Groups Minimum Average Maximum Points ----------------+--------------------------------------------------------- nation | 9 3 39.3 95 7 region | 78 1 4.5 13 7 -------------------------------------------------------------------------- Wald chi2(2) = 25.69 Log likelihood = -1089.411 Prob > chi2 = 0.0000 ------------------------------------------------------------------------------ deaths | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- uv | .0056975 .0137931 0.41 0.680 -.0213364 .0327314 uv2 | -.0058374 .001388 -4.21 0.000 -.0085579 -.0031169 _cons | .1289976 .1581122 0.82 0.415 -.1808966 .4388918 expected | (exposure) ------------------------------------------------------------------------------ ------------------------------------------------------------------------------ Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval] -----------------------------+------------------------------------------------ nation: Identity | sd(_cons) | .4290366 .1101666 .2593735 .7096809 -----------------------------+------------------------------------------------ region: Identity | sd(_cons) | .1956382 .0224569 .1562233 .2449974 ------------------------------------------------------------------------------ LR test vs. Poisson regression: chi2(2) = 1267.13 Prob > chi2 = 0.0000 Note: LR test is conservative and provided only for reference. . VERSUS Stata 11.0: . gen uv2 = uv^2 . xtmepoisson deaths uv c.uv#c.uv, exposure(expected) || nation:|| region: Refining starting values: Iteration 0: log likelihood = -1169.4088 (not concave) Iteration 1: log likelihood = -1156.8957 (not concave) Iteration 2: log likelihood = -1101.6231 Performing gradient-based optimization: Iteration 0: log likelihood = -1101.6231 Iteration 1: log likelihood = -1097.2443 Iteration 2: log likelihood = -1089.4042 Iteration 3: log likelihood = -1089.4016 (not concave) Iteration 4: log likelihood = -1089.4007 numerical derivatives are approximate flat or discontinuous region encountered Iteration 5: log likelihood = -1089.4001 (backed up) numerical derivatives are approximate flat or discontinuous region encountered Iteration 6: log likelihood = -1089.3999 (not concave) numerical derivatives are approximate flat or discontinuous region encountered Iteration 7: log likelihood = -1089.3998 (not concave) numerical derivatives are approximate flat or discontinuous region encountered Iteration 8: log likelihood = -1089.3998 (not concave) numerical derivatives are approximate flat or discontinuous region encountered Iteration 9: log likelihood = -1089.3998 (not concave) numerical derivatives are approximate flat or discontinuous region encountered Iteration 10: log likelihood = -1089.3998 (not concave) numerical derivatives are approximate flat or discontinuous region encountered Iteration 11: log likelihood = -1089.3998 (not concave) numerical derivatives are approximate flat or discontinuous region encountered Iteration 12: log likelihood = -1089.3998 (not concave) numerical derivatives are approximate flat or discontinuous region encountered Iteration 13: log likelihood = -1089.3998 (not concave) numerical derivatives are approximate flat or discontinuous region encountered Iteration 14: log likelihood = -1089.3998 (not concave) numerical derivatives are approximate flat or discontinuous region encountered Iteration 15: log likelihood = -1089.3998 (not concave) numerical derivatives are approximate flat or discontinuous region encountered Iteration 16: log likelihood = -1089.3998 (not concave) --Break-- -----Original Message----- From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Scott Merryman Sent: Thursday, April 22, 2010 8:04 AM To: statalist@hsphsun2.harvard.edu Subject: Re: st: AW: changes in -xi: xtmixed- command from Stata 10 to 11.0 On Thu, Apr 22, 2010 at 7:38 AM, Martin Weiss <martin.weiss1@gmx.de> wrote: > > <> > > A similar thing happens when I run this example in Stata 10.1 MP: > > > ************* > webuse melanoma, clear > gen uv2 = uv^2 > xtmepoisson deaths uv uv2, exposure(expected) || nation: || region: > ************* > > versus this in Stata 11 MP: > > ************* > webuse melanoma, clear > xtmepoisson deaths uv c.uv#c.uv, exposure(expected) || nation:|| region: > ************* > > The Stata 11 incarnation constantly reports "flat or discontinuous region > encountered", while the 10.1 version converges after a couple of steps: That's odd - I get convergence with Stata 11 MP. . webuse melanoma, clear (Skin cancer (melanoma) data) . xtmepoisson deaths uv c.uv#c.uv, exposure(expected) || nation:|| region: Refining starting values: Iteration 0: log likelihood = -1169.4088 (not concave) Iteration 1: log likelihood = -1156.8957 (not concave) Iteration 2: log likelihood = -1101.9749 Performing gradient-based optimization: Iteration 0: log likelihood = -1101.9749 Iteration 1: log likelihood = -1097.2043 Iteration 2: log likelihood = -1089.5644 Iteration 3: log likelihood = -1089.4115 Iteration 4: log likelihood = -1089.411 Iteration 5: log likelihood = -1089.411 Scott * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/ * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/