Stata 15 help for whatsnew9

Additions made to Stata during version 9

This file records the additions and fixes made to Stata during the 9.0, 9.1, and 9.2 releases:

+---------------------------------------------------------------+ | help file contents years | |---------------------------------------------------------------| | whatsnew Stata 15.0 and 15.1 2017 to present | | whatsnew14to15 Stata 15.0 new release 2017 | | whatsnew14 Stata 14.0, 14.1, and 14.2 2015 to 2017 | | whatsnew13to14 Stata 14.0 new release 2015 | | whatsnew13 Stata 13.0 and 13.1 2013 to 2015 | | whatsnew12to13 Stata 13.0 new release 2013 | | whatsnew12 Stata 12.0 and 12.1 2011 to 2013 | | whatsnew11to12 Stata 12.0 new release 2011 | | whatsnew11 Stata 11.0, 11.1, and 11.2 2009 to 2011 | | whatsnew10to11 Stata 11.0 new release 2009 | | whatsnew10 Stata 10.0 and 10.1 2007 to 2009 | | whatsnew9to10 Stata 10.0 new release 2007 | | this file Stata 9.0, 9.1, and 9.2 2005 to 2007 | | whatsnew8to9 Stata 9.0 new release 2005 | | whatsnew8 Stata 8.0, 8.1, and 8.2 2003 to 2005 | | whatsnew7to8 Stata 8.0 new release 2003 | | whatsnew7 Stata 7.0 2001 to 2002 | | whatsnew6to7 Stata 7.0 new release 2000 | | whatsnew6 Stata 6.0 1999 to 2000 | +---------------------------------------------------------------+

Most recent changes are listed first.

-------- more recent updates --------------------------------------------------

See whatsnew9to10.

-------- update 20jul2007 -----------------------------------------------------

Ado-files

1. adoupdate did not set the saved result r(pkglist) correctly in all cases. adoupdate should save in r(pkglist) a space separated list of package names that need updating (option update not specified) or that were updated (option update specified). In fact, adoupdate did this except when a package list was specified, and then it returned the package list specified. This has been fixed.

2. _diparm reported a conformability error when one of the equations in a multiple equation specification contained predictors. This has been fixed.

3. fcast produced a syntax error when applied to a VAR with only one lag. This has been fixed.

4. irf cgraph reported an unbalanced-quotes error when there was a space at the end of a quoted subtitle. This has been fixed.

5. irf cgraph produced a syntax error when level() was specified for one of the graphs to be combined. This has been fixed.

6. nestreg, when used with stcox or streg, incorrectly bound standalone variables into blocks with multiple variables when the first block was bound in parentheses. For example,

. nestreg: stcox (x1 x2) x3 x4 x5

was incorrectly interpreted as

. nestreg: stcox (x1 x2) (x3 x4 x5)

This has been fixed.

7. nlcom and testnl returned an error when used after svy jackknife with replicate weights. This has been fixed.

8. nlogit failed to omit observations with missing values in the dependent variable or the level alternative variables. This has been fixed.

9. nlogittree now allows if and in qualifiers.

10. predict was unable to generate scores after heckprobit when the fitted model did not have a dependent variable specified in the selection equation. This has been fixed.

11. stepwise, when used with stcox or streg, incorrectly bound standalone variables into terms with multiple variables when the first term was bound in parentheses. For example,

. stepwise, pr(.4): stcox (x1 x2) x3 x4 x5

was incorrectly interpreted as

. stepwise, pr(.4): stcox (x1 x2) (x3 x4 x5)

This has been fixed.

12. sts graph with the hazard option would return the error message "variable __ub not found" when variable abbreviations were turned off. This has been fixed.

13. suest complained about missing score values when used after heckman and the dependent variable contained missing values. This has been fixed.

14. sunflower failed to draw the final segment of a hexagon's outline. This was only noticeable on Mac computers and has been fixed.

15. svy: tabulate with the subpop() option miscounted omitted strata when the row or column variable contained missing values for all of the subpopulation observations. This has been fixed.

16. testnl reported incorrect standard errors when the estimated parameter values were close to the boundary of the support for the functions specified in the user defined constraints. For example,

. testnl sqrt(_b[x1]^2 - _b[x2])

resulted in incorrect standard errors if the squared value of _b[x1] was very close to the value of _b[x2], leading to the square root of a negative number. This has been fixed.

17. The vce(bootstrap) and vce(jackknife) options were not aware of the noconstant option when presented with indicator variables that were collinear with the constant. This resulted in an incorrectly dropped indicator variable. This has been fixed.

18. xi with the noomit option failed to keep track of one of its generated indicator variables. On subsequent xi calls with the noomit option, xi failed to remove one of the previously generated indicator variables. This has been fixed.

19. xi with the noomit option generated variables without the default _I prefix when the prefix() option was not specified. This has been fixed.

Stata executable, all platforms

20. The Data Editor could not display more than 20 variables at a time. This limitation affected users with very high-resolution displays. This has been fixed.

21. fcast after var produced an error when the VAR contained only one endogenous variable. This has been fixed.

22. Using a Stata matrix function that returns a scalar within an if clause of a regress command caused e(b) and e(V) to be saved incorrectly, thereby causing postestimation commands not to work after regress. This has been fixed.

23. Inverse probability functions invbinomial(), invFtail(), invibeta(), invnFtail(), and invnibeta() could enter an endless loop when the unique solution could not be found due to the limit of machine precision. Now, as with other inverse functions, they return missing in such cases. Also nibeta() now returns 1 if x > 1 and noncentrality parameter L < 1.

24. infile (fixed format) now accepts a %[#]S infmt. %[#]S is similar to %[#]s, but %[#]S preserves leading and trailing spaces; %[#]s does not.

25. infile (fixed format) treated the ASCII character 255 (hexidecimal 0xFF) as an end of file marker and would stop reading the file if it encountered it. This character is now treated just as any other ASCII character.

26. Mata miscompiled the expression y[p] = y. The notable feature of this expression is that y[p] appears on the left-hand side of the expression and y, the same variable, appears on the right. This has been fixed.

27. Mata's bufput() function ignored the offset argument when it copied elements into the buffer. This has been fixed.

28. odbc load incorrectly imported real data as missing values for values that were within 27 of the upper boundary of integer data types. This has been fixed.

29. Returned values from r-class programs could mistakenly include r() values from the last command the ado-file ran under certain rare circumstances. This has been fixed.

30. sreturn list mistakenly listed matrices returned in r(). This has been fixed.

31. svy, when used with regression commands such as regress, incorrectly reported missing standard errors when the subpop() option excluded one or more strata with a single sampling unit. Here, svy incorrectly counted the out-of-subpopulation singleton strata even though they were dropped from the estimation sample. This has been fixed.

32. unabcmd now recognizes the abbreviations for the following formerly built-in commands: cnreg, logit, mlogit, ologit, oprobit, probit, regress, and tobit.

Stata executable, Windows

33. In certain circumstances double buffering could prevent some Stata windows from rendering correctly when used with multiple monitors. This has been fixed.

34. In the Do-file Editor, the Search -- Find Next menu item and its corresponding F3 keyboard shortcut were not working. This has been fixed.

35. The Stata Data Editor displayed an empty system menu under Windows Vista. This has been fixed.

36. Two scroll bars sometimes appeared on the Variables window under Windows Vista. This has been fixed.

Stata executable, Mac

37. The Output Settings dialog is no longer displayed when translating a SMCL file to another format.

38. The scroll bar thumb would not scroll past the 32,767th line. This has been fixed.

39. Clicking multiple times on the Apply button in the Variable Properties window of the Data Editor could crash Stata on early versions of Mac OS X. This has been fixed.

40. Graphs copied to the Clipboard could possibly contain two filled rectangles as the background instead of just one. This has been fixed.

Stata executable, Unix

41. Remotely launching Stata (for example, ssh mymachine xstata-se) would cause Stata to believe it was running in batch mode. This has been fixed.

-------- update 22may2007 -----------------------------------------------------

Ado-files

1. binreg reported the error "option rrr not allowed" when the rr option was used with vce(bootstrap) or vce(jackknife). This has been fixed.

2. ivprobit did not apply if or in restrictions specified by the user when checking for perfect predictors. Also ivprobit would exit with an error if the two-step estimator was used and the reduced-form model had perfect predictors. These issues have been fixed.

3. jackknife reported an error when weights were specified with commands that have multiple equation specifications using the parenthetical notation. This has been fixed.

4. A note has been added to the ksmirnov output to warn users when ties are found in the variable whose distribution is being tested.

5. nlogit with vce(robust) neglected to cluster on the group() variable when computing standard errors. This has been fixed.

nlogit with robust did not have this problem.

6. roccomp could produce different chi-squared tests when monotonic transformations were applied to the ratings. This has been fixed.

7. rreg reported the error "option noheader not allowed" when used with the bootstrap prefix. This has been fixed.

8. svyintreg reported the error "initial values not feasible" for constant-only models. This has been fixed. svy: intreg did not have this problem.

9. svy: tabulate did not always post the correct e(sample) when one or more strata were omitted because they contained no subpopulation members; some observations from the omitted strata may have been left in the estimation sample identified by e(sample) when they should not have been. This has been fixed.

10. svy: tabulate failed to report on omitted strata when options ci or se were not specified. This has been fixed.

11. xtgee reported the error "option level() incorrectly specified" when used with vce(jackknife). This has been fixed.

12. xtlogit, when reporting the number and size of groups, ignored groups with more than 32,740 observations. This has been fixed.

13. xtreg exited with an error message that the "independent variables are collinear with the panel variable" if the scale of the panel variable was too large. This has been fixed.

-------- update 07mar2007 -----------------------------------------------------

Ado-files

1. ctset with the option clear did not clear the ct setting. This has been fixed.

2. estimates table reported a conformability error when used after mlogit when one or more equations contained a space. This has been fixed.

3. icd9 and icd9p were updated to use codes through the V24 update released on October 1, 2006. The descriptions were also updated so that the description of invalid codes that were valid in the past end in a hash mark (#).

4. The matrix input dialog did not create a symmetric matrix correctly when negative values were specified. This has been fixed.

-------- update 16feb2007 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 7(1).

2. cttost exited with an error message if the total number of failures and censored observations in a certain period was equal to the total number of individuals in the study, and if the number of failures was not smaller than the number of individuals in the study two periods before. Now, in this setting, cttost produces the correct output.

3. svy jackknife reported an incorrect subpopulation size and number of subpopulation observations when the prefixed estimation command dropped observations due to missing values. This has been fixed.

-------- update 06feb2007 -----------------------------------------------------

Ado-files

1. lookfor can now search for phrases enclosed in double quotes.

2. predict, scores after ologit and oprobit incorrectly returned missing values for models having one cutpoint. The score variable for the cutpoint contained missing values in observations where the dependent variable equals the second ordered value. This has been fixed.

3. svy: ologit and svy: oprobit produced incorrect standard errors when used to fit models where the dependent variable had only two values. This was caused by a problem with predicted equation-scores for these models, which has been fixed.

Stata executable, Windows

4. (Windows Vista only) Pressing Ctrl-C to copy text from the Command window caused Stata to crash. This has been fixed.

5. Printing the Do-file Editor from the File-Print menu or the main Stata toolbar would print spurious pages. This has been fixed.

-------- update 24jan2007 -----------------------------------------------------

Ado-files

1. hausman failed to correctly recognize equation names in mlogit's estimation results when one or more equation names contained a space. This has been fixed.

2. hausman now exits with an error message if either model was fitted with cluster() or with pweights. The performance of the test can be forced by using force.

3. lrtest exited with an error message if the dependent variable of one of the models being compared contained time-series operators and the restricted model was specified before the unrestricted model in the call to lrtest. This has been fixed.

4. statsby did not respect the level() option. This has been fixed.

5. svy: tabulate failed to report standard errors or confidence intervals if they were requested when the BRR variance estimation method was used. This has been fixed.

6. xtpoisson, re normal applied the noskip option by default; this has been fixed.

-------- update 12jan2007 -----------------------------------------------------

Ado-files

1. adjust has been modified to perform intermediate computations by using double precision. The resulting gain in precision is usually evident in the fourth or fifth decimal place.

