Stata 15 help for whatsnew7

Additions made to Stata during version 7

This file records the additions and fixes made to Stata during the life of Stata version 7:

+---------------------------------------------------------------+ | 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 | | whatsnew8 Stata 8.0, 8.1, and 8.2 2003 to 2005 | | whatsnew7to8 Stata 8.0 new release 2003 | | this file 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. Note: Starred (*) items mean the update was made to the executable.

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

See whatsnew7to8.

----- update 05dec2002 --------------------------------------------------------

1. On-line help and search index brought up-to-date for Stata Journal 2(4).

2. binreg with the by ...: prefix did not restrict the observations based on the by. This is fixed.

----- update 08nov2002 --------------------------------------------------------

1. xpose did not allow specification in the optional _varname variable of variable names of the form "v" followed by a number, such as "v1" or "v2"; this is fixed.

----- update 28oct2002 --------------------------------------------------------

1. codebook misspelled "trailing" in the output; this is fixed.

2. savedresults compare was not comparing matrices; this is fixed.

3. testnl did not allow the g() or r() options when the expressions were enclosed in parentheses (the second syntax of testnl); this is fixed.

----- update 14oct2002 --------------------------------------------------------

1. adjust used after a model specified with an offset() or exposure() that varied within groups defined by adjust's by() option, would arbitrarily choose one of the offset() or exposure() values found within each group in computing predictions. Now adjust requires that offset() or exposure() be constant within the groups defined by adjust's by() option or that adjust's nooffset option be specified.

----- update 02oct2002 --------------------------------------------------------

1. mvreg with collinear x variables correctly dropped the collinear variables from the estimation, but did not correspondingly drop them from the count contained in e(k) and e(df_r) which caused standard errors to be slightly larger than they should have been. This is fixed.

----- update 25sep2002 --------------------------------------------------------

1. swilk, lnnormal would fail when there were 11 or fewer observations; this is fixed.

2. xtabond produced an incorrect e(sample) when the sample was more restricted by lags of the predetermined variables than by lags of the endogenous variables; this is fixed.

----- update 18sep2002 --------------------------------------------------------

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

2. kdensity, parzen when specified with weights did not apply the weights; this is fixed.

3. svytab, ci now produces a better error message when more than one of the cell, count, row, or column options are specified.

4. tabodds, woolf or incorrectly labeled the Woolf confidence interval as being Cornfield; this is fixed.

----- update 09aug2002 --------------------------------------------------------

1. cluster kmeans and cluster kmedians with a binary dissimilarity measure option (for example, matching, Jaccard, ...) and presented with nonbinary data having no zeros, would cause Stata to crash. It now displays an appropriate error message.

2. egen ... cut() with the at() option was updated on 08may2002 so that k instead of k-1 categories were created when k numbers were specified in at(). The original behavior of creating k-1 categories is now reinstated.

Some argue that at() should create k categories, others counter that k-1 is most natural. Either way, the 08may2002 change should not have occurred because StataCorp policy is that updates during a release are to fix bugs and to add new functionality, not to change existing functionality.

3. xtivreg, be with [if exp] or [in range] did not save e(sample); this is fixed.

----- update 02aug2002 --------------------------------------------------------

1. collapse would generate 0 for the standard deviation (sd) of a variable that equaled missing value (.) in all observations even though it would generate missing value for the mean. collapse now generates missing value for the standard deviation, too.

2. ksm, logit by() did not handle the by() graph option correctly; this is fixed.

3. roctab could produce a "not enough memory" error due to temporary variables not being dropped at the end of each iteration of a loop. This is fixed.

4. ssc has new subcommand whatsnew -- just type ssc whatsnew -- that summarizes the packages most recently made available or updated. Output is presented in the Viewer, from where you may click to find out more or install. See help ssc.

----- update 18jul2002 --------------------------------------------------------

Stata executable(*)

The 18jul2002 update is relevant only for 64-bit executables: 64-bit Sun Solaris, 64-bit SGI IRIX, and Tru64/Digital Unix. For all other platforms (including Windows, Mac, and all other Unix), the 11jun2002 executable continues to be the most up to date.

For the 64-bit executables:

1. Stata on 64-bit computers could crash when there was more than 2 gigabytes of data in memory and certain memory reorganizations were necessary. This is fixed.

----- update 11jul2002 --------------------------------------------------------

1. boxcox treated iweights as if they were fweights and would not allow noninteger iweights (see help weights); this is fixed.

2. egen now allows longer numlists in the values() option for the eqany() and neqany() functions.

3. nl now produces more informative error messages.

4. svytab, missing would fail to run when missing values were present; this is fixed.

5. svytab would occasionally override the user specified stubwidth() and cellwidth() options (added in the 21jun2002 update). This is fixed.

----- update 21jun2002 --------------------------------------------------------

1. On-line help and search index brought up-to-date for Stata Journal 2(2).

2. glm previously calculated the Bayesian Information Criterion (BIC) using the model degrees of freedom as a component of the calculation, whereas the more common formula for BIC in the literature uses the residual degrees of freedom. glm now calculates BIC using the residual degrees of freedom.

3. newey now preserves the sort order of the data.

4. svytab has new options: cellwidth(), csepwidth(), and stubwidth(). See help svytab for details.

----- update 11jun2002 --------------------------------------------------------

1. blogit and bprobit with the cluster() or offset() options produced an error message; this is fixed.

2. dstdize ignored the if and in qualifiers when checking the data; this is fixed.

3. egen ... cut(), label gave incorrect value labels as a result of the 08may2002 ado-file update; this is fixed.

4. fracpoly reported a missing value for the deviance when the model fit the data exactly; this is fixed.

5. icd9 check and icd9p check with the generate() option would fail if the operating system's temporary file directory name had a space in it; this is fixed.

6. predict, anscombe produced incorrect results when used after glm for negative binomial and non-Bernoulli binomial models. This is fixed.

7. xtabond: A number of improvements and bug fixes have been made.

A. xtabond did not use the first available moment condition for predetermined variables, causing the estimates to be consistent but less efficient than they could be. This is fixed.

B. xtabond, when there was a gap common to all panels, did not properly calculate the degrees of freedom of the Sargan test. This is fixed.

C. xtabond, robust and xtabond, twostep, with some datasets, could unnecessarily exit with the error "matrix not symmetric", r(505), due to an overly restrictive tolerance. This is fixed.

D. xtabond, noconstant required that exogenous variables be specified, even if predetermined variables were specified. noconstant is now allowed when either exogenous or predetermined variables are specified.

E. xtabond now allows endogenous regressors; see help xtabond for syntax.

F. xtabond is now 17% faster.

8. xtgls has an improved error message.

9. xtivreg: One improvement and one fix has been made.

