Stata 15 help for whatsnew13

Additions made to Stata during version 13

This file records the additions and fixes made to Stata during the 13.0 and 13.1 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 | | this file Stata 13.0 and 13.1 2013 to 2015 | | whatsnew12to13 Stata 13.0 new release 2013 | | whatsnew12 Stata 12.o 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 | | 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 whatsnew13to14.

-------- update 16dec2016 -----------------------------------------------------

1. (Mac) macOS Sierra 10.12.2 introduced a change that caused Stata's Do-file Editor to crash. This has been fixed.

-------- update 25jul2016 -----------------------------------------------------

1. Stata has been updated to account for the leap second that was added by the International Earth Rotation and Reference Systems Service (IERS) on June 30, 2015, and for the leap second that will be added by IERS on December 31, 2016. Stata's datetime/C UTC date format now recognizes 30jun2015 23:59:60 and 31dec2016 23:59:60 as valid times.

2. estat teffects did not correctly compute the standard errors for total and indirect effects between endogenous variables. This has been fixed.

-------- update 10mar2016 -----------------------------------------------------

1. The multilevel mixed-effects estimators gsem, meglm, melogit, meprobit, meologit, meoprobit, mepoisson, and menbreg have the following improvement and fix:

a. gsem, meglm, melogit, meprobit, meologit, meoprobit, mepoisson, and menbreg are now more likely to converge for models fit to data with large group sizes.

b. gsem, meglm, melogit, meprobit, meologit, meoprobit, mepoisson, and menbreg, when used with datasets that had many observations per group, could indicate that the model converged and report results that did not include the large groups in the computations. To determine whether your model was affected, you can use predict to obtain empirical Bayes mean estimates for the random effects and check for zero values in large groups. This has been fixed.

2. After the 12nov2015 update, xtlogit and xtprobit with option vce(robust) or option vce(cluster panelvar), when perfect predictors caused a reduction in the sample size, reported incorrect standard errors. This has been fixed.

-------- update 09dec2015 -----------------------------------------------------

1. fp generate, when qualifier if was specified with an expression containing quotation marks (""), incorrectly exited with error message "invalid expression". This has been fixed.

2. predict after gsem, meologit, or meoprobit, when depvar's outcomes were not contiguous values starting at 1 and when the requested prediction required empirical Bayes estimates, incorrectly exited with error message "could not compute empirical Bayes' means; missing values were returned by the evaluator". This has been fixed.

3. (Mac) The Do-file Editor, when executing a selection of lines and when the selected lines were preceded by lines that contained Unicode characters, executed the wrong lines. This has been fixed.

4. (Mac) In the Update dialog, clicking on the link for manual updates did not open the "Keeping Stata 13 up to date" webpage. This has been fixed.

-------- update 12nov2015 -----------------------------------------------------

1. collapse has the following fixes:

a. collapse with option percent and missing values in a by() variable used the wrong denominator to calculate the percentages. This has been fixed.

b. collapse with option percent and iweights used the unweighted total instead of the weighted total to calculate the percentages. This has been fixed.

c. collapse with option percent and aweights ignored the specified weights. This has been fixed.

2. contrast with the observation-weighted orthogonal polynomial operators pw(numlist). and qw(numlist). did not correctly interpret the numlist as the degree of the polynomial. When either pw. or qw. was specified, numlist was interpreted as the value of the factor variable. This has been fixed.

3. egen cut(varname), group(#), when # > 1000 and when the data were not previously sorted by varname, did not produce # groups. This has been fixed.

4. etregress with aweights incorrectly exited with an error. This has been fixed.

5. forecast solve with option simulate(residuals) returned a Mata error message when used to define an identity before estimates were defined or before a coefficient vector was added. This has been fixed.

6. Stata function regexm(s, re) could crash Stata when one or both arguments were binary strL. This has been fixed.

7. Stata function strproper(s) could crash Stata if the argument was binary strL. This has been fixed.

8. graph export, when used to export to EPS format a graph that had a font with glyphs of more than 2,000 points, could cause Stata to crash. This has been fixed.

9. gsem with intercepts or fixed predictors on some but not all latent variables would fail to converge, even for models that were otherwise statistically identified. Instead, the command reported the message "(not concave)" or "(backed up)" with unchanging log-likelihood values in the iteration log. This has been fixed.

10. The documentation for gsem states that time-series operators are allowed. However, when a time-series-operated depvar and multilevel latent variables were specified, gsem exited with an uninformative error message. This has been fixed.

11. icd9 and icd9p have the following fixes:

a. icd9 check and icd9p check, for a small number of category and subcategory codes, incorrectly indicated that the codes were not defined. This has been fixed. Note that no billable ICD-9-CM codes were affected.

b. icd9 generate and icd9p generate with option description returned a blank description for a small number of codes instead of indicating that these codes were further subdivided. This has been fixed.

c. icd9 generate with option description for the codes 176, 764, 765, V29, and V69 returned unofficial descriptions created by StataCorp that included the text "(Stata)". The description for these codes is now "*", which is the documented indicator that a code has been subdivided into billable codes.

d. icd9 lookup and icd9p lookup, when looking up a small number of category codes and subcategory codes without specifying a range, incorrectly returned note "(no matches found)". For example, typing icd9 lookup 040.4 would trigger this note, but icd9 lookup 040.4* would not. This has been fixed.

