Stata 15 help for whatsnew8

Additions made to Stata during version 8

This file records the additions and fixes made to Stata during the 8.0, 8.1, and 8.2 releases:

+---------------------------------------------------------------+ | help file contents years | |---------------------------------------------------------------| | whatsnew Stata 15.0 and 15.1 2017 to present | | whatsnew14to15 Stata 15.0 new release 2017 | | whatsnew14 Stata 14.0, 14.1, and 14.2 2015 to 2017 | | whatsnew13to14 Stata 14.0 new release 2015 | | whatsnew13 Stata 13.0 and 13.1 2013 to 2015 | | whatsnew12to13 Stata 13.0 new release 2013 | | whatsnew12 Stata 12.0 and 12.1 2011 to 2013 | | whatsnew11to12 Stata 12.0 new release 2011 | | whatsnew11 Stata 11.0, 11.1, and 11.2 2009 to 2011 | | whatsnew10to11 Stata 11.0 new release 2009 | | whatsnew10 Stata 10.0 and 10.1 2007 to 2009 | | whatsnew9to10 Stata 10.0 new release 2007 | | whatsnew9 Stata 9.0, 9.1, and 9.2 2005 to 2007 | | whatsnew8to9 Stata 9.0 new release 2005 | | this file 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 whatsnew8to9.

-------- update 24feb2005 -----------------------------------------------------


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

2. After the update of 09dec2004 fcast graph, irf graph. irf cgraph, irf ograph, and sts graph when used with a monochrome scheme, such as s2mono, and when plotting more than one line, plotted all lines with the same solid style. Each line is now plotted with a unique pattern.

3. graph box was not fully compliant with the new setting to turn off variable abbreviation - set varabbrev off. In the specific case where more than one yvar was specified in combination with the over() option, graph box terminated with an error message; this has been fixed.

4. svytab incorrectly computed the design effects when supplied with the row option with one or more of the se, ci, deff, and deft options; this has been fixed.

-------- update 10jan2005 -----------------------------------------------------


1. streg ignored the level() option at estimation; this has been fixed.

2. streg issued a meaningless error message when noconstant was specified and there were no regressors specified. streg now issues an appropriate error message.

3. rocfit erroneously gave one-sided significance levels for individual parameter tests that were labeled as two-sided. rocfit now gives the correct two-sided significance levels.

Stata executable, all platforms

4. lowess and twoway lowess produced slightly different graphs when run multiple times using the same data and there were ties in the x variable; this has been fixed.

5. tabi with the exact option after the 05oct2004 update sometimes reported an incorrect p-value for Fisher's exact test; this has been fixed.

6. translate with the translator(smcl2log) option caused Stata to exit; this has been fixed.

7. If a matrix expression referred to a variable through observation subscripting, such as matrix X[2,4] = mpg[3], the expression, in rare cases, returned missing, even though the variable contained a nonmissing value in the observation; this has been fixed.

-------- update 20dec2004 -----------------------------------------------------


1. estimates stats displayed correct results, but the returned results matrix r(S) had the ll0 and ll columns switched; this has been fixed.

2. sdtest and sdtesti produced an incorrect p-value for the two-sided F test; this has been fixed. The format of the output has also been improved.

3. statsby now allows string variables in the by() option.

-------- update 17dec2004 -----------------------------------------------------

Stata executable, all platforms

1. anova with the repeated() option, specifying a pattern of repeated measure variables such that no subject had a complete set of observed levels, caused the e(Srep) matrix and the Huynh-Feldt and Greenhouse-Geisser epsilons to be incorrect. The regular ANOVA results and the Box epsilon were not harmed. This has been fixed.

2. clogit now uses better starting values and does not resort the data at log-likelihood evaluation. Both changes result in faster execution times for most problems, notably for large datasets.

3. filefilter now accepts file paths without the need to double-quote the path.

4. (GUI) The label define dialog now allows you to continue adding/modifying labels until you cancel. Stata previously required you to click on the Add or Modify button for each value label you wanted to add.

5. (GUI) The label define dialog displayed the numeric representations of missing values; this has been fixed.

6. odbc load (available for Windows, Mac OS X, and Linux) has two new options, allstring and datestring, which import either all data or just dates as strings; see help odbc.

7. postfile with the every() option specified as every(0) caused Stata to crash; this has been fixed.

8. rmdir is a new command for removing an existing directory (folder); see help rmdir.

9. tabulate with the exact option took an excessive amount of time if the number of rows of a contingency table was greater than the number of columns. In such a case, Stata now works with the transpose of the table to dramatically decrease the computational time for Fisher's exact test.

Stata executable, Windows

10. You may now copy the contents of the Review window to the Clipboard.

11. The Do-file Editor is limited to 30K of text. However, text could be pasted into the editor that would exceed this limit. This has been fixed.

12. dialog VARNAME and VARLIST controls will now display the varlist even if there is a --more-- condition. Stata previously displayed an error message that it was busy.

13. The label define dialog showed only the first 10 characters of a label when it was being modified; this has been fixed.

Stata executable, Mac

14. There is now a console version of Stata/SE for the Mac. See for more information.

15. When running Stata remotely to execute a batch job, Stata now treats the -b option as an -e option to prevent the alert dialog from appearing.

16. Graphs created while Stata is running in batch mode no longer appear on the screen.

17. dialog VARNAME and VARLIST controls will now display the varlist even if there is a --more-- condition. Stata previously displayed an error message that it was busy.

18. In the Do-file Editor, if you found some text and selected Replace..., it ignored the highlighted text and skipped down to the next occurrence and replaced that instead. This has been fixed.

19. The tiling of newly opened Do-file Editor windows has been fixed.

20. If the Viewer's scheme was changed and Default Windowing was selected from the Prefs menu, the Viewer's scheme would not be restored to the default setting. This has been fixed.

21. The Viewer button on the toolbar brought up a prompt for a file to view when it should have just displayed the Viewer; this has been fixed.

Stata executable, Unix (GUI)

22. The Data Editor now allows you to accept changes to the data when exiting, discard changes to the data when exiting, and cancel exiting. It previously did not allow you to cancel exiting.

23. When you refreshed a long file opened in the Viewer, the Viewer would flicker as the file was being refreshed. This has been fixed.

-------- update 09dec2004 -----------------------------------------------------


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

2. graph twoway, graph bar, graph box, and graph dot have a new aspectratio() option to control the aspect ratio of a graph's plot region; see help aspect_option.

3. The default connecting line patterns for the s2mono and s2manual schemes have been improved to take advantage of the increased flexibility available to schemes. The default connecting line patterns drawn by twoway line continue to cycle through dashed and dotted patterns, while those drawn by twoway connected and twoway scatter -- with the connect() option -- are now all solid.

4. Graph margin settings made by the margin() option of graph display would "stick" and apply the same margins to all subsequently drawn graphs until a discard command was issued; this has been fixed.

-------- update 24nov2004 -----------------------------------------------------


1. Documentation for graph scheme files is now available; see help scheme files.

2. The Data menu has been slightly reorganized. The Labels & Notes submenu has been split into two separate entries. Under Labels, value labels have been organized into their own submenu.

3. cc and cci with small numbers in the case-control table and when cornfield was specified entered an endless loop; this has been fixed.

4. cf produced an incorrect message when the dataset contained a variable named _merge; it now produces the correct error message.

5. cf created local macros named after every variable in the dataset. This could cause a naming conflict with the local macros used in cf. This has been fixed.

6. clonevar has improved error messages.

7. duplicates report is now r-class and returns r(unique_value) and r(N). r(unique_value) is the count of unique observations, while r(N) is the total number of observations.

8. graph twoway has a new pcycle() option that specifies the number of plots on a graph before the pstyles recycle back to the first style; see help advanced options.

9. loneway now allows the group variable to be a string.

10. The mfp dialog incorrectly allowed oprobit and ologit; this has been fixed.

11. roccomp with the graph option now supports turning off the default grid lines using the ylabel(,nogrid) and xlabel(,nogrid) options.

12. sampsi now

a. has a new option nocontinuity.

b. displays a note when assumptions for large one- and two-sample tests on proportions are violated.

c. saves r(warning) in returned results.

d. displays an error message when both r() and onesample options are specified.

13. separate with the missing option failed if the variable contained extended missing values (.a, .b, ..., .z); this has been fixed.

14. stcurve added an extra space between the filename and the extension if the filename was quoted in the outfile() option; this has been fixed.

15. svyprop with the subpop() option could incorrectly exit with a "stratum with only one PSU detected" error; this has been fixed.

16. tabodds now interprets base(#) as the category number of expvar when options or base() are used with option cornfield, tb, or woolf.

17. The tobit dialog did not allow only one censoring limit to be specified; this has been fixed.

18. xtgee with the score() option failed to calculate the score if there existed a variable _merge in the dataset; this has been fixed.

19. xtreg , re refused to estimate constant-only models when the data contained unbalanced panels; this has been fixed.

-------- update 06oct2004 -----------------------------------------------------

Stata executable, Windows

1. (Windows GUI only) The 05oct2004 update introduced a problem in the Viewer that caused Stata to crash; this has been fixed.

2. (Windows GUI only) As of the 05oct2004 update, closing Stata while a dialog was open would cause a harmless exception error; this has been fixed.

-------- update 05oct2004 -----------------------------------------------------


1. areg, after the 20aug2004 update, reported a missing value for the F statistic when the model was not full rank, even in cases when it should have reported an F statistic; this has been fixed.

2. bsample without the strata() and cluster() options ignored the if qualifier; this has been fixed.

3. clonevar is a new command that makes an identical copy of an existing variable; see help clonevar. This command is based on work by Nicholas J. Cox, University of Durham.

4. codebook failed if one of the variables had a large number of value labels; this has been fixed.

5. graph pie option plabel( ... , gap()) would be applied to subsequent graphs even if the option were not specified on the subsequent graph command. This has been fixed.

6. graph twoway connected ignored the sort option when a weight was specified; this has been fixed.

7. irf refused to estimate the IRFs after estimating a univariate VAR via var; this has been fixed.

8. lroc now preserves the sort order of the data.

9. mfp has a new option, aic, for selecting models by the Akaike information criterion (AIC); see help mfp.

10. Many commands were updated for better version-control handling of extended missing values. See item #20 below.

Stata executable, all platforms

11. gllamm now has more utility programs internalized in Stata's executable, further speeding up estimation for certain classes of models, including models with composite links and latent class models with class membership determined by covariates. See

12. format date and time modifiers 'c' and 'y' did not omit leading zeros when they should have. They behaved exactly like 'C' and 'Y'. This has been fixed.

13. graph export created invalid PostScript or Encapsulated PostScript files when graph labels were longer than 128 characters; this has been fixed.

14. ibeta(a,b,x) has been improved to 10 digits of accuracy when 0 < a, b < 100,000. Additionally, functions atan(), asin(), and acos() now have greater accuracy.

15. matrix opaccum and matrix glsaccum, with a string group variable group(), caused Stata to crash; this has been fixed.

16. predict, r is documented as the minimal abbreviation for obtaining residuals after many estimation commands, yet in some cases, such as regress with robust standard errors, a minimal abbreviation of re was required. predict now allows r as the minimum abbreviation.

17. set varabbrev is a new command that allows users to turn off variable abbreviation in Stata. See help set varabbrev.

18. tabulate and tabi with the exact option are now significantly faster; see help tabulate.

19. tabulate with both the generate() and missing options generated a dummy variable containing . rather than 1 for observations where varname was equal to missing; this has been fixed.

20. version #, missing ignored the missing option in the standalone case. When used in a one-line manner, such as version #, missing : cmd, it caused an ado-file with # less than 8 to understand the extended missing values. Now, in all cases, the missing option causes Stata to adopt the missing-value behavior of the calling program. That is, an ado-file with version #, missing, when called by a program that understands extended missing values, will also understand extended missing values. version #, missing, when called by a program that does not understand extended missing values, will also not understand extended missing values.

21. window stopbox with long text caused Stata to crash. Stata now accepts text up to 500 characters in each argument, and if the limit is exceeded, a warning message is given.

22. When you pasted into a single cell, the Data Editor could interpret a string containing a comma as two cells; this has been fixed.

23. Specified heights for dialog VARLIST and VARNAME controls are now ignored, and the default height is always used.

Stata executable, Windows

24. You can now paste to the Command window when the Results, Graph, Variables, or Review window is in the front by selecting Paste Text in Command Window from the Edit menu.

25. Variable Properties, Sort, Hide, and Delete menu items have been added to the contextual menu of the Data Editor.

26. dir now handles file sizes greater than 2 GB. The output of dir was changed slightly to better accommodate these large file sizes.

27. The dropdown portion of dialog COMBOBOX controls now resizes based on the longest text contained in their initial list.

28. If you pressed Enter in a dialog without an OK button or pressed Esc in a dialog without a Cancel button, Stata crashed. This has been fixed.

29. A timing issue with the Windows Save As dialog caused an error when you tried to save a dataset to a floppy drive. This has been fixed.

Stata executable, Mac

30. You can now paste to the Command window when the Results, Graph, Variables, or Review window is in the front by selecting Paste Text in Command Window from the Edit menu.

31. Variable Properties, Sort, Hide, and Delete menu items have been reorganized on the contextual menu of the Data Editor.

32. The floating Toolbar can now be resized.

33. If the Toolbar was anchored, and you clicked the Zoom button of a window, the titlebar of the window would be obscured by the Toolbar. This has been fixed.

34. Results and Viewer windows could not be resized to the full width of the screen; this has been fixed.

35. When you printed from the Viewer, the margins would be incorrect causing part of the header not to print and the output to be printed with no left margin. This has been fixed.

36. When printing long documents from the Do-file Editor, Stata crashed or might not print the last page. Both problems have been fixed.

37. Right-justified text drawn in the Graph window sometimes was off by a pixel or two; this has been fixed.

38. The dropdown variety of dialog COMBOBOX controls did not scroll text correctly when long text was selected; this has been fixed.

39. Specifying a height for some types of dialog COMBOBOX controls changed the controls' position. Now, the default height is always used, and the position is not affected.

40. Small Stata crashed after drawing a number of graphs or invoking a sequence of dialogs; this has been fixed.

Stata executable, Unix (GUI)

41. Variable Properties, Sort, Hide, and Delete menu items have been added to the contextual menu of the Data Editor.

42. Specifying a height for some types of dialog COMBOBOX controls changed the controls position. Now, the default height is always used, and the position is not affected.