A. xtivreg now separately checks for collinearity in the varlists of exogenous variables and instruments. This ensures that when an instrument is perfectly collinear with an exogenous variable, it is the instrument and not the exogenous variable which is dropped from the regression.

B. xtivreg, fd did not save e(sample). This is fixed.

10. xtregar, fe displayed temporary variable names when there were time-series operated variables. This is fixed.

Stata executable(*)

11. (All platforms) The maximum number of significant description lines in a .pkg file (see help net) has been increased from 20 to 100. Previously, only the first 20 description lines would be listed. Now, the first 100 will be listed.

12. (All platforms) The lngamma() function now has 8 to 9 digits of accuracy whereas before it had 6 to 7.

13. (All platforms) Internal service routines were modified to support the changes made to xtabond documented in (7) above.

14. (Stata/SE, all platforms) Using the data editor, if you added onto the end of a string variable to make it longer than str80, the variable's contents would be truncated to 80 characters. This is fixed (the maximum length of strings in Stata/SE is 244 characters).

15. (Stata for Unix) Using the Print dialog when no PostScript printer had been defined would cause Stata to crash. Stata now displays an error message describing the problem.

----- update 08may2002 --------------------------------------------------------

1. alpha, if invoked with several variables and one of the variables was dropped because it was constant, resulted in all the subsequent variables also being dropped. This is fixed.

2. destring when called with multiple variables could fail due to changes made in the 04apr2002 update. This is fixed.

3. egen ... cut() with the at() option previously created k-1 categories when k numbers were specified in at(). Values larger than the kth value of at() were mapped to missing. Now k categories are created, and values larger than the kth element of at() are mapped to the kth element of at().

4. mfx produced incorrect standard errors when used on certain predictions after certain commands. This is fixed.

A. mfx calculated standard errors for any statistic calculated by predict after any estimation command, even if that would be inappropriate, such as predictions of residuals. The marginal effects calculated, however inappropriate they might be, adhered to the definition and so were "correct", but the standard errors were incorrect and, in fact, cannot be calculated by ordinary means. In this case, marginal effects are now reported but standard errors are suppressed.

B. mfx calculated standard errors that were incorrect in the following cases:

B.1. Predictions of pr(a,b), e(a,b), and ystar(a,b) after regress, truncreg, cnsreg, heckman, eivreg, intreg, svyintreg, and tobit.

B.2. Predictions of nu0 and iru0 after xtnbreg.

B.3. Predictions of ycond and yexpected after heckman.

B.4. Predictions obtained after streg, dist() frailty() that are unconditional on the frailty.

In all other cases, reported marginal effects and standard errors were correct. This includes use after probit, logit, mlogit, etc.

In the cases listed, the reported marginal effect was correct but the standard error was incorrect; this is fixed. In the case of (B.1), marginal effects for pr(a,b), e(a,b), and ystar(a,b), are reported for all commands, but standard errors will not be reported after regress, cnsreg, and eivreg.

All the cases in (B) concerned the calculation of standard errors when the underlying model had ancillary parameters and they were either not recorded in the coefficient vector (for example, RMSE in the case of linear regression) or recorded in an e() scalar as well as the coefficient vector and the corresponding predict command was written using the e() scalar result. mfx works by perturbing the coefficient vector and recalculating the predictions in order to obtain standard errors. In all of the above cases, the effect of the uncertainty about an ancillary parameter was being ignored in the standard-error calculation of the marginal effect.

Thus, the fixes for the problem were actually made to the corresponding predict commands. In addition, mfx was modified so that it now verifies that results from predict are a function of the coefficient vector only and, if they are not, mfx suppresses the reporting of standard errors.

The cases in (A) were similarly cases of the prediction not solely being a function of the coefficient vector, but in those cases, what was asked for was inappropriate because the standard error substantively depended on other things as well. Since mfx now checks that results from predict are a function of the coefficient vector only, such inappropriate standard errors will no longer be reported.

5. mfx did not clear its estimation results, so if you typed "mfx compute" and then typed "mfx compute, nose", it would appear that the nose option was being ignored. This is fixed.

6. mfx compute, predict(residual) after regress would claim to be ignoring your specification predict(residual) and then in fact not ignore it. This is fixed.

7. predict after streg now labels the new variable better and produces more informative error messages.

8. predict, mean time is now available after streg, dist(gamma). (predict after streg, dist(gamma) previously could not make mean-time calculation).

9. predict, ir when used after zip and zinb produced a syntax error. This is fixed.

10. predict, n nooffset when used after zip and zinb would apply the nooffset option to both the main equation and the "inflate" equation, when in fact it should only be applied to the main equation. This is fixed.

11. xtgls and xtpcse gave wrong standard errors when aweights were specified; this is fixed.

Stata executable(*)

12. tabdisp and table would misalign the value labels above the supercolumns in 3-way tables, and display text such as "{hline -4}", when the value label was very long. This is fixed.

13. The cause of a crash reported with reshape has been fixed. The crash would in fact be invoked by the following sequence of events: (1) a long, very narrow dataset which nearly filled memory was loaded; (2) the data were dropped by drop _all; (3) set obs was then used to create a 0-variable dataset; (4) generate was finally used to create a variable that itself was wider than the sum of the widths of the previously loaded narrow dataset. Stata would crash on the generate as it tried automatically to repartition memory containing 0 variables and _N>0 observations.

14. (Windows) Filenames with paths longer than 128 characters may now be opened in the do-file editor.

15. (Mac) With set slash on, macros containing a file path would be incorrectly expanded; this is fixed.

16. (Mac) The Viewer window would sometimes not properly refresh its contents after the Back button had been pressed if the scrollbar was not at the top of the document. This is fixed.

17. (Mac) In OS X, if the Viewer window was in front of the Results window, it was not possible to make a selection in the area of the Viewer that obscured the Results window. This is fixed.

----- update 04apr2002 --------------------------------------------------------

1. cci could enter an endless loop if one of the four numbers entered was zero; this is fixed.

2. destring now allows an abbreviated newvarlist in the generate() option. For example,

. destring a b c x y z, generate(v1-v6)

generates six variables (v1, v2, ... v6).

3. destring failed when a large number of variables (300+) were specified; this is fixed.

4. ir, by() ird did not give correct standard errors for the standardized rate difference (SRD) for stratified incidence-rate data; this is fixed.

5. With the 09jan2002 update, predict after streg, frailty() ignored the optional user specified storage type and instead created the variable using the default type (float). This is fixed.

6. tabodds, cornfield calculated exact confidence intervals instead of the requested Cornfield confidence intervals; this is fixed.

7. tabodds could produce missing values for some of the odds ratios when the exposure variable had noninteger values; this is fixed.

