Stata 15 help for whatsnew11

Additions made to Stata during version 11

This file records the additions and fixes made to Stata during the 11.0, 11.1, and 11.2 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 | | this file 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 | | whatsnew9 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 whatsnew11to12.

-------- update 01sep2011 -----------------------------------------------------

Ado-files

1. discrim knn when the values of groupvar (specified in option group()) were not contiguous for a group (that is, for example, when all observations for group 1 were not together, all observations for group 2 were not together, etc.) could give incorrect results. This has been fixed.

2. margins was reporting all margins, even the estimable ones, as "(not estimable)" when factor variables were specified in models fit using slogit or stcrreg. This has been fixed.

3. Mata's optimize() with the Nelder-Mead technique and constraints would error-out with a Mata trace log. This has been fixed.

4. sspace when time-series operators were combined with the e. error operator on the dependent variable in an obs_efeq equation produced an error message. This has been fixed.

5. svy now calls option validator programs for user-written estimation commands; see svy_parsing.

6. svy brr with an expression would sometimes report an error about a working variable not found. This has been fixed.

7. svy ivregress reported an invalid-operator error when factor-variables notation was used in the list of instrumental variables varlist_iv but not in the list of exogenous variables varlist1. This has been fixed.

8. var with 10 or more lags of exogenous variables failed to sort the lags in the right order. This has been fixed.

Stata executable, all platforms

9. anova and manova specified with a termlist longer than 503 characters would typically report a variable not found error. This has been fixed.

10. Mata functions fget() and fgetnl(), in the case of files with Windows-style end-of-line characters (\r\n) when a line had a length of a multiple of 512 characters, minus 1, could return an extra blank line upon a repeated call to one of these functions. This has been fixed.

11. The maximum length for the Mata library path was 256 characters. It has been increased to 4096 characters.

12. set memory with memory allocations between 105 GB and 208 GB could cause Stata to crash. This has been fixed.

13. stcox using options efron and vce(robust) could take a long time to compute for large datasets with many tied failure times. This has been fixed.

14. xmlsave might crash Stata if the variables to be saved were sorted. This has been fixed.

15. The text limit of VARLIST controls in dialog boxes has been increased to 2048 characters.

Stata executable, Windows

16. graph text with a vertical alignment of middle or top had superscripts and subscripts shifted slightly vertically. This has been fixed.

Stata executable, Mac

17. How Stata displays its main windows after it has been launched has changed because of changes in behavior in Mac OS X 10.7 (Lion) from earlier versions of Mac OS X.

18. Stata's file dialogs now support opening symbolic links to files or directories.

19. Changes to Mac OS X 10.6.7 could cause printing from the Do-file Editor to crash. This has been fixed.

20. A change in Mac OS X 10.7 (Lion) caused the cursor to disappear when the cursor was on a line that only contained a carriage return in the Do-file Editor. This has been fixed.

21. c(mode) creturn value has been added to the console version of Stata.

22. When printing from the Results window or the Viewer window, the last line would not be printed. This has been fixed.

Stata executable, Unix

23. graph text with a vertical alignment of middle or top had superscripts and subscripts shifted slightly vertically. This has been fixed.

24. c(mode) creturn value has been added to the console version of Stata.

-------- update 19jul2011 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 11(2).

2. estat concordance was not allowed after stcox with bootstrap or jackknife adjusted standard errors. This has been fixed.

3. estat endogenous after ivregress sometimes reported an error when factor variables were specified. This has been fixed.

4. gmm with a string cluster variable produced incorrect standard errors. This has been fixed.

5. svy jackknife now reports a note when subpopulation observations were dropped because of missing values in the model's variables. This note explains why svy jackknife may report that there are insufficient observations to compute standard errors, and gives some advice on how this might be fixed.

6. margins was incorrectly allowing option vce(unconditional) with svy bootstrap and svy sdr. The standard errors in this case were not computed using either bootstrap or SDR methods but were linearized based on the currently svyset design. margins with option vce(unconditional) will now report an error with svyset data if the default vce() is different from linearized or if the currently estimated standard errors are not linearized.

7. mca did not allow a varlist with notation * or _all to represent all variables in the current dataset. This has been fixed.

8. mi estimate now checks for structural variables (variables declared by stset, svyset, tsset, or xtset) that are registered as passive or imputed and issues an error message if there are any.

9. mi import flongsep did not respect spaces in folder names specified within option using(). This has been fixed.

10. mi impute has the following fixes:

a. mi impute ologit failed to detect cases when the simulated posterior estimates of the cutpoints did not satisfy the monotonicity constraint. This could result in the middle categories of the imputation variable being underrepresented. Such situations could arise in the presence of factor variables among the independent variables. This has been fixed.

b. mi impute regress and mi impute pmm with analytic weights produced imputed values with dispersion smaller that it should have been. This has been fixed.

c. mi impute now treats missing frequency weights as ones rather than zeros when displaying the numbers of complete, incomplete, imputed, and total observations. This does not affect the imputation procedure, only the displayed numbers of observations in the imputation table.

11. predict after mprobit did not allow option equation() as a synonym for outcome(). This has been fixed.

12. proportion with the svy brr or svy jackknife prefix and option subpop() that excluded one or more of the categories of a variable in the varlist reported an "insufficient observations to compute jackknife standard errors" error. This has been fixed.

13. proportion, ratio, mean, and total return a better error message when there are no observations.

14. robvar did not allow string variables in option group(). This has been fixed.

15. smooth gave an uninformative error message when no smoother was specified. The message has been improved.

16. svy: tabulate with option ci and bootstrap or SDR standard errors was reporting confidence intervals that were too narrow for the specified level(). This has been fixed.

17. svy: tabulate with option stdize() and one of the replication methods for variance estimation would incorrectly report a syntax error that option stdize() was not allowed. This has been fixed.

18. truncreg did not respect frequency weights when computing the number of observations before truncation, e(N_bf). This has been fixed.

19. xtpoisson with options re and normal did not estimate the constant-only model when option noskip was specified. This has been fixed.

20. xtreg, re incorrectly reported that the GLS estimates assumed a Gaussian distribution for the random effects. This has been fixed.

21. xttest0 reported test statistics to be from a one-degree-of-freedom chi-squared distribution when they were actually from a one-degree-of-freedom 50:50 mixture of a chi-squared distribution with zero. This has been fixed.

-------- update 30mar2011 (Stata version 11.2) --------------------------------

Ado-files

1. bstat ignored if and in restrictions. This has been fixed.

2. In some cases, when text was added to graph bar, graph box, or graph dot graphs using option text() and that text was subsequently edited in the Graph Editor, graph use after graph save would fail with an error message. This has been fixed.

3. In the Graph Editor when using the Grid Edit tool, expanding or contracting cells using "Above 1 cell" could fail to perform the expand or contract operation. This has been fixed.

4. merge, if the master dataset contained a temporary variable, could rarely report that no observations were merged even though they were. This has been fixed.

5. merge with option keep(), in rare situations, could report that it needed more memory to complete the merge than was actually necessary. This has been fixed.

6. mi impute has the following fixes:

a. mi impute pmm would sometimes fail to issue an error when missing imputed values were produced and would automatically proceed with imputation as if option force had been issued. This has been fixed.

b. Contrary to the documentation, mi impute regress and mi impute pmm assumed frequency weights by default rather than analytic weights. The implementation now matches the documentation; analytic weights are the default.

c. mi impute monotone, with the default specification, treated right-hand-side factor variables as continuous in the prediction equations. This has been fixed.

d. mi impute mvn failed with an error message when used with factor variables and an if condition. The error arose when factor variables contained fewer levels in the restricted sample than in the unrestricted sample. This has been fixed.

7. quadchk after xtpoisson with option exposure() resulted in a syntax error. This has been fixed.

8. roctab, when graphing with option specificity, did not sort the data correctly. This has been fixed.

9. svy with stset data now recognizes when it is performing subpopulation estimation with the stcox and streg models even when the subpopulation information was specified in svyset but not option subpop().

10. svy: stcox and svy: streg specified with an if or in condition changed the record selection information stored by stset when it should have left the original stset information unchanged. This has been fixed.

11. xtmixed ignored factor-variable specifications of the form "ibn." (no base level). This has been fixed.

Stata executable, all platforms

12. drawnorm produces different results from previously because of the changes in function rnormal() documented below. Prior results are reproduced under version control.

13. mi impute produces different results from previously because of the changes in function rnormal() documented below. Prior results are reproduced under version control. The statistical properties of the new results are no better or worse than the prior results.

14. rnormal(), the Gaussian random-number generation function in both Stata and Mata, now produces different, better values. Prior results are reproduced under version control.

rnormal() produced sequences that were insufficiently random for certain applications. After setting the seed, the sign of the first random number drawn was correlated with the sign of the first random number that would be drawn after setting a different seed; the sign of the second random number drawn was correlated with the sign of the second random number that would be drawn; and so on. Thus the sequence produced by rnormal() after set seed was not statistically independent from the sequence produced after another set seed command.

This lack of independence made no difference in the statistical quality of results when the seed was set only once, because the lack of independence did not arise. Setting the seed once is typical in many cases, including the running of simulations.

The correlation is of statistical concern when the seed is set more than once in the same problem.

Only the rnormal() function had this problem. None of Stata's other random-number functions, such as runiform(), rbeta(), etc., had this problem.

The problem is fixed, with the result that random-number sequences produced by rnormal() are now different. If you need to re-create previously produced results, use version control and specify a version prior to 11.2 when setting the random number seed with set seed.

15. Help for set seed now includes useful advice on how to set the seed and explains the difference between a random-number generator seed and its state as recorded in c(seed).

16. The way version control is handled for random-number generators has changed. Version control is now specified at the time command set seed is issued; the version in effect at the time the random-number generator (for example, rnormal()) is used is now irrelevant. The situation was previously the other way around.

Under the new scheme, typing

. set seed 123456789 . any_command

causes any_command to use the new, version 11.2 rnormal() function even if any_command is an ado-file itself containing explicit versioning for an earlier release. Thus existing ado-files need not be updated to benefit from the updated rnormal() function.

Similarly, if you wish to run any_command using the prior version of rnormal(), you may type

. version 11.1: set seed 123456789 . any_command

Even years from now, any_command will still use the 11.1 version of rnormal(), and it will do that even if any_command was written for a later release of Stata.

17. Programmers do not need to update their previously written ado-files because of the change in function rnormal(), with one exception. If the ado-file itself contains a set seed command, the ado-file should be updated to use the version in effect at the time the ado-file was called. To do this, early in the code, obtain the version of the caller. Later, use the caller's version when issuing command set seed:

program xxx version ...

syntax ... ... local callersversion = _caller() ... version `callersversion': set seed ... ... end

If set seed appears in a private subroutine of xxx, you must pass callersversion to the subroutine.

If set seed appears in another program that you did not write, execute that program under the caller's version:

program xxx version ...

syntax ... ... local callersversion = _caller() ... version `callersversion': mi impute ..., seed(...) ... end

18. New creturn result c(version_rng) records the version number currently in effect for random-number generators.

19. fvrevar with option tsonly failed to preserve the b(none). (abbreviated bn.) operator attached to factor variables. This has been fixed.

20. intreg with a constant-only specification reported a "too few variables specified" error. This has been fixed.