-------- update 01sep2004 -----------------------------------------------------


1. Online help and the search index have been brought up to date for Stata Journal 4(3).

2. dstdize, with the 20aug2004 update, saved an incorrect r(se) matrix; this has been fixed.

-------- update 20aug2004 -----------------------------------------------------


1. areg now returns e(ll) and e(ll_0), allowing the estimates table command to compute AIC and BIC after areg.

2. areg did not use the number of clusters for the denominator degrees of freedom when the cluster() option was used. The F statistic is now set to missing if test drops any variables when trying to compute the Wald statistic for the model.

3. bootstrap, jknife, permute, simulate, and statsby now set more off before looping.

4. dotplot has the new over() option, which is a synonym for the by() option.

5. dstdize now returns the standard errors and confidence intervals in the matrices r(se), r(lb), and r(ub).

6. graph combine has a new option, altshrink, that provides an alternate sizing of the text, markers, line thickness, and line patterns on the individual combined graphs; see help graph combine.

7. mfx has been changed as follows.

a. mfx is no longer allowed after boxcox or vec.

b. mfx displays links to an explanation when it is unsuitable for computing the marginal effects or standard errors.

c. mfx displays e(cmd2) as the estimation command instead of e(cmd) when e(cmd2) is present.

d. mfx now displays an error message if the eqlist() option is used and equation names contain spaces.

e. mfx leaves r() empty at completion. mfx continues to return results to e().

f. mfx gave incorrect standard errors when using the linear method on models that contained a constant equation before an equation containing variables; this has been fixed.

g. mfx failed to complete if the model contained an equation with all coefficients at zero; this has been fixed.

8. predict, te after fitting a model with frontier in which both the uhet() and vhet() options were specified exited with an error; this has been fixed.

9. tabi produced an appropriate error message but left the returned error code _rc as 0 in some cases; this has been fixed.

10. truncreg refused to fit a constant-only model; this has been fixed.

11. xtdes declared that the observations were not unique for the id by time grid if the width of the participation pattern exceeded the value of the width() option.

-------- update 27jul2004 -----------------------------------------------------


1. _get_diparmopts, a programmer's tool for estimation commands, now handles models with a large number of model parameters. Correspondingly, svar no longer exits with an obscure error message when fitting models with a large number of parameters.

2. archlm now allows a single panel from a panel dataset to be specified with an if or in qualifier.

3. blogit and bprobit now issue a more informative error message when either pos_var or pop_var is not integer valued.

4. predict, after xtlogit, xtprobit, and xtcloglog for fitting random-effects models, ignored the if and in qualifiers; this has been fixed.

5. stcurve, after streg or stcox with pweights, issued an error stating that pweights were not allowed, even though they were allowed when fitting the original model. stcurve has been fixed to allow pweights and handle them properly.

6. xttobit ignored the noskip option; this has been fixed.

-------- update 23jul2004 -----------------------------------------------------

The 23jul2004 update includes extensions to Stata's time-series capabilities, including new commands for fitting and analyzing cointegrated vector error-correction models (VECMs); see help vecintro. Several postestimation commands for use after estimating vector autoregressions (VARs) and structural VARs have been renamed to better reflect their function. These new time-series features are documented in a second edition of the Stata Time-Series Reference Manual; see

A list of the other ado-file and executable updates for this update is included after the description of the time-series updates.

Time-series update

1. The new command vec fits cointegrated vector error-correction models (VECMs) using Johansen's method; see help vec.

2. The new command vecrank produces statistics used to determine the number of cointegrating vectors in a VECM, including Johansen's trace and maximum-eigenvalue tests for cointegration; see help vecrank.

3. The new command fcast, which replaces the old command varfcast, produces and graphs dynamic forecasts of the dependent variables after fitting a VAR, SVAR, or VECM; see help fcast.

4. The new command irf, which replaces the old command varirf, does everything the old command did and more. irf estimates the impulse-response functions, cumulative impulse-response functions, orthogonalized impulse-response functions, structural impulse-response functions, and forecast error-variance decompositions after fitting a VAR, SVAR, or VECM. irf can also make graphs and tables of the results. See help irf.

varirf continues to work but is no longer documented. irf accepts .vrf result files created by varirf.

5. The new command veclmar computes Lagrange-multiplier statistics for autocorrelation after fitting a VECM; see help veclmar.

6. The new command vecnorm tests whether the disturbances in a VECM are normally distributed. For each equation, and for all equations jointly, three statistics are computed: a skewness statistic, a kurtosis statistic, and the Jarque-Bera statistic. See help vecnorm.

7. The new command vecstable checks the eigenvalue stability condition after fitting a VECM; see help vecstable.

8. The new command vecstable and the existing command varstable have a graph option that produces publication-quality graphs to facilitate interpreting and presenting the stability results. See help vecstable and help varstable.

9. The output of the following commands has been standardized for easier understanding: var, svar, vargranger, varlmar, varnorm, varsoc, varstable, and varwle.

Additional ado-files

10. ac, bgodfrey, corrgram, cumsp, dfgls, dfuller, durbina, dwstat, pac, pergram, pperron, wntestb, wntestq, and xcorr now allow a single panel from a panel dataset to be specified with an if or in qualifier.

11. glm, irls quietly ignored ml-specific maximize options; this has been fixed.

12. sw, when used with clogit, did not allow the strata() option as a synonym for group(); this has been fixed.

13. xi: command ... run under version 6 or earlier, produced indicator variables appropriate for the calling version but ignored the calling version when executing command, instead running command under version 7. xi now runs command under the calling version.

Stata executable, all platforms

14. A menu item, 'Back', has been added to the Viewer's contextual menu that appears when you right-click on the window. It does the same thing as the 'Back' button on the Viewer.

15. VERSION in .dlg files now takes an optional list of operating systems after the version number. Any combination of WINDOWS, MACINTOSH, or UNIX may be specified. The dialog will then be displayed only on the specified operating system(s). By default, a dialog may be displayed on any operating system.

16. The maximum number of description lines in a stata.toc file has been increased from 10 to 50.

17. anova, sequential, with the "/" syntax for specifying error terms, produced incorrect F statistics; this has been fixed.

18. clogit has new options robust and cluster(); see help clogit. In addition, clogit has been converted from a built-in command to one that now uses ml. As a result, clogit now supports options that are available to ml-programmed estimators, such as constraint() for linear constraints.

19. When #delimit ; was on, // comment lines without a leading space resulted in an error; this has been fixed.

20. factor, ml with collinear variables entered an endless loop. Now it presents an error message.

21. infix, when incorrectly specified without using, caused Stata to crash; this has been fixed.

22. kdensity and twoway kdensity exited with an uninformative error message when supplied with a variable that did not vary; this has been fixed.

23. label variable and label data, when given a label longer than the maximum allowed label length, truncated the label without warning. A warning message is now displayed.

24. rotate caused Stata to crash if a negative number was supplied in the factor() option; this has been fixed.

25. score, in rare cases, did not set the generated variable to a missing value in an observation where one or more of the variables involved in the preceding pca command had missing values; this has been fixed.

Stata executable, Windows

26. When the Command window has the keyboard focus, you can press the Up or Down cursor keys in combination with the Shift key to scroll through the Results window a line at a time. You can also press the Page Up and Page Down keys in combination with the Shift key to scroll through the Results window a page at a time.

27. shell could not pass quoted arguments to a program installed in a path containing a space; this has been fixed.

Stata executable, Mac

28. The maximum number of controls in Stata's old-style dialog programming language has been raised to 300 to be consistent with the Windows and Unix (GUI) versions of Stata. Stata users are still encouraged to use the new dialog programming language when programming dialogs, as the old-style dialogs have been deprecated.

29. In some cases, the Graph window automatically resized beyond the resolution of the screen, preventing the window from being resized manually; this has been fixed.

30. The zoom control in the window titlebar has been reworked so that the ideal window state for a window that is to be zoomed is optimal for the current window. For example, clicking on the zoom box in the Results window or Viewer will resize the window so that it is 80 characters wide and uses the full height of the screen (adjusted to the nearest line height). Clicking on the zoom control in the Data Editor will result in the editor taking up the whole screen. Clicking on the zoom control in the Graph window will result in the current graph taking up as much of the screen as possible while maintaining the correct aspect ratio. Clicking on the zoom control again will restore the window to its previous state.

31. Printing from Stata could potentially lead to a crash; this has been fixed.

32. When printing multiple pages from the Do-file Editor, page margins after the first page were smaller on all sides; this has been fixed.

Stata executable, Unix

33. A command that saved a file with a path having tilde "~" as the first character without the replace option overwrote the file without generating an error or warning; this has been fixed.

34. (GUI) If a do-file was executed by starting Stata with command-line arguments, the scrollbar of the Results window would not be in sync with the number of lines outputted in the Results window; this has been fixed.

35. (Console) If the terminal window in which a console version of Stata for Unix was running was resized, Stata did not recognize the new size of the window. Stata now resets the line width and page length if the terminal window in which it is running is resized.

-------- update 01jul2004 -----------------------------------------------------


1. adjust now allows the pr option after binreg; see help adjust.

2. _coef_table is a new programmer's tool for displaying coefficient tables; see help _coef_table.

3. contract has new options cfreq(), percent(), cpercent(), float, and format() based on the user command pcontract written by Roger Newson, King's College, London; see help contract.

4. (Mac only) graph set without any other arguments produced partial output followed by an error message; this has been fixed.

5. The _tab class is a new programmer's tool for displaying tables; see help _tab.

6. xtdes has new option width(#) specifying the maximum width of the participation pattern display. The default is 100; see help xtdes. The width of the pattern determines how many temporary variables xtdes creates. With a large number of time points, xtdes failed when it exceeded the maximum number of variables allowed in Stata. width() limits xtdes to a reasonable width.

Stata executable, all platforms

7. String scalars are now supported; see help scalar. They are limited to a maximum length of a string in a string expression; see help limits.

8. The extended macro function all scalars now allows numeric or string as a prefix to scalars; see help macro.

9. A warning note is displayed at startup if a new executable has been downloaded without a subsequent update swap to replace the old executable with the new one.

10. The limit for the number of dyadic operators in an expression has been increased from 200 to 500 for Intercooled Stata and Stata/SE; see help limits.

11. SMCL has a new directive, {rcenter:text}, that is the same as {center:text}, except when there is an unequal number of spaces to be placed around the centered text. {rcenter:text} places the extra space to the left of the text, moving the text to the right. {center:text} places the extra space to the right of the text, moving the text to the left; see help smcl.

Previously, {center:text} acted inconsistently. With a width argument, it placed the extra space on the right. Without a width argument, it placed the extra space on the left. Now, with or without a width argument, {center:text} places the extra space to the right.

12. Raw SMCL directives were sometimes displayed when help include (.ihlp) files were used; this has been fixed.

13. If there are multiple errors in a dialog programming if command, the first error is now reported rather than the last.

14. stcox, basechazard() produced incorrect estimates of the baseline cumulative hazard when the efron option for ties was specified; this has been fixed.

15. summarize with the format option now respects date formats.

Stata executable, Windows

16. Pressing the alt key in combination with a character did not automatically open the appropriate menu from the main Stata window's menubar in some cases. This has been fixed.

17. Pressing the escape key sets the focus to the Command window when working with the Viewer.

18. Keyboard shortcuts now work when the mouse pointer is positioned over the variables window.

19. (Windows 95/98/NT) When the do-file editor was invoked, it might fill the entire screen or be placed beyond the screen (that is, not be shown); this has been fixed.

Stata executable, Mac

20. In the past, Stata used Internet Config to determine the proper type and creator for new files it created based on the file extension; that is, files with an .html extension would have the creator set to Internet Explorer. If a file does not have a type and creator, Mac OS X determines which application should open the file based on its extension. However, Internet Config and Mac OS X do not always agree on which application should be associated with a file. Stata now allows you to turn on or off Internet Config file mapping with the set icmap command or through the General Preferences dialog. The default setting is for Internet Config file mapping to be off; see help icmap.

21. Canceling a print job from the Viewer or Results window caused error dialogs to be displayed for every page that was to be printed. Stata now presents only one dialog when printing has been canceled.

-------- update 24jun2004 -----------------------------------------------------


1. Online help and the search index have been brought up to date for Stata Journal 4(2).

2. bootstrap, permute, simulate, and statsby use the first 80 characters of command to label the resulting dataset. An error resulted if command was extremely long; this has been fixed.

3. codebook correctly recorded noninteger variables with date formats in r(realdate) but did not display the information in the output; this has been fixed.

4. codebook_problems and labelbook_problems are new help files providing guidance for resolving the various problems diagnosed by the problems option of the codebook and labelbook commands.

5. dfuller has new option drift for testing the null hypothesis of a random walk with drift, corresponding to Hamilton's (1994) case three. The algorithm for calculating MacKinnon's approximate p-values is now more accurate in cases where the p-value is relatively large. See help dfuller.

6. egen ... group(), label exited with an error when group() variables shared the same value label; this has been fixed.

7. The graph dot rwidth() option did not allow *# relativesize specifications; this has been fixed.

8. jknife sometimes did not perform all replications when used with non-eclass commands and datasets with missing values; this has been fixed.

9. ml sometimes produced a missing model F statistic when fitting a model with an offset using survey data; this has been fixed.

10. pperron now calculates MacKinnon's approximate p-values more accurately when the p-value is relatively large.

11. recode variable ..., generate() exited with an error when variable was of type long or double; this has been fixed.

12. twoway area now accepts the option cmissing(n) and interprets it to mean that missing values in the plotted data mark the end of an area and that the next set of nonmissing data begins a new, possibly disjoint, area.

13. var with panel data did not allow the by prefix command; this has been fixed.

14. xi now allows the alternative (undocumented) syntax:

xi term(s) [, prefix(string)]

15. xtline did not allow the recast() option; this has been fixed.

16. Some commands failed if the system used a temporary directory where the path contained spaces; this has been fixed.

-------- update 18may2004 -----------------------------------------------------

The 18may2004 executable and ado update includes 60 changes and improvements. We recommend that you read the entire numbered list below. Highlights:

o The Windows interface has been improved. The improvements include the ability to have the Review and Variables windows inside or outside the main Stata window; see #26 below and help revwindow. Also see items #25-47 below.

The 18may2004 executable will not change the current settings of existing users of Stata. To take advantage of some of the new improvements, you must change your window settings from the new Windowing tab on the General Preferences dialog or select Default Windowing from the Preferences menu.

o Windows and Mac dialog variable fields have a drop-down list containing the current variables. See #42a and #48 below.

o help now allows include files and accepts spaces and trailing parentheses so that requests such as help log(), help mat acc, and help two conn send you to the help for the log() function, matrix accum command, and twoway connected command, respectively. See #10 below.

o anova is now faster and returns more items in e(). See #2 and #3 below and help anova.

o merge and append now deal with notes. See #18 below and help merge and append.