----- update 22mar2002 --------------------------------------------------------

Stata executable(*)

1. (All platforms) set more has new permanent option that will cause Stata to remember the more setting between sessions. For instance, you can type "set more off, permanent" so that (1) more is turned off and (2) Stata will subsequently come up with more turned off.

2. (All platforms) Stata's reaction time to pressing the Break key, and the refreshing of windows during long calculations, is greatly improved.

3. (Windows) Under Windows 98, Stata could not allocate more than 256 megabytes to the data area; this is fixed.

4. (Windows) Under Windows XP, users sometimes could not start Stata from the Start menu. This is fixed.

5. (Windows) Under Windows 2000 and Windows XP, clicking on a Stata window to make it active would sometimes not result in the window bar turning blue (that is, the window would not be activated, even though the window did get the keyboard focus and so allowed input). The most noticeable side effect was the window would not allow the highlighting of text for copying into other windows, and this was most noticed in the Command window. This is fixed.

6. (Mac) set slash is now better documented and, if you "set slash on" slashes (/) in filenames will be understood everywhere as the directory separator.

7. (Mac) Icons compatible with OS X are now included.

8. (Mac) graph would sometimes produce text that should have been vertically aligned but was not; this is fixed.

9. (Mac) window menu clear will no longer result in multiple instances of the same menu item appearing.

----- update 15mar2002 --------------------------------------------------------

1. On-line help and search index brought up-to-date for Stata Journal 2(1).

2. arima may now be used with the by prefix command.

3. arima now allows estimation and prediction using samples that are wholly contained within one panel of a panel dataset (a dataset that has been tsset with both panel and time identifiers). This allows arima to be used in loops over panels. Previously, unless the condition option was specified, arima refused to estimate any models on panel data.

4. ci, poisson previously treated missing values in the varlist as representing zero events when the exposure() option was specified and the corresponding exposure variable was not also missing. Now ci excludes those observations.

5. ml and all estimators using ml have a new tolerance option for determining convergence: nrtolerance(). Convergence is declared when g*inv(H)*g' < nrtolerance(), where g represents the gradient vector and H the Hessian matrix; see help maximize.

6. reg3 produced an extraneous horizontal line when the noheader option was specified; this has been fixed.

7. sts generate would produce the error message "_merge already defined" when the user had previously used the merge command and had not dropped the _merge variable. That was because, in its machinations, sts generate used the merge command itself. sts generate still does this, but it no longer uses the variable name _merge to mark the outcome of its internal merge, meaning that if you have done a previous merge, sts generate will not complain that _merge is already defined, nor will this fix result in the value of an existing _merge variable having its values changed.

stci had the same behavior as sts generate because it used sts generate as a subroutine. stci no longer cares whether the variable _merge exists in the dataset.

8. xi could fail when the command line contained quotes; this is fixed.

----- update 27feb2002 --------------------------------------------------------

1. icd9 and icd9p have been updated to use the V19 codes; V18 codes (see the 11jun2001 update) and V16 codes were previously used. V16, V18, and V19 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.

2. merge no longer has a limit on the number of key (matching) variables; the previous limit was 10. This change was implemented in the 01feb2002 executable update, but was not mentioned.

3. mkmat is now faster.

4. rchart and shewhart now allow [if exp] and [in range] as claimed in [R] qc.

5. stcurve has two new options, unconditional and alpha1, for use after fitting frailty models with streg. unconditional plots curves that are unconditional on the frailty. alpha1 plots curves that are conditional on a frailty value of one. See help stcurve for details.

----- update 01feb2002 --------------------------------------------------------

1. Stata/SE was released, a new product that allows more variables, longer strings, and larger matrices.

Stata/SE is not a new release of Stata and so you should not be worried that you will be out-of-date if you do not upgrade. Stata/SE 7 is a bigger version of Stata 7. If any of the limits of Intercooled Stata have gotten in your way, Stata/SE provides the solution.

See help stata/se for more information and point your browser to for purchase information.

2. biprobit, in the unusual case where robust variance estimates were requested and the parameter /athrho was constrained, would produce an extraneous message that the constraint was dropped. The message was incorrect, and the constraint was included in the estimation. This is fixed.

3. bstat could exit with an unhelpful error message in certain cases when calculating the endpoints for the bias-corrected CI; this is fixed.

4. corr2data and drawnorm could produce the wrong correlation structure when the covariance matrix specified in the cov() option contained equation names; this is fixed.

5. ssc now produces the same informative error messages for those using a proxy server as those who do not.

6. stset now preserves the sort order of the data.

7. xtpcse, when specified with the options correlation(psar1) and hetonly, and when the matrix size was declared smaller than the number of panels, would issue an error message and refuse to estimate the model. This is fixed.

Stata executable(*)

8. Intercooled Stata has been updated to read the new Stata/SE .dta file format automatically if the dataset does not contain strings longer than str80 or more than 2,047 variables.

9. set matsize and set memory have new option permanent to set the value and remember it between sessions.

10. anova could cause Stata to crash when estimating a model with more than 255 terms (a*b counts as one term no matter how many levels of a or b there are). This is fixed.

11. The text for the error messages associated with return codes 900 (no room to add more variables), 901 (no room to add more observations), 902 (no room to add more variables due to width), 903 (no room to promote variable), and 908 (matsize too small) has been improved. If you are curious, you can see the improved text by typing error 900, ..., error 903, and error 908 in the Command window.

12. Stata for Mac will now properly display line breaks in help files created under Windows.

----- update 09jan2002 --------------------------------------------------------

1. heckman, twostep did not save e(sample). This affected no outputted results and is fixed.

2. predict after streg, frailty() has two new options, alpha1 and unconditional. alpha1 generates predictions conditional on a frailty equal to 1. unconditional generates predictions that are unconditional on the frailty.

Previously, when fitting a frailty model with streg, frailty() without the shared() option, predict would give unconditional predictions, and one may now override this default by using the new alpha1 option.

Previously, when fitting a shared frailty model with streg, frailty() shared(), predictions from predict were conditional on a frailty value of 1, and one may now override this default by using the new unconditional option.

See help streg.

3. predict after streg, dist(llogistic) ancillary() or after streg, dist(gamma) ancillary() anc2() sometimes gave predictions based on erroneous estimates of the ancillary parameter(s). The estimated ancillary parameters should vary with the data, yet the parameters were sometimes held fixed at their values for _n==1 when the predictions were calculated. This is fixed.

4. predict after streg handles if and in conditions more efficiently (it runs faster).

5. svydes, svymean, svyprop, svyratio, svytab, and svytotal now preserve the sort order of the data.

----- update 20dec2001 --------------------------------------------------------

1. ivreg no longer lists perfectly collinear variables in e(insts).