21. margins and other estimation commands that allow factor-variables notation crashed Stata when multiple duplicate variables were specified in a single interaction term. This has been fixed.

22. Mata did not allow Mata class scalar or matrix to be declared as external; a "sysfail 52" error was reported. This has been fixed.

23. Mata function st_addvar() added "str245" variables to Stata even though the maximum length string variable is str244. The variables were not really str245, but numeric. st_addvar() did this when called with a real scalar first argument equal to 245. The function now aborts with error in this case.

24. Mata function st_view() with factor variables in the j argument ignored the selectvar argument when filling in the levels of the factor variables, this led to the full set of levels being included instead of the levels present in the selected sample. This has been fixed.

25. ml and Mata's moptimize() with type e1, e2, lf1, and lf2 evaluators and matafavor set to speed sometimes resulted in a nonconvergent iteration log full of (backed up) messages. This specifically occurred when some of the equation-level score values were equal to zero, causing a bad gradient vector calculation. This has been fixed.

26. The graph produced by palette symbolpalette, when exported to PostScript, had the names of the square symbols shifted down slightly. This has been fixed.

27. mean, proportion, ratio, and total now report a missing value for the standard error when the effective sample size for the point estimate is based on a single value. Prior to this change, these commands would report a zero-valued standard error; this previous behavior will not be preserved under version control.

28. prais with any of the robust vce() options erroneously included the intercept in the model F statistic. This has been fixed.

29. regress with option hascons and option vce(robust) or vce(cluster ...) erroneously reported an omitted intercept. This has been fixed.

30. With Stata/MP, mlmatbysum produced wrong results if the global macro $ML_w was empty. This has been fixed.

31. The Data Editor has the following modifications and fixes:

a. Variables are no longer abbreviated in the location bar.

b. Single cell paste is now handled as a manual edit and thus issues a Stata command.

c. Copying and pasting string data from within the Data Editor could result in extra characters getting pasted if the source string was copied from a shorter-string type. This has been fixed.

Stata executable, Windows

32. Loading a saved window preference did not set the font in the Command window until Stata was restarted. This has been fixed.

Stata executable, Mac

33. Graphs exported as PDF files are now exported with increased resolution.

34. The point symbol was previously drawn as a 1/72" x 1/72" rectangle which is the minimum size of a screen pixel. Stata now draws the point symbol as a 1/216" x 1/216" rectangle when printing for better looking results.

35. The Do-file Editor has the following new features and fixes:

a. The Do-file Editor can now optionally display line numbers when printing text. Text is now line-wrapped to fit the printed page.

b. Syntax highlighting highlighted as a comment entire lines that started with the comment delimiter */. Only text up to and including the comment delimiter */ should be highlighted as a comment.

36. Dragging a variable or command from the Variables and Review windows, respectively, would insert the currently highlighted variable or command instead of the item that was dragged. This has been fixed. Dragging and dropping a command from the Review window to the Command window now inserts the command instead of replacing the contents of the Command window.

37. The Data Editor allowed text to be pasted while in browse mode. This has been fixed.

38. Stata now prompts you to save unsaved data when opening new data from the Open Recents menu.

39. Selecting File > Save on a modified graph that has been already saved no longer prompts the user for a filename. Stata now uses the saved filename.

40. Mac OS X 10.7 (Lion) might not include the font Monaco. When a Stata window's font setting includes a font that does not exist on the system, Stata will use the default Mac OS X fixed-width system font instead. You will not have to change the font settings yourself but this is why Stata might be using a different font the first time you run Stata on a Mac upgraded to Lion.

Stata executable, Unix

41. Saving a windowing preference using command window manage prefs save did not work unless you already had a windowing preference saved. This has been fixed.

-------- update 21mar2011 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 11(1).

-------- update 24feb2011 -----------------------------------------------------

Ado-files

1. estat summarize ignored the c. operator in interactions, treating all variables in an interaction as factor variables. This has been fixed.

2. margins with option vce(unconditional) with robust results from mlogit reported a conformability error. This has been fixed.

3. Mata's moptimize() with a gf0 evaluator, grouped data identified with the moptimize_init_by() setting, and weights reported a conformability error while taking numerical derivatives. This has been fixed.

4. When the eigenvector computation for Mata functions symeigensystemselecti() and symeigensystemselectr() did not converge, the functions incorrectly generated a "conformability error". This has been fixed.

5. mean, proportion, ratio, and total with two or more variables and more than 400 subpopulations implied by option over() incorrectly labeled the estimated summary statistics in the estimation table. The first 400 subpopulation estimates were correctly associated with the first variable; all remaining subpopulation estimates for the first variable were associated with the second variable. A similar mislabeling of subpopulation estimates happened to the second and third variables, if specified. This has been fixed.

6. ml with a gf0 evaluator, option group(), and weights reported a conformability error while taking numerical derivatives. This has been fixed.

7. predict with option working after glm produced working residuals that were incorrectly multiplied by the reciprocal of the derivative of the link function instead of just the derivative itself. This has been fixed.

8. svy: clogit using a replication method reported an incorrect population-size estimate. It reported the sum of the sampling weights from each observation instead of the sum of one weight value from each group. This has been fixed.

9. svy jackknife with a poststratification adjustment and subpopulation estimation reported a subpopulation size computed from the sum of unadjusted sampling weights instead of the poststratification adjusted sampling weights. This has been fixed.

-------- update 04jan2011 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 10(4).

2. arch, when used with capture, sometimes failed to return the error code of the model. This has been fixed.

3. estat imtest with option preserve reported an invalid syntax error when the model contained factor-variable notation. This has been fixed.

4. estat imtest with models containing interactions of factor variables with many base levels exited with an error, usually because it attempted to create too many temporary variables or exceeded the maximum matrix size. While this can still happen, estat imtest will now ignore all base-level factors and omitted variables in its internal calculations.

5. gsort sorted missing values (.a, ..., .z, and .) in ascending order when a descending sort was specified. This has been fixed.

6. Mata's moptimize_result_V_robust(), when used with evaluator type gf and a matrix set with moptimize_init_by(), reported the error "scores returned by evaluator not conformable with the dataset" even when the scores matrix was conformable with the number of groups. This has been fixed.

7. ml check with one of the gf evaluator types incorrectly failed in the first test, checking the value of the log-likelihood scalar. This has been fixed.

8. nl and nlsur in the unusual case where the same linear combination was specified more than once within a single equation estimated unique parameters for each specification of the combination. They now estimate a single set of parameters that are used everywhere the combination is specified.

9. nlogit issued an error when commas were used in the constraints numlist. This has been fixed.

10. pwcorr and spearman incorrectly reported a p-value when rho was missing. This has been fixed.

11. sspace issued an error for models with more than 100 parameters. This has been fixed.

12. svy: tabulate with option row (or column) and direct standardization where one of the standard strata does not intersect with a row (or column), instead of reporting the estimated row (or column) proportions, reported values that were scaled by the row (or column) total. Usually, the row (or column) proportions sum to 1, but in this case they cannot. Now svy: tabulate reports the row (or column) proportions as they were estimated, revealing where the sum of the row (or column) proportions do not add up to 1.

13. xttest0 reported p-values that did not account for the null hypothesis being on the boundary of the variance parameter space. This has been fixed.

-------- update 10nov2010 -----------------------------------------------------

Ado-files

1. areg with option vce(bootstrap) or vce(jackknife) stored a temporary variable name in e(datasignaturevars) instead of the variable specified in option absorb(). This has been fixed.

2. biprobit exited with an invalid syntax error when option iterate() was specified. This has been fixed.

3. Mata's optimize() now allows negative values to be set in optimize_init_nmsimplexdeltas().

4. tobit with robust variance estimates with a reduced rank did not update the e(rank) result. This has been fixed.

5. unzipfile exited with an error when the zipfilename contained a space. This has been fixed.

6. xtpoisson with option fe reported an "o." operator not allowed error when a time-series-operated variable was dropped because of collinearity. This has been fixed.

Stata executable, Mac

7. Line numbers in the Do-file Editor did not draw at the correct location. This has been fixed.

-------- update 04nov2010 -----------------------------------------------------

Ado-files

1. There is now a dialog for command clonevar.

2. estat mfx after asclogit would exit with an error when the alternative variable was numeric without value labels. This has been fixed.

3. graph, by() would exit with an error when any of the variables specified in option by() had value labels or string values that contained commas. This has been fixed.

4. lincom and nlcom now accept options cformat(), pformat(), and sformat(); see estimation options.

5. Mata's deriv() function with the verbose setting off still displayed notes when it encountered missing values, or flat or discontinuous regions. This has been fixed.

6. ml maximize ignored option negh. This has been fixed.

7. prais with option vce(robust), vce(cluster clustvar), vce(hc2), or vce(hc3) included the constant in the overall model F test. This has been fixed.

8. svy: tabulate with option level() reported the error "option level() not allowed" when the supplied value did not match the value in c(level). This has been fixed.

9. twoway lfit, twoway lfitci, twoway qfit, and twoway qfitci now support time-series operators on both yvar and xvar.

10. xtreg with factor variables and option sa reported a "variable not found" error when the panels were not balanced. This has been fixed.

Stata executable, all platforms

11. heckprob with missing values in the indepvars where depvar was missing or depvar_s was equal to zero would sometimes report the error "Hessian is not negative semidefinite". The missing values were incorrectly being used in calculating the Hessian matrix. This has been fixed.

12. Mata's moptimize() with the Nelder-Mead technique and evaluator types lf, lf0, lf1, and lf2 exited with an error reporting an incorrect number of arguments. This has been fixed.

13. Mata's moptimize_result_post() with the Nelder-Mead technique reported the error "matrix has missing values". Because the Nelder-Mead technique does not provide a variance matrix, e(V) will not be posted, with one exception: when constraints are specified, a zero matrix will be posted to e(V).

14. Mata's moptimize_util_vecsum() used with a constant only equation and weights could crash Stata. This has been fixed.

15. In rare cases, merge with option update could cause a crash. This has been fixed.

16. odbc has new option connectionstring(), which allows you to specify a connection string instead of a DSN to connect to a database. See odbc for more information on this option.

17. Using a subexpression in a regular expression could cause a crash. This has been fixed.

18. xtabond, xtdpd, and xtdpdsys, in the rare case where GMM-style instruments are created from a single lag of a variable, produced estimates that were based on an incorrect instrument matrix. This has been fixed.

Stata executable, Windows

19. When using the Do-file Editor to run a selection, and if running the selection caused a --more-- condition, the subsequent running of a selection from the Do-file Editor could cause Stata to generate the error message "Encountered a sharing violation while accessing xxxx", where xxxx was the name of a temporary file. This has been fixed.

Stata executable, Mac

20. The diamond or plus symbols would be incorrectly rendered in graphs drawn to the screen, exported, or printed. This has been fixed.

21. Truetype font support in graphs exported as EPS files has been enabled in both GUI and console versions. The console version now searches the standard Mac locations for fonts.

22. Resizing a checkbox groupbox control in a programmable dialog could result in the control being clipped. This has been fixed.

-------- update 30sep2010 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 10(3).

2. The icd9 databases have been updated to use the latest version (V28) of the ICD-9 codes, which will officially become effective on 1 October 2010.

3. ivregress exited with an error when option vce(cluster) contained a string variable. This has been fixed.