2. glm's Newey-West standard-error calculation, when used with the Anderson kernel function, failed to account for this kernel's being nonstandard. The Anderson kernel is nonstandard because (1) it permits a noninteger lag, that is, a bandwidth and (2) the kernel weights are nonzero outside the bandwidth/lag range. glm now accounts for both of these nonstandard behaviors.

3. gprobit computed incorrect weights for the GLS stage of the computations. As a result, the coefficient estimates were inefficient. This has been fixed.

4. graph now better supports variable labels that are by default used to label axes and legend keys. Labels with unbalanced parentheses and brackets are now supported, as well as labels with separately double-quoted strings. As with text appearing in graph options, each double-quoted section of text now appears on its own line in the title or label. Results for options xtitle(), ytitle(), legend(label()), and legend(order()) are similarly improved.

5. pac exited with an error when used with the option yw if varabbrev was set to off. This has been fixed.

6. stcox and streg now exit with an error when option vce(bootstrap) or vce(jackknife) is specified when weights were stset. Prior to this update, weights corresponding to the resampled observations were used as is for computing the replicates.

bootstrap does not allow weights.

jackknife allows weights but currently cannot handle them correctly for stcox and streg.

7. suest allowed survey estimation results that used poststratification to be mixed with estimation results that did not. This has been fixed.

suest can be used only with survey estimation results that are consistent with the poststratification information in the current svysettings.

suest will now verify that the poststratification information is consistent between svyset and the specified estimation results.

suest will exit with an error message when non-svy estimation results are specified and poststratification variables have been svyset.

8. svy, in the 21nov2006 update, was changed to use encode on string-valued stratum and sampling unit variables. This change restricted the number of sampling units to no more than 65,536 if the variable that identifies sampling units was a string variable. This upper-limit restriction has been removed.

To decrease the amount of overhead, consider replacing string variables that identify the strata and sampling units with encoded numeric variables.

9. svy: tabulate failed to drop categories of the row and column variables that were exclusively defined in omitted strata. This has been fixed.

10. xtrc exited with an error when used with more than 1,000 panels. This has been fixed.

Stata executable, all platforms

11. in now accepts F as a synonym for f as shorthand for the beginning of the in range. L has been added as a synonym for l as shorthand for the end of the in range.

12. Mata has the following new features and fixes:

A. New function _diag() replaces the principal diagonal of a matrix with a specified vector or scalar.

B. New (undocumented) functions trace_AVBV() and trace_ABBAV() compute the trace of special matrices used for spatial statistical calculations.

C. Appending rows to a matrix ([M-2] op_join) in Mata is now faster.

D. Matrix multiplication, A*B, is now faster when A contains zeros.

E. cross(X, Z) is now faster when X contains zeros.

F. select(X, v) exited with a conformability error when X was a row vector and v was a scalar; this has been fixed.

G. sizeof() could not be used; including sizeof() in code resulted in the error "'sizeof' found where almost anything else expected" during compilation. This has been fixed.

13. Two new functions have been added for use with Stata plugins: sstore() and sdata(). sstore() stores string data in Stata and sdata() reads string data from Stata; see http://www.stata.com/plugins/.

14. The random-effects panel-data estimators xtlogit, re; xtprobit, re; xtcloglog, re; xtpoisson, re normal; xtintreg; and xttobit could ignore large panels when fitting the model. This was caused by an internal numerical underflow, although no warning message was presented, and this is now fixed. What follows is a description of the bug and how to determine whether the bug affected any models you have fitted.

The above commands use adaptive Gaussian quadrature to estimate panel-level likelihoods, where each is the integral of some function (call it G) times the normal density. The function G is the panel-level likelihood conditional on the random effects and thus is itself a product of observation-level densities of the observations within that panel. Typically, but not necessarily, densities are between 0 and 1, and in fact they are between 0 and 1 for binary-data models. In any case, the product of these densities over a large panel (a panel with many observations) can be small. If the values got too small, they were treated as if they were exactly zero. The net effect was that such panels were ignored in the estimation of the parameters.

This problem arose in datasets with large panels, that is, datasets with many observations per panel. How large a panel needed to be in order to be affected depended on the average observation-level density for the panel. If this density (probability in a binary model) averaged 10%, underflow would occur in panels containing 300 or more observations. If this density averaged 50%, underflow would occur in panels containing 1,000 or more observations. At 90%, the underflow would occur with panels containing 6,500 or more observations. Since most users use these models to estimate for large N, small T data, this large-T problem did not affect many users.

Also, regardless of panel size, calculations for determining the optimal Gaussian quadrature points were also subject to a similar underflow problem. This has also been fixed. The net result of this fix, however, is merely that quadrature points are now more precisely estimated, resulting in better-approximated log-likelihoods. The gain in precision is slight, usually being in the fourth or fifth decimal place of the estimated regression parameters.

15. Stata/MP with more than two processors had a bug in tobit and cnreg. Let N_PROCS be the number of processors with N_PROCS>2, and let N_OBS be the number of observations for the model. When N_OBS/100>1 and N_OBS/100<N_PROCS, Stata/MP made a mistake in determining the actual number of processors to use. The mistake could crash Stata or produce wrong results. This is fixed.

16. When exporting a graph with the Stata logo as a PostScript file, Stata would specify the wrong font name for the Stata logo. This caused applications such as Adobe Illustrator or Distiller to complain that a font wasn't found. This bug is not serious since those applications would substitute a font for the incorrectly named font. This has been fixed.

Stata also now outputs characters outside the range of 7-bit ASCII characters in PostScript's preferred octal format. Users should see no difference when their PostScript or EPS files are rendered or printed.

Stata executable, Windows

17. A process spawned by winexec would inherit and hold open any file handles previously opened by Stata (such as an open log file) even if Stata later closed those file handles. Stata now does not allow a process spawned by winexec to inherit Stata's file handles.

Stata executable, Mac

18. You may now change the font in the Results, Viewer, Command, Data Editor, Do-file Editor, Review, and Variables windows. You may select a fixed-width font only in the Results, Viewer, Data Editor, Do-file Editor, and Variables windows.

Stata continues to allow you to change the font in the Graph window.

19. Stata previously output text from graphs and SMCL exported to PostScript by using the incorrect font encoding. This caused characters outside the 7-bit ASCII range to not translate properly when using an external application's PostScript interpreter. This has been fixed.

-------- update 30nov2006 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 6(4).

Stata/MP executable, 64-bit Itanium Windows

2. The 64-bit Itanium Windows Stata/MP executable could have problems completing an update executable command; this has been fixed.

-------- update 21nov2006 -----------------------------------------------------

Ado-files

1. bstat exited with an error indicating a name conflict when the stat() option contained a matrix with equation names. This has been fixed.

2. cnreg, logit, mlogit, ologit, oprobit, probit, regress, and tobit are now noticeably faster (up to five times faster) when used within large loops, such as statsby with datasets that have many small by-groups.

3. histogram with an if condition that excluded one of the categories in the by() option resulted in an extra empty plot when option addlabels or kdensity was specified or resulted in an error when option normal was specified. This has been fixed.

4. nestreg reported confidence intervals for the default confidence level even when a different confidence level was specified in the level() option of the prefixed command. This has been fixed.

5. quadchk exited with an error when used to check models using the offset() option. This has been fixed. quadchk also ignored weights and the collinear and constraints() options. This has been fixed.

6. svy now identifies the rare case when a sampling unit bridges adjacent strata in multistage designs. svy treats all sampling units as nested within strata and units from prior stages.

7. svy: heckman excluded from the estimation sample observations with a missing value in the dependent variable when it should not have. This has been fixed.

8. svy jackknife reported incorrect values for the sample size, population size, and number of strata when some of the strata were omitted. This has been fixed.

9. svy linearization is now two to five times faster. svy: mean, svy: proportion, svy: ratio, and svy: total are considerably faster when the over() option identifies many subpopulations.

10. svy: mean, svy: proportion, svy: ratio, and svy: total reported missing standard errors due to strata with one sampling unit even when a subpop() option excluded the offending strata. This has been fixed.

11. svy: mean, svy: proportion, svy: ratio, and svy: total failed to omit strata that contain no subpopulation members. This has been fixed.

12. svy: tabulate twoway, when option row or column is specified, uses the ratio estimator to compute cell proportions relative to the row or column margins. For direct standardization, option stdize(), a better estimator is available: the standardized ratio estimator. This standardized ratio estimator is now used in place of the simple ratio estimator when stdize() is specified.

13. svy: tabulate incorrectly reported zero standard errors when used with survey data that had strata with one sampling unit. Standard errors and confidence intervals are now suppressed with a note explaining why when the se or ci option is specified in such cases.

14. xtcloglog and xtpoisson with the re and normal options would ignore constraints on lnsig2u. This has been fixed.

15. xtintreg, xtlogit, xtprobit, xtpoisson with the re and normal options, and xttobit all accepted the collinear option but exited with an error when used with a constraint to define a collinear model. This is now fixed.

16. xtintreg and xttobit ignored constraints on sigma_u and sigma_e. This has been fixed.

17. xtline has new i() and t() options. The i() option allows the panels (groups) to be identified using a string variable. The t() option allows a variable with noninteger values and repeated values within group.

The important effect of this is that xtline can now be used to easily create two-way graphs with separate plots for each level of a categorical variable. For example,

. xtline mpg, t(displacement) i(rep78) overlay

creates five line plots of mpg versus displacement, one for each of the five levels of rep78.

These plots can be rendered as other twoway plottypes by adding a recast() option. For example, adding recast(scatter) to the command above will produce five scatterplots rather than five line plots.

Stata executable, all platforms

18. In Mata's ghk() function, when the covariance matrix is not positive definite, the ghk() Cholesky routine will pivot out the offending row and column of the covariance matrix. The ghk() function failed to permute the corresponding upper bound of integration. This has been fixed.

Stata executable, Windows

19. Stata for Windows now supports Microsoft Windows Vista.

Stata executable, Mac

20. For efficiency, when resizing a graph window, Stata draws a preview of the graph rather than completely redrawing the graph. However, the preview could be drawn at the wrong size if the window is quickly resized by jerking the mouse. This has been fixed.

21. When there is a --more-- condition in the Results window, Stata will continuously poll for a key press to clear the --more-- condition, which causes the processor usage to be very high. Stata now does not poll as often when there is a --more-- condition, causing the processor usage to be very low.

22. Some graph markers and objects would not be drawn with an outline in the Graph window when using the Quartz 2D graphics engine. This has been fixed. This bug did not affect EPS and PS files.

23. The console version of Stata/MP for Intel-based Macs could not correctly determine whether an executable update was available. This has been fixed.

24. The Do-file Editor will now treat a tab character as multiple spaces when printing a do-file. The Do-file Editor previously treated a tab character as one space when printing.

25. Stata now creates the Stata license file as stata.lic instead of Stata.Lic after license initialization. This distinction allows Mac OS X systems using the UFS file system to read the license file because the UFS file system is case sensitive. The default file system for Mac OS X is HFS+ and is not case sensitive, and it will accept both stata.lic and Stata.Lic. Current Stata users who are using the UFS file system can simply rename the file Stata.Lic to stata.lic. Current Stata users who are using the HFS+ file system do not need to rename Stata.Lic to stata.lic.

Stata executable, Unix (GUI)

26. Clicking a link in the Results window to open a help file in the Viewer could cause Stata to crash. This has been fixed.

-------- update 27oct2006 -----------------------------------------------------

Ado-files

1. For bootstrap and jackknife, if a user specified a cluster() option before the colon and a conflicting cluster() option after the colon, the second cluster() option was being ignored. An error message is now issued.

2. If a user-written e-class command did not set e(cmd), bootstrap, jackknife, svy brr, and svy jackknife exited with an error message. This has been fixed.

3. If a URL longer than 44 characters was used to open a dataset, codebook's problems option reported an error. It no longer does so.

4. If a value label had unbalanced quotes, codebook stopped with an error message. This has been fixed.

5. ivprobit did not set e(depvar) when the twostep option was used. This has been fixed.

6. screeplot's addplot() option failed with multiple plots. This has been fixed.

7. svy: regress when typed in the abbreviated forms svy: regr, svy: regre, and svy: regres (with one s), failed to adjust standard errors for mean squared error of the regression. This has been fixed.