2. pac (and corrgram since it calls pac) would overwrite any existing e() results with the results from the regression computed within pac. Now the previous e() results are restored.

3. treatreg would report a "conformability error" and fail to run when there were perfect predictors in the treatment equation; this is fixed.

4. xtreg now issues a more appropriate error message when the panel variable is collinear with the independent variables.

5. xtregar, fe did not save e(rmse); this is fixed.

----- update 11dec2001 --------------------------------------------------------

1. streg has new option shared(varname) for fitting parametric shared frailty models, analogous to random effects models for panel data: the frailty value is common among groups of observations and randomly distributed across groups. streg could and still can estimate frailty models where the frailties are assumed to be randomly distributed at the observation level (specify option frailty() without the new option shared()). See help streg.

2. On-line help and search index updated to include reference to more of the material available at

3. adjust now allows the pr option when used after svylogit, xtlogit, svyprobit, and xtprobit.

4. arima with the unusual combination of a model with no covariates and with options from(armab0) and noconstant specified, would issue an error message and refuse to estimate the model; this is fixed.

5. findit without the saving() option now places the file _finditresults.smcl in the directory specified by the global macro FINDIT_DIR if defined or the current directory if not.

6. linktest now works after stcox and streg.

7. nlogit with the by prefix command would present results for the complete sample after presenting the results for each by group. Now it presents only the results for each by group.

8. roccomp, graph now accepts the l2title() graph option. Additionally, roccomp, graph binormal now allows graph's pen() option.

roccomp, graph used to assign a key to the 45-degree line in the graph. It no longer does, instead providing keys only for the actual curves.

9. roctab has a new option specificity to graph sensitivity versus specificity, instead of the default sensitivity versus (1-specificity); see help roctab.

10. treatreg, twostep did not save e(sample); this is fixed.

11. prais could, due to collinearity, drop the dependent variable; this is fixed.

12. zip and zinb now accept the from() option.

----- update 07dec2001 --------------------------------------------------------

Stata executable(*), Windows

1. The 06dec2001 fix for the on-going problem in which the mouse could not be used to select text in edit fields under Windows 2000 and XP created a problem for Windows 95 and 98; that is fixed.

----- update 06dec2001 --------------------------------------------------------

1. ml unnecessarily left behind matrices named ML_CT and ML_Ca after estimation with constraints().

ml now saves the constraint matrix in internal matrix Cns which can be obtained by typing "matrix name = get(Cns)".

ml display now displays any constraints associated with the estimates.

Note, there was nothing wrong with any results produced by ml; these changes merely affect how results are stored and displayed.

Stata executable(*)

2. In downloading from the SSC Archive, Stata would occasionally (some would say often) produce the error message "I/O error" and you would have to try again. This is fixed. (Stata now watches for the problem and reestablishes the connection.)

3. In expressions, FORTRAN syntax such as "1d+2" was not understood to mean 100 despite the documentation's claim otherwise. Now 1d+2 and even 1d2 are understood.

4. foreach and forvalues now respond more promptly to the Break key.

5. ivreg, beta produced random values for the betas (standardized coefficients); this is fixed. Coefficients, standard errors, t-statistics, etc., were all correct.

6. mlogit previously printed any constraints before the iteration log; it now prints them after the log and before the coefficient table. Note, there was nothing wrong with any results produced by mlogit; this change merely affects how results are displayed.

7. syntax allows new syntactical element anything; see help syntax.

8. If you are programming in the dialog boxes created by window dialog, you now can repopulate a combo box by changing the contents of the macro associated with it.

9. (Windows) The permanent, more elegant fix promised in the 27oct2001 update for the on-going problem in which the mouse cannot be used to select text in edit fields under Windows 2000 or XP has been found and implemented. If you have set cmdtitle off in order to work around the bug and want the title back, set the cmdtitle back on.

10. (Windows) Multiple instances of Stata running at the same time are now clearly marked in their title bar with an instance number.

11. (Windows) Results and Viewer windows now have contextual menus; right-click when you are in the window to try it.

12. (Windows and Mac) set rmsg could report the decimal portions of timings incorrectly; this is fixed.

13. (Mac) You can now select all the contents of the Results or Viewer windows by selecting Select All from the Edit menu.

14. (Mac) Any errors that occur when printing are now reported. Previously, Stata relied on the Finder to report printing problems which, in some cases, it did not.

15. (Mac) Clicking in the Variables window will now overwrite selected text in the Command window.

16. (Unix GUI) Printing the Results or Viewer windows through the Print dialog box would sometimes fail; this is fixed.

17. (Unix GUI) window menu did not allow appending to the default Stata menus when the keyword sysmenu was enclosed in quotes; this is fixed.

18. (Unix GUI) If a minimized window was selected from the Window menu, Stata sometimes crashed; this is fixed.

----- update 14nov2001 --------------------------------------------------------

1. On-line help and search index brought up-to-date for Stata Journal 1(1).

2. loneway with the exact option produced correct results, but displayed spurious smcl commands in a table header; this is fixed.

3. ml model, waldtest(0), rather than executing, produced an error message; this is fixed.

4. nl now saves the number of iterations in e(ic).

5. nlogit used with weights did not correctly check for constant weights within groups; this is fixed.

6. roctab, detail could, under exceedingly rare conditions, produce incorrect results; this is fixed. Additionally, a display format was changed to better display noninteger values.

7. ssc is a new command that lists and installs packages (and files) from the Statistical Software Components (SSC) archive, often called the Boston College Archive, and provided by

ssc is based on archutil by Nicholas J. Cox, University of Durham, and Christopher Baum, Boston College. The reworking of the original was done with their blessing and their participation.

8. stcurve, hazard is no longer allowed after stcox since the graph produced was not a true hazard curve.

9. tabodds' display format was changed to better display noninteger values in the table.

10. xtivreg now saves the list of instrumented variables in e(instd) and the list of instruments in e(insts).

----- update 26oct2001 --------------------------------------------------------

Stata executable(*), all platforms

1. matrix makeCns, a programmer's command, now allows specifying constraints as a matrix or as a list of constraint numbers; see help matconst. This new matrix feature is also inherited by most commands that allow the constraints() option, such as cnsreg. In those commands, the constraints() option may now be specified with a list of constraint numbers, as previously, or the name of a matrix.

2. mat_put_rr is a new programmer's command that directly specifies the Rr matrix to be used in the next replay call of test; see help mat_put_rr.

3. post has improved error messages to deal with too many or too few expression arguments.

4. program drop (and automatically generated program drops) could cause subprograms to be mistakenly dropped when the subprogram had the same name as the main program in the dropped module. This is fixed.

5. search now has new option sj which is a synonym to the existing option stb that limits the search to the Stata Journal and Stata Technical Bulletin.