4. margins with unconditional variance estimates and survey subpopulation estimation results with some of the independent variables containing missing values in observations where the sampling weights were zero would report the error "unconditional standard errors derived assuming full estimation sample; indepvars dropped observations from the estimation sample". This has been fixed.

5. mean, proportion, ratio, total, and svy: tabulate would report an "option not allowed" error when one or more of the following settings were in effect:

set showemptycells off set showbaselevels all set showbaselevels off set showomitted off

This has been fixed.

6. ml has the following fixes:

a. Subcommand footnote was not recognized. This has been fixed.

b. Subcommand report reported a Mata trace log when used with an evaluator that did not compute the Hessian matrix. This has been fixed.

c. Option group() (and Mata's moptimize() with a group variable set using moptimize_init_by()) and a type gf evaluator exited with an error, usually resulting from a conformability error. This has been fixed.

d. With option noclear, the estimation sample identifier was not generated as a byte variable. This would result in a "type mismatch" error when mleval was called. This has been fixed.

7. nestreg, when used with intreg, exited with an error instead of performing as expected. This has been fixed.

8. robvar was treating observations with missing values in option by(groupvar) as another group when it should have excluded those observations from the analysis. This has been fixed.

9. svy: tabulate with option deff or deft and either option row or option column would compute the marginal SRS variances incorrectly, resulting in incorrect DEFF and DEFT values in the row or column margins. This has been fixed.

10. The use with options dialog box now allows you to select variables from the dataset on disk that you want to load into Stata.

11. _vce_parse marked out all observations if vce(cluster clustvar) was specified with clustvar being a string variable. This has been fixed.

12. xtmixed, xtmelogit, and xtmepoisson, when used with prefix commands such as statsby and bootstrap, issued a syntax error if options were specified within more than one model equation. This has been fixed.

13. xtsum would terminate with an error when string variables were included in the variable list. This has been fixed.

Stata executable, all platforms

14. anova and manova with unbalanced data and interaction terms with empty cells could fail to properly fit the specified model, yielding incorrectly low values for sums of squares and their corresponding degrees of freedom. This has been fixed.

15. cluster measures now accepts option Gower.

16. In the format dialog box, one of the sample quarterly formats was incorrect. This has been fixed.

Stata executable, Windows

17. Scrolling in the Viewer window did not work correctly when viewing very large log files. This has been fixed.

Stata executable, Mac

18. The currently selected variable in the Variables Manager is now scrolled into view if necessary when using the arrow button to navigate through the list of variables.

19. Programmable dialogs on the Mac did not support resizing of controls. This has been fixed.

Stata executable, Unix

20. Using a dialog box varlist control on some 32-bit Ubuntu machines caused Stata to crash. This has been fixed.

-------- update 01sep2010 -----------------------------------------------------

Ado-files

1. estat phtest with option detail, after stcox with robust/clustered standard errors, produced rho test statistics and p-values that did not take into account that robust standard errors were used in the original model. Predictions of scaled Schoenfeld residuals, after stcox with robust/clustered standard errors, were similarly affected. Both issues have been fixed.

2. graph twoway, in the extremely rare case where a variable being graphed took on only a single value and when that same value was specified as the single label in option xlabel() or ylabel(), exited with an error and did not draw the graph. This has been fixed.

3. margins now reports a warning when option expression() does not contain a reference to predict() or xb().

4. margins with a factor variable in option dydx() failed to properly assign the omit operator to the column elements in r(b) that corresponded to the base level. This has been fixed.

5. The Populate button on the Options tab for the merge dialog box did not work if the path to the using dataset had a space in it. This has been fixed.

6. ml commands with a large number of independent variables in an equation occasionally reported an error about extra parameters when option init() or command ml init was used. This has been fixed.

7. ml reported an extraneous Mata trace log when it could not compute numerical derivatives for a model fit with constraints. This has been fixed.

8. nlogit with option collinear will now allow the same variable to be specified in multiple level equations.

9. Nonstationary sspace models using method(dejong), the default method, could terminate with an "invalid subscript" error. This has been fixed.

10. stcurve with options at1(), at2(), ..., at10() produced an error message if the dataset contained a _merge variable. This has been fixed.

11. sts graph with options by and censored() could produce an incorrect survival curve if one individual was censored just before another individual failed. This has been fixed.

12. testparm would incorrectly report the error "no such variables" when used with mlogit results where the first equation was the base outcome. This has been fixed.

13. tsset with option noquery was not posting r(tdelta). This has been fixed.

14. xtmixed, when used to model a residual-error structure with exactly two parameters -- for example, distinct residual variances for males and for females -- in models with no random effects, would report a p-value for the likelihood-ratio (LR) comparison test that was half of what it should have been. This has been fixed.

Stata executable, all platforms

15. Expressions with factor-variable notation like

1.d - 2.d

where the variable name was the letter d or the letter e (lowercase or uppercase) would cause an "invalid operator" syntax error. This has been fixed.

16. merge 1:m with option update, in the rare case when 1) there was a mismatch in the value of a variable with the same name in both master and using datasets, 2) when that value was missing in the master dataset, and 3) when that value varied in multiple observations of the using dataset, could use the first updated value from the using dataset in the subsequent matched observations. This has been fixed.

17. Mata functions st_matrixcolstripe_fvinfo() and st_matrixrowstripe_fvinfo(), when used with matrices containing an interaction with only continuous variables, would crash Stata. This has been fixed.

18. Mata functions tokenget() and tokengetall(), in the case where multiple characters were being treated as white-space characters, could fail to properly return tokens. This has been fixed.

19. Mata functions tokenget() and tokengetall(), in the case where multiple characters were being used as parsing characters, could fail to properly return tokens. This has been fixed.

20. Stata/MP was overoptimistic in parallelizing the Mata colon operator. Stata/MP would sometimes run much slower than Stata/SE with some Mata colon operations on small matrices. This has been fixed.

Stata executable, Windows

21. haver is now supported in 64-bit Stata for Windows.

22. With tabbed graphs enabled when using the Graph Editor, it was possible to switch away from the graph currently being edited to another graph. This behavior was unsafe and in rare cases could lead to unsaved edits. The Graph Editor now requires that you complete your edits to a single graph before editing a graph in another tab.

Stata executable, Mac

23. As a Viewer window was being resized, Stata for Mac automatically refreshed the content. However, quickly clicking the Viewer window's size control could cause the Viewer to get in a state where further clicks in the Viewer caused the window to resize. The Viewer now only refreshes a Viewer window after the mouse has been released off the size control.

24. Syntax highlighting in the Do-file Editor could crash Stata if it encountered the // comment delimiter near the end of a do-file. This has been fixed.

25. When using the up or down cursor keys to navigate the Data Editor, the previously selected cell still displayed a selection box when the Data Editor was scrolled vertically. This has been fixed.

Stata executable, Unix

26. Using the keyboard shortcut Ctrl+w in the Command window caused Stata to hang. Stata for Unix (GUI) now ignores this shortcut for the Command window and allows the default key binding to take effect.

-------- update 21jul2010 -----------------------------------------------------

Ado-files

1. graph box and graph hbox, when the variable being graphed contained only two unique values, could in rare cases identify both values as outliers when only one value was an outlier. This has been fixed.

2. logistic, when called under version control less than 11, would report an "option ... not allowed" error whenever any of the following settings were in effect:

set showbaselevels all set showbaselevels on set showemptycells off set showomitted off

This has been fixed.

3. mean, proportion, ratio, and total with two or more variables in option over() and one of vce(bootstrap), vce(jackknife), svy bootstrap, svy brr, svy jackknife, or svy sdr failed to properly identify the replicated point estimates. With vce(bootstrap), this resulted in the error message "insufficient observations to compute bootstrap standard errors". In the other cases, the subpopulation point estimates were labeled with generic integer indices without prefix _subpop_. This has been fixed.

4. mi impute uses listwise deletion when predictors (or weights) contain missing values, which leads to a reduced number of observations being used during the estimation task. mi impute now reports a note to notify users when this happens.

5. mi impute mvn has the following new features and fixes:

a. mi impute mvn with a large number of variables is now significantly faster.

b. With more than 53 imputation variables, mi impute mvn could mistakenly treat adjacent missing-value patterns as a single pattern. In such cases, this would often result in an r(498) error, "imputed data contain missing values", unless option emonly or force was specified. If option emonly was specified or option force was used to proceed with imputation anyway, mi impute mvn produced EM parameter estimates and imputed values that were based on overinflated tallies of missing-value patterns. This has been fixed.

c. In cases where predictors (right-hand-side variables) contained missing values, mi impute mvn did not account for the reduced numbers of observations during computation and thus produced incorrect imputed values. This has been fixed.

6. Mata's moptimize() function moptimize_result_coefs() was documented but did not exist. This has been fixed.

7. ml plot exited with a Mata run-time error the first time it was run after ml model. This has been fixed.

8. ml score exited with an error message when used after an lf-family or gf-family evaluator. This has been fixed.

9. prais exited with an error message if option vce() was specified. This has been fixed.

10. svy mistakenly used observations where the poststrata variable contained a missing value. This has been fixed.

11. svy with multistage data mistakenly used observations that contained missing values in a stratum variable in any but the last stage. This has been fixed.

-------- update 01jul2010 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 10(2).

2. Mata's moptimize() and optimize() with the Nelder-Mead technique reported the unhelpful error message "Hessian is not negative semidefinite" when it should have reported "missing values returned by evaluator". This has been fixed.

3. xtpoisson with options fe and vce(robust) failed when factor variables were used. This has been fixed.

-------- update 16jun2010 -----------------------------------------------------

Ado-files

1. estat bootstrap, when reporting the default bias-corrected or the optional bias-corrected and accelerated confidence intervals (CIs) and in the rare case where the standard errors (SEs) were so large relative to the coefficient that the dataset was not truly identifying the parameter estimates, produced CIs that were overly conservative. Specifically, the problem occurred only when the ratio of the SEs and the parameter estimates was on the order of 10 million (1e7) or greater. This has been fixed.

2. ml trace reported a syntax error even when supplied with the correct syntax. This has been fixed.

3. svy would incorrectly report that option baselevels was not allowed when set showbaselevels was set to on. This has been fixed.

4. svy would incorrectly report that option allbaselevels was not allowed when set showbaselevels was set to all. This has been fixed.

5. svy: mean, svy: proportion, svy: ratio, and svy: total with vce(bootstrap) and both option subpop() and option over() would report an invalid syntax error if there were two or more overgroups. This has been fixed.

6. svy: tabulate with options cv and percent would report coefficients of variations that were too large by a factor of 100. This has been fixed.

Stata executable, all platforms

7. outsheet could crash Stata if a date format and value label were assigned to a variable. This has been fixed.

8. _pctile with option percentile() containing a value within 1e-15 in relative difference to 100 would cause Stata to crash. This has been fixed.

9. svy estimation commands with independent variables in multiple equations (such as svy: mlogit) and svyset poststratification variables would report incorrect standard errors in all but the first equation or would report the error estimates post: matrix has missing values. This has been fixed.

Stata executable, Mac

10. Stata would crash if the Variable Properties window was used to navigate to an empty column in the Data Editor. This has been fixed.

11. The Data Editor could display drawing artifacts when there were few variables and the scrollbar was quickly set to its maximum value. This has been fixed.