1. codebook, labelbook, and numlabel now deal better with labels containing left-quote (`) characters. See #15 below.

Stata executable, all platforms

2. anova with the repeated() option and the replay of all runs of anova are now much faster.

3. anova now returns additional items in e() -- e(sstype), e(term_#), e(ss_#), e(df_#), and e(F_#) for those terms where an F statistic is calculated. When the test of a term does not involve residual error, the following are also returned e(errorterm_#), e(ssdenom_#), and e(dfdenom_#).

4. char rename, a low-level programmer's command, created characteristics with duplicate names. "char rename old new" did this if a characteristic with the same name appeared in both old and new. This has been fixed; char rename now issues an error message in such cases. A new option, replace, indicates that, in such cases, old[name] should replace new[name].

5. The Data Editor now has the following fixes and enhancements:

a. A single value may be pasted into the edit field. Doing so in the past renamed the currently selected cell's column.

b. When a single cell is selected, right-clicking on any other cell with data will change the selection to the new cell before presenting a contextual menu.

c. When scrolling to the right, Stata now scrolls into view a newly selected cell that is partially visible. Previously, if a cell's right edge was even with the editor window's right edge (not including the vertical scrollbar), the Data Editor could not scroll further right; this has been fixed.

d. Pressing Control-Home moves the cursor to the first row and column. Pressing Control-End moves the cursor to the last row and column with data.

Also see #29, #36, #40, #41, and #47 for Windows enhancements and fixes to the Data Editor. See #57 and #58 for Unix enhancements and fixes to the Data Editor.

6. dialog PROGRAM execution subcommand stata does not execute if Stata is running another command. This prevents a dialog from breaking the currently running Stata command.

7. A %g format in rare cases caused the number 99.9999... to incorrectly display as 10; this has been fixed.

8. generate, after clear or drop _all and when creating a string variable with fewer observations in the same position as a string variable that existed in the old dataset, put some of the string values from the old dataset in the new dataset. This has been fixed.

9. gllamm utility programs received cosmetic changes to support the current version of the user-written gllamm command; see Most importantly, with aggregated binomial data, the log-likelihood reported by gllamm is now scaled to reflect the actual probability density.

10. help has the following fixes and enhancements:

a. .hlp files may now include .ihlp files. The syntax is INCLUDE help argument. This includes the contents of the argument.ihlp file into the .hlp file. The INCLUDE help directive must begin in the first column.

b. Trailing parentheses "()" in the requested help argument are treated as a request for help for a function. The trailing parentheses are removed, and "f_" is prepended to the request. Typing help log() is equivalent to typing help f_log and provides help for the log() function, while typing help log provides help for the log command.

c. Spaces and colons in the requested help argument are translated to underscores. For example, help graph intro is equivalent to help graph_intro.

d. Additional aliases have been added to allow command abbreviation. For example, help mat acc is the same as help matrix_accum, help gr matrix is the same as help graph_matrix, and help tw con is the same as help twoway_connected.

e. Entering some of the Viewer commands without arguments produced strange results. For example, under the Unix GUI, entering "search" in the Viewer with no item to search for caused the Viewer to flash until an exhaustive search was finished. Stata now returns help for the search command. For example, entering "search" by itself is the same as entering "help search". Similarly, at the Stata command line, whelp search is equivalent to whelp help search.

11. infile with a dictionary specifying a nonexistent raw file presented a message indicating that the file did not exist even when the using() override option was specified with a file that did exist. Now, if the using() override option is specified and the file it names exists, the message is suppressed.

12. insheet's double option is now the default if the default data type has been set to double. You can specify the nodouble option to prevent this. See help insheet.

13. insheet with a using file containing a row of variable names with no data reported that -1 observation had been read into Stata. This led to problems with other commands, such as describe. The problem has been fixed.

14. label language limited languagename to 8 characters. Now the limit is 24 characters.

15. label save now escapes with a backslash all left-quote (`) characters in labels as they are saved to the do-file so that they will be properly re-created when the do-file is executed.

16. macro extended functions rownames, colnames, roweq, coleq, rowfullnames, and colfullnames now accept references to matrices stored in r() and e(); see help matmacfunc.

17. matrix svd produced different, but still correct, answers on different platforms. One solution had columns that were the negative of the other solution. Now, for each column of V, if the sum of the elements of the column is negative, the column is replaced by its negative, and the corresponding column of U is replaced by its negative.

18. merge and append now incorporate all notes from the using dataset that do not already appear in the master dataset. A new option, nonotes, ignores the notes in the using dataset which was the previous behavior. See help merge and append.

19. monthly() time-series function with "ym" as the second argument produced missing values when months in the first argument were abbreviated; this has been fixed.

20. tabulate entered an endless loop when the linesize was set too small; this has been fixed.

21. In rare circumstances, a hostname cached by Stata's internet-aware features was forgotten by Stata, resulting in an error message when contacting that host. This has been fixed.

22. Rarely, the Variables window displayed its variable names off by one row; this has been fixed.

23. Edit--Copy Table did not eliminate separator lines consisting of "-" and "+" characters if the beginning of the line started with spaces; this has been fixed.

24. The Schemes drop-down list on the Graph Preferences dialog now shows all available schemes. Previously it showed only the original schemes shipped with Stata.

Stata executable, Windows

25. Stata now presents a dialog that allows you to save your data before exiting when there is unsaved data.

26. You now can choose to have the Review and Variables windows inside or outside the main Stata window. You can make the choice through the Windowing preferences in the General Preferences dialog or from the command line; see help revwindow. The default preference places them inside the main Stata window.

27. If the position preference of a floating Variables window, floating Review window, or Do-File Editor window would place the window off the screen, Stata now moves the window completely into view at the closest screen edge when the window is shown.

28. Pressing Control-Tab cycles forward through windows inside the main Stata window. Pressing Control-Shift-Tab cycles through the windows in reverse order.

29. A contextual menu has been added to the edit fields of the Viewer and Data Editor windows that allows Cut, Copy, Paste, Clear, and Select All in the edit field. Clicking on the edit field brings its respective windows to the front.

30. The default font is now Lucida Console. Stata now calculates the proper window sizes for the Results window and Viewer window based on the default font. The Review, Variables, and Command windows are then sized and positioned according to the size and position of the Results window.

31. You may now scroll the Results window when it is the frontmost window by using the Up, Down, Page Up, Page Down, Home, and End keys. Any character input is redirected to the Command window.

32. The Home and End keys now scroll to the top or bottom of the current document in the Viewer window.

33. You can change the Results window scroll buffer size in the Preferences dialog.

34. The Go To Line dialog in the Do-file Editor now shows the current line number, as well as the total number of lines in the editor.

35. If the Data, Graphics, or Statistics submenus of the User menu are empty, the empty submenu is disabled.

36. The Data Editor and Viewer edit fields strip font styles when pasting text so that only plain text is pasted in.

37. Control-R is now a synonym for Page Up and retrieves the previous command from the command history.

38. When the font size was increased in the Variables window, the Target text label's font also changed the next time Stata was launched; this has been fixed.

39. Clicking in a blank area of the Review window entered random values into the Command window; this has been fixed.

40. The Data Editor now supports Scroll Lock. When Scroll Lock is on, pressing the cursor keys scrolls the Data Editor rather than moving the cursor within the edit field when the keyboard focus is on the edit field.

41. The Data Editor only allowed you to copy data if there was more than one cell selected. The Data Editor and Browser now allow single cell copies.

42. dialogs have the following fixes and enhancements:

a. dialog VARLIST and VARNAME controls are now drop-down lists that display the current variables.

b. Keyboard shortcuts for Copy, Paste, and Cut now work within dialogs.

c. Pressing the Return key is the same as clicking the OK button. Pressing Shift+Return is the same as clicking Submit. Pressing the Escape key is the same as clicking Cancel.

d. Pressing the space bar when the keyboard focus is on a radio button now simulates a mouse click on the radio button.

e. Keyboard arrow keys now function with dialog spinner controls.

f. In dialog programming, a .setlabel member function resulted in improperly drawn labels when used with a CHECKBOX control that was specified with the groupbox option; this has been fixed.

43. graph export, when writing large emf files, produced an empty file or a file in which the text on the graph had an incorrect font; this has been fixed.

44. tabi, in rare cases depending on processor, produced incorrect output for Fisher's exact test; this has been fixed.

45. window menu actions had quotes stripped from them; This has been fixed.

46. (Windows XP) You may now disable the Windows XP theme in Stata, allowing some dialogs to load faster. See help xptheme.

47. (Windows XP) The wheel mouse did not work in the Data Editor, Review window, and Variables window; this has been fixed.

Stata executable, Mac

48. dialog VARLIST and VARNAME controls are now drop-down lists that display the current variables.

49. Pressing shift+Return in a dialog is the same as selecting the Submit button.

50. Stata did not properly enable and disable the font size menu based on the frontmost active window; this has been fixed.

51. Opening a do-file when the current contents of the Do-File Editor have not been saved replaced its contents; this has been fixed.

52. Changes to the graph scheme in the Graph Preferences dialog were not saved between Stata sessions; this has been fixed.

Stata executable, Unix (GUI)

53. Control-Break can now be used to interrupt a Stata command.

54. A Print Selection checkbox has been added to the print dialog for the Results and Viewer windows when there is a selection. If the checkbox is checked, only the selected text is printed.

55. The default font for a graph has been changed from Times to Helvetica to be consistent with Stata for Windows and Stata for Mac, which default to sans-serif fonts. This will not affect current users unless Default Windowing is selected from the Prefs menu.

56. If the Data, Graphics, or Statistics submenus of the User menu are empty, the empty submenu is disabled.

57. In the Data Editor, pressing the Home key moves the cursor to the first column in the current row. Pressing the End key moves the cursor to the last column in the current row. (Also see #5d, which discusses the actions of pressing Control-Home and Control-End.)

58. Resizing the Data Editor when there was less than a screen's worth of data caused Stata to stop responding; this has been fixed.

59. The dialog LISTBOX control failed to appear; this has been fixed.

60. Pressing the Page Up key while there was a --more-- condition caused the previous command to be entered into the command buffer; this has been fixed.

-------- update 14may2004 -----------------------------------------------------


1. adjust's by() option is no longer required; see help adjust.

2. ci with the by ...: prefix and both the exposure() and total options failed with an uninformative error message. It now correctly executes.

3. glm, when fitting a binomial model with either a log or identity link, now issues a warning message when the estimated parameters produce predicted probabilities outside their admissible range of [0,1]. The warning is warranted for these two links since the linear predictor may be mapped outside [0,1].

4. glm with family(binomial arg), when arg was greater than one, produced an incorrect BIC value; this has been fixed.

5. ltable's saving() option did not accept filenames or paths containing spaces; this has been fixed. Additionally, the dialog for ltable has been reorganized.

6. mvreg now produces better error messages when the residual degrees of freedom are zero or the residual covariance matrix is singular.

7. nptrend now displays three decimal places for the p-value instead of two decimal places. Also nptrend now preserves the sort order of the data.

8. print generated an error if the filename contained more than one period; this has been fixed.

9. recode with rules involving extended missing values, such as (.a/.y=999), did not treat extended missing values as most people would expect. It now recodes the extended missing values as directed.

10. svynbreg with the dispersion(constant) option displayed incorrect misspecification effects; this has been fixed.

11. tsfill did not fill in the dataset if the size of the gap was larger than the maximum value that can be held by an int; this has been fixed.

12. xtabond included invalid instruments when lags of the predetermined variables were included in the model. This problem only arose when 1 or more lags of the predetermined variables were included in the model. Thus, for q > 0, xtabond , pre(x , lagstruct(q , .)) defined x_{it} as predetermined instead of x_{it-q}, causing xtabond to produce inconsistent estimates. This has been fixed.

13. xtabond included some instruments for predetermined variables beyond the limit specified in the lagstruct() option. Assuming a correctly specified model, these extra instruments increased the asymptotic efficiency at the cost of some finite sample bias. This has been fixed.

-------- update 14apr2004 -----------------------------------------------------


1. cii with two arguments now allows the option binomial to be specified. Exact binomial confidence intervals have always been the default in this case, and now the option can be specified explicitly.

2. graph box and graph hbox did not plot outside values on the side where the adjacent value could not be determined; this has been fixed.

3. graph query listed some available schemes twice, once as schemename and once as scheme-schemename. Each scheme is now listed only once as schemename.

4. graph twoway did not work properly with date and left-justified time-series formats; this has been fixed.

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

6. ivreg option score(), added with the 11mar2004 ado-file update, is now named pscore() to distinguish it as a projected score; see help ivreg.

7. ltable option notab is now notable; see help ltable.

8. ml display with the level() option did not produce confidence intervals using the specified confidence level for auxiliary parameters; this has been fixed.

9. palette color has a new cmyk option specifying that color values be reported in CMYK rather than RGB. The position of the color box, line, and symbols has also been shifted left to allow for longer color names.

10. statsby posted missing values when used with a command that did not allow the [in range] qualifier; this has been fixed.

11. tssmooth ma, window() restricted the span of the filter too tightly for some unbalanced panel datasets; this has been fixed.

12. xtpcse could return a nonsymmetric matrix error message when there were a large number of regressors; this has been fixed.

-------- update 22mar2004 -----------------------------------------------------


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

2. mfx now has the following fixes and enhancements:

a. mfx has new options varlist(), tracelvl(), diagnostics() and nodrop; see help mfx.

b. mfx is now much faster.

c. The subcommand compute is now optional.

d. The at() option now allows numlist and matname following multiple-equation estimation. Entering 1 for the constant term(s) in numlist and matname is now optional.

e. mfx performs a new check for the suitability of marginal effects calculation, and has improved the way it determines suitability for the linear method.

f. mfx did not take into account offsets from the preceding estimation command. Now marginal effect will be evaluated at the mean of any offsets.

g. Occasionally mfx produced an error with time-series operators in the independent variable list; this has been fixed.

h. Some standard errors of discrete variables were not computed due to a version-control problem; this has been fixed.

i. mfx displayed "(no effect)" when the marginal effect was zero. Now mfx displays the zero marginal effect and the standard error, if any.

j. mfx is now more responsive to the break key.

3. stcurve, survival, after streg ..., distribution(weibull) ..., erroneously gave plots of S(t | t>1) rather than S(t); this has been fixed.