There was no problem when you typed svy: regress or its most common abbreviation, svy: reg.

8. Previously, xtgee could produce model results in which the estimated working correlation matrix was not positive definite, as a proper correlation matrix must be. Now, when xtgee reaches a solution for which the working correlation matrix is not positive definite, the command issues an error message stating that convergence has not been achieved.

Also, for exchangeable correlation matrix structures, the estimation algorithm resets the correlation parameter to its minimum bound implied by the maximum panel size when the correlation parameter is infeasible. This reset will help models that can converge to do so, although if the data are incompatible with an exchangeable correlation structure, then convergence cannot be achieved.

-------- update 06oct2006 -----------------------------------------------------

Ado-files

1. areg with option vce(bootstrap) or vce(jackknife) resampled observations instead of the groups identified by the absorb() variable. This has been fixed.

2. estat mfx after asmprobit, in extremely rare cases, gave incorrect standard errors for the marginal effects of discrete case-specific covariates. These standard errors were incorrect only when different individuals faced different alternatives and when the subsample used to compute marginal effects was missing the first alternative. This has been fixed.

Standard errors of marginal effects for both discrete and continuous alternative-specific covariates, as well as those for all continuous covariates, were always correct.

3. Bayesian information criterion (BIC) calculations within glm, with family(binomial) and an explicit binomial denominator, were based on an incorrect total number of observations. This has been fixed.

4. quadchk used with version control for versions before 9 exited with an error. This has been fixed.

5. rolling returned an error message when used with panel data and an if qualifier that disqualified an entire panel. This has been fixed.

6. svyset with the noclear option resulted in one sampling stage even when the original settings had more than one stage. This has been fixed.

7. xtgee, when used with family(binomial), issued an "estimates diverging" error when used with fractional dependent variables containing values equal to one. This has been fixed.

8. xtset, when both panel and time variables were specified, misreported unbalanced panels as either strongly balanced or weakly balanced. This has been fixed.

Stata executable, all platforms

9. pctile and xtile, when dividing the data into more than 1,001 quantiles, would occasionally put data into the wrong quantile because of small rounding errors. This has been fixed.

10. svy reported missing standard errors when there were strata with one certainty unit. This has been fixed.

Stata executable, Windows

11. Pressing Page Up/Down in the Command window no longer sounds a system beep.

12. The Value labels list box from the label define dialog could obscure part of the Define label names list box on systems where the window appearance setting was changed. This has been fixed.

Stata executable, Mac

13. Stata prevented windows from being dragged above the menu bar on Macs with multiple monitors. This has been fixed.

-------- update 28aug2006 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 6(3).

2. clogit standard errors were not correct if pweights were used without option cluster(). This problem has been fixed.

3. cloglog and xtcloglog reported error messages when the collinear option was specified even when constraints were specified that properly identified the model. This problem has been fixed.

4. graph bar, in rare cases, produced bars based on incorrect values when a variable was repeated for a statistic and followed by another statistic and variable. This problem has been fixed.

5. histogram, kdensity, twoway histogram, twoway kdensity, twoway lfit, and twoway qfit produced unusable graphs when the by() option identified one or more groups that resulted in an empty graph. This problem has been fixed.

6. linktest, when run after glm, used a Stata 6 version of glm rather than the modern one. This problem has been fixed.

7. ml commands using fweights with options technique(bhhh) or vce(opg) produced incorrect standard error estimates. This problem has been fixed.

8. ml exited with an unhelpful error message when checking for convergence and the likelihood evaluator resulted in an unstable or asymmetric Hessian matrix. This problem has been fixed.

9. nestreg returned an error message when svy omitted strata due to zero weights or subpopulation estimation. This problem has been fixed.

10. The function evaluator version of nl exited with an error if the user tried to pass any options to the function evaluator program. This problem has been fixed.

11. nlogit's robust standard-error calculation failed to take into account the within-choice-set correlation inherent to nlogit models. This problem has been fixed.

12. pcamat, factormat, corr2data, and drawnorm now allow singular correlation (or covariance) structures. pcamat, factormat, corr2data, and drawnorm have new option forcepsd that maps a matrix intended to be correlations (covariances) into a positive-semidefinite singular correlation (covariance) matrix.

13. ranksum now allows by() to contain a string variable.

14. _rmdcoll's nocollin option has been renamed to normcoll. The nocollin option will continue to work as a synonym for normcoll but will become undocumented.

15. stepwise dropped variables collinear with the constant even when the noconstant option was specified in the prefixed command. This problem has been fixed.

16. svy jackknife sometimes reported the number of subpopulation observations and subpopulation size even when the subpop() option was not specified. This problem has been fixed.

17. table's format() option was not applied to the frequency column when the frequency was greater than nine digits. This problem has been fixed.

18. testnl with an expression that included a comma produced an error message. Expressions that are bound in parentheses now allow commas. For example, testnl _b[x]=M[1,3] is still not allowed, whereas testnl (_b[x]=M[1,3]) is now allowed.

19. tsset and xtset now indicate whether the panels are balanced.

Stata executable, all platforms

20. asmprobit sometimes incorrectly marked the estimation sample when invalid observations were dropped. This problem has been fixed.

21. graph twoway plot types pcarrow, pcbarrow, and pcarrowi may now be drawn on plot regions with log scales or reversed scales.

22. Once logit, logistic, or probit reported an error 902 (no room to add more variables due to width), the error would continue being reported on subsequent attempts even after the memory had been set to a value large enough to perform the estimation. This problem has been fixed.

23. Mata function findexternal() is updated to allow returning pointers to functions as well as matrices, vectors, and scalars.

24. svy brr, svy jackknife, svy: mean, svy: proportion, svy: ratio, and svy: total did not omit strata for which every observation contained a zero sampling weight. This problem has been fixed.

Stata executable, Windows and Mac

25. When Stata checked if an update was available from Stata's website but no updating was performed, it displayed a note that an update was available after Stata was launched. In rare cases, Stata's update status was not synchronized with Stata's website. This problem has been fixed.

-------- update 31jul2006 -----------------------------------------------------

Ado-files

1. Estimation commands called with options vce(bootstrap) or vce(jackknife) now report a note when a variable is dropped due to collinearity.

2. asmprobit when used with vce(bootstrap) or vce(jackknife) now requires that basealternative() and scalealternative() options be specified.

3. heckman, twostep's dialog incorrectly displayed a button for launching the svyset dialog. This problem has been fixed.

4. mlogit, mprobit, and slogit when used with vce(bootstrap) or vce(jackknife) now require that the baseoutcome() option be specified.

5. nlogit issued a "too many values" error if the number of groups exceeded tabulate's row limit; see limits. This problem has been fixed.

6. nlogit reported incorrect degrees of freedom for the LR test of homoskedasticity if constraints were applied to the regression parameters. This problem has been fixed.

7. ologit sometimes reported a missing model F statistic when it should have reported a missing model LR statistic. This problem has been fixed.

8. rolling failed to compute results with commands that generate new variables. This problem has been fixed.

9. svy now omits strata that have zero-valued sampling weights in every observation.

10. tabstat with the by() option truncated the word Total when the by() variable's name and value label were fewer than 5 characters. This problem has been fixed.

Stata executable, all platforms

11. The following commands now have the collinear option: arch, arima, asmprobit, biprobit, clogit, cloglog, frontier, gnbreg, heckman, heckprob, hetprob, intreg, ivprobit, ivtobit, mlogit, mprobit, nbreg, nlogit, poisson, _rmcoll, _rmdcoll, slogit, streg, treatreg, truncreg, xtcloglog, xtfrontier, xtintreg, xtlogit, xtnbreg, xtpoisson, xtprobit, xttobit, zinb, zip, ztnb, and ztp.

12. _parse expand and _parse canonicalize now have the gweight option for parsing global weights.

13. datasignature did not always calculate the correct checksum when using the fast option on Power PC-based Mac, AIX, Solaris, HP/9000, HP/Itanium, and Irix computers. This problem has been fixed.

14. Mata's allof() and anyof() functions did not allow pointer or structure arguments and, in fact, could crash when such arguments were specified. This problem has been fixed, and the functions may be used with pointer or structure arguments.

Stata executable, Windows

15. The ability to submit multiple commands to the Command window has been reinstated.

16. Review window selections are now retained.

17. Under rare conditions, programmable dialogs could show up with a black background. This problem has been fixed.

Stata executable, Unix (GUI)

18. The previous update of Stata introduced a problem where printing a Viewer window would cause the Command edit field to not accept any keyboard input. This problem has been fixed.

19. When building a list of fonts for display in the Graph Preferences dialog, Stata queries the font server for a list of fonts and expects the fonts to be specified in the X Logical Font Description (XLFD) format. If a font was not in the XLFD format, Stata crashed. Stata now ignores fonts that are not returned in the correct format.

-------- update 13jul2006 -----------------------------------------------------

Stata executable, Windows

1. Certain windowing layouts could cause Stata to incorrectly display windows upon starting up. For example, a pinned Variable pane could cause a portion of Stata's background to not be drawn properly. This problem has been fixed.

-------- update 11jul2006 -----------------------------------------------------

Ado-files

1. clogit now accepts pweights.

2. tetrachoric's default algorithm for computing tetrachoric correlations has been changed from the Edwards and Edwards estimator to a maximum likelihood estimator. The Edwards and Edwards estimator performed poorly with skewed data. tetrachoric's features are now similar to those of spearman and ktau. Standard errors and two-sided significance tests are now included. A frequency adjustment when one cell has a zero count is now available with the zeroadjust option. This change has been made without version control; to get the old behavior, the edwards option can be used.

3. xtmixed has new option collinear, which tells xtmixed to retain collinear variables within random-effects equations.

-------- update 06jul2006 -----------------------------------------------------

Ado-files

1. clear has been updated to clear the new timers described below for profiling ado and Mata code.

2. heckman's dialog incorrectly displayed a button for launching the svyset dialog. This problem has been fixed.

3. lowess ignored certain suboptions of the by() option. This problem has been fixed.

4. predict, scores after streg, distribution(exponential) computed scores incorrectly when t0 was not 0. This problem has been fixed.

5. survey estimation commands that allow the technique() option reported incorrect standard errors when technique(bhhh) was specified. This problem has been fixed.

6. svy: tabulate oneway reported an error message when specified with a constant variable. This problem has been fixed.

7. tsline misread or would not accept certain datelist specifications in the tlabel() and tline() options. This problem has been fixed.

8. twoway fpfitci did not draw the confidence interval when the option estcmd() was used to specify some estimation commands. This problem has been fixed.

Stata executable, all platforms

9. Stata/MP, the parallel-processing flavor of Stata, is now available for 64-bit Solaris on x86-64 hardware. The supported platforms for Stata/MP are Windows (32-bit, 64-bit Itanium, and 64-bit x86-64), Linux (32-bit, 64-bit Itanium, and 64-bit x86-64), and Solaris (64-bit SPARC and 64-bit x86-64).

10. The internal utilities for gllamm now support inverse-probability weighting schemes.

11. Mata has the following new functions:

A. Mata functions minindex() and maxindex() return the indices of minimums and maximums (including tied values) of a vector, v.

B. Mata function timer() is a new undocumented function for profiling the execution time of Mata code.

12. Mata had the following fixes:

A. Mata functions crossdev() and quadcrossdev() aborted with error if the centering variables contained missing values. Now the functions return an all-zero result.

B. Mata functions rowshape() and colshape() would not let you change certain types of null matrices to other types. For instance, you could not change a 1 x 0 matrix into a 0 x 0. This problem is fixed. A side effect of the fix is that you must recompile any programs that use colshape(). You do not need to recompile programs that use rowshape(), however.

C. Mata function st_strscalar() did not return results from Stata's c(), such as st_strscalar("c(byteorder)"). This problem has been fixed. Users are reminded that another way to obtain c-class values is via Mata's c() function: c("byteorder").

13. Inserting data via odbc insert to an existing database could, depending on the underlying SQL code generated, cause SQL syntax errors. This problem has been fixed.

14. The regular expression functions regexm(), regexs(), and regexr() could cause Stata/MP to crash; this problem has been fixed.