12. ivtobit with models having more than one endogenous variable specified is now more likely to converge.

13. margins, in the unusual case where xtlogit, re or xtlogit, pa was also specified with option noconstant and either option vce(bootstrap) or option vce(jackknife), incorrectly reported the error message "conformability error". This has been fixed.

14. The documentation for meologit and meoprobit states that they support time-series operators for depvar. However, when a time-series-operated depvar was specified, these commands exited with an uninformative error message. This has been fixed.

15. pctile newvar = exp, nquantiles(#), when # > 1000 and when the data were not previously sorted by exp, correctly computed the quantile values but did not store the values in the first #-1 observations of the dataset. This has been fixed.

16. predict with option latent after gsem, when used to fit a model with two or more latent variables that were each predicted by observed variables, incorrectly exited with error message "variable _cons not found". This has been fixed.

17. predict after gsem with multinomial outcomes incorrectly exited with an error message if the specified prediction involved the prediction of a latent variable. This has been fixed.

18. predict with stub* and option pr, after using ologit or oprobit to fit a model without covariates, incorrectly exited with error message "varname required". This has been fixed.

19. prtest and prtesti, when used for a two-sample test of proportions, reported the two-sided p-value of the test with the label Pr(|Z| < |z|). The label should have been Pr(|Z| > |z|). This has been fixed.

20. svy when used to fit a regression model to data from a multistage survey design that 1) did not have a sampling unit or stratum variable in the final stage and 2) had FPC variables in all stages could exit with error message "fpc for all observations within a stratum must be the same", depending on the sort order of the data. This has been fixed.

21. svy: tabulate, when value labels for a tabulated variable contained quotation marks, exited with error message "conformability error". This has been fixed.

22. teffects nnmatch, in the unusual case that fweights were specified with option vce(robust), produced standard errors that were too small. This has been fixed.

23. xtlogit and xtprobit with option vce(robust) or option vce(cluster panelvar), when perfect predictors caused a reduction in the sample size, incorrectly exited with error message "calculation of robust standard errors failed". This has been fixed.

24. Contrary to the documentation, xtpoisson, re with option normal did not accept time-series operators in depvar or indepvars. This has been fixed.

25. xtregar has the following fixes:

a. xtregar with option re, in the rare case of a negative autocorrelation coefficient, produced standard errors that were too large. This has been fixed.

b. xtregar with option fe, in the rare case of a negative autocorrelation coefficient, produced standard errors that were too large and a point estimate for the constant term that was biased toward 0. This has been fixed.

26. (Mac) In the Find dialog, selecting "Match whole word only" while searching in the Do-file Editor caused Stata to correctly find the search term in the current search but incorrectly save the setting. In subsequent searches, results only returned text that started with the requested search term rather than the whole word. This has been fixed.

27. (Mac) When a table with multiple rows was copied from the Results window as HTML, the HTML output would appear as a table with just one row. This has been fixed.

28. (Mac) Stata graphs could incorrectly render to the screen the first point of polygons, such as arrowheads. The first point was shifted from where it should have been drawn by one pixel. This has been fixed.

29. (Mac) Graph dialogs on Mac OS X 10.10, when combined with this specific sequence of steps -- 1) select a variable from a variable control; 2) switch to a tab that contains no active text edit fields; and 3) click on the OK button without making any other changes -- displayed a warning message and, in rare cases, caused Stata to crash. This has been fixed.

30. (Mac) view browse did not open https URI schemes or URLs that were enclosed in quotes. This has been fixed.

31. (Mac GUI) The System Integrity Protection feature of OS X 10.11 (El Capitan) prevents users from making modifications to the /usr/bin directory. Because Stata can no longer write to the /usr/bin directory, Stata now creates a symbolic link to the console version of Stata/MP or Stata/SE in the /usr/local/bin directory if the user selects "Install Terminal Utility..." from the Apple menu. Stata will create the /usr/local/bin directory if it does not exist.

-------- update 03jun2015 -----------------------------------------------------

1. ereturn display, ml display, and _coef_table, when reporting estimation results from a multiple-equation model that included an interaction term in more than one equation, would sometimes fail to recognize some base levels of interaction terms. In this case, base levels were labeled "(omitted)" instead of "(base)". The estimated coefficients and related statistics were correct; the error affected only the labels.

For example, suppose we have factors A and B, each with levels 1 and 2. Then the coefficient table from

. sureg (y1 = A##B) (y2 = A#B), allbase

would mislabel elements _b[y1:1.A#2.B] and _b[y1:2.A#1.B] as "(omitted)" instead of as "(base)". This has been fixed.