-------- update 03jun2010 (Stata version 11.1) --------------------------------

What's new in release 11.1 (compared with release 11.0)

Highlights for release 11.1 include the following:

o Multiple imputation. The mi command now officially supports fitting panel-data and multilevel models. See mi estimation.

o Truncated count-data models. New commands tpoisson and tnbreg fit models of count-data outcomes with any form of left truncation, including truncation that varies by observations.

o Mixed models. Linear mixed (multilevel) models have new covariance structures for the residuals: exponential, banded, and Toeplitz. See xtmixed.

o Probability predictions. predict after count-data models, such as poisson and nbreg, can now predict the probability of any count or any count range.

o Survey bootstrap. Estimation commands can now estimate survey bootstrap standard errors (SEs) using user-supplied bootstrap replicate weights. See svy bootstrap.

o Survey SDR weights. Successive difference replicate (SDR) weights are now supported when estimating with survey data. These weights are supplied with many datasets from the U.S. Census Bureau. See svy sdr.

o Concordance. estat concordance adds a new measure of concordance, Gönen and Heller's K, that is robust in the presence of censoring.

o Survey GOF. Goodness-of-fit (GOF) tests are now available after probit and logistic estimates on survey data. See svy estat.

o Robust SEs. Cluster-robust SEs have been added to xtpoisson, fe.

o Survey CV. The coefficient of variation (CV) is now available after estimation with survey data. See svy estat.

o Estimation formatting. Numerical formats can now be customized on regression results. You can set the number of decimal places for coefficients, SEs, p-values, and confidence intervals using either command-line arguments or the set command. See estimation options and set cformat.

Settings have also been added to control the display of factor variables in estimation tables. These display settings include display of empty cells, display of base levels, and omitting variables excluded because of collinearity. See set showbaselevels and query output.

o Stata/MP performance. Stata/MP, the parallel version of Stata, has several performance improvements: many panel-data estimators are even more parallelized, improved parallelization of estimations with more than 200 covariates, improved tuning of MP on large numbers of processors/cores.

o Clipboard improvements. Clipboard support in the Data Editor has been enhanced. Copies to the Clipboard now retain variable formats and other characteristics of the data when pasting within Stata.

o Windows XP. The amount of memory available to Stata has been increased on 32-bit Windows XP.

o Do-file Editor. Syntax highlighting, bookmarks, and other Do-file Editor features have been added to Stata for Mac.

o ODBC. ODBC support has been added for Oracle Solaris (Sun Solaris).

o Dialog boxes. Dialog boxes on Unix now have varlist controls that allow you to select variables from a list of variables in your dataset.

Ado-files

1. datasignature, when used under version control with a version less than 10, was not saving the resulting data signature in r(datasignature). This has been fixed.

2. _diparm reported that it could not find the _cons coefficient for the specified equation when it was marked as omitted with the o. operator. This has been fixed.

3. estat concordance provides a new concordance measure, Gönen and Heller's K concordance coefficient, which is robust to censoring unlike the conventional Harrell's C index. The new measure is available by using option gheller.

4. estat cv is a new postestimation command available with svy estimation results. estat cv reports a table of coefficients of variation for each coefficient in the current estimation results.

5. estat gof is now available after svy: logistic, svy: logit, and svy: probit. estat gof reports a goodness-of-fit test for logistic and probit regression models that were fit using complex survey data.

6. estat summarize has the following fixes:

a. It reported zero values for levels of factor variables that were less than the base level as set by fvset. This has been fixed.

b. It reported incorrect values for interactions involving factor variables when the level of the factor variable was less than the base level as set by fvset. This has been fixed.

7. estat vif produced incorrect results when used with a subset of the data. This has been fixed.

8. estimates replay would issue the error last estimates not found when used after mi estimate, post: regress. This has been fixed.

9. expandcl with an if restriction would also add observations even where the if restriction was false. This has been fixed.

10. In rare cases, gmm would exit with an error message if option xtinstruments() was used and the data contained many gaps. This has been fixed.

11. heckman has the following fixes:

a. When the two-step estimator was used and weights were specified, heckman ignored the weights and produced unweighted two-step estimates. Now an error message is produced indicating that weights are not allowed with the two-step estimator.

b. When the maximum likelihood estimator was used and the BHHH maximization technique was requested, heckman exited with an error message even though the BHHH technique is allowed. This has been fixed.

12. margins, when used after glm with either a power or an opower (odds power) link, would erroneously report that the default prediction is a function of possibly stochastic quantities other than e(b). margins now proceeds with the prediction.

13. mean, proportion, ratio, and total with option over() are now faster, especially when used with replication-method option vce(bootstrap) or vce(jackknife) or with prefix command svy brr or svy jackknife.

14. mi estimate has the following new features and fixes:

a. mi estimate now officially supports estimation of many panel-data and multilevel models with multiply imputed data; see mi estimation for a list of supported commands.

b. When the number of tested coefficients in the model test exceeded model degrees of freedom, mi estimate reported a minimum-degrees-of-freedom test based on a reduced set of coefficients. It now reports a missing value and provides a link to the description of why the joint model test of all coefficients could not be performed.

c. mi estimate reported nonmissing multiple-imputation standard errors when some of the within-imputation standard errors were reported to be missing. This has been fixed.

d. mi estimate, post would incorrectly set macro e(cmd) to the contents of macro e(cmd2) if that macro was set by the estimation command used with mi estimate. This has been fixed.

15. mi test used the rank of the VCE matrix instead of its inverse to determine model degrees of freedom. In rare cases where, computationally, these ranks did not coincide, mi test reported incorrect model degrees of freedom. This has been fixed.

16. In rare cases, ml and Mata's moptimize(), when rescaling the parameters during a search for starting values, reported a Mata run-time error. This has been fixed.

17. ml reported a conformability error with a Mata trace log when having difficulty computing numerical derivatives for methods lf, lf0, and gf0. This has been fixed.

18. predict after gnbreg, nbreg, poisson, xtgee, xtnbreg, xtpoisson, zinb, and zip has two new options: pr(n) and pr(a,b). Postestimation command predict varname, pr(n) stores the probability Pr(y = n) for the specified model in varname, where n can be a number or a variable. Postestimation command predict varname, pr(a,b) stores the probability Pr(a < y < b) for the specified model in varname, where a and b can be a number or a variable.

19. predict after sspace and dfactor now allow dynamic forecasting when the estimation's dependent variables included time-series operators.

20. Dynamic forecasts from predict, dynamic() after sspace could be incorrect if the dependent variables in the observation equations were not in the same order as listed in the dataset and were included on the right-hand side of an observation equation.

21. predict after xtmixed with a model containing an empty random-effects structure would produce an error message if predicted fitted values (option fitted), residuals (option residuals), or standardized residuals (option rstandard) were out of sample. This has been fixed.

22. predict with option iru0, when used after xtnbreg, re, created a new variable when the estimated value of r was less than one and the expected value was undefined. An appropriate error message is now produced.

23. rocfit with option continuous could produce fewer groups than specified. This has been fixed.

24. The dialog box for rolling concatenated options clear and keep when specified together. This has been fixed.

25. sqreg coefficient labels for quantiles less than 0.1 were missing the leading zeros. This has been fixed. For example, quantile(0.025) is now labeled as q025 instead of q25.

26. sspace did not allow the wildcard character * in a varlist specification. This has been fixed.

27. svy has new option dof() for specifying the design degrees of freedom. This option overrides the default calculation, df = N_psu - N_strata.

28. svy bootstrap is a new prefix command for performing bootstrap variance estimation for survey data. svy bootstrap requires that the bootstrap replicate weights be svyset.

29. svy jackknife with svyset iweights and suboption multiplier() of jkrweight() incorrectly computed the design degrees of freedom. This has been fixed.

30. svy sdr is a new prefix command for performing successive difference variance estimation for survey data. svy sdr requires that the replicate weights be svyset.

31. svy: tabulate has new option cv, which specifies that svy tabulate display the coefficient of variation for each cell in the table.

32. svy: tabulate now reports the error matsize too small much earlier when option stdize() identifies more groups than matsize would allow.

33. svyset has the following new features:

a. New option bsrweight() sets bootstrap replicate weights, and new option bsn() sets the variance multiplier for mean bootstrap replicate weights. New option vce(bootstrap) sets the default method for variance estimation to be the bootstrap.

b. New option sdrweight() sets successive difference replicate weights. New option vce(sdr) sets the default method for variance estimation to be the successive difference replication method.

c. New option dof() specifies the design degrees of freedom. This option overrides the default calculation, df = N_psu - N_strata.

34. tnbreg is a new command that implements truncated negative binomial regression for any nonnegative truncation point. Truncation can be specified as a number or a variable, which allows for varying truncation values. Additionally, options have been added to return conditional and unconditional probabilities for the truncated negative binomial regression model.

35. tpoisson is a new command that implements truncated Poisson regression. Truncation can be specified as a number or a variable, which allows for varying truncation values. Additionally, options have been added to return conditional and unconditional probabilities for the truncated Poisson regression model.

36. xtmixed now supports three new residual variance-covariance structures: banded, toeplitz, and exponential; see xtmixed for details.

37. With version set to less than 11.1, xtnbreg, re returns xtn_re in e(cmd2) and xtnbreg, fe returns xtn_fe in e(cmd2). As of version 11.1, xtnbreg instead returns e(model), containing re, fe, or pa to indicate which model was specified.

38. xtpoisson now returns e(model), containing re, fe, or pa to indicate which model was specified.

39. xtrc has the following fixes:

a. xtrc, when used with factor-variable notation, produced erroneous results if any variables were dropped because of collinearity while performing OLS regression on each of the panels. In other cases, xtrc exited with an error message if any variables were dropped because of collinearity while performing those regressions. Both issues have been fixed.

b. xtrc drops panels with fewer observations than the number of indepvars but failed to report the proper error message when the entire estimation sample was dropped. This has been fixed.

Stata executable, all platforms

40. ereturn display is now allowed after regress.

41. estimation commands have the following new features:

a. Three new display options control formatting within the coefficient table. cformat() controls the formatting of coefficients, standard errors, and confidence limits. sformat() controls the formatting of test statistics. pformat() controls the formatting of p-values. These options override the default output format for individual estimation commands.

b. Three new settings control formatting within the coefficient table. set cformat controls the formatting of coefficients, standard errors, and confidence limits. set sformat controls the formatting of test statistics. set pformat controls the formatting of p-values. These settings override the default output format for all estimation commands.

c. Three new settings control what is shown in the coefficient table. set showbaselevels controls the output of base levels for factors and interactions. set showemptycells controls the output of empty cells for factors and interactions. set showomitted controls the output of omitted coefficients.

42. Factor-variable interactions were not parsed properly. The specific conditions that triggered this were as follows:

a. You specified a simple interaction (a single # sign) between two or more factor variables.

b. The first variable in the interaction had the value zero as one of its categories.

c. The first specification in the interaction had a base level that is not the default of zero (the lowest level for the first variable).

d. At least one of the remaining variables in the interaction had a base equal to the lowest-valued category for that variable, whether explicitly specified or taken as the default.

When the above conditions occurred, Stata attempted to omit an extra cell in the interaction. Sometimes, the cell would be omitted altogether; other times, Stata produced a coefficient for that cell, but it produced missing standard errors and confidence intervals. Either way, the model fit was thrown off because the cell's coefficient was not properly estimated. Almost all estimation commands were affected by this bug, with regress being one notable exception. This has been fixed.