15. To prevent output from wrapping, return list, ereturn list, and sreturn list now display only the first line of long string results. The result looks better and is more easily read. If a report is truncated, two dots are added to the end of the part shown.

16. In tabulate twoway, cells with zero counts contributed missing (.) instead of zero to the value of the likelihood-ratio chi-squared test. This problem has been fixed.

17. timer is a new undocumented command for profiling the execution time of ado-code.

18. Small Stata (Windows and Mac) could crash when the extended macro function properties was used in a program. This problem has been fixed.

Stata executable, Windows

19. The Command window now has keyboard shortcuts reinstated from Stata 8. Using Ctrl-Backspace to delete a whole word is an example.

Stata executable, Mac

20. The contextual menu item Copy Review Contents to Do-file Editor for the Review window was enabled even when the Review window was empty. This problem has been fixed.

21. The Font Size submenu from the Prefs menu was not enabled when the Variables window was selected. This problem has been fixed.

22. Using the Quartz drawing engine for drawing graphs could render some non-ASCII characters as outlined boxes. To avoid this limitation of the Quartz text renderer, you may use ATSUI (Apple Type Services for Unicode Imaging) to render text in graphs by using set use_atsui_graph on.

Stata executable, Unix (GUI)

23. Closing a Viewer window while a modal dialog was up would cause Stata to crash. Stata now prevents you from closing the Viewer until the warning dialog is closed.

24. A GTK warning message could appear from the console Stata was started from if a dialog had a dropdown list combo box. This problem has been fixed.

25. After the 17may2006 executable update, the setposition member program did not work correctly for certain types of programmable dialog controls. This problem has been fixed.

-------- update 20jun2006 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 6(2).

2. areg exited with an error when if or in was properly supplied after the absorb() option. This problem has been fixed.

3. egen's median() function failed when presented with expressions containing double quotes. This problem has been fixed.

4. estat hettest has new options iid and fstat. By default, estat hettest performs a score test for heteroskedasticity that assumes that the regression disturbances are normally distributed. iid causes estat hettest to perform a score test for heteroskedasticity that drops the normality assumption. fstat causes estat hettest to perform an F test for heteroskedasticity that drops the normality assumption.

5. estat failed to work after stcox, shared(). This problem has been fixed.

6. Mata functions mean(), variance(), quadvariance(), meanvariance(), quadmeanvariance(), correlation(), and quadcorrelation() previously required two arguments, the second being w, the weight. The second argument is now optional and, if omitted, unweighted estimates (equivalent w=1) are returned.

7. Mata functions variance(X,w), quadvariance(X,w), meanvariance(X,w), and quadmeanvariance(X,w) aborted with an error if X or w contained all missing values. Now the functions return an all-missing result.

8. nl exited with an error if a parameter's name was a substring of another parameter's name and the parameter whose name was longer was declared first. This problem has been fixed.

9. nlogit did not use weights when computing the initial estimates and null likelihoods used to test the regression coefficients and homoskedasticity. This problem has been fixed.

10. roctab, with the detail option, reported incorrect values of sensitivity when the classification probabilities corresponding to different levels of the classification variable were equal. This problem has been fixed.

11. svy: mlogit exited with an invalid-syntax error when the dependent variable had value labels with spaces in them. This problem has been fixed.

12. treatreg now accepts the from() option.

13. tsline reported an invalid-number error when the time variable had a %d format and a time axis option was specified using a date string. This problem has been fixed.

-------- update 17may2006 -----------------------------------------------------

Ado-files

1. estat vif after regress has a new option, uncentered, that computes uncentered variance inflation factors.

2. estat vif after regress reported invalid results with options vce(bootstrap) or vce(jackknife) or with the svy prefix. estat vif now reports an error message.

3. estimates table did not label estimates properly when equation names contained spaces. This problem has been fixed.

4. glm would exit with an error message if an HAC covariance matrix was requested and the list of regressors included time-series operators. This problem has been fixed.

5. glm reported incorrect deviances and deviance diagnostics when fitted to proportion data with family(binomial 1). This problem has been fixed.

6. graph twoway now allows more than 100 variables to be plotted even when those variables are from the same observations. Previously the only way to plot more than 100 variables was to ensure that some of the variables were plotted over a different range of observations. One plot is still limited to 100 or fewer variables.

7. The dialog box for graph twoway did not generate the correct options for affecting the rendition of lines when used with twoway lfitci, twoway qfitci, or twoway fpfitci. This problem has been fixed.

8. ivprobit and ivtobit exited with an error message when regressors were dropped because of collinearity, when the list of regressors exceeded about 500 characters, or when time-series operators were used. These problems have been fixed.

9. nlogit now allows the robust option for models that nest deeper than three levels.

10. qreg, and similar quantile regression commands, sometimes dropped the dependent variable when it equaled a linear function of the covariates. This problem has been fixed.

11. svyset now has the fay(#) option.

12. svy: tabulate twoway reported a one in the marginals when the column and row options were specified, even when given data that resulted in a zero in the marginals. This problem has been fixed.

13. table now issues an improved error message when the sd statistic is used with pweights.

14. xi omitted a group when called with the noomit option and syntax i.varname1*varname2 or i.varname1|varname2. This problem has been fixed.

15. xtreg, fe cluster() did not report the proper degrees of freedom or p-value for the model F test. This problem has been fixed.

16. xtset is a new command for declaring panel data. This command is meant to eventually replace iis and tis. For panel data with a time variable, xtset and tsset are equivalent. View the xtset command as a new convenience command, something you do not really need but might like to use anyway.

Stata executable, all platforms

17. datasignature is a new command that many users will find useful. datasignature reports a short string (a signature) that you can use to detect whether your data have changed. See "Examples of interactive use" under Remarks in the help file.

18. insheet, when given a file with more than the maximum allowed number of variables, could cause Stata to crash. This problem has been fixed.

19. irf create now estimates dynamic-multiplier (DM) functions and cumulative DM functions, as well as their standard errors. The irf reporting commands were modified to display the new DM estimates.

20. Mata functions select() and st_select() are new. They select rows or columns of a matrix on the basis of a criterion.

21. Mata incorrectly executed statements of the form "v[i] = name" and "v[i,j] = name" when v and name were structures. Rather than assigning a copy of structure name to v, Mata assigned name itself, so that subsequent changes in name changed v[i] and v[i,j]. This problem has been fixed. You do not need to recompile Mata code.

22. Mata inconsistently classified "" as being greater than nonempty strings (such as "a"), whereas Stata classifies them as being less than nonempty strings. Mata has been modified so that it agrees with Stata: "" < "a".

23. Mata's >= operator returned an incorrect result when used with strings and the string to the left of the operator was greater than the string to the right of the operator. This problem has been fixed.

24. signestimationsample and checkestimationsample are new programmer commands for coordinating estimation and postestimation ado-files.

25. Classes can now define an .oncopy member program to perform operations when a copy of an object is created. This is an advanced concept; see classman section 4.8 Advanced initialization, .oncopy.

26. For Stata/MP, Stata/SE, and Intercooled Stata, the limit on the number of dyadic operators in an expression has been increased from 500 to 800, and the limit on the number of numeric literals in an expression has been increased from 150 to 300. See limits.

27. New functions _byn1() and _byn2() are available within a byable(recall) program. They return the beginning and ending observation numbers of the by-group currently being executed. See byable for more information.

Stata executable, Windows

28. The Data editor on Windows ME could freeze Stata. This problem has been fixed.

29. The dropdown print menu from the toolbar print button could get corrupted when displaying multiple graphs. This problem has been fixed.

30. Stata could crash if scheduled to run at a preset time when there was no user logged in to Windows. This problem has been fixed.

31. When invoking a dataset via the recent files menu in Stata that happened to also be on a shared network drive, an extra slash was prepended to the path. This problem has been fixed.

32. Stata/MP on Windows XP loaded a wrong version of the common controls DLL. Various abnormal GUI behaviors, such as the dialog backgrounds becoming black, could occur if XP Themes were used. This problem has been fixed.

Stata executable, Mac

33. xmlsave now uses only double quotes as tag attributes when saving to the Excel doctype. This change circumvents a bug in Excel for Mac when reading XML data files generated from Stata.

Stata executable, Unix (GUI)

34. When a programmable dialog has the keyboard focus, pressing Enter simulates clicking the OK button if one exists, pressing Escape simulates clicking the Cancel button if one exists, and pressing Shift and Return simulates clicking the Submit button if one exists.

35. The Command window is no longer resized if the main Stata window is resized vertically. The Review and Variables windows are no longer resized if the main Stata window is resized horizontally. You may still drag the splitter controls to resize the windows.

36. Sometimes, the Command window lost the keyboard focus, preventing users from inputting any commands unless they clicked on the body of the main Stata window. This problem has been fixed.

37. Exporting a graph as TIFF would result in a graph with the width and height swapped. This problem has been fixed.

38. Resizing a Graph window could cause its graph to not draw at the final window size if the graph could not finish drawing while it was being resized. This problem has been fixed.

39. Resizing the Data Editor also caused the Data Editor's toolbar and edit field to resize vertically. This problem has been fixed.

40. Right-clicking in a Viewer window and selecting Find in Viewer... would not display the Find bar. After selecting File > Log > View... to display the Choose File to View dialog, clicking the Browse... button and selecting a filename would not be entered into the File or URL: edit field. Both problems have been fixed.

41. Selecting Prefs > Manage Preferences > Load Preferences > Open... could crash Stata. This problem has been fixed.

42. If any modal dialog caused Stata's Results window to scroll before the dialog could close completely, the first line to be obscured by the top of the dialog would not be updated after it scrolled. This problem has been fixed.

43. shell under Linux could fail if more than half of the memory available on the system was allocated to Stata. This problem has been fixed.

-------- update 14apr2006 -----------------------------------------------------

What's new in release 9.2 (compared with release 9.1)

Highlights for release 9.2 include the following:

o Stata/MP is a new parallel-processing version of Stata that can perform calculations in parallel on computers with 2, 4, 8, 16, and even 32 processors or cores. Stata/MP has all the capabilities of Stata/SE but can perform many analyses much faster by taking advantage of special code to perform symmetric multiprocessing (SMP). See Stata/MP and http://www.stata.com/statamp/ for more information.

o Mata now has structures, which will be of special interest to those writing large, complicated systems. See item 14 A below.

o Mata now engages in more thorough type checking, and produces better code, for those who explicitly declare arguments and variables. See item 14 B below.

o hsearch is a new command that searches help files. It is an alternative to the search command. You will want to try hsearch.

We recommend that you set version to version 9.2 at the top of new do-files and ado-files. See help version.

Ado-files

1. bootstrap and jackknife exited with an error when an appropriate eform_option was specified. This problem has been fixed.

2. bsample returned incorrect frequency weight values (in the weight() option) when called with a resample size less than the selected sample size. This problem has been fixed.

3. cloglog and xtcloglog have new option eform.

4. corr2data was inadvertently setting the random-number generator seed to the current value of sortseed. This problem has been fixed.

5. glm and xtgee labeled generic exponentiated coefficients as "ExpB" and "e^coef". For consistency, both labels have been changed to "exp(b)". This change applies only to models where exponentiated coefficients do not already have meaningful names, such as hazard ratios.

6. hsearch is a new command for searching the Stata help files on your computer.

7. ml complained that the variable ML_subv already exists when called twice interactively with the subpop() option. This problem has been fixed.

8. ml display has new option showeqns.

9. sts graph did not respect value labels when the by() and separate options were combined with hazard. This problem has been fixed.

10. svy jackknife with the mse option after svyseting iweights with strata() resulted in incorrect standard errors. This problem has been fixed.

11. xthtaylor no longer requires the user to specify both endogenous time-invariant variables and endogenous time-varying variables; users may now specify endogenous time-invariant variables and/or endogenous time-varying variables.

Stata executable, all platforms

12. asmprobit with unbalanced data and the BHHH optimization technique, tech(bhhh), did not work properly and typically did not converge. This problem has been fixed.

13. Date values in Stata should be integers. Given noninteger values, however, date functions and date formats rounded toward 1jan1960. Now they always round down (toward the past).