4. stcurve, after streg, used by default the observed time values in the data as the x-variable in the resulting plot, yet if option range() were specified, stcurve plotted on an evenly spaced grid. stcurve now always plots over an evenly spaced grid, producing smooth-looking curves even in small samples.

5. truncreg now has the following fixes and enhancements:

a. truncreg, marginal, when used with dummy variables, calculates marginal effects for these variables, even though it is often more desirable to obtain the discrete change in the prediction as the dummy variable changes from 0 to 1. When dummy variables are present, a message is now displayed describing how the discrete change(s) may be obtained using mfx.

b. truncreg, marginal now suppresses the display of a marginal effect for the constant (intercept) term, producing output more consistent with mfx.

c. truncreg, marginal gave incorrect standard errors for the estimated marginal effects; this has been fixed.

d. truncreg, marginal, with offset(), fitted the base model, including the offset, but ignored the offset when marginal effects were calculated; this has been fixed.

e. truncreg, with offset(), displayed the offset variable in both displayed equations, even though the offset applies only to the first. truncreg now displays the offset variable only in the first equation.

-------- update 11mar2004 -----------------------------------------------------


1. cluster dendrogram with the labels() option and either the cutvalue() or the cutnumber() option mislabeled the bottom leaves of the dendrogram; this has been fixed.

2. estimates table with the label option and with the varwidth(#) option greater than 32 truncated the displayed variable labels to 32. It now truncates the variable labels to # characters.

3. Graphics dialogs did not allow some title_options and textbox_options to be specified unless text was explicitly supplied. Some statistical graphs and graph commands that accept by_options have default text which could not be changed using graphics dialogs. Graphics dialogs now allow title_options and textbox_options, even if text is not explicitly entered.

4. histogram, with both the by() and normal options, produced the same normal density graph for each by-group; this has been fixed.

5. ivreg now has the score() option; see help ivreg.

6. mfx after svyheckman did not produce standard errors for some of the predict() options; this has been fixed.

7. monochrome schemes now provide better default shading for sunflower plots.

8. predict after glm failed when using the modified option; this has been fixed.

9. predict with the dfbeta option after regress did not label the created variable; this has been fixed.

10. predict with the u or e options exited with an error after xtivreg when the varlist contained time-series operators; this has been fixed.

11. varlmar, with some sample restrictions, included data from the excluded sample in the auxiliary regressions instead of replacing this data with zeros; this has been fixed.

12. zinb with the zip option and no inflation variables, inflate(_cons), produced an error; this has been fixed.

-------- update 19feb2004 -----------------------------------------------------


1. bstat ignored the elements of a matrix supplied to the stat() option if the column names of the matrix did not match the variable names in the bootstrap dataset; this has been fixed.

2. codebook produced an error message when it encountered a labeled variable containing all missing values; this has been fixed.

3. estimates table, when used with the keep() option, required full specification of the coefficients to be kept; that is, eqname:varname, sometimes caused a confusing error message when only varname was specified. Now, if you specify varname within keep(), estimates table assumes you are talking about varname from the first equation.

4. graph, using the textsizestyle quarter_tiny, produced a warning message and incorrectly sized text; this has been fixed.

5. hausman has a new warning message when the rank of the differenced covariance matrix is less than the number of coefficients being tested.

6. histogram with the normal option produced a normal density curve that did not cover the full range of the bars in the histogram; this has been fixed.

7. histogram sometimes plotted an extra bin to the left of what should be the minimum class bin when the input variable was of the type float; this has been fixed.

8. palette color, when issued after Stata started or after a discard command and before issuing any graph commands or other palette commands, showed the RGB values for the specified color to be "255 255 255", regardless of the specified color. This has been fixed.

9. predict ..., hat, when used after glm, gave incorrect values of leverage (hat diagonals) when a noncanonical link was used; this has been fixed.

10. predict ..., mean time, when used after streg ..., dist(gamma), gave incorrect results when the estimated kappa was less than zero due to the integral diverging with kappa < 0. Now, missing values are generated, and a note is presented indicating the problem.

11. set dp comma caused several subsequent commands to fail; this has been fixed. The problem was that statements such as local vv : display _caller() created vv equal to "8,2" (notice the comma instead of decimal point). The way to overcome this is to instead use the statement local vv : display string(_caller()).

12. svygnbreg, svyheckman, svyheckprob, svyintreg, svynbreg, and svypoisson ignored if and in specified in the subpop() option; this has been fixed.

-------- update 30jan2004 -----------------------------------------------------


1. cf failed when the filename contained spaces; this is fixed.

2. lincom did not allow the or, irr, and rrr options after some svy commands when they should have been allowed; this has been fixed.

3. ltable, after the 20jan2004 update, ignored some of the twoway_options when producing a graph; this has been fixed.

4. sts graph with the tmin() option did not restrict the graph appropriately; this has been fixed.

5. twoway, scatter, line, tsline, and xtline failed if the caller's version were set to 7 or earlier; this has been fixed.

Stata executable, all platforms

6. generate, used with time-series operators referring to the variable being created (for example, generate newvar = l.newvar), produced all missing values rather than issuing an error message; this has been fixed.

7. graph now has support for CMYK output to PostScript and Encapsulated PostScript (EPS) files. This support includes the ability to convert RGB color values to CMYK color values on export and to specify either RGB or CMYK values wherever colorstyles are accepted, including options to graph and in scheme files. The new graph export cmyk(on) option allows either PostScript or EPS files to be created with CMYK color specifications. CMYK can be made the default conversion using graph set ps cmyk on and graph set eps cmyk on; see help graph set.

Printing presses require CMYK color separations, and CMYK support is expected to be of greatest use to those using Stata graphics in books or other publications intended for large-scale printing.

8. kdensity has new option epan2 providing an alternate Epanechnikov kernel; see help kdensity. Accordingly, sts graph and stcurve now allow kernel(epan2) for specifying this kernel.

9. postfile set the created dataset data label to the same as that of the dataset in memory. The dataset label is now set to "".

10. sortpreserve, specified as an option to a user-written program that both dropped some observations and added others (something valid for a program to do), caused Stata to enter an endless loop when it tried to restore the sort order of the modified data; this has been fixed.

Stata executable, Mac

11. The Log button on the Toolbar did not reflect the status of the current log file; this has been fixed.

12. Mac OS X 10.3.2 introduced a bug in Apple Type Services for Unicode Imaging (ATSUI) that sometimes caused text to be rendered incorrectly in the graph window. Stata now bypasses ATSUI and draws text directly using the Quartz drawing engine.

13. Mac OS X 10.3 introduced a bug where rectangles drawn in the graph window also drew extra lines. The 15dec2003 update of Stata avoided this problem by drawing a rectangle as a path of four connected lines. Mac OS X 10.3.2 fixed the problem, so Stata now draws a rectangle as it did before the 15dec2003 update.

14. In dialog programs, depending on the order in which they were created, checkbox groupboxes within other checkbox groupboxes did not respond to mouse clicks; this has been fixed. Also, a potential problem where groupboxes and frames could be obscured by larger groupboxes and frames has been fixed.

15. Translating a SMCL file to a PDF file presented an error dialog for every line in the SMCL file and did not create the PDF file; this has been fixed. In addition, when there is a printing error, the error dialog is only presented once for each unique error.

16. As-is graphs were saved with the wrong file permissions when saved from the File menu or the Graph Window's contextual menu; this has been fixed.

17. Stata does not support Unicode characters. However, Stata attempts to open a file even if the path or filename contains Unicode characters. Previously, a failure to convert Unicode to ASCII could cause Stata to crash; this has been fixed.

Stata executable, Unix

18. You can now enable or disable replacing existing Stata datasets when Save is selected from the File menu or activated by a shortcut. From the General Preferences dialog, check or uncheck the checkbox "Overwrite existing data when saving".

19. Pressing the Back button in the Viewer changed the pointer to a hand rather than an arrow; this has been fixed.

20. (Linux only) odbc failed to load the iODBC library on some systems; this has been fixed. You no longer need to set LD_LIBRARY_PATH to work around the problem.

-------- update 20jan2004 -----------------------------------------------------


1. adjust after an estimation command that used the exposure() option failed with a message such as "offset (ln(exposure)) not constant within by()", even when the exposure() variable was constant; this has been fixed.

2. corr2data and drawnorm option cstorage(), added on 06jan2004, read matrix elements by column instead of by row (as indicated in the hlp files); this has been fixed.

3. ltable has two new options, overlay and saving(); see help ltable.

4. mfp ran its component estimation commands under version 6, sometimes causing incompatibilities; this has been fixed.

5. rreg cleared e() before parsing its syntax so that statements such as rreg ... if e(sample) failed; this has been fixed.

-------- update 06jan2004 -----------------------------------------------------


1. Online help and the search index have been brought up to date for Stata Journal 3(4).

2. bootstrap, jknife, permute, and statsby exited with an error when used with a multiple-equation estimation command having no independent variables; this has been fixed.

3. codebook, labelbook, and uselabel have been enhanced to handle multiple-language variable and value labels (added in the 09sep2003 update; see help label_language).

4. codebook has a new option, languages(), for selecting the languages to be reported in the codebook; see help codebook.

5. corr2data and drawnorm now support triangular specification of the correlation or covariance matrix. See the discussion of the new cstorage() option in help corr2data and drawnorm.

6. corr2data has a new option, seed(); see help corr2data.

7. estimates store name failed if variable _est_name already existed; this has been fixed.

8. graph's [x|y]label(alternate) options were not respected after the 03dec2003 update; this has been fixed.

9. graph bar failed when three over() options were specified and each over() option included a sort() suboption. This also happened when yvaroptions() with a sort() suboption was combined with one or more over(, sort()) options. This has been fixed.

10. graph bar, graph box, and graph dot produced an error message when used with a by() option that included a suboption containing a quoted string consisting of a single blank character (for example, graph bar mpg, by(rep78, note(" "))). This has been fixed.

11. graph display's scheme() option failed after the 03dec2003 update; this has been fixed.

12. graph matrix failed with any of the following options: ysize(), xsize(), name(), scheme(), saving(), fxsize(), or fysize() after the 15dec2003 update; this has been fixed.

13. graph pie, in rare cases, did not allow the over() and by() options to be combined after the 03dec2003 update; this has been fixed.

14. graph twoway options tline() and ttext() caused an error when the time variable contained a %d format; this has been fixed.

15. lookfor failed when left single quotes were part of a variable label; this has been fixed.

16. merge with no arguments (an illegal syntax) froze the operation until break was pressed after the 15dec2003 update; this has been fixed.

17. ml has two new subcommands, ml hold and ml unhold, see help ml_hold. We acknowledge the helpful suggestions of Mead Over, The World Bank.

18. recode now produces more informative error messages if value labels are not expected.

19. svmat with the names(col) command now displays an informative error message when one of the matrix column names is _cons.

20. uselabel failed with long value labels; this has been fixed.

-------- update 15dec2003 -----------------------------------------------------


1. arch and arima help have been updated to document the vce() and technique() options, which are preferable to the older method of specifying hessian, opg, bhhh, bfgs, dfgs, dfp, nr, bhhhbfgs(), or bhhhdfp(). These older options continue to work but are now undocumented.

2. Estimation command dialogs now have an improved layout.

3. graph box and graph hbox did not allow you to label the outside values using the same variable being graphed; this has been fixed.

4. maximize help has been updated to document vce(oim), which is an alias to the older vce(hessian) option, which continues to work.

5. ml exited with an error when no dependent variables were specified; this has been fixed.

6. orthog now accepts time-series operators; see help orthog.

7. svyset now accepts the srs option, indicating that the data come from a simple random sample; see help svyset.

8. All svy estimation commands allow an extended syntax for the subpop() option; the extended syntax is

subpop([varname] [if exp] [in range] [, srssubpop])

srssubpop can still be specified outside the subpop() option (this old syntax will no longer be documented). This extension is also available in svymean, svytotal, svyratio, svyprop, and ml.

Stata executable, all platforms

9. by with the illegal syntax by (variables) : cmd now stops and returns an error message and a 198 return code. Previously, it produced incorrect results.

10. c() has the following new items:

c(Wdays) "Sun Mon ... Sat" c(Weekdays) "Sunday Monday Tuesday ... Saturday"

See help creturn.

11. filefilter, when presented with identical filenames for oldfile and newfile and with the replace option, deleted the file and returned an error message. Now, it returns an error message and does not affect the file.

12. list *, after running an estimation command, displayed an extra variable named "(e)"; this has been fixed.

13. merge now accepts multiple using files, and a nosummary option has also been added; see help merge.

14. numlist now allows a longer input numlist.

15. odbc can now connect to IBM Red Brick databases. This fix involved removing any unnecessary spaces from the connection string, even though ODBC syntax allows spaces.

16. odbc insert now converts Stata missing values to NULL prior to insertion.

17. rotate now has a stricter convergence tolerance. Previously, the undocumented ltolerance() option defaulted to ltolerance(0.0001). It now defaults to ltolerance(0.00000001). You can specify ltolerance(0.0001) to obtain the old setting.

18. tabulate reported frequencies without commas (for example, 12345 rather than 12,345) even when there was still room in the table to include commas. tabulate now reports numbers with commas if there is room in the table.

19. trace has a new setting, set tracehilite, that highlights a specified pattern in the trace output. See help trace.

20. test after anova crashed Stata when presented with the illegal syntax test term / (that is, nothing following the "/"). Now, it displays a syntax error message and gives a 198 return code.

Stata executable, Windows

21. When the overall magnification on the Printer tab of the Graph Preferences was changed from 100%, fonts did not print at the right size; this has been fixed.

22. (Windows XP) dialog combo boxes that were supposed to allow multiple selections did not always allow more than one selection; this is fixed.

Stata executable, Mac

23. odbc now uses runtime linking to ensure the latest Apple runtime support for ODBC; see help odbc.

24. Mac OS X 10.3 introduced a bug that causes some of Stata's graphs to display extra lines when the Graph window is of a certain size or smaller. Stata now avoids using the OS X drawing command that causes the extra lines to be drawn.

25. If there is more than 32K of text in the Clipboard, Stata will no longer attempt to copy the data into its local scrap area when Stata is brought to the foreground.

Stata executable, Unix

26. Scrollwheel support has been added to the Unix GUI.

27. The Data Editor now supports the Page Up and Page Down keys for scrolling.

28. log did not recognize ~/ to mean the user's home directory; this has been fixed.