Stata executable(*), Windows, dated 27oct2001

1. A workaround has been introduced for an on-going problem in which the mouse cannot be used to select text in the Command window under Windows 2000 or XP. The workaround is to type set cmdtitle off. Doing this makes the mouse work but causes the window title to disappear, meaning moving the window or changing its font is impossible. Typing set cmdtitle on will bring back the title bar along with the mouse problem.

We are working on a permanent, more elegant fix. For more information, see

----- update 17oct2001 --------------------------------------------------------

1. bitest, prtest, ttest, and sdtest now allow == to be used wherever = is allowed in their syntax. The use of == in these commands is more consistent with Stata's general syntax that treats == as indicating comparison and = as meaning assignment.

2. cc and cci would not show part of the results when the exact option was combined with the cornfield, woolf, or tb options. This is fixed.

3. testnl could, on the rare occasion that the derivative with respect to a coefficient is nearly zero, try to post a missing value to e(b). This is fixed, and a warning message is displayed instead.

4. xtgls reported too large of an average number of observations per group in the header of its output; this is fixed.

----- update 12oct2001 --------------------------------------------------------

Stata executable(*), all platforms

1. In the 01oct2001 update, quietly and while could not be used on the same command line interactively. Rather than typing

. quietly while {

you had to type

. quietly { . while ... . }

This is fixed.

Stata executable(*), Mac

1. Under OS X there is a new menu item, Bring All to Front in the Window menu. This item brings all Stata windows to the front.

2. The Enter key on the numeric keypad is now recognized as a carriage return.

3. When clicking a variable in the Variables window, a space is now output before the variable name rather than after. The 01oct2001 update claimed to have fixed this, but did not.

4. In OS 10.1, selecting Minimize from the Window menu after all windows had already been minimized resulted in a crash. This is fixed; selecting Minimize now restores the Results window.

5. Stata now properly uses Internet Config to map nonnative Stata file extensions to their appropriate applications. For example, saving a file with the extension .html will set the file's type and creator to your default Internet browser. File extension mappings can be set via the File Exchange control panel.

6. The original Mac Menu Manager API did not allow "/" in menu titles and so Stata changed slashes to backslashes ("\"). The new Menu Manager does allow slashes, and so the substituting of backslashes in menu titles has been removed.

Stata executable(*), Stata for 64-bit Sun Solaris

1. Stata for 64-bit Sun Solaris has been under beta test. Stata for 64-bit Sun Solaris is now in production and fully supported as of this update.

----- update 01oct2001 --------------------------------------------------------

Ado-files, all platforms

1. predict, dyn() after arima would sometimes claim that some missing values of the prediction were produced when in fact the predictions were correctly produced. This is fixed.

2. xtgls, panel(corr) would fail when the path name for temporary files included spaces. This is fixed.

Stata executable(*), all platforms

1. display, under rare circumstances, could cause Small Stata to crash. This is fixed.

2. graph, oneway box could cause Stata to crash when the variable labels were very long. This is fixed.

3. matrix score, when the matrix variables had time-series operators, could, in extremely rare instances use an incorrect time variable. This could happen if the tsset time variable was changed immediately before matrix score and the data were previously tsset. In that case, the previously set time variable would be used. This is fixed.

mleval had the same problem and is also fixed.

4. The new net sj command makes loading files from the new Stata Journal easier. The Journal begins publication fourth quarter, 2001, so net sj does not do anything useful right now. You can, however, experiment with the command using volume 0 issue 0, written as 0-0. "net sj 0-0" shows the files available. "net sj 0-0 xx0001" shows a description for the files associated with reference number xx0001. See help net.

The command net stb is now documented; see help net. net stb works similarly to net sj: Typing "net stb 61" shows the contents of STB-61 and typing "net stb 61 dm91" would show the software associated with dm91.

5. translate, when translating graph files to PostScript format, would incorrectly translate characters such as á and à. This is fixed.

6. The following should have been documented as of the previous, 17sep2001 executable update: In rare instances it was possible to shadow the definition of an autoloaded subprogram if a program with the same name as the subprogram was defined interactively or as an ado-file. For example, if the ado-file mymain defined the subprogram mysub and mysub was also defined interactively, then the interactively defined version of mysub would be run in all cases, even when called from mymain. This was fixed.

Stata executable(*), Mac

1. When clicking a variable in the Variables window, a space was output after the variable name rather than before it. This is fixed.

2. Stata will now attempt to resize and move its windows if the display is changed to a smaller screen size.

3. update executable, after downloading and when presenting the final instructions for installing the new executable, would sometimes display extra (random) characters when referring to the name of the Stata executable. This is fixed.

Stata executable(*), Unix

1. When clicking a variable in the Variables window, a space was output after the variable name rather than before it. This is fixed.

2. print, when printing graph files to a PostScript printer, would incorrectly print characters such as á and à. This is fixed.

----- update 17sep2001 --------------------------------------------------------

Stata executable(*), all platforms

1. Stata now runs ado-files 6 to 9 percent faster than previously.

2. When copying-and-pasting into the data editor, Stata could become confused as to the character used to delimit cells and paste everything into a single cell. This is fixed.

3. anova could cause Stata to crash when used with no right-hand-side variables. This is fixed.

4. syntax, used in a program to parse user input, did not properly recognize abbreviation rules for options when the option name began with an underscore. Rather than capitalization being taken to indicate minimum abbreviation length, it was taken literally and so required the user to specify the options in uppercase letters. This is fixed.

5. set checksum is now set off rather than on by default. This change affects how Stata copies files over the Internet, including update, net, and any other command used with a filename starting with "http://".

Previously, Stata determined whether a checksum file existed at the site and, if it did, used the contents of that file to verify the error-free download of the original file. To restore the old behavior, you can type "set checksum on".

We have made the default off because (1) we are not aware of any user actually having errors and (2) some sites in Europe use caching to save on I/O to America. The result is that, when we would post a new update at, some sites would update their cache for the files themselves but not the checksum files and so Stata would be fooled into thinking that downloads had an error. This made it impossible for some users to update their Stata.

Stata executable(*), Mac

6. Stata for OS X could crash when clicking on the Delete button in the data editor. This is fixed.

7. There is now a console version of Stata for OS X. For details see

Stata executable(*), Unix

8. window menu append popout sysmenu, rather than appending a new menu to the system menu, mistakenly cleared the system menu first. This is fixed.

----- update 06sep2001 --------------------------------------------------------

1. mfx now respects the version of the calling program when it generates its predictions.

2. The regression diagnostic commands acprplot, cprplot, hettest, lvr2plot, ovtest, rvfplot, and rvpplot have been extended to work after anova.