2. estat mfx, after asclogit or asmprobit with a value label assigned to varname specified in alternatives(), exited with an error if any strings in the value label were longer than 26 characters. This has been fixed.

3. Any estimation command that allowed factor variables would slow down when matsize was set larger than 1,000 even for models with relatively few parameters. This has been fixed.

4. export excel, when exporting variables with any calendar date format other than %td, incorrectly exported the variables as a daily date instead of respecting the weekly, monthly, quarterly, half-yearly, or yearly format. This has been fixed.

5. For variable names of the form d#, D#, e#, and E#, Stata's expression parser misinterpreted some factor-variable specifications such as 1.d1 and 1.e2 as numeric literals instead of the intended indicator variables. This has been fixed.

6. Functions normalden(x,m,s) and lnnormalden(x,m,s), when s was specified as a string value rather than as a numeric value representing the standard deviation, returned the specified string instead of exiting with an error message. This has been fixed.

7. gnbreg, nbreg, and tnbreg now report the pseudo-R-squared when option vce(robust), option vce(cluster clustvar), or pweights are specified.

8. graph, when using a .gph file that was created by a newer version of Stata that required a newer format .gph file and when the graph file name was quoted on the command line, simply declared the file to be not found. graph now issues the message that the format of the file is too new for the running version of Stata.

9. graph bar, when no variable or statistic was specified and when options over() and by() were combined (for example, graph bar, over(rep78) by(foreign)), incorrectly exited with an error. This has been fixed.

10. graph pie, when the number of slices exceeded the defined number of pstyles in a scheme, made all excess slices white and issued uninformative notes. graph pie now behaves like graph bar, graph twoway, and other graph commands by recycling the defined pstyles when the number of slices exceeds the number of defined pstyles. graph pie now also supports the pcycle() option to reduce the number of pstyles used before recycling.

11. import haver has the following fixes:

a. import haver, when aggregating a daily series to weekly, defined weeks as beginning on Mondays. Friday is now treated as the first day of the week.

b. import haver, when a series contained no data, loaded the series as a variable containing all missing values instead of dropping the series from the query. This has been fixed.

c. import haver, describe with option saving() always displayed the series meta-information in the Results window. The meta-information should only be displayed when suboption verbose is specified in saving(). This has been fixed.

12. irf graph with option ci#opts() did not change the rendition of the confidence interval for the #th statistic. This has been fixed.

13. ivprobit for models with more than one endogenous variable specified is now more likely to converge.

14. margins after me commands with one or more factor variables that have their base level fvset incorrectly exited with an error. This has been fixed.

15. mi impute intreg exited with a Mata error "st_store(): 3203 colvector required" when incomplete data contained only one censored observation. This has been fixed.

16. mi impute pmm, mi impute monotone, and mi impute chained now display a note whenever predictive mean matching imputation uses the default one nearest neighbor. This default is arbitrary. The optimal number of nearest neighbors varies from one application to another. Recent simulation studies demonstrated that using one nearest neighbor performed poorly in many of the considered scenarios. We suggest that you choose the number of neighbors appropriate for your data and specify it in option knn() when imputing variables using predictive mean matching.

17. minbound, when the minimum value of the function occurred at the specified left boundary, returned the function value evaluated close to, but not exactly at, the boundary. This has been fixed.

18. ml and Mata function moptimize(), with lf1 specified as the evaluator type, were passing the value 2 instead of 1 for todo while computing numerical second derivatives. This has been fixed.

19. mlexp ignored reporting options associated with factor variables. This has been fixed.

20. mvreg, when replaying results from a model fit using manova, did not display value labels for factor variables. This has been fixed.

21. pause would become stuck in an endless loop if you tried to start an interactive Mata session while Stata was already in a paused state. pause now returns an error message in this case.

22. predict with option pr() after etpoisson did not correctly compute the predicted probabilities. predict failed to integrate over the unobserved component to obtain the final results. This has been fixed.

23. predict with option lnsigma after teffects ra, when the specified prediction created one new variable instead of a group of variables, incorrectly exited with an error. This has been fixed.

24. set coeftabresults off is a setting used to make regress faster when run in tight loops, in simulations, and when bootstrapping. It is now faster still.

Other estimation commands are also affected, but you would rarely notice a difference.

25. regress, when a time-series-operated variable was omitted because of collinearity, reported that a variable named _delete was omitted instead of reporting the name of the omitted variable. This has been fixed.

26. sem has the following fixes:

a. sem, when fitting a model with no observed exogenous variables, with more than one latent variable, and with at least two coefficients constrained to 1 on paths from a latent variable to its observed measurements, incorrectly exited with an error message. This has been fixed.

b. sem, when option covariance() specified more than one covariance term and when starting values were specified for a subset of those covariance terms via the init() option, sometimes incorrectly exited with an error message. Whether an error was reported depended on the order of the terms within the covariance() option. For example, the option covariance(x1*x2 (x1*x3, init(.2))) would trigger this error. This has been fixed.

c. In the SEM Builder, canvas length and canvas width were not properly restricted to be between 1 inch and 20 inches. This has been fixed.