14. Mata has the following new features:

A. Mata now has structures (see [M-2] struct), which will be of special interest to those writing large, complicated systems. Because of the change, the format of .mlib libraries is different. There is no backward-compatibility issue -- you do not need to recompile old code -- you *NEVER* need to recompile old code. However, because of the format change, you will not be able to add new members to old libraries. Libraries cannot contain a mix of old and new code. To add new members, you must first rebuild the old library.

B. Mata now engages in much more thorough type checking, and produces better code, for those who explicitly declare arguments and variables. As before, declaration is optional. However, if you make explicit declarations, old code may not compile because Mata detects problems it previously could not. As an example, assigning a string result to a variable explicitly declared as real is no longer allowed. If you have old code with such problems, either change the variable's declaration to transmorphic or introduce a new variable to fix the problem properly.

C. Mata now includes all 29 Stata date functions; see [M-5] date().

15. Mata had the following fixes:

A. The first time Mata was invoked after launching Stata, if the first thing typed into Mata was a double quote character and there was no matching close quote, Stata crashed. This problem has been fixed.

B. Mata function ascii() returned negative values for the extended ASCII characters rather than returning values in the range 128 to 255. This problem has been fixed.

C. Mata functions fread() and _fread() caused Mata to issue a stack-overflow error. This problem has been fixed.

D. Mata function regexr() did not allow replacement with an empty string (for deletions) and did not allow substitutions on strings shorter than 10 characters. Both problems have been fixed.

E. Mata comparison operator >= was treated the same as > for string comparisons. This problem has been fixed.

16. merge, specified with match variables, caused Stata to crash when the master dataset had no observations. This problem has been fixed.

17. net install and net get now accept a force option, which specifies that the downloaded files replace existing files if any of the files already exist, even if Stata thinks all the files are the same.

18. PNG (Portable Network Graphics) files produced by graph export on 64-bit systems could contain valid image data but an invalid checksum on that data. This could cause some third-party applications to have problems displaying the PNG files. This problem has been fixed.

19. use did not clear the previous sort list when a varlist was specified. This problem has been fixed.

20. xtcloglog, xtintreg, xtlogit, xtpoisson, xtprobit, and xttobit used single precision instead of double precision storage for the log likelihood. This problem has been fixed.

21. xtintreg and xttobit results were not as precise as they could have been because of inaccuracy in the calculation of the derivatives of the log likelihood. This problem has been fixed.

Stata executable, Windows

22. odbc load now imports the ODBC bit data type as Stata byte. Previously bits were imported as Stata integers. This change reduces the memory needed to import data.

23. Alt-Tab functionality and focusing of the Command window has been improved for nondefault layouts.

24. New layouts are now saved properly when issuing the exit command.

25. Under Windows 98, closing the Data Editor caused Stata to crash. This problem has been fixed.

Stata executable, Mac

26. You may now paste multiple lines of text into the Command window. Stata executes each line of text that is terminated by a carriage return. If a line of text does not end in a carriage return, it is simply pasted into the Command window. Any errors in the lines of text that are executed do not prevent subsequent lines from executing, which can cause unintended results. As such, this setting is off by default, which means that only the first line of text is pasted into the Command window and is not executed.

27. Stata now uses very little of the processor while Mata is idling in interactive mode.

28. Stata now prevents multiple lines of text from being pasted into an edit control from a programmable dialog. Only the first line of text is pasted. Pressing the Home or End key in an edit control of a programmable dialog now moves the selection to the beginning or end, respectively, of the edit control.

29. Stata's Do-file Editor tiles new Do-file Editor windows 20 pixels to the right and 20 pixels below the last moved or created Do-file Editor window. However, a new Do-file Editor window could potentially be tiled off the screen if the last window is too close to the right edge or bottom of the screen. Stata now starts over from the top-left corner of the screen if a new Do-file Editor window is too close to the right edge or bottom of the screen. Also, Do-file Editor, Viewer, and Graph windows will now tile properly on multiple displays. Stata previously always tiled new windows on the primary display rather than the display in which the last window was located.

30. Pressing Return in the Data Editor's Edit dialog inserted a carriage return into the currently active edit control. This problem has been fixed.

31. The 20jan2006 executable update, when run on a PowerPC-based Mac, would require a G4 processor or better (this limitation did not affect Stata when run on an Intel-based Mac). Stata will now run on any PowerPC computer that supports Mac OS X just like it did prior to the 20jan2006 update.

Stata executable, Unix (GUI)

32. If the main Stata window was resized to smaller than the default, it still opened at the default window size the next time Stata was launched. This problem has been fixed.

33. Stata now prevents multiple lines of text from being pasted into an edit control from a programmable dialog. Only the first line of text is pasted.

34. The Do-file Editor could crash if there was no carriage return on the last line of the do-file. This problem has been fixed.

-------- update 27feb2006 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 6(1).

2. bstat always reported 95% confidence intervals even when the level() option was specified. This problem has been fixed.

3. cs exited with an error when by() contained more than one by-variable, and option rd was specified. This problem has been fixed.

4. estat dwatson could produce erroneous results when used after a command that performed a regression that included temporary variables. Now an error message is produced.

5. gladder and qladder did not apply user-specified schemes to individual plots. This problem has been fixed.

6. Mata function cat() now executes faster on large files.

7. merge was documented as allowing uniqu as a valid abbreviation for both unique and uniqusing. The help file now documents that uniqu is an abbreviation only for unique.

8. mfx, eyex and mfx, dyex would report negative standard errors for marginal effects if the value of the x variable was negative. This problem has been fixed.

9. ml now saves e(k_eq_model), the number of equations to include in a model Wald test, even when lf0() is specified.

10. permute with the left option did not count all cases where replicate values were less than or equal to the observed value of the test statistic. This problem has been fixed.

11. pksumm sometimes incorrectly reported the error 'follow-up times are different for each patient'. This problem has been fixed.

12. predict after logit and probit reported an error when an independent variable was dropped because of collinearity. This problem has been fixed.

13. predict, after xtcloglog, exited with a file-not-found error. This problem has been fixed.

14. predict, xbu after xtreg, fe failed if any of the regressors contained time-series operators. This problem has been fixed.

15. predictnl after ologit exited with an error when the special predictnl function predict() was specified. This problem has been fixed.

16. recode's min element replaced a variable's value with missing when the if qualifier was specified and no observations met the if criterion. This problem has been fixed.

17. stptime exited with a syntax error when an if qualifier containing strings was specified. This problem has been fixed.

18. svy: tabulate twoway would report tests of independence when the table had a zero margin in a column beyond the number of rows. This problem has been fixed.

19. xtlogit and xtprobit did not accept constraints on [lnsig2u]_cons. This problem has been fixed.

20. pwcorr with perfectly correlated data now returns correct significance values.

-------- update 20jan2006 -----------------------------------------------------

Ado-files

1. nestreg exited with an error when a block of variables was preliminarily dropped from the model specification. This problem has been fixed.

2. stepwise exited with an error when a term consisting of a single variable was preliminarily dropped from the model. This problem has been fixed.

Stata executable, all platforms

3. The maximum number of programmable menu items that can be added to Stata has been increased from 1,000 to 1,250.

4. asmprobit's GHK simulator generated too many uniform numbers in the simulation. This problem has been fixed.

5. Class member programs were not using the version number specified at the top of a class-definition file. This problem has been fixed.

6. confirm format is a new command that allows formats to be verified. Optionally, you can confirm that the given format is of a particular type: numeric, string, date, or ts.

7. An expression evaluated by a command issued in a loop with an `=exp' on the initial loop command line could return missing value (.) even when it should not. This problem has been fixed.

8. filefilter used on files containing embedded NULL characters in certain positions sometimes caused the filters to be ignored for a portion of the file following the NULL before eventually resuming correct filtering. This problem has been fixed.

9. graph box, when specified with only a single variable that did not have a value label, left the y-axis untitled. In such cases, the y-axis will now be titled with the variable name.

10. graph twoway now allows you to create custom axis ticks and labels that have a different appearance (color, size, tick length, etc.) from the other ticks and labels on the axis. Such custom ticks can be used to emphasize points in the scale, such as important dates, physical constants, or other special values. See the new custom suboption of {y|x|t}label() in axis_label_options.

11. graph twoway has new options, yoverhangs and xoverhangs, that attempt to adjust the graph region margins to prevent long labels on the y- or x-axes from extending off the edges of the graph; see advanced_options.

12. graph twoway has a new suboption, norescale, to the options {y|x|t}label(). This suboption specifies that the ticks and labels specified in the {y|x|t}label() option be placed directly on the graph without rescaling the axis or associated plot region for the new values; see axis_label_options.

13. help, under rare conditions when accessing help files across a network, could be slow; this problem has been fixed.

14. inspect output has been slightly modified. All output on the right-hand side has been shifted to the left three spaces so that the "Nonintegers" heading is placed on one line.

15. label define, followed by something that was not a valid label name, caused a memory leak that could result in Stata's incorrectly reporting an 'unexpectedly out of memory' error with a return code of 9710. This problem has been fixed.

16. lnnormal() and lnnormalden() are new functions that provide the natural logarithm of the cumulative normal distribution function and the normal density function, respectively.

17. The new extended macro function adosubdir returns the subdirectory in which Stata would search for a file on the ado-path.

18. Mata has the following new functions:

A. adosubdir() returns the subdirectory in which Stata would search for a file on the ado-path.

B. ghk() provides a Geweke-Hajivassiliou-Keane multivariate normal simulator.

C. halton(), _halton(), and ghalton() compute Halton and Hammersley sets.

D. lnnormal() and lnnormalden() find the logarithm of the cumulative normal and of the density function, respectively.

E. pathsearchlist() returns a row vector of full paths/filenames specifying all the locations Stata will search for a named file along the official Stata ado-path.

19. Mata also has the following changes:

A. Mata's logical negation (not) operator ! treated missing values incorrectly. A missing value is supposed to be treated as meaning true just like any other nonzero value. Thus if x==., !x is 0. Instead, !x returned missing (.). This problem has been fixed.

B. Mata's printf() and sprintf() functions left-justified strings printed with %#s formats and right-justified strings with %-#s formats. This effect was the reverse of what was correct and has been fixed.

C. Mata/Stata interface functions st_local(), st_global(), st_numscalar(), and st_strscalar(), when used to store something more than 65,532 times in a tight loop, could cause a Mata stack overflow error. This problem has been fixed.

D. On 64-bit computers, Mata restricted matrix size such that r*c*8 < 2^31 for regular matrices and r*c*16 < 2^31 for complex matrices even when more than 2 GB of memory was available. This problem has been fixed, and matrices larger than 2 GB may now be used in Mata on 64-bit computers.

20. odbc commands with extended variable lists did not respect quotes or compound quotes to allow nonalphanumeric variable names to be imported. This problem has been fixed.

21. Exporting a graph with the width() or height() options caused Stata to crash if the graph window was closed. Stata now prevents you from exporting a graph when the graph window is closed.

22. _qreg, a programmer's routine, skipped observations when if or in qualifiers were specified. This problem has been fixed.

23. The regexr() function and the Mata regexr() function did not allow a string to be replaced with an empty string. For example:

. display regexr("mystring", "mystring", "")

displayed "mystring" instead of "". This problem has been fixed.

24. tabulate, exact, for computing Fisher's exact test, now allows larger tables, and an enumeration log is displayed. New option nolog suppresses the log.

25. twoway lfit and twoway qfit can now pick up value labels to annotate the x-axis by using the existing option mlabel(, valuelabel); see axis_label_options.

26. A contextual menu item has been added to the Viewer that closes all viewers except the current one.

Stata executable, Windows

27. Multiple do-file editors may now be opened from a single open dialog. The first file goes to the currently open do-file editor, whereas subsequent files open in new editor windows.

28. The most recently used file information was not saved properly when exiting Stata with the exit command. This problem has been fixed.

29. The Variables Property dialog has been modified to set the focus to the variables edit field and to set its text selection to all text.

30. Specifying a width() of less than 115 pixels when exporting a graph would not result in a bitmap image with the specified width. This problem has been fixed.

31. When Alt-Tabbing between windows and the main Stata window, focus will automatically be sent to the Command window. Previously, focus was sent only after alphanumeric keystrokes were detected.