-------- update 03dec2003 -----------------------------------------------------


1. cnsreg now accepts time-series operators; see help cnsreg.

2. graph has improved control over whether the largest and smallest possible grid lines are drawn. This control is provided by improving the actions of the existing suboptions [no]gmin and [no]gmax.

3. graph areas such as boxed titles and legends now recognize suboptions lstyle(none), lwidth(none), lcolor(none), and lpattern(blank) to mean that no line is to be drawn around the area.

4. graph axis labeling options did not allow some legal forms of numlist; this has been fixed.

5. graph pie, in the unusual case when options by() and over() and an if condition were specified, and when no variables were specified, included all over() categories in the legend, even if some were completely excluded by the if condition. Now, such categories are excluded from the legend.

6. graph twoway, when two plots were plotted on a different x or y axis and when the option plotregion(margin()) was also specified, did not apply the specified margin to the second plot. This has been fixed.

7. graph twoway with grids, xlines, and ylines on complicated graphs and with multiple x axes or y axes could, in rare cases, overlay the grids (or xlines or ylines) on top of the plots. Grids, xlines, and ylines now always appear behind the plots.

8. hotelling now mentions and drops collinear variables before computing Hotelling's T-squared. Previously, the T-squared value was correct, but the degrees of freedom, and hence the resulting F test and p value, were incorrect when there were collinear variables.

9. impute has a new option nomissings(); see help impute.

10. iri, csi, cci, and mcci dialogs now allow you to input numbers up to 10,000,000.

11. loneway with aweights and more than 375 groups produced incorrect results; this has been fixed.

12. lookfor is now faster and has more detailed error messages.

13. recode mistakenly reported rule-overlap warnings in some cases; this is fixed.

14. reshape now provides a more detailed error message when the j variable takes on too many values.

15. sts graph has two new options, atriskopts() and lostopts(), that affect how the labels for at-risk and lost observations are rendered; see help sts.

16. sunflower produced graphs with colors even when supplied with a monochrome scheme; this has been fixed.

17. xpose with the varname option when executed three times produced the error message "v1 already defined"; this has been fixed.

-------- update 05nov2003 -----------------------------------------------------


1. ml and all estimators that use ml by default now check convergence using the nrtolerance() convergence criterion: g*inv(H)*g'; see help maximize. This is a true convergence criterion that ensures that the gradient is numerically 0 when scaled by the Hessian -- the shape of the likelihood or pseudolikelihood surface at the optimum. This new default of nrtolerance(1e-5) was added because researchers are programming more likelihoods that have surfaces that are difficult to optimize. It is also an improvement over the gtolerance() criterion that was added in Stata 7 to help ensure convergence of arch and arima models.

In rare cases, this change may cause models that previously stopped iterating without converging to continue iterating and, in extremely rare cases, to continue iterating to the limit of 16,000 if convergence cannot be achieved. nrtolerance() can be turned off with the new nonrtolerance option; see help maximize.

-------- update 31oct2003 -----------------------------------------------------


1. graph bar, graph dot, graph box, and graph pie when used with both options by() and over() and when if conditions were specified included all over categories in the legend, even if some were completely excluded by the if condition. Now, such categories are excluded from the legend.

2. sampsi with the method(ancova) option could return an incorrect result when r1() or r01() was specified with a negative number. This has been fixed.

Stata executable, Windows

3. The change listed in item #49 in the 28oct2003 update had the unintended side-effect of setting the scroll position of the Do-file Editor to the top of the window. In addition, an 'undo' was not possible after pasting text into the editor. This has been fixed.

Stata executable, Mac

4. Due to the 28oct2003 update Stata 8 preferences were mistakenly saved as the file "Stata Preferences" rather than "Stata 8 Preferences"; this has been fixed.

5. Stata failed to uncompress downloaded updates if the path to Stata had a space in it; this has been fixed.

6. The Viewer window would draw text on top of the scrollbar while the window was being resized in Mac OS X 10.3; this has been fixed.

-------- update 28oct2003 -----------------------------------------------------

What's new in release 8.2 (compared to release 8.1)

1. (Windows only) haver use and haver describe are new commands for loading data from Haver Analytics ( database files. See help haver.

2. odbc has a new subcommand, sqlfile, for batch processing of SQL text files; see help odbc.

3. The default axis titles have changed for graph twoway. Previously, if more than one variable was plotted on an axis, the axis showed the labels or variable names of the variables. Now, in such cases, no axis title is drawn, and the variables are identified in the legend by default. The original behavior can still be obtained by setting the version to less than 8.2, for example, typing version 8.1: graph twoway ...; see help version.

4. (Mac only) The Stata for Mac executable format has been changed from a single executable to Mac OS X's preferred executable format, an Application Package. Although you can still launch Stata from the Finder by double-clicking the application from the Stata folder, launching Stata from a terminal or shell script will require a minor change. If Stata is installed in /Applications/Stata, the path to the Stata executable as a single executable was /Applications/Stata/StataSE (for Stata/SE). The path to the Stata executable as part of an Application Package is now /Applications/Stata/

5. (Mac only) A window- and screen-edge snap feature has been added that can be set using the general preferences dialog. When you move or resize a Stata window, it will snap to the nearest edge of the frontmost and closest Stata window or to the edge of the screen. Holding down the Option key while dragging the window disables this feature. The gap between the windows they will snap to can be set.

6. (Mac only) You can move all of Stata's currently open windows at the same time by holding down the Control key while dragging any of Stata's windows. This will also bring all of Stata's currently open windows to the foreground. The screen-edge snap feature can be disabled by also holding down the Option key.

7. (Windows only) In the Do-file Editor, Ctrl+Shift+D has been added as a keyboard shortcut for the menu item "Tools--Do to bottom".

8. Recommendation on setting version at the top of do-files and ado-files: Use version 8.2. version 8.2 adds new features and changes the default behavior of axis titles for graph twoway. (Setting version 8.1 restores graph twoway's old behavior.) See help version.

Other additions and fixes


9. fillin, means, recast, tab1, and tab2 are now a little faster.

10. graph axes with automatically generated or rule-specified tick labels (see help axis_options) could on rare occasions display a tick at the value 0 as an exponential value very close to zero, for example, .11e-219. This is now even more unlikely.

11. graph legends have more precise alignment between keys and their labels. The legends also adjust to accommodate large symbols used as keys.

12. Legends for graphs now accept any valid relative size in the suboptions rowgap(), colgap(), keygap(), symxsize(), and symysize().

13. graph bar, graph dot, graph box, and graph pie have a new option allcategories specifying that the legend is to include all over() groups in all the dataset, not just groups in the sample specified by if and in.

14. graph bar, graph dot, graph box, and graph pie, when used with both options by() and over() when different by groups contained different subsets of over groups, could produce legends that were not correct for all the by graphs. This has been fixed.

15. graph bar with options stack and blabel(total) did not show correct totals. This has been fixed. Note that horizontal bar charts graph hbar always showed correct totals.

16. graph combine options xcommon and ycommon "stuck" when specified; that is to say, subsequent graphs would be drawn as though the option were specified, even if it were not. This has been fixed.

17. graph combine did not respect the rescaling of the original graph axes when this rescaling was done with the range() suboption of the xscale(), yscale(), or tscale() options. It now respects these rescalings.

18. graph dot options linetype(), lowextension(), and highextension() "stuck" when specified; that is to say, subsequent graphs would be drawn as though the option were specified, even if it were not. This has been fixed.

19. graph pie mislabeled the pie slices when the option sort was combined with plabel(... name). This has been fixed. Note that this does not apply to options plabel(... sum), plabel(... percent), and plabel(... "text"), which have always produced correct labels.

20. graph pie when the sort and by() options are specified together now draws individual legends for each graph so that the pie slice colors are correctly identified for each graph.

21. hetprob stopped with an error message when used with the technique() option, for techniques other than Newton-Raphson; this has been fixed.

22. ir rounded the time at risk to the nearest integer, causing a potential loss of precision; this has been fixed.

23. ksmirnov was not preserving the sort order of the data; this has been fixed.

24. lowess stopped with an error when a legend() option was supplied within the by() option; this has been fixed.

25. The dialogs for lowess and spikeplot incorrectly allowed the missing and total options to be used with by; this has been fixed.

26. predict after ologit and oprobit would assign a value of zero to the linear predictor when there were no model degrees of freedom. predict now sets the linear predictor to missing in this case to avoid misleading results.

27. statsby failed if the number of statistics to be collected were larger than approximately 65. This limit has been increased so that the number of statistics allowed by statsby is virtually unlimited.

28. stmc's compare() option now has a minimum allowed abbreviation of c(). This matches stmc's companion command stmh.

29. The 09sep2003 update of sts graph overrode the 03jun2003 update that had the yscale(log) option only graphing nonzero values. This fix has been restored.

30. sunflower has a new option nosinglepetal, see help sunflower.

31. twoway rbar and twoway rarea now fully respect line patterns set by the blpattern() option.

32. xtabond refused to estimate AR(1) models with only three data points; this has been fixed.

33. xtabond incorrectly labeled the output when predetermined variables were dropped because of collinearity; this has been fixed.

34. xtdes gave an error message when presented with data where the time variable included negative values; this has been fixed.

35. xtreg, re produced erroneous results if the same independent variable were specified twice; this has been fixed.

Stata executable, all platforms

36. encode has new noextend option that modifies the existing label() option; see help encode.

37. ereturn matrix no longer allows you to post matrices named b or V. The correct method for posting these matrices is with the ereturn post or ereturn repost commands.

38. graph pie in the degenerate case of only one slice drew a line from the center to the top. This has been fixed.

39. A new macro extended function, all, allows programmers to obtain a list of currently defined global macros, scalars, or matrices; see help macro.

40. `macval(.a.b.c)', when specified at the top level of macro expansion, causes the class reference .a.b.c to be macro expanded only once, so that after expansion the result is not macro expanded. This extends the functionality of the expansion operator macval() to class system references.

41. merge's uniqusing option failed to issue an error message when the using data contained nonunique values in the match variable(s) if the master data also contained nonunique values in the match variable(s); this has been fixed.

42. odbc describe caused a "cursor error" with Microsoft ODBC drivers; this has been fixed.

43. odbc dialog connection options now allow previously saved usernames and passwords to be loaded by the MySQL driver.

44. odbc insert caused internal buffer overflow in the SQL statements used to feed data to the driver; this has been fixed.

45. search when executed in the Viewer on rare occasions incorrectly reported no results; this has been fixed.

46. If an invalid syntax element such as xyz(,abc) were specified in the varlist portion of a command, Stata could crash. More specifically, if what appeared to be an option that included a comma were specified where a varlist was expected and the invalid command was parsed with the syntax command, Stata could crash. This has been fixed.

47. update now clears all cached programs from memory. This ensures that the updated programs will be loaded instead of Stata remembering the older version.

48. Note to programmers: For holding on to the caller version, use local ver : display _caller() instead of local ver = _caller(). The first method places "8.2" in local macro ver, while the second places "8.199999999999999". A subsequent call to version `ver' would fail with the second method.

Stata executable, Windows

49. Rich text copied into the Do-file Editor could change the font and color in the Do-file Editor; this has been fixed.

50. When a user tried to paste text into the Command window, if the Viewer was open but not in front, sometimes the text could be pasted into the Viewer's edit field. This has been fixed.

Stata executable, Mac

51. The Toolbar may be a floating window or may be anchored to the menubar. The advantage of making the Toolbar a floating window is that it takes up less room on the screen and can be moved around. A Toolbar submenu containing the current hide and show options for the Toolbar as well as the option to make the Toolbar float or anchor has been added to the Window menu.

52. Some keyboard shortcuts were remapped to be more consistent with similar menu items. The keyboard shortcut for the Preferences dialog was remapped to "Apple-," to be consistent with other applications.

53. Some of the window preferences were misnamed; this has been fixed.

54. Executing commands from the Do-file Editor and immediately typing caused Stata to crash; this has been fixed.

Stata executable, Unix

55. The Do-file Editor previously did not understand ~/ to mean the user's home directory (that is, doedit ~/; this has been fixed.

56. Stata's Do-file Editor no longer ignores a request to cancel closing its window when there is an unsaved document and the window is in the process of being closed by a request to exit Stata.

-------- update 30sep2003 -----------------------------------------------------


1. dotplot now uses the variable format for the y axis if it is common among all the specified variables.

2. lincom after svy estimation now saves the degrees of freedom in r(df).

3. lnskew0's dialog has been redesigned to better handle the level() option. Previously, level(95) was ignored while other values of level() were accepted.

4. sunflower is a new command that produces sunflower density-distribution plots; see help sunflower. This command is based on earlier work by William D. Dupont and W. Dale Plummer, Vanderbilt University, who provided helpful suggestions for this version.

5. Dates containing spaces, slashes, or commas must be bound in parentheses when specified in the time-axis graph options tscale(), tlabel(), tmlabel(), ttick(), tmtick(), and ttext(); see help datelist.

-------- update 26sep2003 -----------------------------------------------------

Stata executable, Mac only

1. Stata's update feature has been modified to handle the download and extraction of Mac OS X Application Packages. Future updates to Stata will ship as a Mac OS X Application Package rather than as a single executable. The advantages of this format include faster startup, better performance, and the ability to download future Stata updates as a compressed file making updating faster.

2. In some cases, Stata crashed if run in batch mode; this has been fixed.

3. Extended macro function dir did not recognize ~/ to mean the user's home directory; this has been fixed.

-------- update 15sep2003 -----------------------------------------------------


1. graph hbar and graph bar, horizontal after the 12sep2003 update produced an error message when the text() option was specified; this has been fixed.

2. xtline now documents option overlay which replaces overlayed.

-------- update 12sep2003 -----------------------------------------------------


1. graph twoway has two new plottypes for plotting time-series data, tsline and tsrline; see help tsline. Also, all plottypes automatically produce better label and tick values for variables having date formats or time-series formats.

2. graph twoway has seven new options useful when plotting time-formatted variables: tscale(), tlabel(), tmlabel(), ttick(), tmtick(), tline(), and ttext(); see help axis_options, added_line_options and added_text_options. With these options you may directly specify date literals, such as 12sep2003 or 1990q2 to identify positions.

3. xtline is a new command for plotting panel data; see help xtline. xtline allows either overlaid or separate graphs to be produced, by panel.