43. margins with abbreviated names in options at(), dydx(), dyex(), eydx(), and eyex() picked non-factor variables that matched the abbreviation, even when a factor variable was intended. These options now check for perfect matches and report an error for ambiguous abbreviations.

44. matrix dissimilarity with options gower and variables and with the number of observations exceeding matsize crashed Stata. This has been fixed.

45. mlogit failed to flag the indepvars in the base equations as omitted. This has been fixed.

46. summarize with option format formatted the Std. Dev. as a date if the variable you specified had a date format, even after the 11feb2010 update. This has been fixed.

47. svy: total would randomly but rarely produce e(V_srssub) and e(V_srssubwr) matrices filled with zeros instead of with the correct variance estimates. This has been fixed.

48. tobit now accepts time-series operators.

49. xtpoisson with option fe now supports vce(robust).

50. xtprobit with option intmethod(aghermite) would not allow factor variables in Stata/MP. This has been fixed.

51. In rare circumstances, accessing the Internet from within Stata could cause Stata to crash. This has been fixed.

Stata executable, Windows

52. set memory on a 32-bit Windows XP computer with Service Pack 3 installed may now be able to allocate more memory, depending on your computer's particular configuration.

53. For executable updates, update swap will now get called automatically when the update completes. This behavior has been the default for Windows Vista and Windows 7 with User Account Control (UAC) enabled and will now be extended to include all versions of Windows. The new behavior will take effect on subsequent executable updates.

54. The Do-file Editor has the following fixes:

a. When a long do-file was opened, the first line was scrolled off the screen, so the first visible line was the second line. This has been fixed.

b. When doing a "Replace All", if the "Regular Expression" checkbox was checked, Stata might get into an infinite loop depending on the regular expression being replaced. This has been fixed.

55. Opening PDF documentation from Stata could cause any open log files to become locked until Adobe Reader was closed. This has been fixed.

Stata executable, Mac

56. The Do-file Editor has the following new features:

a. Syntax highlighting. See the Do-file Editor preferences in the General Preferences dialog for syntax highlighting options.

b. Line numbers.

c. Bookmarks for lines. Bookmarks allow you to navigate between lines that have been bookmarked. To add a bookmark for a line, click on the line number for the line. To remove a bookmark from a line, click on the line's marker. Click on the Next Bookmark toolbar button to move the cursor from its current position to the next bookmarked line. Click on the Previous Bookmark toolbar button to move the cursor from its current position to the previous bookmarked line.

57. The Do-file Editor no longer opens files already open in the Do-file Editor in multiple Do-file Editor windows. If you attempt to open a file that is already open, the file's Do-file Editor window is brought to the front.

58. The 09mar2010 update introduced a bug where clicking on the Open button of the Graph window's toolbar caused Stata to crash. This has been fixed.

59. When loading a previously saved set of preferences, the Results window would not be resized and moved to the saved window size and location if the saved Results window font was different from the current font. This has been fixed.

60. Mac OS X would set the current directory for the Open and Save dialogs to Stata's preferences directory after saving preferences. Stata now sets the current directory for the Open and Save dialogs back to the previous directory after saving preferences.

Stata executable, Unix

61. In release 10.04 of the Ubuntu distribution of Linux, the data in Stata's memory could become corrupt. This was caused by a change in the behavior of a low-level call in an operating system library in the Ubuntu distribution of Linux. This was restricted to only the Ubuntu 10.04 distribution of Linux. This has been fixed.

62. A filter search in the Variables Manager did nothing. This has been fixed.

-------- update 20apr2010 -----------------------------------------------------

Ado-files

1. As of the 09mar2010 update, the speed of Stata/MP for many panel-data estimators was improved by reducing the unparallelized regions of the commands. Faster commands include xtgee, xtgls, xthtaylor, xtintreg, xtlogit, xtnbreg, xtpcse, xtpoisson, xtprobit, xtreg, xtregar, and xttobit.

2. xtunitroot ips would exit with an error message if the panels contained too few time periods to compute p-values for the Z_t-tilde-bar test statistic. This has been fixed.

3. The dialog box for infix did not specify a default file extension for a dictionary. This has been fixed.

4. The dialog box for controlling paired-coordinate arrowhead properties failed to work. This has been fixed.

Stata executable, all platforms

5. anova and other commands specified with multiple interactions split between two or more i.(...) groups that shared two or more variables, and one or more of them containing parentheses-bound terms in their specification, would occasionally omit some of the interaction terms or even crash Stata. This was a rare occurrence except with the slash operator, /, in anova and manova. This has been fixed.

Stata executable, Mac

6. Programmable dialogs that used an empty file filter to open files caused Stata to crash. This has been fixed.

-------- update 25mar2010 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 10(1).

2. estimates use would not recognize valid estimation results that were estimates stored with set dp comma in effect. This has been fixed.

3. margins with continuous variables in option dydx() and with set dp comma in effect would report a cryptic invalid name error. This has been fixed.

4. The odbc load dialog did not bring up the file chooser dialog for Excel files and MS Access Database data source names, even after the 11feb2010 update. This has been fixed.

-------- update 09mar2010 -----------------------------------------------------

Ado-files

1. The display of results for most estimation commands is now faster because _coef_table and ereturn display are faster. This speedup will be most noticeable when running many estimation commands on small samples.

2. describe, in addition to its other capabilities, can now produce a dataset containing the description. describe does this when new option replace is specified.

3. ds, temporarily moved to "previously documented", is now restored to being an official, documented command of Stata. To be clear, ds is as fully up to date and fully supported as any other official command of Stata.

4. graph hbar, graph hbox, and graph dot did not honor option yline(). This has been fixed.

5. graph pie could fail with an error when options by() and over() were combined and when the number of categories in the over-groups differed across by-groups. This has been fixed.

6. histogram with a reference to an r() result in the if condition, such as

. summarize mpg, detail . histogram mpg if mpg < r(p50)

would result in an incorrect graph, even after the 11feb2010 update. This has been fixed.

7. mi estimate, when used with mlogit or mprobit where the base outcome was the first level of the depvar, reported the note "(base outcome)" instead of the specified transformations. This has been fixed.

8. ml check, when used with lf0, lf1, or lf2 evaluator types, would incorrectly report a failure caused by the evaluator not setting the log-likelihood scalar in the first test. This has been fixed.

9. ologit and oprobit failed to report the offset variable in the coefficient table when there were no indepvars. This has been fixed.

10. svy: tabulate oneway and svy: tabulate twoway, when used with missing values in one or both of the direct standardization variables, specified in options stdize() and stdweight(), would report inflated values for the cell observations. This has been fixed.

All other cell statistics were properly computed in this case.

11. tsset and xtset have new option noquery.

12. xtrc incorrectly marked some elements of column names of e(b) as omitted based on the sample identified by the last panel. This only affected the coefficient table if option noomitted was specified. This has been fixed.

Stata executable, all platforms

13. Expressions involving compound double quoted strings that contained a single right quote could in rare cases cause the string to be misinterpreted. This has been fixed.

14. fmtwidth() did not recognize centered string formats. This has been fixed.

15. margins was incorrectly reporting "(not estimable)" with estimation results from commands like heckman and heckprob that allow certain missing-value patterns in the independent variables. This has been fixed.

16. Mata has the following fixes:

a. _editmissing() called thousands of times could cause Mata to produce a stack overflow error message. This has been fixed.

b. fmtwidth() did not recognize centered string formats. This has been fixed.

c. printf() was not correctly handling date-time formats containing "\n", "\t", and "\r". This has been fixed.

d. mata mlib add reported an invalid opcode error when attempting to add a class definition to an mlib file. This has been fixed.

17. test with the syntax

test [eq]element

or

test [eq]:element

would report the error

[eq]o.element not found

when element was omitted in the first equation but not in equation eq. This has been fixed.

18. test failed to recognize the Break key when working with estimation results with a large number of elements in e(b). This has been fixed.

19. Previously, pasting into the Data Editor was not allowed with a filter applied for observations. This restriction has been loosened so that the user can override this behavior.

Stata executable, Windows

20. On Windows XP, File > Open... on the main Stata window and the Do-file Editor displayed the old-style Windows "File Open" dialog. This has been fixed.

21. Switching from another window to Stata should automatically focus Stata's Command window; however, this feature was unintentionally disabled in the 11feb2010 executable update. This has been fixed.

Stata executable, Mac

22. You can now select all text in the Review window by selecting the Edit > Select All menu item when the Review window has the keyboard focus.

23. The default behavior for the Open dialog is to allow only files that Stata recognizes to be opened. A file format pop-up button has been added to the Open dialog that allows you to limit the opening of files to all or just a subset of the file types that Stata recognizes. You can also set the Open dialog to open any file and any file type that Stata does not recognize in the Do-file Editor. The file format pop-up button does not show all file types that Stata recognizes but just the ones that are most commonly opened by Stata users.

24. The Do-file Editor's Save dialog now includes a file format pop-up button that allows you to specify the file extension to use in the filename.

25. The 11feb2010 update added the ability to open a do-file from the Finder in the Do-file Editor instead of executing it. However, the current working directory was changed to the location of the do-file. This has been fixed.

26. Navigating the Data Editor with the cursor keys would not work after entering or changing a value in the main edit field. This has been fixed.

27. When using the Make Text Bigger or Make Text Smaller menu items to change the font size of the Results or Viewer windows, the windows are automatically resized to maintain the same number of rows and columns currently visible. However, the windows would move slightly from their original positions while they were being resized. This has been fixed.

28. When exporting graphs to PostScript or Encapsulated PostScript (EPS) files, extended ASCII characters would not be outputted correctly and could appear blank in the resulting files when viewed. This has been fixed.

Stata executable, Unix

29. odbc is now supported on Solaris.

30. The Data Editor now has a preference for disabling visual gradients. Enabling this preference can increase the speed at which the Data Editor refreshes its display when running Stata over a network.

-------- update 11feb2010 -----------------------------------------------------

Ado-files

1. arima, on very rare occasions, would exit with a "matrix not symmetric" error. This has been fixed.

2. bsample, when used with option cluster(), would exit with an incorrect error message if the requested sample size was greater than the number of clusters. Now a better error message is produced.

3. fracplot and fracpred, dresid now produce an appropriate error message when used after mfp: stcrreg or fracpoly: stcrreg.

4. gmm, if the minimum lag requested in option xtinstruments() was at least two more than the greatest lag of the dependent variable in the regression equation and if a constant term or other exogenous regressors were included in the model, would exit with an error message or report incorrect results for dynamic panel-data models. This has been fixed.

5. graph has the following fixes:

a. graph with option by(byvar), when byvar was a string that contained an unquoted comma, would fail with an error. This has been fixed.

b. graph trimmed trailing spaces from variable labels when those labels appeared in a graph. These spaces are no longer trimmed.

c. graph could be extremely slow at handling variables whose value labels had extremely large gaps in the values being labeled, for example, 1 and 1,000,000. This has been fixed.

d. graph did not correctly draw axis lines when option yscale(noextend) or option xscale(noextend) was specified. Although the axis line itself was not drawn correctly or sometimes was not drawn at all, the tick marks and their labels were all correct and correctly positioned. This has been fixed.

e. graph hbar, graph hbox, and graph dot did not honor suboption angle() of option ylabel(). This has been fixed.

f. graph matrix, when graphs were drawn with a monochromatic scheme (such as s1mono or s2mono), outlined the markers with navy lines. The markers are now outlined in the same gray with which the markers are filled.