27. Stata matrix row and column names parsing logic, when presented with interactions where the order of factor variables was not the same in each term, did not properly keep track of the levels. For example,

. matrix colna b = 0.foreign#1.rep78 2.rep78#1.foreign

would result in colnames

. matrix colna b = 0.foreign#1.rep78 2.foreign#1.rep78

instead of the intended

. matrix colna b = 0.foreign#1.rep78 1.foreign#2.rep78

This has been fixed.

28. stcurve, when plotting curves for a model that included an interaction between a factor variable and a continuous variable and when option at() set the continuous variable to a value that was not observed in the data, incorrectly reported error message "# is not a valid level for factor variable varname". This has been fixed.

29. Time-series operators have the following fixes:

a. Seasonal difference time-series operator S# did not appropriately translate the zero-period difference (S0). Depending on the command with which it was specified, S0 was translated to S1 or resulted in a variable containing only 0 values. In all cases, typing S0.var is now equivalent to typing var; this translation is the same as that used for the other time-series operators.

b. Time-series operators, when used with a dataset with more than 1 billion observations and when the time series contained gaps, could cause Stata to crash or to return incorrect results. Whether this occurred or not depended on certain relationships between the location of the gap, the number of observations, and the operator used. This has been fixed.

30. twoway scatter, twoway line, twoway connected, twoway area, twoway bar, twoway spike, twoway dropline, twoway dot, twoway rarea, twoway rbar, twoway rspike, twoway rcap, twoway rcapsym, twoway rscatter, twoway rline, and twoway rconnected now support directly graphing the columns of a Mata matrix. For the sake of speed and space, the matrix data are passed directly to the graph, not brought into the Stata dataset.

The basic syntax is

twoway scatter matamatrix(matname) [namelist]

This feature is not officially documented, but the syntax is documented in help twoway mata.

31. xtgls issued a note when the number of panels was greater than the number of periods. The note incorrectly stated the opposite, that the number of periods was greater than the number of panels. This has been fixed.

32. xtivreg with option vce(bootstrap) or option vce(jackknife) exited with an error when an independent variable was included after the endogenous variable and instruments were specified in parentheses. For example, the placement of variable not_smsa in

. xtivreg ln_wage c.age##c.age (tenure = union south) not_smsa, vce(bootstrap)

would trigger this error. Variables can now be specified after the expression in parentheses.

33. Executing an operation using the https:// protocol when an internal content buffer was less than 100 bytes would cause Stata to exit with return code r(1). This typically happened only when the results to be returned by the operation were very short. This has been fixed.

34. (64-bit platforms) Mata functions fseek() and ftell(), when used to search for a position larger than 2GB in a file that was also larger than 2GB, returned a negative error code. This has been fixed.

35. (Mac) import excel, when importing long variable names, incorrectly truncated the variable names to 33 characters when it should have truncated them to 32 characters. This has been fixed.

36. (Mac) translate with translator smcl2pdf and when the destination file was specified either without an absolute path or without a path beginning ~/ resulted in an empty destination file. This has been fixed.

37. (Mac) xshell did not launch xterm when used on Mac OS X 10.8 or newer because XQuartz installs the xterm executable in a location other than where Stata is expecting. This has been fixed.

38. (Mac) The Do-file Editor has the following fixes:

a. Line numbers were not correctly resized when the Do-file Editor was zoomed in or zoomed out. This has been fixed.

b. Using the Find dialog and selecting "Replace all" might not replace all the text specified in "Find what:" if 1) the text to be replaced occurs more than once; 2) the length of the text specified in "Replace what:" is longer than the text that is being replaced; and 3) the text that is being replaced comes very close to the end of the document.

39. (Mac) Stata graphics no longer attempts to render a 0 line width path but will instead ignore the drawing command. In addition, Stata no longer sets the minimum line width to 1 pixel when rendering paths on the screen. Users with Mac OS X may notice that there is no longer any variation in how 0 width lines are rendered across output devices such as monitors and printers.

40. (Unix GUI) If a user changed themes on the computer, the Preferences dialog box continued to use a black font even if the theme requested another color. This has been fixed.

41. (Unix GUI) The SEM Builder's Estimation menu, when used to request exponentiated coefficients after fitting a generalized SEM, opened an incorrect dialog box. This has been fixed.

42. (Linux/Unix) In the Stata class system, if Stata was launched in batch mode, destructors were not invoked when Stata exited. This affected only users who had created classes that used destructors to manage computing resources not internal to Stata. This has been fixed.

-------- update 17apr2015 -----------------------------------------------------

1. Online help and the search index have been brought up to date for Stata Journal 15(1).

2. estat esize, when a noncentrality parameter could not be calculated, incorrectly reported a lower limit of 0 and upper limit of 1 for the confidence intervals of eta-squared and omega-squared. A missing lower limit and missing upper limit are now reported.