Stata executable, Mac

32. You can select multiple files from the Open dialog and Do dialog. Selecting multiple do-files from the Open dialog will open all files in separate Do-file Editor windows. Selecting multiple do-files from the Do dialog will do do-files in the order that they were selected. Selecting multiple datasets and graphs will open only the first file selected.

33. Horizontal scroll wheel support has been added to the Data Editor and Do-file Editor. You may now use an Apple Mighty Mouse scroll wheel to scroll in any direction.

34. The menu item Tools > Back and its keyboard shortcut Cmd-[ have been added to the Viewer to match the keyboard shortcut for Back in Safari. However, Cmd-[ had been previously reserved for Edit > Shift Left (to shift text one tab stop to the left in the Do-file Editor). To avoid any conflicts, the keyboard shortcuts for Shift Left and Shift Right have now been changed to include the shift key (press Cmd-Shift-[ to shift text left and Cmd-Shift-] to shift text right).

35. Stata now displays the current number of columns and rows of the Results, Viewer, and Do-file Editor windows as they are being resized.

36. Stata no longer displays window transition effects when it is in the background.

37. Pressing the keyboard shortcut Cmd-W did not close the Review window. This problem has been fixed.

38. In some cases, clicking the Back button in the Viewer would result in the Viewer toolbar's being overwritten with the Viewer's output. This problem has been fixed.

39. In some cases, a single line in the Results window would appear chopped after output had scrolled. This problem has been fixed.

40. Typing exit from the Command window exited Stata without prompting to save any unsaved changes in the Do-file Editor. This problem has been fixed. Type exit, clear to exit Stata without prompting.

41. Stata(console) for Mac failed to perform an update swap after an executable had been downloaded. This problem has been fixed. However, current Stata(console) for Mac users will have to manually download the updated executable with the fix from http://www.stata.com/support/mac/.

Stata executable, Unix (GUI)

42. Stata now displays the current number of columns and rows of Viewer windows as they are being resized.

43. Pressing Ctrl-W now closes the Viewer.

-------- update 12jan2006 -----------------------------------------------------

Ado-files

1. cnsreg has the new mse1 option and accepts iweights.

2. Estimation commands with model Wald tests composed of more than just the first equation now save the number of equations in the model Wald test in e(k_eq_model).

3. kwallis now exits with an error when there are no observations to perform the calculations or when fewer than two populations are specified.

4. ml now saves the number of equations used to compute the model Wald test (from the waldtest(#) option) in e(k_eq_model).

5. ml display after ml-based user-written Stata 9 estimation commands that were prefixed with svy did not display the coefficient table header properly. This problem has been fixed.

6. ml score has the new missing option.

7. pkexamine has the following changes:

A. pkexamine with an if condition that excluded the observation with zero time and concentration did not use the proper zero time and concentration in the computations. This problem has been fixed.

B. pkexamine no longer adds an observation with zero time and zero concentration when the dataset has an observation with nonzero concentration at time zero.

C. pkexamine exited with an error message when fit() contained the total number of observations. This problem has been fixed.

8. predict after ologit and oprobit reported an error when the correct number of new variable names was specified for computing multiple equation-level score variables from a constant-only model. This problem has been fixed.

9. qreg did not set e(wtype) and e(wexp) when weights were used. This problem has been fixed.

10. suest after intreg reported an error when there were missing values in one or both of the dependent variables. This problem has been fixed.

11. suest produced incorrect variance estimates when used with cnsreg. This problem has been fixed.

-------- update 13dec2005 -----------------------------------------------------

Ado-files

1. Errors in a couple of help files have been corrected.

-------- update 12dec2005 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 5(4).

2. adoupdate is a new command for keeping the user-written ado-files you have installed from The Stata Journal, the SSC, and other sources up to date.

3. bootstrap did not pass the nodrop option to jackknife when both the nodrop and bca options were specified. This has been fixed.

4. estat exited with an error message after arima if the condition option was specified. This has been fixed.

5. glm ignored the exposure() option. This has been fixed.

6. histogram occasionally reported one fewer bin than the number produced in the graph. This has been fixed.

7. jackknife reported a "last estimates not found" error when called as a prefix to a command without arguments. This has been fixed.

8. jackknife reported an "options not allowed" error when called as a prefix to a command that does not allow options even when no options were specified. This has been fixed.

9. lrtest output inappropriately wrapped to a new line if the names of the tested models were long. This has been fixed.

10. nl now uses the format /param in listing the parameters in the coefficient table.

11. nl, when used with weights, reported an incorrect sum of weights and residual deviance. This has been fixed.

12. recode's max element treated missing values as the maximum value of a variable if the variable contained all missing values. This has been fixed.

13. stphplot did not display the legend properly in certain circumstances. This has been fixed.

14. suest now has option svy for applying the current svyset characteristics to the variance calculation.

15. The 15sep2005 update changed some low-level graphics calls. In rare cases, the previous low-level graphics calls were not properly interpreted by some HP printer drivers, causing the entire graph to be printed black. We have confirmed with users who had experienced the problem that the 15sep2005 update allows these drivers to render the graphs correctly.

This problem also occurred when copying and pasting into Microsoft Office on the Mac platform. The 15sep2005 update fixed this problem.

-------- update 11nov2005 -----------------------------------------------------

Ado-files

1. nestreg exited with an error when used with a survey estimation command other than svy: regress. This has been fixed.

-------- update 10nov2005 -----------------------------------------------------

Ado-files

1. adjust after ivprobit can now calculate adjusted probabilities when the maximum likelihood estimator is used.

2. bootstrap with the bca option exited with error when the user specified names for bootstrap statistics. This has been fixed.

3. estat bootstrap reported an invalid subcommand error when used after mean, proportion, ratio, and total even when vce(bootstrap) was specified. This has been fixed.

4. ds sometimes failed when certain values were specified together in the indent() and skip() options and the linesize was smaller than needed. This has been fixed.

5. glm saved an invalid e(power) when user-defined links used the SGLM_p macro for non-standard purposes (such as holding a variable name instead of a number). This has been fixed.

6. ivreg ignored the beta option. This has been fixed.

7. nestreg is a new prefix command for reporting nested model comparison tests (also known as block or hierarchical regression analysis).

8. The dialog for stepwise has been modified so that only the weights that are accepted for a specific command are presented.

9. predict exited with an error after xtgee with the link(nbinomial) option. This has been fixed.

10. xtdes exited abruptly if the time variable had fewer than two distinct nonmissing values. Now a more informative error message is issued.

-------- update 20oct2005 -----------------------------------------------------

Ado-files

1. merge, when specified with no using files, gave an incorrect error message. This has been fixed.

2. merge failed if any of the using files were located on a network drive. This has been fixed.

3. nl has received the following enhancements. The previous behavior of nl is preserved by setting version to one earlier than 9.1.

A. mfx can be used after nl to obtain marginal effects.

B. nl now supports the vce(vcetype) option. Allowed vcetypes include bootstrap, jackknife, robust, and three heteroskedasticity- and autocorrelation-consistent (HAC) variance estimators.

C. predict after nl now allows you to obtain the probability that the dependent variable lies within a given interval, the expected value of the dependent variable conditional on it being censored, and the expected value of the dependent variable conditional on it being truncated. These predictions assume that the error term is normally distributed.

D. The iteration log now reports the new sum of squares after each iteration is complete instead of the old value.

E. The coefficient table now reports each parameter as its own equation, analogous to how ml reports single-parameter equations.

F. nl no longer reports an overall model F test because the test that all parameters other than the constant are jointly zero may not be appropriate in arbitrary nonlinear models.

G. The value of the log-likelihood function assuming i.i.d. normal errors is saved in e(ll), and likelihood-ratio tests can be performed after estimation using lrtest.

4. qreg's wlsiter(#) option with # > 1 did not change the coefficients compared with when # was 1. This has been fixed.

5. stepwise incorrectly exited with an error when adding back a term that was previously removed for backward stepwise regression (when using options pr() and pe()). This has been fixed.

6. stepwise incorrectly exited with an error when performing forward hierarchical selection (when using options pe() and hierarchical). This has been fixed.

-------- update 05oct2005 -----------------------------------------------------

Ado-files

1. asmprobit with the cluster() option set e(wtype) to pweight causing an error in estat mfx after asmprobit. This has been fixed.

2. egen's rowtotal() function, when passed a variable list containing a wildcard followed by numbers, such as "*00", sometimes calculated an incorrect sum due to the inclusion of a temporary variable. This has been fixed.

3. svy brr with the fay() option produced incorrect variance estimates. This has been fixed.

4. tabodds with the or and cornfield options failed when the var_case (case variable) contained missing values. This has been fixed.

-------- update 19sep2005 -----------------------------------------------------

Ado-files

1. codebook has new option dots for use with the compact option. dots displays a dot for each variable processed before displaying the compact codebook table. This will interest those working interactively with large datasets. Additionally, the compact option, introduced in the 15sep2005 update, displayed counts for string variables before displaying the correct compact codebook table. The spurious output is now suppressed.

2. graph export, when specified with an extremely small pixel size for the width() or height() options sometimes caused Stata to crash. Stata now limits the pixel size range to be between 8 and 16,000.

3. xtpoisson sometimes labeled the variables in the output incorrectly when variables were dropped from the estimation. This has been fixed.

-------- update 15sep2005 -----------------------------------------------------

What's new in release 9.1 (compared with release 9)

Highlights for release 9.1 include the following:

o Multiple log files may be opened simultaneously. See item 39 below.

o Survey linearization is now 2 to 100 times faster, the larger multiplier applying to large datasets with many sampling units. See item 46 below.

o The string length limit has been increased to 244 for Intercooled and Small Stata to match Stata/SE. See item 32 below.

o Up to 300 estimation results may be stored. See item 34 below.

o Mata has a new permutation function making it easier to program exact permutation tests. See item 41 D below and the examples in help mata cvpermute().

o Mata has several other new features, including new string, regular expression, and binary I/O functions. See item 41 below.

o Graphs exported to PNG or TIFF bitmap format may now have their sizes specified, allowing the resolution to be greater than screen resolution. See item 36 below.

o There is a new command for computing marginal effects after an alternative-specific multinomial probit model. See item 3 B below.

o There is now a console version of Stata/SE for the Mac. See item 63 below.

o Collecting statistics across a by list using the statsby command is now allowed in more situations, including with postestimation commands. See item 21 below.

o Fay's adjustment is now allowed with balanced repeated replications (BRR) for complex survey data. See item 25 A below.

We recommend that you set version to version 9.1 at the top of do-files and ado-files. version 9.1 improves the default behavior for permute and xtreg. See help version.

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 5(3).

2. Some maximum likelihood estimation commands (such as arima, nbreg, streg, and xtnbreg) ignored constraints when there were no predictors in the model. This has been fixed.

3. asmprobit has the following changes.

A. asmprobit has a new default covariance parameterization and a new option, structural. The new default parameterization estimates the covariance of the alternatives differenced with the base alternative. For a J-alternative model, the differenced covariance is a J-1 x J-1 matrix with J(J-1)/2 - 1 estimable parameters, where the variance of the latent errors for the scale alternative minus the base alternative is fixed to 2.

structural specifies that the structural J x J covariance parameterization be used, which sets J-1 of the correlations to zero and fixes the base and scale variances to 1.

B. estat mfx is a new postestimation command for computing marginal effects after asmprobit. The Postestimation / Marginal effects or elasticities menu displays the estat mfx dialog after an asmprobit has been fitted.

C. When called with pweights, asmprobit incorrectly stored iweights instead of pweights in e(wtype). This has been fixed.

D. asmprobit did not allow the constant-only model. Now it does.

E. Computing the likelihood for cases with only two alternatives had the alternative probabilities reversed. If all cases had two alternatives, the signs of the estimated coefficients were reversed. If only some cases had two alternatives, estimates were inaccurate. This has been fixed.

F. predict after asmprobit mislabeled the predictions when one or more alternatives were excluded throughout the prediction dataset. This has been fixed.

4. biplot with the option separate by default always places both subgraphs on common x- and y-scales. Previously, the scales sometimes differed.

5. bootstrap exited with an error, complaining about time-series operators in the expression when an expression contained a numeric literal with a decimal. This has been fixed.

6. bootstrap and jackknife exited with an obscure error message when used with xtmixed. This has been fixed.