-------- update 09sep2003 -----------------------------------------------------

The 09sep2003 executable and ado update includes 45 changes and improvements. We recommend you read the entire numbered list below. Highlights include,

o Stata can now read and write files in the format required by the U.S. Food and Drug Administration (FDA) for new drug and device applications (NDAs). See #13 below and see help fdasave.

o Stata now allows multiple variable, value, and data labels, such as labels in different languages; see #22 below and see help label_language.

o Stata now has multiple-comparison tests after anova and manova. In addition, specifying contrasts is now easier. See #28 below and see help testanova and help manovatest.

o A new file-editing/file-filtering command makes it possible to replace one set of characters with another; this may be done on files of unlimited size and on both binary and ASCII files. See #14 below and see help filefilter.

o Stata now supports TIFF files; see #16 and #17 below and see help graph_export.

o Under Stata for Windows, the Review and Variable windows may now be moved together with the main Stata window by holding Ctrl while dragging the main Stata window with the mouse; See #31 below.

Ado-files, all platforms

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

2. cf is now faster. The improvement results from code provided by David Kantor, Institute for Policy Studies, Johns Hopkins University.

3. graph color setting options, such as mcolor(red*.3), color(blue*.5), and lcolor(navy*.8), that used the *# syntax to change the intensity of a named color would cause that intensity change to stick. For example, after option mcolor(red*.3) was specified, the color red would be shown as a light red in all subsequent graph commands until clear or discard was typed to reset the graph system. This has been fixed.

4. graph bar and graph box now respect the blpattern() suboption of the bar() option.

5. graph box would not plot outside values when the interquartile range was zero; this has been fixed.

6. graph twoway options connect() and cmissing() would "stick" when specified. That is to say subsequent graphs would be drawn as though the option were specified, even if it were not. This has been fixed.

7. sts graph, hazard, when used with by(), would graph estimated hazards over the range of observed failures for the entire data. As a result, the graphs produced were sometimes misleading, since the smoothed hazard estimate for a particular by() group is valid only over the failure range for that group. sts now graphs the estimated hazard over the failure range for each group.

8. tabstat documented and computed statistics(cv) as variance/(mean^2). tabstat now documents and computes it as sd/mean to match the standard definition for the coefficient of variation.

9. tabstat misaligned the header if a value less than 8 was specified in varwidth(); this has been fixed.

10. varsoc failed when using pre-estimation syntax if maxlags() was not specified; this has been fixed.

Stata executable, all platforms

11. class exit issued an error message if a number with a leading decimal point was specified as the return value, for example, .456. This led to error messages and failure to draw graphs if any text to be drawn were smaller than 1 unit, typically for very small fonts or text with a single narrow character such as "i". This has been fixed.

12. cluster dissimilarity option Hamann has been added as the preferred alias for the previously documented Hamman. The correct spelling is with two ns and one m. Both spellings are now allowed.

13. fdasave, fdause, and fdadescribe are new commands for saving and using files in the format required by the U.S. Food and Drug Administration (FDA) for new drug and device applications (NDAs). These commands are designed to assist people making submissions to the FDA, but the commands are general enough for use in transferring data between SAS and Stata. The FDA format is identical to the SAS XPORT Transport format; see help fdasave.

14. filefilter is a new command for copying an input file to an output file while converting a specified ASCII text or binary pattern to another pattern; see help filefilter.

15. graph now produces better label and tick values for variables having a date formats or time-series formats.

16. graph export now supports TIFF files. The files are limited to the resolution of the display device and the size of the Graph window. When a TIFF file is exported using an 8-bit display device, the TIFF file will be in 8-bit format.

17. graph export now supports TIFF previews for EPS files through the preview(on) option; the graph window must be open so that it can be used to create the preview.

18. EPS files produced by graph export did not print properly when placed in some Adobe applications such as PageMaker and InDesign, although they would print and could be viewed by other applications; this has been fixed.

19. EPS files exported from Stata graphs now contain information describing the font that must be downloaded to the printer. This assists desktop publishing applications when printing embedded Stata EPS files using a font that may not exist on a particular printer.

20. EPS graph files that used the small square symbol could contain invalid PostScript code; this has been fixed.

21. infix and infile with a dictionary had difficulty handling end-of-line characters from files created under different operating systems. Stata for Windows and Stata for Unix did not read raw text files with Mac end-of-line characters. Stata for Mac read a raw text file with DOS end-of-line characters as having two end-of-line characters. Both problems have been fixed.

22. label has new subcommand language that lets you create and use datasets that contain different sets of data, variable, and value labels. A dataset might contain one set in English, another in German, and a third in Spanish. Another dataset might contain one set of short labels and another of long labels. Either way, you can switch between the label sets as desired. Up to 100 sets of labels are allowed. See help label_language.

23. odbc describe produces a more detailed error message when the table does not exist.

24. odbc exec now allows lengthy SQL statements; see help odbc.

25. odbc load has a new sqlshow option for debugging SQL communication with the driver; see help odbc.

26. The maximum number of sersets has been increased from 100 to 2,000; see help serset. This means that complex graphs and combined graphs can be drawn with many more plots (lines or scatters).

27. tabulate now accents the é when labeling Cramér's V (obtained using the V option). The value continues to be returned in r(CramersV) (sans accent).

28. test after anova and manova has new options mtest(), test(), and showorder; see help testanova and manovatest. mtest() provides adjustment for multiple tests. test() allows easier specification of contrasts and other tests involving the coefficients of the underlying regression model. showorder shows the order of the columns in the underlying regression model.

29. _tsnatscale is a new programmer command for obtaining nice label or tick values for time series; see help _tsnatscale.

Stata executable, Windows

30. The Review and Variable windows may be jointly opened or closed by pressing Ctrl-9.

31. The Review and Variable windows may be moved together with the main Stata window by holding down the Ctrl key while dragging the main Stata window with the mouse.

32. The default actions associated with double-clicking on various Stata files no longer include /m1 on the action command line. Thus, the value of set memory ..., permanently will be respected. You may still edit the action command line for all Stata file types to override the default memory allocation. If you wish Stata to use the new default associations, choose Restore File Associations from Stata's Prefs menu.

33. The path passed to Stata when a Stata file was double-clicked was sometimes converted to old-style 8.3 notation before being displayed in Stata. This did not affect functionality but was visually unappealing. Windows now passes Stata nicely formatted, full paths. This takes effect if you reset Stata's default file associations by selecting Restore File Associations from Stata's Prefs menu.

34. The Viewer edit field now allows text to be pasted into it.

35. The Print item in the contextual menu in the Viewer did not print the Viewer's contents; this has been fixed.

36. (Windows XP) The visual style of Stata for Windows now matches the currently selected Windows theme.

37. (Windows XP) On some Dell platforms, Stata would fail to launch due to another process being frozen. If frozen processes exist, Stata now works around this and still launches.

Stata executable, Mac

38. An "Edit--Copy Table as HTML" menu item has been added.

39. Graphs saved or copied in the PICT format now provide true rotated text via PICCOMMENTS. For applications that do not understand PICCOMMENTS and don't know to ignore them, there is an option to turn off this feature: set piccomments off or unchecking the "Use PICCOMMENTS in PICT output to draw rotated text" checkbox in the Graph Preferences dialog.

40. Stata has been enhanced to support Panther, Apple's forthcoming OS X upgrade.

41. Running Stata remotely in batch mode caused Stata to crash when drawing the progress bar on Stata's icon in the Dock. Before attempting to draw the progress bar, Stata now determines whether it is being run remotely by checking that the environment variable REMOTEHOST is not empty.

42. The Search menu was never enabled; this has been fixed.

Stata executable, Unix

43. (console mode) Underscores associated with the smcl directive {title:...} did not match the indentation of the title; this has been fixed.

44. (Linux) update swap now uses gunzip rather than uncompress.

45. (Solaris) license positions were sometimes tracked incorrectly; this has been fixed.

-------- update 22aug2003 -----------------------------------------------------


1. aorder gave a system-limit error if some of the variable names had the letter "d" or "e" after an initial stub and number; this is fixed.

2. bootstrap, jknife, permute, simulate, and statsby have a new trace option that displays a trace of the execution of the command being operated on.

3. Graphics options and suboptions that accept a compassdirstyle now respect arguments that are clock directions -- numbers 0 through 12.

4. The legend() option of graph now honors the height() and width() suboptions.

5. graph box can now label outside values; see the marker label options suboptions of marker() and help graph_box.

6. graph box issued an error message and did not draw the graph if any of the variables had value labels that labeled missing values only; this is fixed.

7. graph twoway scatter now has a jitterseed() option that sets the random-number seed for jittered points; see help scatter.

8. graph twoway now honors the xlabel(, valuelabels) option even when more than one y axis is specified. Previously, if more than one y axis were specified, the valuelabels suboption was ignored and numbers were shown in place of the associated value labels on the axis.

9. ml now has a vce(oim) option that is the preferred alias for the existing vce(hessian) option which continues to work; see help ml.

10. separate now has a shortlabel option that specifies shorter variable labels are to be created; see help separate.

11. statsby is now faster. The improvement is based on code provided by Michael Blasnik, Blasnik & Associates.

-------- update 13aug2003 -----------------------------------------------------


1. bootstrap, jknife, permute, simulate, and statsby would exit with an error when supplied with long expressions; this is fixed.

2. ci and cii now produce more accurate Poisson exact confidence intervals.

3. intreg with the het() option would, on rare occasions, issue an error message saying there were no observations; this is fixed.

4. separate now preserves the sort order of the data.

5. svar displays a better error message when the specified model passes the order condition but fails the rank condition.

6. tostring, a command for converting numeric variables to string, is now an official Stata command; see help tostring.

7. varirf create would fail to estimate the impulse response functions after some structural VAR models with long-run constraints. This has been fixed.

-------- update 15jul2003 -----------------------------------------------------


1. ci and cii have new options exact, wilson, agresti, and jeffreys for computing different types of binomial confidence intervals. The default behavior of computing exact confidence intervals in the binomial case is unchanged. See help ci for details.

2. codebook now allows if exp and in range qualifiers; see help codebook.

3. findfile, when used with the all option, saved results in the return list with unbalanced double quotes; this is fixed.

4. graph has improved axis titles when time-series operators are applied to a variable that does not have a label. Previously the title did not include the variable name, only the time-series operator(s). The title now includes ts.varname where ts are the time-series operator(s) and varname is the variable name.

5. graph bar has new options over(, reverse) and yvaroptions(reverse) that specifies the categorical scale be reversed to run from maximum to minimum rather than the default minimum to maximum. See help graph_bar.

6. Several of the graph by() suboptions could "stick", meaning that they would be applied to subsequent graphs drawn with the by() option even though the options were not specified subsequently. Suboptions that could "stick" were xtitles, ytitles, edgelabel, individual, ixaxes, iyaxes, ixtitles, iytitles, ixticks, iyticks, rescale, xrescale, and yrescale. This has been fixed.

7. graph twoway mband for plots with only two observations now connects the two points in the data. Previously no line was drawn.

8. graph twoway mband and graph twoway mspline now have better default number of bands.

9. graph twoway mspline for plots with only two unique values of the x variable now plots the two unique points and the medians for the y variables at those two points. Previously a clearly incorrect line was drawn.

10. hausman has new option df() for controlling the degrees of freedom for the Hausman test; see help hausman.

11. hausman could incorrectly state which model was consistent and which inconsistent-but-efficient in the message presented at the bottom of the output table; this is fixed.

12. ml when used with an if or in restriction, could produce the error "tempvar not found" in rare instances; this is fixed.

13. When the nrtolerance() option of maximum-likelihood estimators (such as heckman) or ml maximize was specified, the tolerance could be declared met even if the Hessian was not positive definite. Now the Hessian must be positive definite as a precondition for the nrtolerance() convergence criterion.

14. palette linepalette now honors the scheme() option.

15. recode, in the case of rule overlap, used the last matching rule, rather than the first matching rule as documented; this is fixed. In addition, recode is now faster.

16. Scheme files may now have *! comments. This allows their version to be displayed by which.

17. svymarkout is a new programmer's command for marking observations for exclusion based on survey characteristics set by svyset; see help svymarkout.

18. twoway__function_gen could produce values of "x" greater than indicated in the range() option; this is fixed.

-------- update 01jul2003 -----------------------------------------------------

What's new in release 8.1 (compared to release 8.0)

1. ml has lots of new features; see help ml and see the book Maximum Likelihood Estimation with Stata, 2d Edition (Gould, Pitblado, and Sribney 2003).

a. ml can now perform BHHH, DFP, and BFGS optimization techniques with user defined likelihood functions. The default technique remains modified Newton-Raphson.

b. ml has a new variance estimator -- outer product of the gradients (OPG) -- to accompany the two previously available -- the Hessian-based estimator and the robust estimator. The new vce() option along with the existing robust option specify which is used:

vce(hessian) specifies the Hessian-based variance estimator.

vce(opg) specifies the outer-product-of-the-gradients variance estimator.

robust specifies the Huber/White robust estimator.

c. ml handles irrelevant constraints, if specified, more elegantly. Previously, irrelevant constraints caused an error message. Now they are flagged and ignored.

ml model uses the new command makecns when supplied with the constraint() option. makecns uses matrix makeCns to generate a constraint matrix, and displays a note for each constraint that causes a problem; see help makecns.

(Prior to version 8.1, ml would ignore the constraint() option if there were no predictors in the first equation. This behavior is preserved under version control. Under version 8.1, ml applies constraints under all circumstances.)

d. Constraints now imply a Wald test for the model chi-squared test. For those who wish to perform likelihood-ratio tests for models with constraints, see help lrtest.

e. mlmatbysum is a new programmer's command for use by method d2 likelihood evaluators to help define the negative Hessian matrix in the case of panel-data likelihoods; see help mlmatbysum.

2. Plugins (also known as DLLs or shared objects) written in C can now be incorporated into and called from Stata. Point your browser to for more information and documentation.

3. sts graph has new options cihazard and per(#). cihazard draws pointwise confidence bands around the smoothed hazard function. per() specifies the units used to report the survival or failure rate. See help sts for details.

4. (Windows, Mac, and Linux) odbc has new commands odbc insert and odbc exec for writing data to an ODBC data source. Positioned updates are also possible using the odbc exec command. See help odbc for more information.

5. The constant and current-value class c() has the following new items.

c(alpha) (lowercase letters) c(ALPHA) (uppercase letters) c(Mons) "Jan Feb ... Dec" c(Months) "January February March ... December"

See help creturn.

6. outfile has new option missing that preserves all missing values when the comma option is also specified. Previously, if you specified comma, system missing values (.) were changed to null strings ("") and extended missing values (.a, .b, ..., .z) were retained. See help outfile.