3. estat stdize after sem could, with nearly unidentified models, sometimes report the warning "variance matrix is nonsymmetric or highly singular" and report missing standard errors and test statistics. This has been fixed.

4. etpoisson, heckoprobit, heckprobit, nbreg, poisson, tnbreg, and tpoisson, when a time-series-operated variable was specified in option offset() or option exposure(), ignored the option and fit the model without an offset or an exposure. This has been fixed.

5. glm with family(binomial varname) did not exclude observations for which varname was missing. This typically resulted in a model that would not converge. This has been fixed.

6. graph, when using a .gph file that was created by a newer version of Stata that required a newer-format .gph file, simply declared the file specification to be invalid. graph now issues the message that the format of the file is too new for the running version of Stata.

7. gsem and meglm with family(binomial varname) did not exclude observations where varname was missing. This typically resulted in log-likelihood values labeled as "(not concave)", leading to the error "Hessian is not negative semidefinite". This has been fixed.

8. gsem, mecloglog, meglm, melogit, menbreg, meologit, meoprobit, mepoisson, meprobit, xtologit, and xtoprobit specified with constraints or factor variables would sometimes exit with a Mata trace log when the starting values were not feasible. Stata now produces an informative error message.

9. lv failed to ignore string-valued variables in the varlist when determining the estimation sample. This has been fixed.

10. margins has the following fixes:

a. Specifying any factor variable as asbalanced when that factor variable is interacted with other factor variables in models fit using gsem, mecloglog, meglm, melogit, menbreg, meologit, meoprobit, mepoisson, meprobit, xtologit, and xtoprobit produced asobserved rather than asbalanced marginal predictions. This has been fixed.

b. margins after sem incorrectly exited with an error message when endogenous covariates were specified in option at(). margins now fully supports endogenous covariates in option at().

c. margins with option subpop() and weighted contrast operator gw., hw., jw., pw., or qw. applied the overall-sample frequencies instead of the subpopulation-sample frequencies to the weighted contrasts. This has been fixed.

11. marginsplot now allows an optional using filename. The data in filename are assumed to come from the data saved by margins's saving() option. This feature is not documented with marginsplot; instead see margins saving.

12. melogit and meprobit with option binomial(varname) failed to exclude observations where varname was missing. This typically resulted in log-likelihood values labeled as "(not concave)", leading to the error "Hessian is not negative semidefinite". This has been fixed.

13. power twoproportions with option test(fisher) reported an incorrect value of actual alpha when the specified proportion in the control group was greater than the proportion in the experimental group. This has been fixed.

14. predict after gsem did not allow depvar to be abbreviated in option outcome(). This has been fixed.

15. rolling with option keep(varname) did not keep the correct values of varname unless the time variable specified in tsset was exactly equal to the observation number. This has been fixed.

16. streg with distribution(exponential) did not allow option from(). This has been fixed.

17. sts test with option fh() reported incorrect sum of ranks and test statistics except when fh(0 0) was specified. This has been fixed.

18. sts test with option strata() combined with either option peto or option fh() reported incorrect sum of ranks and test statistics. This has been fixed.

19. suest with option vce(cluster clustvar) failed to identify missing values in clustvar within the estimation sample. suest now exits with the error message "cluster variable clustvar is not allowed to be missing within the estimation sample".

20. testnl with option mtest and only a single testable constraint would produce an error instead of dropping untestable constraints and testing the remaining constraint. This has been fixed.

21. xtcloglog, xtlogit, xtpoisson, and xtprobit with options vce(robust) and noconstant specified in the default random-effects model produced a conformability error. This has been fixed.

-------- update 19dec2014 -----------------------------------------------------

1. Online help and the search index have been brought up to date for Stata Journal 14(4).

2. forecast solve using estimates from arima with no covariates exited with an error message. This has been fixed.

3. gsort, when arranging observations to be in descending order of a numeric variable, replaced the results in r() even though gsort is not an r-class command. gsort now preserves existing results in r().

4. mfp would not allow abbreviated variable names to be entered in the options zero() and catzero(). This has been fixed.

5. total with option over() and without prefix svy sometimes failed to estimate the variance matrix and reported the warning "variance matrix is nonsymmetric or highly singular". This most commonly occurred when over() was combined with vce(cluster clustvar). This has been fixed.

6. tsreport, when used with data that included repeated time values and had not been either xtset or tsset with a panel variable, returned an uninformative error message. This has been fixed.

-------- update 07nov2014 -----------------------------------------------------

1. collapse with option percent now reports percentages rather than proportions.

2. estat summarize after gsem ignored all arguments and options. This has been fixed.

3. graph bar and graph dot with option percent now report percentages rather than proportions.

4. ksmirnov, exact could report an incorrect p-value because of numerical precision for the two-sample exact test when the number of observations in at least one group was not equal to a power of 2. In such cases, the reported p-value was too small. This has been fixed.