6. histogram with a reference to an r() result in the if condition, such as

. summarize mpg, detail . histogram mpg if mpg < r(p50)

would result in an incorrect graph. The contents of r() were being changed before the if condition was used to determine the sample. This has been fixed.

7. ivregress and gmm, if the quadratic-spectral kernel was used and the time series contained gaps, would calculate heteroskedasticity and autocorrelation-consistent (HAC) weight and variance-covariance matrices incorrectly. This has been fixed.

8. makecns, when called under version control between 8 and 10.1, would truncate the constraint list to 244 characters. This only affected models that employed a large number of constraints. This has been fixed.

9. margins now reports a note when option continuous is turned on because a single-level factor is specified in option dydx() or eydx().

10. Mata has the following new features and fixes:

a. moptimize() now has a long and a short name associated with the evaluator types it supports. Some of the short names for the evaluator types have been renamed. The old names will continue to work but are going undocumented. Click here to see how moptimize() was previously documented.

Short name Long name Old name ------------------------------------------------ d0 derivative0 d1 derivative1 d2 derivative2 d1debug derivative1debug d2debug derivative2debug gf0 generalform0 v0 gf1 generalform1 v1 gf2 generalform2 v2 gf1debug generalform1debug v1debug gf2debug generalform2debug v2debug q0 quadraticform0 q1 quadraticform1 q1debug quadraticform1debug ------------------------------------------------

b. moptimize() has three new evaluator types to replace e1 and e2. The e-type evaluators will continue to work but are going undocumented. Click here to see how moptimize() was previously documented.

Short name Long name ------------------------------ lf0 linearform0 lf1 linearform1 lf1debug linearform1debug lf2 linearform2 lf2debug linearform2debug ------------------------------

c. optimize() now has a long and a short name associated with the evaluator types it supports. Some of the short names for the evaluator types have been renamed. The old names continue to work but are going undocumented. Click here to see how optimize() was previously documented.

Short name Long name Old name ------------------------------------------------ d0 derivative0 d1 derivative1 d2 derivative2 d1debug derivative1debug d2debug derivative2debug gf0 generalform0 v0 gf1 generalform1 v1 gf2 generalform2 v2 gf1debug generalform1debug v1debug gf2debug generalform2debug v2debug q0 quadraticform0 q1 quadraticform1 q1debug quadraticform1debug ------------------------------------------------

d. optimize() with the default optimize_init_colstripe() setting and optimize_init_tracelevel() set to "params", "gradient", or "hessian" reported a conformability error after the first iteration. This has been fixed.

11. mfp and fracpoly now respect the Break key.

12. mi impute monotone has the following fixes:

a. mi impute monotone produced the syntax error "duplicates found" when the length of the list of names of the specified imputation variables exceeded 244 characters. This has been fixed.

b. mi impute monotone, when the logit method was used, would display an extra empty line. This has been fixed.

13. ml has the following new features and fixes:

a. ml now has a long and a short name associated with the evaluator types it supports. Some of the short names for the evaluator types have been renamed. The old v0, v1, and v2 evaluator types will continue to work but are going undocumented. Click here to see how ml was previously documented.

Short name Long name Old name ------------------------------------------------ d0 derivative0 d1 derivative1 d1debug derivative1debug d2 derivative2 d2debug derivative2debug gf0 generalform0 v0 ------------------------------------------------

b. ml with a model specification that contained multiple equations with the same name would report a cryptic error message after attempting to fit the model. ml model now catches the error early and reports a better error message.

c. ml, when supplied with an indepvarlist containing more than one duplicate, such as

. sysuse auto . logit foreign turn turn turn

would complain about duplicate entries. This has been fixed.

d. ml, when called under version control between 8 and 10.1, would truncate the constraints in constraint(numlist) to 244 characters. This only affected models that employed a large number of constraints. This has been fixed.

14. nlcom, for expressions involving the _b[eq:name] notation, reported an "invalid name" error. This has been fixed.

15. The odbc load dialog did not bring up the file chooser dialog for Excel files and MS Access Database data source names. This has been fixed.

16. previously documented has been updated. It now contains links to help files for ml for Stata 10 and early Stata 11 (prior to the new evaluator names). It also contains links to help files for optimize() and moptimize() for early Stata 11 (prior to the new evaluator names).

17. putmata and getmata are new Stata commands that make it easy to move data from Stata to Mata and vice versa.

18. _rmdcoll, when option expand was specified, would incorrectly report that option expand was not allowed. This has been fixed.

19. sts test, if the variable used had a value label that had been dropped but the value-label name was still attached to the variable, issued an error. This has been fixed.

20. xtmelogit and xtmepoisson now use a more sophisticated method of shuffling starting values when estimating the random effects as a component of the model log likelihood. As a result, many models that previously failed with an "initial values not feasible" error may now be fit error free. Also, random-effects predictions are now less prone to convergence failures.

21. xtmixed, when used to fit crossed-effects models, is now more computationally efficient. When constructing crossed-effects design matrices, xtmixed's likelihood evaluator now looks up variables by position, where previously it looked up variables by name. The resulting speed gain is more significant the larger the clusters.

22. xtpois and xtclog were replaced by xtpoisson and xtcloglog, respectively. However, xtpois and xtclog continued to work and behaved as if xtpoisson or xtcloglog was called under version control for Stata 8. xtpois and xtclog are now synonyms for the current version of xtpoisson and xtcloglog.

23. zipfile, if there was not a space between the comma in option saving() and suboption replace, issued an error. This has been fixed.

Stata executable, all platforms

24. Base-level information on interactions sometimes got lost after some postestimation commands, such as nlcom and testnl. This only affected the look of the coefficient table. This has been fixed.

25. Under very unusual conditions, valid ado-programming class assignments could fail with the message "= assignment of a reference is not allowed". This has been fixed.

26. Stata functions have the following fixes:

a. binormal(), in rare instances, returned missing values with reasonable input arguments. This has been fixed.

b. rbinomial(), when 15 < n*(1-p) <= 125, p > 0.5, and when calling rbinomial(n, p) followed by rbinomial(n, 1-p), would generate incorrect variates. The second call would generate variates from a binomial(n,p) distribution. This has been fixed.

c. rnormal(), if the standard deviation parameter was zero, returned missing values. Now it returns zeros.

27. graph export, when exporting a graph to PostScript or EPS, would not display bold or italic text if the text was being rendered in a default PostScript font. This has been fixed.

28. lincom and test, when repeatedly used with the new factor-variable notation, would eventually cause Stata to slow down. This has been fixed.

29. logit and probit failed to detect perfect predictors that were base-level virtual variables for factor variables or interactions. This has been fixed.

30. Mata's st_sdata() and st_sview() did not allow varlists in the string elements of the j argument, a feature that was added to st_data() and st_view() in Stata 11. This has been fixed.

31. matrix rownames and matrix colnames, when presented with a string that looked like a variable with time-series operators, such as sd(e), would sometimes report the "invalid name" error. This has been fixed.

32. ologit (and most other ml-based estimation commands), when used with an interaction involving a factor variable with a continuous variable squared, would report the error "multiple 'o' operators attached to a single variable are not allowed within an interaction" when the base level was omitted because of lower-ordered terms. This has been fixed.

33. regress has the following fixes:

a. regress with noninteger-valued iweights was using the integer-truncated degrees of freedom in calculating the mean squared error and variance-covariance values. This has been fixed.

b. regress, when used with a varlist entirely bound in parentheses (including the depvar), such as

. regress (mpg turn)

would report a "system limit exceeded" error. This has been fixed.

34. reg3, regress, xtivreg, xtpcse, and xtreg marked omitted variables with the 'o.' operator, even when called under version control less than 11 (and no factor-variables notation was used). This has been fixed.

35. sspace has the following fixes:

a. The sspace coefficient standard errors for nonstationary models using option method(dejong) were incorrect. Although the De Jong likelihood is used for nonstationary models by default, this bug occurred only if option method(dejong) was specified explicitly. This has been fixed.

b. sspace may have had difficulty converging for nonstationary models using option covstate(unstructured) with option method(hybrid), the default, or option method(dejong). This has been fixed.

36. strtoname() now respects extended ASCII characters. Now all strings that are accepted as names by confirm names and syntax will remain unchanged when passed to strtoname().

37. summarize has the following fixes:

a. summarize with the by prefix and aweights or iweights would ignore the weight specification after the results from the first by group were displayed. This has been fixed.

b. summarize with options detail and format, if the specified variable had a date format, formatted the Std. Dev., Variance, Skewness, and Kurtosis as dates. This has been fixed.

38. svy, when used with certainty PSUs, was not including the observations corresponding to the certainty PSUs in the e(V_srs) and e(V_srswr) calculations, affecting the output from estat effects. This has been fixed.

39. tabulate, summarize() now uses a two-pass formula (instead of a one-pass formula) to compute standard deviations with quad precision. This produces more-accurate standard deviations.

40. testparm with interactions of continuous variables with factor variables with fvset base levels failed to capture all the requested regression coefficients in the Wald test. This has been fixed.

41. Stata's varlist parsing logic would sometimes incorrectly identify a single-letter variable name as a time-series or factor-variable operator. For example, i.(i(1/5).b) caused a syntax error. This has been fixed.

42. The Data Editor has the following fixes:

a. Pasting into the Data Editor from an Excel spreadsheet that contained empty columns caused Stata to always treat the first row as data, even if the first row otherwise contained valid variable names. This has been fixed.

b. In the Data Editor, double-clicking on a column heading for an existing variable should open the Variable Properties dialog; however, if the variables were filtered, the dialog box for generating a new variable might have opened instead. This has been fixed.

Stata executable, Windows

43. sleep could exhibit excessive CPU load during sleep intervals of less than 501 milliseconds. This has been fixed.

44. Stata did not respect the preference for persistent viewers. This has been fixed.

45. Under very specific circumstances, Stata could repeatedly prompt the user to elevate privileges when trying to apply updates on Windows Vista or Windows 7. This problem could only occur if User Access Control (UAC) has been disabled and if the user attempting to apply the update is not part of the Administrators group.

46. The Do-file Editor has the following fixes:

a. Selecting Edit > Find > Balance Braces excluded the open and close braces from the selection, which caused the subsequent Do the selection operation to produce a "matching close brace not found" error. This has been fixed.

b. In the Do-file Editor Preferences dialog, Restore Factory Defaults sometimes failed to correctly restore default settings for code folding. This has been fixed.

47. Some circle and square marker symbols could appear as ovals or rectangles in graphs. This has been fixed.

48. The Graph window has the following fixes:

a. Resizing or maximizing the Graph window could cause Stata to appear slow or unresponsive when many graphs were tabbed into the same Graph window. This has been fixed.

b. The keyboard shortcut for Copy (Ctrl+C) did not work. This has been fixed.

Stata executable, Mac

49. You can now select multiple files in the Open dialog. Stata will open the files in the order that it receives them from the Mac, not necessarily in the order that you have selected the files. When selecting multiple do-files, Stata will open all the do-files in the Do-file Editor. When selecting multiple datasets, Stata will use each one in the order it receives them from the Mac; however, because Stata can only have one dataset in memory at a time, it does not make sense to select multiple datasets in the Open dialog.