7. codebook has new option compact that produces a compact codebook with each line showing a variable name, number of observations, number of unique observations, mean, minimum, maximum, and variable label.

8. The collapse dialog has been improved.

9. estimates table with a long namelist labeled only the first few columns of estimation results. This has been fixed.

10. fracpoly and mfp, when used with glm, ran an older version of glm. This has been fixed.

11. graph when the range of values in the y- or x-dimension was very close to machine precision, but not exactly 0, sometimes entered an infinite loop, requiring that processing be manually broken. This has been fixed.

12. graph combine with options ycommon and xcommon in rare cases did not produce combined graphs with the intended common axes. This has been fixed.

13. graph set did not recognize the PICT format for setting graphics exporting options. This has been fixed.

14. hausman now supports the existing sigmaless and new sigmamore options after xtreg. These options are particularly useful when comparing fixed- and random-effects regressions because they improve results in small to moderate samples by ensuring that the difference covariance matrix will be positive definite.

15. svy jackknife did not correctly count the number of observations in the estimation sample when the sampling weight variable contained zeros. This has been fixed.

16. logit, probit, and dprobit allowed aweights, even though they were not documented and the resulting calculation had no meaningful interpretation. When version is set higher than 9.0, these commands now issue an error message when used with aweights.

17. The matrix input dialog now allows users to enter symmetric matrices by filling in the lower triangle.

18. mean, proportion, ratio, and total exited with a syntax error when supplied with the over() option with more than 400 subpopulations. This has been fixed.

19. mprobit has new option probitparam that scales the coefficients in the same way as probit. Also, mprobit coefficients had reversed signs when there were two alternatives (a probit model). This has been fixed.

20. permute now uses two random uniform variables to generate Monte Carlo permutations of the values of the permute variable. permute must be called under version control in order to reproduce old results.

21. statsby has several enhancements.

A. statsby now works with postestimation commands.

B. statsby now works with commands that support neither if nor in (although it will be slower with such commands).

C. statsby has a new option forcedrop that forces all data to be dropped except the group samples. This allows statsby to work with user-written commands that completely ignore if and in qualifiers.

D. statsby no longer clears estimation results when the command being run is not an estimation command.

E. statsby is now faster when used with most time-series datasets.

22. stci exited with an error message when one or more twoway_options were supplied, even though the emean and graph options were specified. This has been fixed.

23. stcurve exited with an error message when options at(), at1(), and at10() were used together. This has been fixed.

24. stepwise produced a "variable not found" error message when a variable name was an abbreviation of another variable name found earlier in the variable list. This has been fixed. Also, sw with the Stata 8 syntax sometimes incorrectly exited with an error message if the estimation command dropped a variable due to nonestimability. This has been fixed.

25. svy has several improvements.

A. svy brr now allows the fay() option when using replication weight variables. svy brr also saves the value of the Fay's adjustment in e(fay).

B. svy: tabulate now accepts string variables.

C. svy did not correctly set e(sample) when there were missing values in the independent variables. This has been fixed.

D. svy: mlogit, when supplied with a dependent variable with value labels containing dots (.) or colons (:), incorrectly exited with an error message. This has been fixed.

E. svy: proportion, when used with variables having value labels with the right quote character ("'"), incorrectly exited with an error message. This has been fixed.

F. svy: tabulate exited with an unhelpful error message when mata did not have enough memory to perform the necessary calculations. A better error message is now displayed.

G. svy: tabulate oneway, when used with the missing option, dropped the correct "Total" margin and labeled the row for the missing value as the "Total". This has been fixed.

H. Version 8 survey estimation commands (svyregress, svylogit, ...) now provide a better error message when used with data that are svyset under version 9.

I. svyset did not return the name of the sampling weight variable in r(wvar) as documented in [SVY] svyset. This has been fixed.

26. tsline sometimes incorrectly exited with an error when the tlabel() option was supplied with integers but no date strings. This has been fixed.

27. varwle displayed missing values for some tests when the lags were not consecutive. This has been fixed.

28. xi ignored the noomit option when it was called with terms after the comma. This has been fixed.

29. xtmixed, when used with the maximize option iterate(), did not issue the appropriate warning at the bottom of the output if the model failed to converge. This has been fixed.

30. xtreg has the following improvements:

A. xtreg, fe adjusted the VCE for the within transform when cluster() was specified. The cluster-robust VCE no longer adjusts for the within transform unless the new dfadj option is specified.

B. xtreg, fe and xtreg, re produced cluster-robust VCEs when the panels were not nested within the clusters. In some cases this VCE is consistent, and in others it is not. You must now specify the new nonest option to get a cluster-robust VCE when the panels are not nested within the clusters.

C. xtreg, re reported different between R-squared values from one run to another on identical datasets. This has been fixed.

31. xtregar, fe mislabeled variables when variables from the middle of the list of indepvars were dropped due to collinearity. This has been fixed.

Stata executable, all platforms

32. Intercooled and Small Stata now have the same maximum string length (244) as Stata/SE. See limits.

33. encode encoded in reverse order for some platforms. This has been fixed.

34. estimates store and _estimates hold now allow up to 300 estimation results to be stored. The previous limit was 20.

35. gettoken's bind option did not work properly if the first character of the string to be parsed was "(" or "[". This has been fixed.

36. graph export has new options width() and height() for specifying the width and height of a graph in pixels when exporting to PNG and TIFF bitmap formats. If only the width or height is specified, Stata exports the graph with the requested dimension while maintaining the correct aspect ratio.

37. include is a new command similar to do and run for executing do-files, but it differs in that any local macros, changed settings, etc., created by executing the file are not dropped or reset when execution concludes. Rather, results are just as if the commands appeared in the session or file that included the file.

38. itrim() is a new string function that replaces multiple, consecutive, internal spaces with a single space in strings.

39. log can now start multiple log files simultaneously. See log for full details.

40. logistic, logit, and probit, when used with the svy prefix or with options vce(bootstrap) or vce(jackknife), had the following fixes:

A. They did not display a note indicating when there were completely determined successes or failures. This has been fixed.

B. They did not display a note indicating the presence of perfect predictors. This has been fixed.

C. They did not correctly report on variables that were dropped due to collinearity. This has been fixed.

41. Mata has several new functions:

A. bufput() and bufget() assist in performing binary-formatted I/O.

B. byteorder() is equivalent to Stata's byteorder() function.

C. Existing function cat() now takes optional second and third arguments specifying the beginning and ending lines of the file to read.

D. cvpermute() returns all permutations of the values of a column vector V, one at a time. If V=(1\2\3), there are six permutations: (1\2\3), (1\3\2), (2\1\3), (2\3\1), (3\1\2), and (3\2\1). If V=(1\2\1), there are three permutations: (1\1\2), (1\2\1), and (2\1\1).

E. regexm(), regexr(), and regexs() are regular expression functions corresponding to the Stata functions with the same name.

F. stataversion() returns the version of Stata that you are running. statasetversion() allows setting the version.

G. stritrim() replaces multiple, consecutive, internal spaces with a single space in strings.

H. strtoreal() converts strings to numeric values.

42. Mata function cat() produced an extra empty line in the rare case when the input stream had "\r\n" and the "\r" landed at the end of the internal buffer. This has been fixed.

43. ologit and oprobit, when used with the svy prefix or with options vce(bootstrap) or vce(jackknife), did not display a note indicating when there were completely determined observations. This has been fixed.

44. print @Graph with no Graph window open sometimes caused Stata to crash. This has been fixed.

45. print filename.gph, where filename.gph was not a Stata format graph file or was a live-format Stata graph file, sometimes caused Stata to crash. This has been fixed.

46. svy linearization is now faster, especially for large datasets with many sampling units.

47. xmluse with the firstrow option, when reading Excel datasets where the second row did not have sufficient type information, resulted in an error message saying that a variable couldn't be created because it already existed. This has been fixed.

48. xtabond produced consistent but inefficient estimates when there were gaps in the time series. This has been fixed.

49. The overlaid twoway graph dialog has a new tab Y-Axis (2) for affecting the rendition of a second y-axis.

50. Axis tabs on all graph dialogs produced incorrect option arguments for grid-line patterns. This has been fixed.

Stata executable, Windows

51. set doublebuffer is a new command that enables or disables double-buffering in the Results, Viewer, and Data Editor windows. Disabling double-buffering may be necessary if output from the Results window appears very slowly.

52. (64-bit versions of Windows with a 64-bit version of Stata) odbc loaded only a subset of the requested data. This has been fixed.

53. permute, when called with the using option with an absolute path in Windows (as in C:\mystuff\outperm.dta), issued an error message instead of replaying the results stored in the supplied dataset. This has been fixed.

54. If profile.do was located on a UNC path (that is, \\computername\path...), Stata prepended an extra backslash (that is, \\\computername\path...) and caused an error. This has been fixed.

55. On certain Windows versions, most notably Windows 98, opening the Viewer could crash Stata. This has been fixed.

56. Tooltips have been added to some items in the Preferences dialog to aid in readability.

57. The Variables Properties dialog from the Data Editor sometimes did not select the correct value label for the current variable. This has been fixed.

58. The Do-file Editor was incorrectly staggered from the correct position when it was the initial editor opened. This has been fixed.

59. A runtime exception sometimes occurred when the Search dialog was opened. The exception was never critical and always allowed for execution to continue. The cause of the exception has been fixed.

60. The initial dimensions of the Results window were sometimes too small when using the initial, default settings. The Results window was not fully utilizing the available space. This has been fixed.

61. Runtime assertion errors could occur when creating Viewers, Graphs, or Do-File editors. These have been resolved.

62. When a variable control in a dialog has the focus, subsequent clicks inside the Variables window will send variables to the dialog control, as was the case in version 8.

Stata executable, Mac

63. There is now a console version of Stata/SE for the Mac. See http://www.stata.com/support/mac/ for more information.

64. Stata now attempts to resolve aliases to the Stata directory and the Stata executable upon startup.

65. The mouse pointer did not change to a hand when the pointer was over a link. This has been fixed.

66. If you were running Stata from Mac OS 10.4 (Tiger), some lines would appear bold when you made a selection from the Results window or Viewer. This has been fixed.

67. A keyboard shortcut for File > Save As... has been added.

-------- update 05jul2005 -----------------------------------------------------

Ado-files

1. aorder, when used on a variable list containing variable names with certain combinations of the letters 'c', 'd', 'e', or 'f', could exit with a 'system limit exceeded' error. This has been fixed.

2. icd9 and icd9p have been updated to use the V22 codes; V21, V19, V18, and V16 codes were previously used. V16, V18, V19, V21, and V22 codes have been merged so that icd9 and icd9p work equally well with old and new datasets. See help icd9 for a description of icd9 and icd9p; type "icd9 query" and "icd9p query" for a complete description of the changes to the codes used.

3. ml display now behaves as it did in Stata 8 when called under version control.

4. suest allowed estimation results from ivreg when it should have produced an error message. This has been fixed.

5. split, when passed parse("") as an option, failed with an error. split now understands that parse("") is equivalent to parse().

6. svy sometimes identified a larger estimation sample in e(sample) than was actually used. This has been fixed.

7. xtgee, family(binomial) is now equivalent to xtgee, family(binomial 1) for nonbinary dependent variables as documented in the manual.

Stata executable, all platforms

8. Find in the Viewer has been improved. Find is no longer performed through a modal dialog but instead is performed from a Find bar that is located above the Viewer's status bar. In addition to finding a search term and selecting it, the Find bar can also highlight all instances of the search term. Every Viewer window has its own Find bar. To show or hide a Viewer's Find bar, click its Find button, press Ctrl-F (Mac users press Cmd-F), or press F3.

Stata always starts a find operation from the current selection. If there is no selection, Stata starts at the beginning of the document. To perform a find at the beginning of a document when there is text selected, click on the Viewer content area to clear the selection before performing a find, or press Ctrl-F3. If the Find bar's edit field has the keyboard focus, you may also press Ctrl-Enter to do the same.

The following keyboard shortcuts are available when the Find bar is visible.