5. levelsof, when used with a variable containing all missing values, returned an uninformative error message when it should have returned "no observations". This has been fixed.

6. testnl has the following fixes:

a. testnl, when used with survey data, incorrectly reported an F statistic by default instead of a chi-squared statistic. This has been fixed.

b. testnl, when used with survey data, ignored the specified option df(). This has been fixed.

-------- update 09oct2014 -----------------------------------------------------

1. collapse now supports the (percent) statistic specifying that the percentage of observations within each by() group be placed in the collapsed dataset.

2. graph bar and graph dot now handle categorical data better. Graphs of frequencies and percentages of observations within over() groups can be obtained using statistics (count) and (percent). For example,

graph bar (percent), over(grpvar)

creates bars measuring the percentage of observations in each level of grpvar.

graph bar, over(grpvar)

is an abbreviated syntax that does the same.

3. istdize has the following improvements:

a. istdize now stores results in r().

b. istdize now returns an error message

o if option by() is not specified and if the numbers of cases stored in variable casevar_s are not the same (excluding missing values) for all observations;

o if option by() is specified and if the numbers of cases stored in variable casevar_s are not the same (excluding missing values) for all observations in a group defined by the by() variables; or

o if option by() is specified and when the numbers of cases stored in variable casevar_s are missing for all observations in one or more groups defined by the by() variables.

4. predict after svy: glm now allows statistic response, the difference between the observed and fitted outcomes.

5. Online help and the search index have been brought up to date for Stata Journal 14(3).

6. clogit with an unusually large number of positive values of the outcome per group could produce a numeric overflow and cause Stata/MP to crash. This has been fixed.

7. contrast now works with estimation results that do not contain e(V). This means that margins with contrasts and option nose will no longer issue an error message for such estimation results.

8. creturn list and c(k) incorrectly reported a value that was one larger than the number of variables in the dataset whenever e(sample) was present in the current estimation results. This has been fixed.

9. The Data Editor has the following fixes:

a. In the Data Editor, editing a cell containing a $ would result in the text following the $ being dropped because the $ and following text were incorrectly treated as a global macro. This has been fixed.

b. The 06may2014 update modified the way the data view was scrolled when filters were added or removed in the Data Editor. This behavior change could make it difficult to locate the current cursor position after a filter was removed. The behavior has now been restored to the behavior that existed before the 06may2014 update.

10. estimates describe, when e(cmdline) contained quotes, exited with an error message. This has been fixed.

11. expoisson and exlogistic, in rare cases where the model conditioned on a variable with many distinct values, would produce so many enumerations that Stata could crash. This has been fixed.

12. After the 06may2014 update, expressions using ambiguous abbreviations of variable names resulted in error message "... not found" instead of "... ambiguous abbreviation". This has been fixed.

13. heckoprobit incorrectly posted options stdp and stdpsel to e(marginsok) instead of e(marginsnotok). This has been fixed. margins now issues an error message if these predictions are specified in option predict().

14. ivpoisson incorrectly posted option residuals to e(marginsok). This has been fixed. margins now issues an error message if this prediction is specified in option predict().

15. In the Java-Stata API Specification, com.stata.sfi.Data.getFormattedValue() did not display commas for Stata formats for which it should have displayed commas, such as %8.0gc. This has been fixed.

16. log close logname, where logname was a name that was greater than 32 characters in length, could crash Stata. This has been fixed.

17. lrtest unnecessarily required the dependent variables be specified in the same order for all fitted models. This has been fixed.

18. margins has the following fixes:

a. margins with option over() or within() ignored generate() expressions within option at(). This has been fixed.

b. margins with multiple at() settings and with set emptycells drop would incorrectly exit with the error message "conformability error". This has been fixed.

c. margins with option vce(unconditional) failed to restrict computation of scores to within the estimation sample. On rare occasions, out-of-sample scores could not be computed, resulting in error message

cannot compute vce(unconditional); predict after ... could not compute scores

This has been fixed.

d. margins with default vce(delta) and svy estimation results reported a different sample size when if e(sample) was specified if the survey weights variable contained zero values within the estimation sample, because zero-weighted observations were not counted. This has been fixed.

19. Mata function selectindex(v), when v was a vector with no elements, could cause Stata to crash. This has been fixed.

20. Mata function selectindex(v), when v was created using st_view(), incorrectly returned the index values of zero-valued vector elements. This has been fixed.

21. ml with option svy reported an unadjusted model F test instead of an adjusted model F test. This has been fixed.

22. ml display, ereturn display, and _coef_table, when a scalar was posted to e(b), reported the error message "point estimates required", even when the point estimates were also present in matrix e(b). This has been fixed.

23. power did not allow the specification of two-sample options such as n1(), n2(), and nratio() for user-defined power methods. This is now allowed if you also specify new option twosample with power. See power userwritten for more details.

24. predict after teffects nnmatch, atet biasadj() would incorrectly exit with error message "conformability error". This has been fixed.