50. You can now navigate combobox and varlist controls in Stata's programmable dialogs with the keyboard. When the keyboard focus is set to a combobox or varlist control, you can press the down arrow key to display the drop-down list, use the up and down arrow keys to navigate the drop-down list, press the Return key to select an item from the drop-down list, or press the Escape key to dismiss the drop-down list.

51. Rendering graphs has been improved:

a. Text rendering for graphs drawn to a window has been improved so that text with multiple fonts, superscripts and subscripts, and Greek letters will render with a consistent baseline.

b. When exporting a graph as a PDF, rendering connected lines has been greatly improved. Rendering graphs to the screen has also been improved.

c. Some circle and square marker symbols could appear as ovals or rectangles in graphs. This has been corrected.

d. Because of changes in Mac OS X 10.6, options width() and height() had no effect on graphs exported to a bitmap format using graph export. This has been fixed while still maintaining compatibility with Mac OS X 10.5.

52. The Do-file Editor has the following new features and fixes:

a. Do-files opened from the Finder can now be opened in the Do-file Editor instead of executing them. To enable this setting, check the "Edit do-files opened from the Finder instead of executing them" checkbox on the Do-file Editor tab in the Preferences dialog. Any do-file named Stata.do will still execute in Stata and automatically set the current working directory to the directory that the Stata.do file is contained in regardless of whether this setting is enabled.

b. The contextual menu for the Do-file Editor displayed an incorrect keyboard shortcut for executing a do-file. This has been fixed.

53. Pressing any function key in the Data Editor would clear the contents of the current cell and enable editing. The correct behavior is for the F2 key to enable editing in the current cell and to move the text cursor to the end of the current cell's contents; all other function key presses are to be ignored. This has been fixed.

54. The Variables window columns would sometimes resize so that all columns would appear, even if the user intentionally resized the columns to hide some columns such as the Type and Format columns. This has been fixed.

55. In the Variables Manager, if the Reset or Apply button had been clicked on first, the variable properties would not immediately update after clicking on an arrow to navigate between variables. This has been fixed.

56. A command that took some time to execute would not appear in the Results window until it had completed. Stata for Mac now immediately displays the command in the Results window.

57. Listbox controls in Stata's programmable dialogs always defaulted to the first item in the listbox instead of the last selected item when a dialog was reopened. This has been fixed.

Stata executable, Unix

58. The Command window now allows the use of the arrow keys in combination with the Control key to navigate the cursor by word or to move the cursor to the beginning or the end of a line.

59. The Data Editor has the following fixes:

a. Pressing any function key in the Data Editor would clear the contents of the current cell and enable editing. The correct behavior is for the F2 key to enable editing in the current cell and to move the text cursor to the end of the current cell's contents; all other function key presses are to be ignored. This has been fixed.

b. The Data Editor would only use its own private Clipboard data, even when text was copied in another application. This has been fixed.

c. The Data Editor could crash when right-clicking within the Data Editor window. This has been fixed.

-------- update 08dec2009 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 9(4).

2. estat effects and estat lceffects incorrectly reported zeros for meff and meft statistics when one or more of the model coefficients contained the new factor-variable operators. This has been fixed.

3. estat overid, when used after ivregress 2sls, incorrectly reported a heteroskedasticity-robust score statistic if cluster-robust standard errors were reported at estimation time. Now a message indicating that the robust test is not available is displayed in that case.

Also, estat overid, when used after ivregress liml, reported an incorrect Basmann statistic and denominator degrees of freedom. This has been fixed.

4. gmm reported incorrect point estimates and standard errors when an HAC weight matrix was used and the time series contained one or more gaps in the first j observations, where j is the number of lags used to compute the HAC matrix. If a non-HAC weight matrix was used in conjunction with an HAC covariance matrix, the standard errors would be incorrect in this situation. This has been fixed.

5. heckman for some datasets would take a much longer time computing starting values compared with when it was called under version control less than 11. This has been fixed.

6. ivregress reported incorrect standard errors when an HAC covariance matrix was used and the time series contained one or more gaps in the first j observations, where j is the number of lags used to compute the HAC matrix. In addition, if the GMM estimator was used with an HAC weight matrix, the point estimates would also be incorrect in this situation. This has been fixed.

7. lrtest, when used to compare models fit with xtmixed, xtmelogit, or xtmepoisson, calculated an incorrect test degrees of freedom if the fixed-effects specifications of these models contained factor variables and if the number of omitted coefficients differed between models. This has been fixed.

8. margins with options dydx() and asbalanced and using the chain rule to compute derivatives with respect to the model coefficients produced slightly incorrect standard errors. Some numerical derivatives were yielding a zero when they should have resulted in nonzero values. This has been fixed.

9. mean, proportion, ratio, and total now run faster when run quietly because they now skip the code used to formulate the table of estimation results. Previously, only the actual display of the estimation table was suppressed. You will notice the speed increase the most when using prefix commands such as bootstrap and svy jackknife, because they execute estimation commands repeatedly and quietly.

10. merge used with unsorted data would sometimes fail to sort the data and thus issue the error message that the master or using data were unsorted. This has been fixed.

11. mfp and fracpoly now support stcrreg.

12. mfp and fracpoly used incorrect degrees of freedom for the fractional polynomial models when used with mlogit (as of Stata 11) or with glm, which led to incorrect p-values reported for the deviance tests. As a result, in the presence of binary predictors, mfp: glm reported a missing p-value for the deviance test for binary predictors and could mistakenly exclude the binary predictors from the final model. The above issues have been fixed.

13. ml display, when the neq() option was specified, was incorrectly applying eform to all equations. It should have applied it only to the first e(k_eform) equations (e(k_eform) is 1 by default). This has been fixed.

14. nlsur exited with an error if a cluster-robust VCE was requested and the cluster variable was a string variable. This has been fixed.

15. probit and logit did not respect the minimum abbreviation, nocon, for the noconstant option. This has been fixed.

16. _rmcoll2list was incorrectly truncating the varlist result r(blist). This has been fixed.

17. _rmdcoll did not accept the expand option as documented. This has been fixed.

18. roctab, binomial reported incorrect binomial confidence intervals when the estimate of the area under the receiver characteristic curve was 1. This has been fixed.

19. tabodds, or now reports missing values for the confidence interval for the base category when one of woolf, tb, or cornfield is specified.

20. xtivreg produced erroneous results if the same variable was specified both as the dependent variable and in the list of additional instruments. Now an error message is produced.

21. xtmixed now allows the time-indexing variable to contain values equal to zero when modeling the residual-error structure as an unstructured matrix. Previously, the time-indexing variable was restricted to be a positive integer.

22. xtreg, fe exited with an error if a cluster-robust VCE was requested and the cluster variable was a string variable. This has been fixed.

-------- update 21oct2009 -----------------------------------------------------

Ado-files

1. icd9 had its databases updated to use codes through the V27 update released on 1 Oct 2009.

2. ivregress, in certain cases involving time-series operators with variables bound in parentheses, reported a "parentheses unbalanced" error. This has been fixed.

3. margins with option eydx() or eyex() (regardless of whether option atmeans or at() is also specified) would report the error "default prediction is a function of possibly stochastic quantities other than e(b)" for some models producing negative predicted values using the current data. This has been fixed.

4. The At tab on the margins dialog box now lists the overall statistics in the statistics combo box.

5. Mata function optimize() failed to replace the old parameter values with new ones specified in optimize_init_params(). This has been fixed.

6. Mata function optimize_evaluate() reported the "attempt to dereference NULL pointer" run-time error when used without first calling optimize(). This has been fixed.

7. regress did not allow all display options to be abbreviated when replaying results. This has been fixed.

8. The dialog box for stcurve failed to output options specified on the Legend tab. This has been fixed.

9. xtline dropped observations that contained missing values in the variables being plotted. This was not intended and has been fixed.

10. predict after xtmixed now allows out-of-sample predictions that are based on estimated random effects, provided that the out-of-sample observations correspond to groups that are represented in the estimation sample. The one exception is standardized residuals (option rstandard), which are statistically appropriate only within the estimation sample.

11. xtpcse with aweights and factor variables included in the model reported an error message. This has been fixed.

Stata executable, all platforms

12. graph export could crash when exporting PostScript graphs using a nondefault font that could not be found on the computer. This has been fixed.

13. label language ..., delete, when the language name being deleted was the current dataset language name, would leave some labels in the dataset from the deleted language name. This has been fixed.

14. Mata class matrices were not allowed in optional argument lists of Mata functions. This has been fixed.

15. Mata function eltype() now returns "class" for Mata class matrices and "classdef" for Mata class functions.

16. _ms_extract_varlist with option noomit ignored generic factor terms whose base level was fvset. This has been fixed.

17. regress ran slower than usual for small datasets with a small number of independent variables. This has been fixed.

18. _return failed to clear matrix row and column names from dropped r() results. This would cause set memory to complain that "Stata matrices have not been cleared" and prevent clear matrix from fully removing all of Stata's matrices. This has been fixed.

19. testparm dropped factor terms whose base level was fvset. This has been fixed.

20. The Data Editor has the following new features and fixes:

A. Double-clicking on the column heading for an existing variable now opens the dialog for variable properties.

B. Live filtering can now be disabled.

C. It was not possible to paste variable names longer than 80 characters. Now Stata will accept text longer than 80 characters, attempt to form a valid Stata variable name, and apply the first 80 characters to the variable's label.

21. Stata's old programmable dialog system was mistakenly disabled on Windows and Unix(GUI) platforms in the 26aug2009 update. This functionality has been restored.

Stata executable, Windows

22. The Do-file Editor has the following new features and fixes:

A. New option "Replace all in selection" restricts the replacement of all occurrences to the currently selected text.

B. When a "Replace all" was requested of a string, s, with a string s1 that contained s, an infinite loop happened. This has been fixed.

C. "Balance Braces" under the "Find" submenu in the "Edit" menu did not work if the close brace was the last character of the file. This has been fixed.

D. "Match Brace" under the "Find" submenu in the "Edit" menu selected the section of code between the focused brace and its matching brace. It now selects only the matching brace.

E. Syntax highlighting settings could not be restored to factory defaults. This has been fixed.

F. When a line was selected by clicking on the line number, the selected line and the following line were both executed. This has been fixed.

G. Under certain conditions and when auto-indentation was enabled, the Do-file Editor computed the wrong indentation length for the current line based on the indentation length of the previous line. This has been fixed.

23. Applying the "Factory Settings" with some very specific windowing arrangements could crash Stata. This has been fixed.

24. If the Review window was not initially shown because of a tabbed configuration, commands were not added to the Review window. This has been fixed.

Stata executable, Mac

25. In Mac OS X, PDF is the preferred format for images copied to the Clipboard. However, many legacy applications, such as Microsoft Office 2004, do not support PDF images from the Clipboard. Stata copies an image in both PDF and bitmap to the Clipboard for compatibility with both modern and legacy applications. Some modern applications, such as PowerPoint 2008, will still mistakenly paste in a bitmap image even if the preferred PDF image is available from the Clipboard. To avoid this behavior, you can now disable the inclusion of a bitmap image when copying an image to the Clipboard by opening the Preferences dialog and unchecking the "Include a bitmap image in addition to a PDF image" checkbox or by setting include_bitmap to off.