7. _svy_mkdeff, _svy_mkmeff, _get_eformopts, and _get_diparmopts are new programmer utilities.

8. Recommendation on setting version at the top of do-files and ado-files: Use version 8.1. version 8.1 adds new features and changes the behavior of ml for the better. (Setting version 8.0 restores ml's old behavior.) See help version.

Other additions and fixes


9. nlogit now uses the BFGS optimization algorithm by default.

10. nlogit ignored the robust option when the specified model had more than 3 levels. The robust option is no longer allowed when the specified model has more than 3 levels.

11. zip and zinb with the vuong option now report an error if constraints are also specified.

Stata executable

12. insheet did not honor the double option; this is fixed.

13. (Windows) insheet incorrectly identified very large integer values as fitting within type long instead of double; this is fixed.

14. _pctile will now compute up to 1000 percentiles (previously the limit was 20); see help _pctile. In addition, pctile (no underscore) takes advantage of _pctile's new higher limit, and is more precise in determining the boundaries between the requested percentiles.

15. (Mac) The Browse... dialog from the Viewer has been changed to allow selection of either files or folders.

16. (Mac) update swap has been altered to resolve a problem some users were having in updating to a new executable.

17. (Mac) Clicking the toolbar switch button in the Do-file editor to show the toolbar when it was previously hidden would not properly refresh the area between the text and the toolbar; this is fixed.

-------- update 25jun2003 -----------------------------------------------------


1. Online help and search index brought up to date for Stata Journal 3(2).

2. The drop and keep dialog did not enable the drop and keep checkboxes on Mac platforms; this is fixed.

3. ltable ignored the ciopts() option; this is fixed.

4. tsappend exited with an error when applied to panel data if the panel() option was not specified; this is fixed.

5. xtregar, fe did not print warning messages for variables dropped due to collinearity; this is fixed.

-------- update 17jun2003 -----------------------------------------------------


1. cumul has a new equal option; see help cumul.

2. ds has new options that allow you to list variables that match certain criteria. The resulting list is returned in r(varlist), which can then be used in a subsequent command. The new options for ds are not, has(), not(), insensitive, detail, and indent(). See help ds for details.

3. graph suboptions label, nolabel, tick, and notick of the ylabel(), xlabel(), ytick(), and xtick() options were not honored if by() was also specified; see help axis_label_options. This has been fixed.

4. graph box no longer draws a symbol for an outside value at the median when all observations for a variable (or the by group) for a box are constant.

5. xpose now uses the file command instead of the outfile and infile commands and thus is better able to retain precision for variables of type double when the promote option is specified.

-------- update 03jun2003 -----------------------------------------------------


1. bootstrap, jknife, permute, simulate, and statsby exited with an error when the user-supplied command contained double quotes; this is fixed.

2. graph set ps fontface and graph set eps fontface returned an error, and graph set eps displayed settings for ps rather than for eps; this is fixed

3. graph twoway histogram and histogram could produce one less bin than it should have, resulting in the last bin falling short of the maximum value of the given variable; this is fixed.

4. gsort now has option mfirst for specifying that missing values are to be placed first in descending orderings rather than last.

5. jknife stopped with an error when it should have posted a missing value for a requested statistic; this is fixed.

6. roctab could produce ROC graphs with different connecting lines with repeated calls using the same data even though the plotted points were the same; this is fixed.

7. sts graph with the yscale(log) option now graphs only the nonzero values.

8. svytab reported an error when the stubwidth() option was supplied with a value greater than 32 and a value label also exceeded 32 characters. The upper limit is now mentioned in the help file, and svytab uses 32 when supplied with a stubwidth greater than 32.

Stata executable

9. factor, ipf has new option citerate() that controls the number of iterations for reestimating the communalities; see help factor.

10. hexdump incorrectly reported the count of the end-of-line character for Mac, reporting instead the count for Unix; this is fixed.

11. lngamma() function is now computed with approximately 16 digits of accuracy.

12. set memory #g, with #>4 on 32-bit computers, now produces a more descriptive error message.

13. test after manova could cause Stata to crash when presented with invalid syntax; this is fixed.

14. update is faster. Stata no longer waits for the web server to respond if the requested file is 0 bytes.

15. temporary file names remain unique longer. The total number of unique names that are cycled through has been increased from 1.3 million to 1.5 billion.

16. version has new option born() and issues more descriptive error messages. In addition, the help file has been updated to be more understandable; see help version.

17. (Windows) An "Edit--Copy Table as HTML" menu item has been added.

18. (Windows) Pie slices with a 0-length arc drew a complete circle; this is fixed.

19. (Mac) Shift-clicking in the Do-file Editor now selects text from the insertion point to the text location where you click the mouse.

20. (Mac) The ability to undo or redo multiple actions has been added to the Do-file editor.

21. (Mac) Styled text from the clipboard no longer is pasted into the Do-file editor as styled text rather than mono-styled text.

22. (Mac) The General Preferences dialog has a new option for bringing all open Stata windows to the front when Stata is activated. This prevents Stata's windows from overlapping with other application windows when Stata is brought to the front.

23. (Mac) In later versions of the OS X operating system, a hidden progress bar incorrectly draws content that was previously over the control's boundaries. Stata now works around this bug and displays the progress bar as indeterminate when establishing a connection (no progress bar was shown previously when establishing a connection).

24. (Windows, Mac, and Linux) odbc could load values incorrectly when the variables were explicitly specified in an order that did not match that of the underlying data; this is fixed.

25. (64-bit Solaris GUI) Some keystrokes could be omitted or corrupted. Stata now links against a newer version of GDK in which this is fixed.

26. (HP Unix GUI) Stata is now approximately 30% faster.

-------- update 12may2003 -----------------------------------------------------


1. estimates, under rare conditions, would not restore e(sample); this is fixed.

2. estimates table now allows you to specify more models. It wraps the table if necessary. estimates table also has new options

equations() to match equations by number rather than by name.

coded to display the table in a symbolic compact format in which numbers are represented by a single *. This enables the display of a large number of models in a single table.

modelwidth() to set the number of characters for displaying model names.

See help est_table.

3. estimates table produced an error message if the table included many coefficients or the label option was specified with models, such as ologit, that contain special names in the coefficient vector name stripe (such as _cut# with ologit); this is fixed.

4. graph display labeled graph bars with zero height when redisplaying a graph that was drawn using graph bar, stack or graph hbar, stack even though blabel() was not specified; this has been fixed.

5. graph twoway function renamed a variable that started with x to "x" if there was only one such variable; this is fixed.

6. graph twoway histogram reported an error when more than 999 bins were involved; this is fixed.

7. graph twoway fpfit and graph twoway fpfitci now produce smaller .gph, .ps, .eps, and .wmf files.

8. lrtest and suest now work correctly for models having zero parameters (for example, the null model for Cox regression).

9. recode with an if condition involving strings that contained spaces produced an error; this is fixed.

10. svyheckman, svyheckprob, and svynbreg computed confidence intervals using the wrong degrees of freedom for some parameters: athrho, lnsigma, and lambda for svyheckman; athrho for svyheckprob; and lnalpha and lndelta for svynbreg. This is fixed.

11. svytab has a new notable option that suppresses the header and table and presents only the test statistics, if any; see help svytab.

12. vwls returned the error "no observations" when the computed variances were large; this is fixed.

13. xtpcse did not allow time-series operators when options hetonly or independent were specified; this is fixed.

-------- update 29apr2003 -----------------------------------------------------


1. graph bar and graph box ignored weight specifications when users specified suboptions relabel(), sort, or total within options over() or yvaroptions(); this is fixed.

2. graph display has a new scale() option that allows all text, symbols, and line widths to be rescaled when a graph is displayed again.

3. predict with the mse option, if used after arima with the condition option, now produces mean-squared errors that are based on initializing the presample MSE vector of the state vector with the expected long-run MSE. This is the same initialization method used to produce conditional estimates. Previously, the presample MSE vector was initialized with expected values derived from the estimated parameters, which is the initialization method used to produce unconditional estimates. The new method is more consistent with the conditional method of estimation.

-------- update 24apr2003 -----------------------------------------------------


1. egen newvar = group() with the label option would write over the value label named newvar if the value label already existed. Now, it prompts you to use the lname() option to specify a name for the value label.

2. tabodds with weights large enough to cause an overflow of the storage type int would produce incorrect results. Now, storage type long is used, and the problem is fixed.

Stata executable

3. The * comment indicator can now be used in .dlg and .idlg files.

4. The find string from the Viewer window's Find dialog would sometimes be converted to lower case; this is fixed.

5. Some operating systems allow read-only files to be overwritten if the directory is not read only. Stata no longer allows read-only files to be overwritten, regardless of the directory protection.

6. After set dp comma

graph now honors the setting and uses commas in place of decimals when labeling axes.

graph no longer draws extra-wide legends that overwrite other graph objects.

serset extended macro functions now return numbers with decimal points instead of commas to avoid problems in subsequent computations.

++ and -- macro operators, when the value of the macro is not an integer, now return numbers with decimal points instead of commas to avoid problems in subsequent computations.

7. (Windows) graph export of a .wmf file would fail if the path and filename were over 80 characters; this is fixed.

8. (Windows and Unix) window manage maintitle is a new command that allows users to set the main title of the Stata window; see help window.

9. (Mac) Keyboard shortcuts have been added for Edit->Copy Table, Tools->Search->Replace and Find Next, and Tools->Search->Match.

10. (Mac) If the log menu became disabled, it would not be reenabled when Stata could again process log commands; this is fixed.

11. (Mac) The Find button in the Viewer would not move when the window was resized; this is fixed.

12. (Mac) You can now save a graph as a PNG (portable network graphics) file from the Save dialog.

13. (Mac) Stata previously ran at a low scheduling priority when in batch mode. The scheduling priority has been raised, increasing the execution speed.

14. (Mac) The -h startup option has been added. This option does not start Stata but shows the syntax diagram for invoking Stata.

15. (Mac) The -q startup option now suppresses the Stata logo as well as the initialization messages.

-------- update 16apr2003 -----------------------------------------------------


1. The submenu for cross-sectional time-series under the statistics menu has been restructured to improve access to the dialog boxes. (This will take effect once you exit and restart Stata.)

2. egen newvar = group() with the label option did not check to see if a value label named newvar already existed before creation. Now an error message is presented indicating that you should use the lname() option to provide a name for the new value label.

3. findfile produced an error message if the path specified in the path() option contained spaces; this is fixed.

4. graph when specified with the option scale() now causes linewidths to scale in proportion to the scaling of fonts and symbols. This also applies to the automatic scaling by graph, by() and graph combine.

5. Some error messages in graph have been improved.

6. histogram and twoway histogram are now much faster for large datasets.

7. levels is a new command that displays a sorted list of the distinct values of a variable. This command is especially useful for looping over distinct values of a variable with (say) foreach; see help levels.

8. ml did not correctly compute the tolerance for the nrtolerance() option when the nr or bhhh options were also specified. This could cause arch and arima incorrectly to continue iterating beyond convergence. This is fixed.

9. odbc, as of the 10apr2003 executable update, is available for Mac OS X and Linux systems that use the iODBC Driver Manager. ODBC, an acronym for Open DataBase Connectivity, allows data to be imported into Stata via an ODBC data source. More information on configuring ODBC for Mac and Linux can be obtained from the FAQ located at

10. The stcox dialog has been improved.

11. tabstat with the by() option ignored the date format when displaying the by() variable; this is fixed.

-------- update 10apr2003 -----------------------------------------------------

Stata executable

1. Graphics are now approximately 100% faster -- they run in about half the time. Some graphs demonstrate more than a 100% speedup, some less. Speedups are greatest on the second and subsequent graph commands (second after Stata has been started or refreshed by typing clear or discard).

2. The Stata Viewer now has the ability to search for text within the window. Click on the find icon which looks like a pair of binoculars. This icon is at the top right of the Viewer.

3. #delimit ; followed by a compound double quoted string containing an odd number of regular double quotes would cause an error due to matching the beginning and ending compound double quotes incorrectly; this is fixed.

4. graph export now supports PNG (portable network graphics) format, especially useful for posting graphs on the Internet.

5. graph save could, on rare occasions, produce severely misdrawn graphs when preceded by "graph7 ... , saving()". This is fixed.

6. graph ... , saving(... , asis) could, on rare occasions, cause Stata to crash. This occurred only when saving asis-format graphs; see help gph_files. This is fixed.

7. Help file aliases contained in the official help_alias.maint file are now resolved before attempting to locate the .hlp file along the ado path. This ensures that the official Stata .hlp files are found before user-written .hlp files.

8. regress, beta after anova produced nonsense results for the normalized beta coefficients. regress, beta is no longer allowed after anova.

9. After set dp comma, in-line = macro-expansion operators did not work in most contexts because the results it produced would have commas for decimal points and Stata would not understand that as input. In-line = operators now ignore the dp setting and all is fixed.

10. Graphs exported to PS or EPS format now use round line caps on connected lines.

11. (Unix) Connected lines drawn to the Graph window now use round line caps. The Windows and Mac versions already use round line caps.

12. (Mac) Unconnected lines such as grid lines and axes now use butt line caps rather than round line caps.

13. (Windows) Keyboard shortcuts to the menu items Do and Run in the Do-file Editor have been added. The shortcuts are ctrl-d and ctrl-r.

14. (Mac) A keyboard shortcut to the menu item Run for the Do-file Editor has been added.

15. (Mac) Do-file editor preferences were corrupted when the preference dialog was opened; this is fixed.

16. (Mac) The preferences dialogs have been rewritten as one single dialog that can be accessed from Stata's menus or from the application menu.

17. (Mac) When saving a new dataset, Stata now correctly defaults to a filename of Untitled.dta.

18. (Mac) Stata now treats text that is dragged-and-dropped into the Command window the same as text that is pasted -- it truncates the text to the first line because the Command window only accepts one line of text.

19. (Mac and Unix) The contents of the Review window can now be copied to the Clipboard using the Review window's contextual menu.

20. (Unix) The Review window's contextual menu remained permanently disabled if the menu was ever opened while the Review window was empty; this is fixed.

21. Stata for IRIX is now approximately 30% faster.

22. Stata for 64-bit Solaris is now approximately 10% faster.

-------- update 02apr2003 -----------------------------------------------------


1. bstat produced an error message when the stat() or accel() options were specified with a matrix; this is fixed.

2. graph bar when specified with both the over() and asyvars options and with an if or in qualifier, could produce bars that did not match those in the legend; this is fixed.