25. pwcompare now works with estimation results that do not contain e(V). This means that margins with options pwcompare and nose will no longer issue an error message for such estimation results.

26. _rmcollright with a weight expression would incorrectly exit with error message "weights not allowed". This has been fixed.

27. sem did not allow option init() to be used within options means(), variance(), or covariance() when the constraint operator @ was also specified. This has been fixed.

28. serset sort and serset summarize, when there was a strL variable in the dataset, could cause Stata to crash. This has been fixed.

29. smooth now labels the generate() variable.

30. suest, when it encounters results from gmm or gsem, now reports a more informative error message.

31. svy: sem with options group() and method(ml) (the default) did not produce correct linearized standard errors unless the data were already sorted by the group() variable. The score values used in the linearized variance estimator were not put in the proper observations, resulting in different standard-error estimates depending on the sort order of the dataset. This has been fixed.

32. test with an expression using a factor variable with a level value greater than 32,767 would incorrectly exit with the error message "regressor ... not found". This has been fixed.

33. xtnbreg with an offset() variable containing extremely large values could exit with the unhelpful error message "matrix not found". This error message was caused by a numerical overflow caused by the offset() variable. Now xtnbreg will report the following error message: "could not find feasible starting values".

34. xtreg, fe would mistakenly identify independent variables with a constant lag difference as constant within panel. This affected only the value of the test statistic reported below the results table as "F test that all u_i=0". This has been fixed.

35. (Windows) Exporting a graph to a PDF file that contained text with a size of 0 would cause Stata to crash. This has been fixed.

36. (Windows) Creating thousands of graphs could eventually cause Stata to fail to render graphs on the screen. This has been fixed.

37. (Windows and Mac) Calling Stata in batch mode and passing a do-file without specifying command do would result in Stata assuming doedit as the command by default. Stata will now always assume do as the command in batch mode.

38. (Mac) Stata's code signature has been updated for compatibility with OS X 10.9.5 and newer.

39. (Mac) In the power dialog box, clicking on the button for the score test comparing a proportion with a reference value opened the wrong dialog box. This has been fixed.

40. (Mac and Unix console) Calling Stata in batch mode and passing a do-file without specifying command do would result in the log filename getting prepended with a quote. This has been fixed.

41. (Unix GUI) On some Linux distributions, the output from a command could appear twice in the Results window. This has been fixed.

-------- update 06aug2014 -----------------------------------------------------

1. The SEM Builder now allows you to provide a matrix or list of initial values on the Maximization tab of the SEM estimation options dialog.

2. contrast with option level() would incorrectly exit with the error message "option level() not allowed". This also affected margins with option level() and contrast operators. This has been fixed.

3. graph matrix with a varlist specification containing _all or * would include graphs of each variable against the binary temporary variable that identifies the sample. This has been fixed.

4. marginsplot with a single variable specified in option xdimension() ignored suboptions labels() and elabels(). This has been fixed.

5. predict after gsem and me would incorrectly return zero-valued empirical Bayes' mean estimates when intmethod(mvaghermite) was not a viable integration method. In this case, gsem and me will now report a note below the coefficient table acknowledging that the results were produced using nonadaptive quadrature, and predict will issue the error message "could not compute the empirical Bayes' means".

6. predict after gsem has been improved to better handle missing-value patterns in the family(gaussian) outcomes. Before this fix, structural paths between family(gaussian) outcomes would cause predict to incorrectly return missing values in observations where any one of the family(gaussian) outcomes were missing even if that outcome was not a predictor for the outcome of interest.

7. sem has the following fixes:

a. sem did not allow option init() to be used with terms containing the constraint operator @. For example, the following command would return an error message stating that @ is an invalid name:

. sem mpg <- (turn@a, init(.5)) (trunk@a, init(.5))

This has been fixed.

b. sem with option method(mlmv) and collinear variables would exit with a Mata trace log. This Mata bug has been fixed. sem will now report that the starting values are not feasible when it is unable to fit the saturated model in this situation.

8. statsby, in the rare case where one or more by() groups contained a huge number of observations (greater than 16,777,216), could result in a dataset with missing values for the last group, and it could possibly result in extra observations filled in with missing values. Furthermore, statistics for these large groups and those following may not be based on the intended observations from the original dataset. This has been fixed.

9. teffects psmatch with option caliper() would produce an error message for some covariate patterns when attempting to compute robust standard errors for the ATET estimator. This has been fixed.

-------- update 03jul2014 -----------------------------------------------------

1. margins always produced zero-valued marginal effects for time-series-operated variables when working with sem or gsem results. This has been fixed.

2. sem with vce(robust) or vce(cluster ...) and time-series operators on one or more variables iterated to convergence but exited with the error "time-series operators not allowed" instead of reporting the estimation results. This has been fixed.

3. (Mac) The 26jun2014 update introduced a bug where closing a Do-file Editor with unsaved changes would not prompt the user to save the changes and would put Stata into a state where it could not be exited. This has been fixed.