26. graph export could crash when exporting PostScript graphs in the rare case of a nondefault, TrueType font that could not be converted to PostScript. This has been fixed.

27. The 26aug2009 update introduced a bug in the Graph Editor where changing a selection in a pulldown menu of the graph toolbar would have no effect on the graph being edited. This has been fixed.

28. Opening the Data Editor with an if condition could crash Stata. This has been fixed.

Stata executable, Unix

29. Right-clicking within the Data Editor window could cause it to crash. This has been fixed.

30. Stata does not support Unicode and must convert Unicode characters inputted into Stata to ASCII. Stata now displays a warning when it cannot convert Unicode to ASCII and accepts the text it could convert.

31. Copy Table has been added to the contextual menu for the Results window.

-------- update 30sep2009 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 9(3).

2. areg with interactions in the indepvars would report an invalid operator error if one of the virtual interaction variables was omitted because of collinearity. This has been fixed.

3. clogit for some models would take a much longer time computing starting values compared with when it was called under version control less than 11. This has been fixed.

4. dotplot with option bar produced an error message and did not draw the graph. This has been fixed.

5. estimates use would report a Mata run-time error when attempting to restore estimation results containing a matrix with zero rows or columns. This has been fixed.

6. margins is now faster when marginal effects of factor variables are computed.

7. margins with options eydx() and eyex() would report that "prediction <=0, eydx() not available" when it should not have. This has been fixed.

8. margins with nl results and a marginal effects or elasticity option (for example, dydx(), eyex(), eydx(), or dyex()) would report an error about not finding a variable in the list of covariates, even when option variables() was specified. This has been fixed.

9. mi impute logit used a single uniform random number to obtain all imputed values within an imputation rather than using observation-specific uniform random numbers. This has been fixed.

10. ml and Mata's moptimize() with multiple techniques would report the Mata run-time error "subscript invalid" when searching for improved initial values. This has been fixed.

11. ml init and ml with option init() allowed only factor-variable notation in column names of matrices. Now the name=# syntax supports factor-variable specifications.

12. ml plot did not allow factor-variable notation. This has been fixed.

13. _ms_unab is a new undocumented programmers' tool that unabbreviates matrix stripe elements by using the variable names in the current dataset.

14. Mata function optimize() used with evaluator d0 or v0 when the numerical derivatives could not be computed would sometimes crash Stata. This has been fixed.

15. postrtoe is a new undocumented command for moving results stored in r() into e().

16. stci, when used with prefix by, option by(), if, or in, produced correct confidence intervals but reported standard errors that were based on a survivor-function estimate that did not restrict the sample. This has been fixed.

17. sts graph, hazard ignored value labels of the variable specified within option by() when displaying the graph legend. It also did not respect the variable label of the by-variable in the note displayed when option separate was used. These issues have been fixed.

18. sunflower with option addplot() would only render the plots in option addplot(). This has been fixed.

19. svy regression models would incorrectly report missing standard errors for multistage designs that had a strata variable svyset in stage k+1 but not in stage k. svy was incorrectly determining that every stratum in stage k contained only one sampling unit. This has been fixed.

20. svy: tabulate allowed option over() when it should not have. This has been fixed.

21. xtgee for some models would take a much longer time computing starting values compared with when it was called under version control less than 11. This has been fixed.

-------- update 14sep2009 -----------------------------------------------------

Ado-files

1. Almost all Stata estimation commands now run faster when run quietly because they now skip the code used to formulate the table of estimation results. Previously, only the actual displaying of the estimation table was suppressed. You will notice the speed increase the most when using prefix commands such as bootstrap and fracpoly, because they execute estimation commands repeatedly and quietly.

2. gmm exited with an error message if you used the moment-evaluator program version, specified derivatives, and used the xtinstruments() option. This has been fixed.

3. mi impute mvn and mi impute monotone dialogs did not allow multiple imputed variables. This has been fixed.

4. reshape issued an r(181) error if your j variable had a value label and you reshaped from long to wide and back to long, but you added the string option when reshaping back to long. This has been fixed.

5. stcox, when used with options tvc() and texp(), now issues an error message when the expression specified in texp() is not a proper function of analysis time, _t.

6. xthtaylor with option vce(bootstrap) returned an error message when the model contained collinear terms. This has been fixed.

7. xtlogit with option fe returned a conformability error when a model containing collinear variables was run under version 10.1 or earlier. This has been fixed.

8. xtmixed, when used to model standard linear regression (no random effects) with heteroskedastic residuals, has been modified to avoid calculations involving very large matrices.

Stata executable, Mac

9. Some graphs would render to the screen with a stroke or fill color of black for some graph objects. This has been fixed.

10. Clicking a link in the Viewer that is supposed to go to a named destination within a help file would sometimes not go to the correct destination within the help file. This has been fixed.

-------- update 26aug2009 -----------------------------------------------------

1. The 18aug2009 Stata 10.1 update items 4, 15, 19, 25, 26, 30, 43, 59, 65, 66, and 69 have now been applied to Stata 11. The other items from that update were applied to Stata 11 prior to its initial release.

Ado-files

2. ereturn display exits with an error if scalar e(k_eq) contains a value that is not equal to the number of equations in e(b). Prior to Stata 11, ereturn display was oblivious to the scalars in e(). The old behavior is now preserved under version control.

3. margins, when used with svy results that employed option subpop() and when some of the independent variables contained missing values in observations outside the subpopulation sample, incorrectly refused to estimate margins under two additional conditions.

a. When option subpop() was also specified on command margins and when option vce(unconditional) was specified on command margins, margins incorrectly reported an error that observations were dropped from the estimation sample. This has been fixed.

b. When option subpop() was not specified on command margins, margins would incorrectly report all margins as "(not estimable)", even for margins that were estimable. This has been fixed.

4. Mata function moptimize_init_eq_colnames() would only work with equations with a single predictor. This has been fixed.

5. Mata function optimize() with constraints and optimize_init_tracelevel() set to "params", "step", "gradient", or "hessian" would report a conformability error in Mata. This has been fixed.

6. mi import ice with option automatic and when the dataset being imported contained string variables exited with a "type mismatch" error. This has been fixed.

7. ml model dropped constant-only equations that were not supplied with an equation name. Thus the specification

. ml model lf myeval (y = x) ()

was treated as

. ml model lf myeval (y = x).

This has been fixed.

8. predict with both option smooth and option rmse() after sspace with method(hybrid) or method(dejong) and with a nonstationary state-space model produced root mean squared error values that were too small. This effect was less pronounced for later values in the series. This has been fixed.

9. regress with prefix by reported an error message in addition to the usual warning when the last displayed group contained no observations. This has been fixed.

10. xtmixed, when used to estimate the residual variance structure in a model with no random effects, would issue a Mata run-time error if either factor variables were specified or variables were omitted because of collinearity in the fixed-effects portion of the model. This has been fixed.

11. xtunitroot would exit with various obscure error messages or report "no observations" if the variable being tested contained missing values for the first time period for all panels, even if the test could be conducted using subsequent time periods' data. This has been fixed.

Stata executable, all platforms

12. anova with option repeated(), when option dropemptycells was not specified and when there were empty cells in the between-subjects error term, reported values too large for the between-subjects error term, e(N_bse), and the Huynh-Feldt corrections, e(hf#), and the p-values based on these correction factors were consequently incorrect. This has been fixed. The Greenhouse-Geisser and Box corrections and the p-values based upon them were correct.

13. anova with ill-conditioned data suffered loss of precision. This has been fixed.

14. Calling browse or edit commands from a do-file where the command specified a varlist or restricted the observations using if or in showed the entire dataset inside the Data Editor. This has been fixed.

15. clist failed to recognize the data in memory. This has been fixed.

16. Function colnumb() failed to match an equation specification on columns containing time-series operators or factor variables. This has been fixed.

17. margins with a nonlinear prediction after ologit, oprobit, or any other model without an intercept in one of its equations would attempt to use the chain rule for covariate patterns that it should not have, resulting in a subscript invalid error from Mata. This has been fixed.

18. matrix colnames failed to unset equation names when used with an unadorned colon character, as in matrix coleq b = :. This has been fixed.

19. The quoted option of odbc insert did not work. This has been fixed.

20. suest with the same factor variable but different levels across multiple estimation results would report the factor-variable name in an incorrect position in the regression table. This has been fixed.

21. In the Data Editor, pasting external Clipboard data did not support variable names that were longer than 32 characters. In this situation, Stata will now attempt to form a valid variable name and store the original long name as the variable's label.

22. In the Data Editor, editing the contents of a string variable, which resulted in the variable's string width increasing, caused the Data Editor to display the new width improperly. This has been fixed.

23. The Window menu now has an entry for the Variables Manager.

Stata executable, Windows

24. Windows Metafiles (WMF) created by graph export or copied to the Clipboard had incorrectly positioned text. This has been fixed.

25. graph export to PNG and TIFF formats using option width() crashed Stata. This has been fixed.

26. window menu refresh and window menu clear sometimes caused the restore icon for the Results window to disappear. This has been fixed.

27. In the Data Editor, deleting a variable that was used to filter observations sometimes caused Stata to crash. This has been fixed.

28. In the Variables Manager, unpinning the Variable Properties window caused the buttons for managing value labels and notes to not work. This has been fixed.

Stata executable, Mac

29. If the Data Editor was already open when data was loaded, Stata sometimes crashed. This has been fixed.

30. Stata will no longer resave an unmodified do-file when the do-file is executed from the Do-file Editor.

31. 64-bit Mac: The Do-file Editor now automatically adds an end-of-line delimiter at the end of a do-file if one does not already exist. Stata requires that an end-of-line delimiter exist on all lines in a do-file that are to be executed.

32. After changing the aspect ratio of a graph from the Graph Editor, the graph window would not automatically resize to the new aspect ratio. This has been fixed.

33. The Make Text Bigger/Smaller menu items were disabled if the content area of a Viewer was clicked. This has been fixed.

Stata executable, Unix

34. You can now double-click on a snapshot in the Data Editor snapshot dialog to restore the snapshot.

35. The Variables Manager reset button issued a warning before resetting the variables properties. This has been fixed.

-------- update 14aug2009 -----------------------------------------------------

Ado-files

1. Online help and the search index have been brought up to date for Stata Journal 9(2).

2. estat phtest after stcox reported an error message when factor variables were included in the model. This has been fixed.

3. estat summarize after nlsur exited with an error if option variables() was not specified with nlsur. This has been fixed.

4. estimates table reported the error "nothing found where name expected" when splitting a wide table. This has been fixed.

5. ml display failed to recognize Stata 10 estimation results that were restored using estimates use. This has been fixed.

6. _robust ignored the weight specification. This affected only user-written commands that used _robust directly. It did not affect any official Stata estimators, nor did it affect any user-written estimators that used ml or Mata functions optimize() or moptimize(). This has been fixed.

7. xtmixed, when used to fit models with no random effects and with heteroskedastic residual variances, would perform an LR test (comparing the mixed model with standard linear regression) that failed to acknowledge that the mixed model had no random effects.

If the heteroskedasticity occurred over exactly two groups, xtmixed reported a chibar(01) test statistic when instead it should have reported a standard chi-squared with one degree of freedom. If there were more than two groups, xtmixed labeled the chi-squared test as conservative when, in fact, the test was precise. This has been fixed.

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

See whatsnew10to11.

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


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