3. savedresults drop would produce an error message and fail to operate; this is fixed.

4. vif would occasionally fail to provide output for all applicable variables when some of the variables had been dropped from the model due to collinearity; this is fixed.

----- update 24aug2001 --------------------------------------------------------

1. On-line help and search index brought up to date for STB Reprints Vol. 10. Stata related FAQs found at have also been added to the search index.

2. findit now allows the word "for" to be included among the search words.

3. ksm now allows the by() option of graph.

4. xtregar would not report results when the time-variable was included in the regression and the lbi option was specified. xtregar now explicitly prohibits the inclusion of the time-variable in the regression.

----- update 14aug2001 --------------------------------------------------------

1. bs, used with certain user-supplied commands, would sample from the entire data set even when some of the observations were not originally used. Now bs checks if the user-supplied command is e-class and, in that case, resamples only the observations within the estimation subsample and, for other commands, bs displays a warning message.

2. gompertz and stphtest underwent slight output format corrections.

3. jknife, when used on an e-class command, now excludes observations not in e(sample) from the leave-one-out estimators.

4. nlogit reported an "unbalanced data" error when the values of the group() variable exceeded float precision. This has been fixed.

5. predict, nu0 and predict, iru0 used after xtnbreg has been corrected to account for the estimated dispersion parameters.

----- update 08aug2001 --------------------------------------------------------

Stata executable(*), all platforms

1. insheet has new delimiter("char") option that allows you to specify an arbitrary character that separates values; see help insheet.

2. graph, histogram has a new percent option for placing percentages on the vertical axis; see help gr7hist.

3. update now saves in r() various information for use by programs; see help update.

4. When drawing graphs, it was possible for differing line thicknesses of different pens not to be noticed if the color did not also change; this is fixed.

5. args x "a" (note the incorrect double quotes), in addition to producing the appropriate error message, would cause a memory leak so that the user eventually also saw the error message "out of memory" and was thus forced to exit and restart Stata.

6. (Windows) log using with a UNC name (that is, \\computername\path) would prefix the UNC name with the current directory and then refuse to open the file. This is fixed.

7. (Unix) Whenever the print dialog was invoked, there was the possibility of a small memory leak, although the problem was never observed. It is fixed.

8. (Mac) Stata now supports Mac OS X.

----- update 30jul2001 --------------------------------------------------------

1. predict after estimation by arch with the option archmexp() issued an error message and would not produce predictions; this has been fixed.

----- update 23jul2001 --------------------------------------------------------

1. glm would give a syntax error when both the trace and irls options were specified. Also, the eform option was ignored for some family() specifications. These problems have been fixed.

2. lookfor now returns the list of matching variables in r(varlist).

3. reg3 with the constraints() option did not set e(sample); this has been fixed.

4. xtgee and xtlogit have a new nodisplay option that suppresses the header and table of coefficients.

----- update 17jul2001 --------------------------------------------------------

1. binreg now respects the init() option.

2. findit is a new command that finds and lists sources of information on Stata and Stata commands already installed on your computer or available on the web. findit is Stata's most thorough, most complete search command. The results include (1) official help-files installed on your computer, (2) FAQs available at the Stata website, (3) material published in the STB and the Stata Journal, and (4) user-written programs and help files available over the web.

3. reshape long could produce incorrect results when (1) variable names for the j() identifier were longer than 8 characters, (2) the j() identifiers were string variables (string option specified), and (3) you were converting the data from wide to long. This is fixed.

4. sampsi now includes both tails when calculating power for a two-sided test.

5. svyintreg followed by predict with the pr() or e() options now labels the generated variable in the same way as is done by predict after intreg.

----- update 03jul2001 --------------------------------------------------------

1. areg with robust variance calculations could produce a nonpositive definite variance matrix. This is not an error, but sometimes areg would stop and produce an error message instead of posting the estimates. This has been fixed.

2. ivreg with the noconstant and first options would display first-stage results which included _cons; this is fixed. Note: there were NO errors in the final results presented by ivreg.

3. xi now uses natural numeric coding in more cases when creating variable names. Previously, if the variable had values larger than 99, xi would create its own coding to shorten the created variable names. Longer variable names make this no longer necessary.

----- update 26jun2001 --------------------------------------------------------

Stata executable(*), all platforms

1. gettoken sometimes presented an odd error message; the text of the message is improved.

2. graph using, when combining multiple graphs into one, could sometimes use the symbol from the first graph in subsequent graphs (it could do this when, in the subsequent graphs, the symbol was supposed to be invisible); this is fixed.

3. mlogit, if it ever had the error "no room to add more variables" or "system limit exceeded", would continue to repeat that error until another estimation command was run.

4. regress will no longer allow the beta and cluster() options to be combined.

5. translate could sometimes fail to read Stata graph (.gph) files if they were very large; this is fixed.

6. In the viewer, one of the net clickables would issue a syntax error if you clicked on it (because a comma was omitted); that is fixed.

7. (Mac) Internal changes have been made to update to accommodate the forthcoming Mac OS X version of Stata (which will be available as a free update).

----- update 21jun2001 --------------------------------------------------------

1. On-line help and search index brought up to date for STB-61.

----- update 11jun2001 --------------------------------------------------------

1. icd9 and icd9p have been updated to use the V18 codes; V16 codes were previously used. V16 and V18 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.

----- update 04jun2001 --------------------------------------------------------

1. boxcox now accepts iweights.

2. boxcox with the lrtest option now correctly reports the sigma from the unconstrained model. Previously, boxcox would report the sigma from the last constrained model estimated.

3. egen's cut() function could, on rare occasions, produce inappropriate missing values due to the loss of precision from using local macros instead of scalars for the cut points; this is fixed.

4. egen's diff(varlist) function failed when the varlist was over 75 characters long. The failure resulted from attempting to apply a variable label that was too long. Now, if varlist is over 75 characters, no variable label is applied.

5. for is now "version transparent". Previously, if you typed "for ...: cmd ...", cmd would be executed as if you had first typed "version 6". If cmd behaved differently under version 6 rather than version 7, for would execute the version-6 version of cmd. This is fixed. Now cmd is executed according to whatever version you have set, just as you would expect.

6. glm, jknife cluster() now works with weights.

7. reg3 with option ireg3 and sureg with option isure now display the message "convergence not achieved" if the specified or maximum number of iterations is exceeded. Also, to be consistent with ml, the default number of iterations when ireg3 or isure are specified has been increased to 16,000.

8. statsby called with an empty expression list now gives a more informative error message.

9. table would produce a syntax error for some valid contents() option specifications; this is fixed.

----- update 08may2001 --------------------------------------------------------

Ado-files, all platforms

1. egen functions called using _all as the varlist would include an unanticipated temporary variable among the variables; this is fixed.