3. graph pie refused to draw the graph if the data for any pie slice was zero. graph pie now allows a pie slice to be zero, as long as at least one slice in the pie is positive.

4. graph twoway with option yaxis(2), could cause a very slight skewing of points plotted on the second y axis toward the center of the x dimension of the graph; this is fixed. The skewing was only noticeable when points should have been plotted at the same x value.

5. ipolate now creates the new variable containing interpolated values using storage type double instead of float.

6. ml model did not apply the subpop() option correctly in cases where there was no weight variable; this is fixed. This affected svypoisson, svynbreg, svygnbreg, svyheckman, and svyheckprob.

7. sttocc would not produce any case observations when all observations failed or were censored at the same time, that is, when the variance of _t was zero; this is fixed.

-------- update 19mar2003 -----------------------------------------------------


1. cf is now an r-class command and returns r(Nsum), the number of differences found. This addition was made in the 14mar2003 update, but was not mentioned then.

2. graph bar and graph hbar in the rare case where two or more string variables were specified in over() options and the asyvars option was specified, could produce bar colors that did not match the colors in the legend. This is fixed.

3. graph pie when the option over() was specified, did not respect if conditions or by groups in determining the size of the pie slices; this is fixed.

4. sttocc could produce inappropriate matches when the number of observations exceeded 32,740; this is fixed.

-------- update 14mar2003 -----------------------------------------------------


1. Online help and search index brought up-to-date for Stata Journal 3(1).

2. histogram and twoway histogram would not include the minimum value in the first bin if the minimum value could not be exactly represented in numerical precision; this is fixed

3. rocgold, graph excluded missing observations casewise, while rocgold, summary excluded them on a pairwise basis. This inconsistency has been removed. rocgold now uses casewise deletion of missing observations for both graph and summary.

-------- update 10mar2003 -----------------------------------------------------


1. bootstrap with the bca option now gives a more informative error message when it can not compute acceleration.

2. bsample with the cluster() option became noticeably slower as a result of options added in Stata 8. This is no longer the case.

3. Easy histogram dialog has been modified to remove the xscale(log) and yscale(log) options from the Axes tab since these options are rarely used with histograms.

4. gladder now has a nonormal option to suppress drawing the overlaid normal density. The gladder dialog has been modified to include this new option.

5. (Windows) graph set mf fontface did not properly pass along double quotes around the specified font face; this is fixed.

6. histogram, easy histogram, and gladder dialogs have been modified to include the percent option which was added in the 25feb2003 update.

Stata executable

7. clist is a new command similar to list, clean; clist is in fact the list command that appeared in Stata prior to Stata 8, options and all. Some users needed the old command for backward compatibility purposes; list continues to be the preferred command. clist may be abbreviated cl; see help clist.

8. list has new option fast, which is in fact a synonym for option nocompress. list normally does a dry run to determine how best to format the display. nocompress prevents that and, with very large datasets, this can save a little time before the output appears. fast is a more suggestive name for the option in this case. See help list.

9. flist is a new command equivalent to list, fast or, equivalently, list, nocompress. flist may be abbreviated fl; see help list.

10. use, merge, and append, used with old-format datasets stored by Stata 7 or earlier, could change missing values stored in byte or int variables to nonmissing values. Missings in bytes could be converted to 101 and missing in ints could be converted to 32,741. This would happen only if the byte or int variable also contained nonmissing values greater than 100 or 32,740 respectively and if the data were ordered in a certain way. This is fixed.

11. (Windows) graph export did not accept the fontface() option when exporting an Enhanced Metafile (emf); this is fixed.

-------- update 25feb2003 -----------------------------------------------------


1. Five "easy" graph dialogs have been added to the previous eleven made available with the 10feb2003 update. These "easy" graph dialogs have been designed so that only the most essential options are exposed. The full-featured graph dialogs are still available when more complex graphing is required. The new easy graph dialogs are

Regression fit Pie chart (by variables) Pie chart (by category) Histogram Scatterplot matrix

2. The datasets used in the Stata manuals are now easily obtained starting with help dta_contents. Links are provided for describing a dataset before loading it, using a dataset, and viewing the associated help file.

3. archlm gave a misleading error message that said the command only works after regress and newey. In fact, it will not work after newey and only works after regress. The error message has been corrected.

4. graph with the twoway, bar, hbar, box, hbox, pie, or matrix subcommands may now be specified with variables that are labeled with a string that contains unbalanced parentheses. Previously, when a variable had such a label, an error message, unbalanced parentheses, would be issued and the graph was not drawn.

5. graph ... , xtitle(). The {x|y}title() options would not allow the words "if" or "in" to appear in the text of the title unless the text was quoted. These words may now appear in unquoted text. Note that unbound parentheses must be protected by quoting the text.

6. graph matrix , diagonal() now allows blank strings, for example, "" or `""', to be specified in the diagonal() option. Previously this produced an error.

7. histogram and twoway histogram have a new option percent that labels the vertical axis in percentages.

8. intreg would give an inappropriate error message when the matsize was too small. The error message has been improved.

9. svyprop would ignore if and in; this is fixed.

10. svytest, an out-of-date command superseded by test, had a problem with extra quotation marks when version was set less than 8; this is fixed.

11. twoway histogram now allows histogram to be abbreviated down to hist, as documented in [G] graph twoway histogram.

12. twoway lfitci, twoway qfitci, and twoway fpfitci have a new option nofit that prevents the prediction from being plotted.

13. twoway rline now allows rline to be abbreviated down to rl, as documented in [G] graph twoway rline.

-------- update 10feb2003 -----------------------------------------------------


1. Eleven "easy" graph dialogs have been added. These "easy" graph dialogs have been designed so that only the most essential options are exposed. The full-featured graph dialogs are still available when more complex graphing is required. The following easy graph dialogs are available:

Scatter plot Dot chart Connected scatter plot Box plot Line graph Horizontal box plot Area graph Function graph Bar chart Overlaid twoway graphs Horizontal bar chart

2. The Misc. tab has been reorganized to be more intuitive for the following full-featured graph dialogs:

Bar chart Box plot Horizontal bar chart Horizontal box plot Dot chart

3. bootstrap now gives an improved error message when supplied with a statistic that results in a missing value when computed using the entire dataset.

4. bootstrap would ignore the nopercentile option; this is fixed.

5. egen ... ends(), tail would return the whole string instead of the empty string when there was not a match; this is fixed.

6. graph with the option {x|y}label(..., valuelabels) did not show the labels for axis ticks when the numeric value of the tick was outside the range of the x (y) data. Instead, the numeric value was shown. Labels, when available for the numeric tick value, are now always shown when option valuelabels is specified.

7. graph options {y|x}title() and {y|x}tick() were implemented and documented with the same minimal abbreviation {y|x}ti(). This sometimes lead to syntax errors when option {y|x}title() was specified with the minimal abbreviation. The minimal abbreviation for {y|x}tick() has been changed to {y|x}tic().

8. graph bar when specified with temporary variables as either the y variables or in the over() option would sometimes issue an error and refuse to draw the graph; this is fixed.

9. graph box when specified with both the capsize() and yscale(log) options would distort the length of the whisker caps; this is fixed.

10. graph export failed to recognize the .wmf and .emf suffixes due to a problem introduced in the 30jan2003 update; this is fixed.

11. graph twoway rcap would sometimes draw spurious caps on the connecting spikes when the graphed data included missing values; this is fixed.

12. inbase, asked to convert 0 to another base such as 2 (inbase 2 0), would loop endlessly (until break was pressed); this is fixed.

13. lrtest when run under version control with version less than 8 would not leave behind the r() results; this is fixed.

14. mkassert rclass, saving() saved the r() results from the command that opened the file instead of the r() results of the previous command; this is fixed.

15. rvfplot incorrectly exchanged the labels for the x and y axes; this is fixed.

16. scatter with the jitter() option combined with the option yscale(reverse) or xscale(reverse) produced a graph with no plotted points. Option jitter() may now be combined with option {y|x}scale(reverse).

17. xtabond now labels the dependent variable in the estimation table D.depvar instead of just depvar. Likewise, e(depvar) is D.depvar instead of depvar. This change was made for consistency with the labeling of the right-hand-side variables.

18. xtabond now saves e(bnames_ud) and e(depvar_ud). e(bnames_ud) is the names of the right-hand-side variables in undifferenced form. e(depvar_ud) is the name of the dependent variable in undifferenced form.

19. xtclog, which was renamed to xtcloglog in Stata 8, returned an e(cmd) or e(cmd2) of "xtcloglog" even when run under version control. It now returns "xtclog" when version is set to 7 or less, and "xtcloglog" otherwise. In all cases, the xtcloglog command returns "xtcloglog" in e(cmd) or e(cmd2).

-------- update 30jan2003 -----------------------------------------------------


1. graph combine did not always respect noticks and nolabels settings on the x- and y-axes of the graphs being combined. This is fixed.

2. graph combine sometimes refused to accept quoted filenames for the graphs to combine; now it does.

3. (Mac) graph export now correctly recognizes the .pct suffix.

4. graph twoway refused to accept the text() option when the graph had more than one x- or y-axis; this is fixed.

5. gsort failed when performed on a no-observation dataset and a descending sort was requested. This is fixed.

6. The Stata Journal graphics scheme had too large of a default graph size; this is fixed. The new size has the same aspect ratio as the default graph size for the s2color scheme.

7. svytest, an out-of-date command superseded by test, failed to allow variable name abbreviations; this is fixed.

8. tab1's dialog would not accept multiple variables in the "Categorical variable(s)" field; it now does.

Stata executable

9. Expressions that contained a variable name that

1. Started with the letters a-f 2. continued, containing only the letters a-f or digits 0-9 3. ended in x

would generate a syntax error when the variable was immediately followed by a + or - followed by a digit or the letters a-f. For instance, ". generate newvar = fx+2" would produce a syntax error. This is fixed. The problem was that expressions such as "fx+2" were incorrectly confused with hexadecimal literals such as "1.0fx+2".

10. (Windows) In the 17jan2003 update, the default clipboard format was changed from WMF (Windows Metafile) to EMF (Enhanced Metafile) format. Older Windows computers did not support EMF format, so a preference has been added to the Clipboard tab of the Graph Preferences dialog box to allow users to choose between EMF and WMF format. The default clipboard format for modern versions of Windows (XP/2000/NT) is EMF. The default format for older versions of Windows (ME/98) is WMF.

11. (Windows) set reventries did not accept the permanently option; now it does.

12. (Mac) Stata now uses OS X's Quartz 2D drawing engine for drawing and printing graphs. Copying graphs to the clipboard is still limited to the PICT format.

13. (Mac) Reverse vertical and reverse horizontal text are now displayed properly when using the QuickDraw graphics engine for drawing graphs. This only affects graphs copied to the clipboard or saved as PICT files.

14. (Mac) graph export filename.pict would incorrectly report there was an error exporting a graph as a PICT file even though there was no error; this is fixed.

15. (Mac) Graphs did not honor their aspect ratio until the user resized the window manually; this is fixed.

16. (Mac) The precedence behavior for the Review and Variables windows can be set; see help varwindow.

17. (Mac) The buttons on the Viewer window and Data Editor have been changed to use the Small System font.

18. (Mac) display _request() would prompt the user for input, ignore the input, prompt the user again, and then accept the input. This is fixed.

-------- update 23jan2003 -----------------------------------------------------

Stata executable (Windows only)

1. odbc incorrectly stored strings in variables that were 127 characters wider than necessary; this is fixed.

2. The Review and Variables windows now maintain their position independently of the main Stata window. If you install this as an update, the first time you run the new executable the Review and Variables windows may not appear or may appear in odd positions. Select Default windowing from the Prefs menu and then reposition the windows as desired.

3. update swap would return an error if there was a blank in the name of the folder in which Stata is installed; this is fixed.

4. set smalldlg, introduced in the 17jan2003 update, did not accept the permanently option; now it does.

-------- update 21jan2003 -----------------------------------------------------


1. avplots, gladder, qladder, shewhart, and varfcast graph did not drop temporary graphs that they created; this is fixed.

2. graph twoway function and graph twoway kdensity, when the data were sorted by a string variable and there were less than 300 observations, would change the data in memory. This is fixed.

-------- update 17jan2003 -----------------------------------------------------


1. graph twoway scatter with very large datasets under Stata/SE now produces graphs much faster.

Stata executable

2. gllamm, a user written program, runs faster due to some helper routines which are now included as part of Stata's executable. Type search gllamm, all to locate the gllamm command.

3. graph now allows the {c ...} smcl directive to be used in graphical text. For example,

. scatter y x, title("Crecimiento demogr{c a'}fico")

results in the title Crecimiento demográfico.

4. preserve and save, emptyok, given a multiple-observation dataset with no variables, would create an unreadable dataset; this is fixed. This happened only in the rare case of having observations with no variables.

5. (Windows) The default clipboard format has been changed from WMF (Windows Metafile) to EMF (Enhanced Metafile) format. This was done to ensure that items such as filled polygons could be pasted into other applications.

6. (Windows 98/ME) Some Windows 98/ME computers may suffer from a memory limitation that prevents them from opening some of the larger Stata dialog boxes, such as those for graphics commands. Smaller dialog boxes have been implemented and may be selected as the default by typing set smalldlg. See help smalldlg for more information.

In addition, if you ran into this limitation, Stata would refuse to open dialog boxes subsequently. This is fixed.

7. (Mac) The Viewer window could be slow in loading some help files; this is fixed.

8. (Mac) The windowing positioning logic for floating windows was modified to avoid a bug present in all versions of the Mac OS X operating system up to and including 10.2.4.

9. (Mac) Old-style programmable dialogs, such as those used by the StataQuest, could not be displayed due to a change in Mac OS X; this is fixed. Nonetheless, users are strongly encouraged to program dialogs using the new dialog language; see help dialogs.

-------- update 14jan2003 -----------------------------------------------------


1. graph and any commands using graph would issue an error message and fail if Stata was installed in a directory that contained a space anywhere in the full pathname of the directory. This is fixed.

2. hetprob would not work when supplied with the difficult option; this is fixed.

3. lowess would put the default title, subtitle and note on each graph when used with the by() option; this is fixed.

4. svygnbreg and svynbreg would not work when supplied with the irr option; this is fixed.

5. twoway graphs that included a right axis drew the labels on that axis using horizontal text (while the text on the left axis was drawn vertically). Both axes are now drawn using vertical text by default. This affects graphs drawn with the s1 and s2 families of schemes.

6. twoway histogram would not allow the number of bins to exceed the number of observations; this is fixed.

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

See whatsnew7to8.


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