Shortcut Keyboard Focus Action -------------------------------------------------------------------- Ctrl-F Any part of Viewer Show/hide Find bar Return Find bar edit field Find next Shift-Return Find bar edit field Find previous Ctrl-Return Find bar edit field Find from beginning of document Ctrl-Shift-Return Any part of Viewer Find from end of document F3 Any part of Viewer Find next Shift-F3 Any part of Viewer Find previous Ctrl-F3 Any part of Viewer Find from beginning of document Ctrl-Shift-F3 Any part of Viewer Find from end of document Ctrl-G Any part of Viewer Find next Ctrl-Shift-G Any part of Viewer Find previous --------------------------------------------------------------------

Note: Mac users should press the Cmd key instead of the Ctrl key.

9. The dialog programming documentation has been updated for Stata 9. In addition, several bugs that were present in the dialog system have been fixed.

A. The stata command did not echo the command in the Results window as documented. This has been fixed.

B. Spaces that preceded text in a put command were not always handled properly. This has been fixed.

C. target(stata hidden) was not accepted with u-action buttons as documented. This has been fixed.

D. The require command did not work with heavyweight i-actions. This has been fixed.

E. The VERSION statement did not work correctly when multiple operating systems were specified. This has been fixed.

F. onchange(i-action) did not work with BROWSE controls. This has been fixed.

10. Do-files that generate a graph and later produce an error, only showed the error message but did not produce the error code. This has been fixed.

11. Time-series graphs were limited to 100 tick marks. This has been fixed.

12. ktau computations are now faster.

13. Mata's object files (.mo and .mlib files; see [M-1] how) are now stored a little differently. This means that if you have any personal .mo or .mlib files, you must rebuild them from your source code; see [M-3] mata mosave and [M-3] mata mlib. Because this change is being made so close to the original release date, it is not being made under version control.

14. Mata could mistakenly issue a subscript-out-of-range error for the statement d[i]=x when i was a vector. Mata did this when any element of i exceeded length(x) instead of when any element exceeded length(d). This has been fixed.

15. Mata produced a stack-overflow error, r(3998), in a rare case. This has been fixed.

16. quadchk has new option nofrom. By default quadchk starts from the previous estimation results. Adaptive quadrature is sensitive to starting data, so specifying nofrom will often get slightly closer results.

17. After set varabbrev off an expression using _b[some_abbreviated_varname] worked when it should not have. This has been fixed.

18. xtlogit, re, xtprobit, re, xtcloglog, re, xtpoisson, re normal, xtintreg, and xttobit now allow up to 195 integration points in the intpoints() option.

19. xtlogit, xtprobit, xtcloglog, xtpoisson, xtintreg, and xttobit when specified with the intmethod(ghermite) option now allow the constraints() option for specifying constraints.

20. The set level extended macro function (documented in Stata 7 and earlier versions) failed to work in Stata 9. This has been fixed.

Stata executable, Windows

21. cd to a UNC path (that is, cd \\computername\path...) returned c(pwd) as an empty string rather than the correct path. This has been fixed.

22. In Stata's various file dialogs, such as File--Open, if a user attempted to open or save a file across a UNC path (that is, \\computername\path...), Stata would prepend an extra backslash (that is, \\\computername\path...) and cause an error. This has been fixed.

23. haver failed to load datasets that contained more than 2048 variables. This has been fixed. Now, the maximum is determined by the maxvar setting.

24. Stata for Windows incorrectly enumerated multiple instances of Stata. This has been fixed.

25. The main Stata window remained hidden when launched until the commands in profile.do completed. This has been fixed.

26. When a Viewer was launched upon startup of Stata (whether through shortcuts, double-clicking smcl files, etc.) the Viewer never appeared. This has been fixed.

27. Graph window positions were not properly loaded from a named preference set. This has been fixed.

28. Copy (Ctrl-C) and Print (Ctrl-P) shortcuts failed to work in the Graph window. This has been fixed.

29. Resizing a floating Results window could result in the output not matching the dimensions of the window. This has been fixed.

30. The Results window, Viewer, and Data Editor now double-buffer their contents to prevent them from flickering when resized or redrawn.

31. When the Variables or Review windows were clicked and text was sent to the Command window, the right-click menu did not properly allow for Cut/Undo/Copy operations. This has been fixed.

Stata executable, Mac

32. There is a new option to draw smooth fonts with QuickDraw and the Quartz graphics engine rather than drawing smooth fonts using Quartz directly. This allows for slightly faster, yet potentially less accurate, rendering of text output. To enable or disable the option, check or uncheck the Draw text using QuickDraw checkbox from the General Preferences dialog or set use_qd_text {on | off}; see set use_qd_text. When this option is enabled, the minimum font size for smoothing is read from the System Preferences, not Stata's preferences. This enhancement requires Mac OS X 10.2 or higher.

33. Some of Stata's menus that contain hierarchical menus do not display an arrow to indicate that a menu item contains a hierarchical menu while running Stata in Mac OS X 10.4 (Tiger). This is an issue in Tiger, not in Stata. Apple is aware of this and has indicated that it will be fixed in a future software update to Tiger.

34. Due to a change in how the default PATH environment variable is defined in Tiger from previous versions of Mac OS X, Stata now searches the user's home directory regardless of whether it is defined in the PATH variable or not. This allows Stata to determine if a profile.do file exists in the user's home directory when running Stata on Tiger.

35. In Mac OS X 10.4.x (Tiger), Stata files could not be dropped on Stata's Dock icon so that they will open in Stata. This has been fixed.

36. Setting the y resolution of graphs copied to the Clipboard using the Graph Preferences dialog had no effect. This has been fixed.

-------- update 21jun2005 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 5(2).

2. adjust failed when an if condition involved literal strings. This has been fixed.

3. ml now documents that ml display uses e(k_eform) to determine how many equations to exponentiate when supplied with an eform_option.

4. predict with the pr option, when used after probit with the svy prefix or the vce(bootstrap) or vce(jackknife) options, did not produce the predicted probability values. This has been fixed.

-------- update 17jun2005 -----------------------------------------------------

Ado-files

1. areg failed when attempting to fit a constant-only model. This has been fixed.

2. asmprobit uses either clogit or mlogit for initial estimates. These estimates are now scaled by the standard deviation of the extreme value distribution, c(pi)/sqrt(6).

3. asmprobit does not attempt to refine the standard deviation/correlation estimates if the first call to ml (using the Cholesky factored variance/covariance parameterization) does not converge. In this case, a second call to ml is made to reparameterize the estimates, but iterate(0) is used to prevent another nonconvergent search.

4. asmprobit, mprobit, and slogit used inappropriate error codes. This has been fixed.

5. biplot now allows plotting of any two dimensions, not just the first two, with the new dim() option. biplot also now allows you to plot negative column (variable) arrows with the new negcol option. Graph options for the negative column arrows are specified with the new negcolopts() option. New options norow and nocolumn suppress the row points or column arrows.

6. nlogit now allows groups to have different sets of alternatives, so unbalanced groups are now allowed. nlogit now saves e(alt_min), e(alt_max), and e(alt_avg) providing the minimum, maximum, and average number of alternatives per group.

7. cluster measures complained if a measure was not provided. It now defaults to using the L2 measure, as documented.

8. dotplot, when supplied with an if or in restriction, produced graphs with an incorrect range for the frequency axis. This has been fixed.

9. _get_gropts, a programmer's command, now recognizes the altshrink option as a graph combine option when the getcombine option is specified.

10. newey failed if a model contained variables with time-series operators and one of those variables was dropped due to multicollinearity. This has been fixed.

11. newey, like regress, showed "(dropped)" in the table of coefficients for a dropped variable. Now, like most other commands, it displays "note: variable dropped due to collinearity" at the top of the output.

12. predict with the pr option, when used after logistic, logit, or probit with the svy prefix or the vce(bootstrap) or vce(jackknife) options, produced values of the linear prediction instead of the predicted probability. This has been fixed.

13. predict varlist, reffects after xtmixed now produces an appropriate error message and exits with an error code if any of the variables in varlist already exist.

14. svy estimation commands, when string variables were used to identify strata or sampling units, incorrectly counted the corresponding number of strata or sampling units resulting in incorrect variance estimates. This has been fixed.

15. svyopts did not work as advertised in the Stata Press book, Maximum Likelihood Estimation with Stata, 2nd edition (Gould, Pitblado, and Sribney 2003). This has been fixed.

Programmers interested in getting their estimation commands to work with the new svy prefix command should see the Remarks section of program properties.

16. svy: regress and svy: ivreg exited with an inappropriate error message when the sampling weights summed to a number less than 1. Although this has been fixed, we note that the sum of the sampling weights should sum to the population size (or at least estimate the population size) for the DEFF design effect to be valid when using an FPC in the first stage. DEFT is invariant to the scale of sampling weights.

-------- update 25may2005 -----------------------------------------------------

Ado-files

1. asmprobit help now documents that the options basealternative() and scalealternative() may contain a number, a label value, or a string.

2. biplot did not label row markers properly when the mlabel() option was specified and a subsample of the data was used. This has been fixed.

3. cabiplot and caprojection now allow row and column marker labels to be specified. This is accomplished with the mlabel() suboption of rowopts() and colopts().

4. estimates replay failed when called under version 8.2 version control. This has been fixed.

5. glm, mdslong, merge, predict after mds, and rolling failed when the temporary file directory path contained spaces. This has been fixed.

6. graph bar and graph dot, after the 04may2005 update, displayed an extra set of bars when the sort suboption was specified with the over() option. This has been fixed.

7. graph twoway error messages have been improved when too many or too few variables are specified for a plot.

8. icd9[p] check with the generate() option did not generate a variable if all codes were valid. This has been fixed.

9. mprobit and slogit help now document that the option baseoutcome() may contain a number or a label value.

10. nl when used with the pre-version-9 syntax accepted but ignored the robust, cluster(), hc2, and hc3 options. Now specifying one of those options with the old syntax will result in an error message.

11. predict failed after asmprobit with a string alternatives() variable. This has been fixed.

12. predict after xtreg, fe produced the linear prediction when options such as residuals or cooksd were specified. An error message is now issued instead.

13. stepwise incorrectly exited with an error for backward stepwise regression (when using options pr() and pe()). This has been fixed.

14. xtmixed sometimes failed when given an if condition that used value labels, for example, if foreign == "Domestic":origin. This has been fixed.

15. xttobit erroneously treated observations for which the dependent variable was missing as being right-censored instead of ignoring them. This has been fixed.

-------- update 04may2005 -----------------------------------------------------

Ado-files

1. asmprobit models with unstructured covariances produced an error message during the reparameterization phase of the ml optimization if the model definition for the variance/correlation parameters exceeded 80 characters. This has been fixed.

2. bootstrap has a group() option, which was not documented in the manual. See bootstrap for the updated online documentation.

3. clogit performed an extra, but unnecessary, calculation, which caused it to run slowly on large datasets. This has been fixed.

4. clogit uses logit to generate starting values when none are specified. For some models, these starting values resulted in nonsensical clogit log-likelihoods. clogit now checks that starting values make sense before using them.

5. graph bar and graph dot no longer require user-provided names when a variable is repeated with more than one statistic. The dialogs for these commands have been redesigned to take advantage of the new, simplified syntax.

6. When managing multiple graphs or renaming graphs when used from disk, Stata sometimes displayed the wrong graph in a Graph window. This has been fixed.

7. pause executed subsequent commands under version 6 control. It now executes them under the control of the caller's version.

8. roctab produced a different standard error for the area under the curve when a monotonic transformation was applied to the rating. This has been fixed.

9. screeplot has been enhanced to produce scree plots automatically after manova, canon, ca, and camat.

10. stsum sometimes reported percentiles as the next ordered survival time when the desired fraction exactly equaled the empirically observed fraction. This would be similar to saying that the median of (1,2,5,7,9) is 7 rather than 5. stsum rarely made this mistake. This problem has been fixed.

11. svy dropped observations from the estimation sample that had zero as the value of the sampling weight. This has been fixed.

12. xtreg, be and xtreg, fe reported a missing value for the model F statistic when the vce(bootstrap) option was specified. They now report the correct model Wald chi-squared statistic.

13. xtreg, re reported a missing value for the model chi-squared statistic when the vce(jackknife) option was specified. xtreg, re now reports the correct model F statistic.

-------- previous updates -----------------------------------------------------

See whatsnew8to9.

-------------------------------------------------------------------------------


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index