2. print issued an error if filename had spaces in it; this is fixed.

3. The xi update of 13apr2001 would produce an error message when executed after the user had manually deleted some variables created using previous versions of xi; this is fixed.

Stata executable(*), all platforms

1. file is a new command that allows programmers to read and write both ASCII text and binary files; see help file. In addition, two new regular functions and one new macro-extended function were added for use when reading and writing binary files.

New numeric function byteorder() returns 1 or 2 depending on whether the computer writes the most or least significant byte first.

New string function char(#) returns the ASCII character corresponding to numeric code #, 0<#<256.

New extended macro function length returns the length in characters of a local or global macro.

Use of all of these new functions is described in help file, although the new length extended function is useful in other contexts, too.

2. list and correlate now respect linesize. list also has new option noheader.

3. Under certain circumstances, concatenating strings using the plus operator (for example, display stra+strb, or gen str80 newstr = stra+strb) when length(stra)+length(strb)>80, could produce unanticipated output and even cause Stata to crash. This is fixed.

----- update 23apr2001 --------------------------------------------------------

1. mfx produced incorrect answers when used after ologit or oprobit and one or more of the independent variable names contained underscores. The problem only arose when one or more of the variable names contained underscores. The problem is fixed.

2. mvdecode's mv() option now allows a numlist. In addition, certain English grammar errors were fixed in the output of mvdecode and mvencode.

3. svymlogit would produce an error message when the dependent variable had a value label and one of the label values had a period in it; this is fixed.

----- update 13apr2001 --------------------------------------------------------

1. mfx will now use pweights or iweights when calculating the means or medians for the atlist following an estimation command that used pweights or iweights. Previously, only fweights and aweights were supported.

2. mhodds would report missing values when large fweights were used; this is fixed.

3. ranksum is now faster and has a new option, porder, that estimates P(x1>x2).

4. svylc would give a conformability error when the data had value labels that contained spaces; this is fixed.

5. xi now saves in characteristic _dta[__xi__Vars__Prefix__] the prefix names for the corresponding variables listed in the characteristic _dta[__xi__Vars__To__Drop__]. This may be helpful to programmers.

6. xtivreg now reports the first stage regression for re and be models when option first is specified and there are time-series operators in the varlist.

7. xtregar, fe now allows aweights and fweights.

----- update 06apr2001 --------------------------------------------------------

1. xtivreg now estimates models that combine the first and ec2sls options.

2. predict after xtivreg now correctly predicts the individual level effects mu_i and the idiosyncratic errors e.

3. xtlogit now calculates rho using logistic variance of (_pi^2)/3 instead of 1.

----- update 05apr2001 (29mar2001) --------------------------------------------

(On 05apr2001, an update to the Stata executable dated 29mar2001 was made available.)

Stata executable(*), all platforms

1. Stata now allows setting the size of the scrollback buffer for the Results window. The number may be set between 10,000 and 500,000 bytes using the set scrollbufsize command; see help scrollbufsize. Previously, the value was fixed at 10,000 bytes.

2. Stata also has a new set varlabelpos command that controls the positioning of variable labels in the Variables window; see varlabelpos.

3. The smcl directive {center:...} now also allows the syntax {center #:...} for centering text in a field of width #.

4. outfile has a new runtogether option for use by programmers.

5. In all commands, options requiring a string as an argument previously required the string contain no more than 512 characters; that limit is now 67,784 characters (3,400 Small Stata). In addition, for such options, specifying the option without arguments was previously considered an error (return code 198); it no longer is.

6. The maximum number of dialog controls is increased from 55 to 85.

7. graph, twoway sometimes failed to display a key at the top of a graph when it should have; this is fixed. In addition, with the connect(||) or connect(II) options and symbols other than symbol(ii), graph, twoway would show the first variable in the key as if it were connected by straight lines. That is fixed, too.

8. Stata could save an incorrect .gph file. This only happened with large graph files having a particular pattern, and then only when saving the file by selecting File then Save Graph. Directly using the saving() option of graph did not have this problem. The problem is fixed.

9. translate and print did not align the top-right-corner table-drawing character correctly when converting to PostScript; this is fixed.

10. _rmcoll would run out of memory after being issued 40 times with no other intervening Stata commands; this is fixed.

11. Verification of checksums is now temporarily turned off during net search.

Stata executable(*), Windows

1. Stata now looks for the environment variable STATATMP in addition to the environment variable TEMP for the location of the directory for storage of temporary files. STATATMP takes precedence over TEMP.

2. Pressing Alt-F4 will now close the Graph window if it is in front of the Results window rather than closing all of Stata. If the Results window is in front; Stata will close as before.

Stata executable(*), Mac

1. net could cause a crash under rare conditions; this is fixed.

2. Graph output settings have been modified to work better with MS Word.

Stata executable(*), Unix

1. Stata now looks for the environment variable STATATMP in addition to the environment variable TMPDIR for the location of the directory for storage of temporary files. STATATMP takes precedence over TMPDIR.

2. Stata would not properly initialize the line width for the Result window until after it was manually resized; this is fixed.

3. Stata would enter an endless loop if a checkbox in a programmable dialog box was checked; this is fixed.

----- update 28mar2001 --------------------------------------------------------

1. On-line help and search index brought up to date for STB-60.

2. jknife and statsby displayed correct results (and generated the correct variables), but did so in a nonintuitive order when _b or _se with multiple entries were specified in the exp_list. This has been fixed.

3. kdensity now returns results in r() even when the nograph option is specified and the generate() option is not.

4. sample now gives a more informative error message when the requested sample percent is outside the range of 0 to 100.

5. xpose has three new options: format and format(%fmt) (which are mutually exclusive) and promote. The format option finds the largest numeric display format in the pre-xposed data and applies it to the xposed data. The format(%fmt) option sets the xposed data to the specified format. The promote option causes the transposed data to have the most compact numeric data type that preserves the original data accuracy.

6. xtintreg would report an error when used without predictors; this is fixed.

----- update 09mar2001 --------------------------------------------------------

1. aorder, used with long variable names (longer than 8 characters), would sometimes refuse to order the variables and instead issue the error "system limit exceeded"; r(1000). This is fixed.

2. mfx, predict(), with two or more arguments specified in the predict() option, would ignore the second and subsequent arguments; this is fixed.

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

4. xtgee, family(binomial) link(logit) refused to estimate models containing perfect predictors; it now attempts to estimate such models (producing very large or very small coefficients on the offending variables).

5. xtregar now estimates a rho-adjusted constant and predict after xtregar has been fixed to produce the correct u_i and residuals.

6. xttobit and xtintreg have new predict options. Option pr0(a,b) produces the probability of the dependent variable being uncensored -- Pr(a<y<b). Option e0(a,b) produces the expected value of the dependent variable conditional on censoring -- E(y | a<y<b). Option ystar(a,b) produces the expected value of the dependent variable truncated at the censoring point(s) -- E(y*), y* = max(a,min(y,b)).

----- update 01mar2001 --------------------------------------------------------

1. ltable ignored the xlabel() and ylabel() graphing options; this is fixed.

2. predict, residuals eq() used after mvreg calculated -X*b rather than y-X*b; this is fixed.

----- update 26feb2001 --------------------------------------------------------

1. glm, family(gaussian) with the offset() option gave incorrect standard errors; this is fixed.

2. predict, residuals used after rreg produced the message "invalid syntax", r(198), rather than predicting the residuals; this is fixed.

3. predict used after rreg no longer produces predictions for options stdf, stdr, cooksd, rstandard, and rstudent. These are options for linear models that are not strictly appropriate for the estimates produced by rreg.

4. sample has a new count option that allows samples of the specified number of observations (rather than percent) be drawn. In addition, sample now allows the by varlist: prefix as an alternative to the already existing by(varlist) option; both do the same thing.

5. xtpcse when specified without covariates -- a constant only model -- reported "last test not found", r(302), and refused to estimate the model. xtpcse will now estimate the constant-only model.

----- update 19feb2001 --------------------------------------------------------

1. cluster dendrogram would not read the (dis)similarity range information in the rare case when an additional "other" field was set after the range; this is fixed.

2. glogit used long integers for the number of successes and failures. This caused a loss of precision with extremely large populations. Now glogit uses doubles.

3. reg3 would sometimes produce a spurious message concerning missing values being generated; this is fixed.

4. sample, used after set seed, did not always reproduce the same (pseudo) random sample (the sample, in all cases, was random); the problem is fixed. For additional details, see the technical note at the end of help sample.

----- update 06feb2001 --------------------------------------------------------

1. On-line help and search index brought up to date for STB-59.

2. poisgof has a new option, pearson, to request the Pearson chi-square statistic.

3. alpha with the item option would display the negative estimate when a negative estimate was computed; it now displays missing value. As previously, alpha still returns in matrix r(Alpha) results when option item is specified. Because Stata matrices cannot contain missing values, the value 1e+30 is returned in r(Alpha) to indicate missing.

In addition, alpha's generate() option would label the new variable as standardized even when it was not; this is fixed.

Finally, the 17jan2001 update of alpha caused the reverse() option to display an uninformative message and then ignore that reverse() had been specified; this is fixed.

4. strate now drops temporary variables when a break is pressed.

----- update 05feb2001 --------------------------------------------------------

Stata executable(*), all platforms

1. forvalues x = 1(.2)2 (forvalues with noninteger increment) now works properly with set dp comma.

2. graph, by() had difficulty (could actually crash) when the by() variable had value labels and the value labels exceeded 32 characters. This is fixed.

3. graph no longer by default constructs keys for Hi-Lo charts (connect(||) and connect(II)).

4. insheet under version control (version set to 6 or earlier) now reproduces the naming behavior of earlier versions of Stata. When version is set to 6 or earlier, only the first 8 characters of the names are considered.

5. Matrix expressions (for example, matrix name = expression) that include the function nullmat() could cause unnecessary memory to be consumed, which memory would not be freed until Stata was exited. This is fixed.

6. stata.toc files (the files read by net from remote websites) now allow 150 links to be listed rather than 100.

----- update 29jan2001 --------------------------------------------------------

1. glm with a user specified link ignored the eform option. This is fixed.

2. mfx treated dummy variables containing missing values as continuous. In addition, under certain circumstances when one variable name was a proper subset of another, mfx would not report results for both variables. These problems are fixed.

3. stptime, strate, and statsby failed if the computer's temporary directory name contained spaces (typical on many Mac computers). This is fixed.

4. stptime ignored the jackknife option. This is fixed.

5. tabstat with the variance option failed. This is fixed.

6. xi would fail when specified with an interaction term containing a variable with a name of length one. This is fixed.

----- update 23jan2001 --------------------------------------------------------

Stata executable(*), all platforms

1. label values varname lblname, under exceedingly rare conditions, would not adjust the width of varname's %fmt display format. This is fixed.

2. set more off in a do-file would leave --more-- turned off even after the do-file completed. This is fixed.

3. set linesize # will no longer allow setting the linesize to #>255.

Stata executable(*), Windows

1. Stata now properly launches via double-clicking on filenames containing spaces.

2. Colored lines in graphs are now properly copied to the Clipboard under Windows 2000/NT (and all other versions of Windows).

Stata executable(*), Mac

1. A number of small bugs in the interface have been fixed including in the scrolling-variables window, the do-file editor, etc. SMCL files now appear in the Finder with the appropriate icon and are bound with double-quotes when opened from the Finder.

----- update 17jan2001 --------------------------------------------------------

1. alpha now allocates 12 rather than 8 characters for the display of variable names and its output is sensitive to linesize.

2. sts graph has new option lstyle() which allows specifying the style of the connected line. For instance, "sts graph, by(drug) lstyle([l] [-] [-.])" would graph the Kaplan-Meier survivor function for each drug, using a solid line ([l]) to connect the curve for the first drug, a dashed line ([-]) for the second, and dash-dotted line ([-.]) for the third.

3. stsplit, the text of certain error messages has been improved.

4. tabstat has new option varwidth(#) specifying the number of characters used to display variable names. In addition, the new options statistics(variance) and statistics(semean) displays the variance and the standard error of the mean.

5. testnl now allows typing "testnl exp=exp=...=exp" to test whether three or more expressions are equal.

6. xtivreg refused to estimate between models in which there were more parameters than groups; that is fixed. In addition, xtivreg with the first option mislabeled the intercept in the display of the first-stage fixed-effects model. The intercept is now correctly labeled _cons.

7. xtregar, fe refused to estimate models when there was an effect that did not vary within group; this is fixed.

----- update 11jan2001 --------------------------------------------------------

1. streg, dist(exponential) and ereg did not honor standard maximization options such as iterate(#), nolog, etc.; see help maximize. This is fixed.

2. streg, dist(weibull) ancillary(varname) did not show the coefficients of the ancillary equation even though the model was correctly estimated. This is fixed.

3. joinby now accepts filenames with embedded spaces, assuming such filenames are enclosed in double quotes.

----- update 08jan2001 --------------------------------------------------------

1. gphprint (a Stata 6 command that did something similar to print) would issue the error "logo() invalid" when translating to the Windows Metafile format or the Mac PICT format. This is fixed.

2. notes now allows the individual notes to include SMCL directives.

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

See help whatsnew6to7.


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