-------- update 26jun2014 -----------------------------------------------------

1. Online help and the search index have been brought up to date for Stata Journal 14(2).

2. The standard reporting options for estimation commands now include options noci and nopvalues. noci suppresses confidence intervals from being reported in the coefficient table. nopvalues suppresses p-values and their test statistics from being reported in the coefficient table.

3. bootstrap with option bca and when using an estimation command that included factor variables issued a warning about missing acceleration estimates and did not compute bias-corrected and accelerated (BCa) confidence intervals because of zero-valued constraints on base levels and omitted predictors. This has been fixed so that BCa confidence intervals are produced for the other estimates.

4. bootstrap and bsample with option idcluster() and a sample with more than 16,777,216 total observations failed to include all observations from some of the sampled clusters. This has been fixed.

5. forecast solve when obtaining forecast standard deviations or variances with option simulate() and when a specific prediction had not been specified using option predict() of forecast estimate did not produce correct standard deviations or variances for the estimators xtreg, fe, xtreg, re, xtpcse, or xtivreg. This has been fixed. Note that the forecasts themselves were correct.

6. gsem has improved parsing to handle elements within an interaction term that were specified with different variable ordering. For example,

. sysuse auto . gsem mpg <- 0.foreign#c.turn#c.trunk 1.foreign#c.trunk#c.turn

was ignoring the 1.foreign#c.trunk#c.turn element of the interaction term when it should have produced the same result as

. gsem mpg <- foreign#c.turn#c.trunk

This has been fixed.

7. import delimited will now completely ignore any binary zeros in the text-delimited file on disk. Binary zeros are not valid in text data. Upon completion, import delimited will now report the number of binary zero bytes that have been ignored, if any.

8. import excel with option firstrow imported the last character in a variable name incorrectly when the variable name was longer than 32 characters. This has been fixed.

9. jackknife, when option n() was supplied with an expression dependent upon r() results other than r(N), would exit with the error message "number of obs. ... evaluated to missing in full sample". This has been fixed.

10. logistic with option coeflegend reported the odds ratios. This behavior was misleading because the legend elements correspond to coefficients when used in expressions. logistic with option coeflegend now reports the estimated coefficient values.

11. margins has the following fixes:

a. margins did not allow certain varlist specifications within options dydx(), dyex(), eydx(), and eyex(). For example, with the following regression

. sysuse auto . regress mpg turn trunk i.foreign

margins would exit with the following error messages:

. margins, dydx(t*) 't*' not found in list of covariates

. margins, dydx(f*) 'f*' not found in list of covariates

. margins, dydx(i.foreign) i: operator invalid

These varlist specifications are now supported in margins.

b. margins ignored factor-level restrictions in the marginlist for factor variables that had a base preference specified in the margins command or the fvset command. This has been fixed.

12. marginsplot has the following improvements and fixes:

a. Axis labels have been improved when contrasts are present in the margins results.

b. Axis labels now show factor-variable values rather than value labels after margins with option nofvlabel.

c. marginsplot now checks for at() variables that are present as both a factor variable and a regular variable in the estimation results. For example,

. sysuse auto . regress mpg foreign rep78#foreign . margins, at(rep78=(3/5))

treats foreign as a regular variable in the main effect but as a factor variable in the interaction with rep78. marginsplot does not support this and will exit with an error identifying foreign as the offending variable.

The solution to this situation is to refit and specify the offending variable exclusively as a factor variable or exclusively as a regular variable. For the above example, this means specifying the linear regression as

. regress mpg i.foreign rep78#foreign

13. Mata function lnnormalden(x,sd) did not return a missing value in some cases where it was expected to, for instance, when x was missing or extremely large. This has been fixed.

14. odbc query and odbc describe issued the error message "Memory allocation error" when connecting to a MonetDB database. This has been fixed.

15. proportion with option over() reported missing proportion values for categories with no observations within the groups defined by over(). proportion will now report a zero value. The old behavior is not preserved under version control.

16. prtest varname with option by() failed to check that varname was binary for the two-sample test of proportions. If the specified variable contained values other than 0 or 1 and the mean of varname in each group was in the range of 0 to 1, prtest would produce incorrect proportions and tests of differences, but it did not produce an error message. This has been fixed.

17. pwcompare with option mcompare(tukey) no longer reports a warning about unbalanced factors. The method implemented for mcompare(tukey) is commonly known as the Tukey-Kramer method, which is an extension to Tukey's honest significant difference method that accounts for unbalanced factors.

18. pwcorr produced a table containing missing values when numerical overflow occurred. The command now produces a numerical overflow error message.

19. qreg has the following fixes:

a. qreg reported twice the sum of the absolute weighted deviations. This rescaling of the objective function did not affect the other results. The sum of the absolute deviations is now reported.

b. In rare cases in which some of the specified weights were very large and others were relatively small, numeric imprecision could accumulate and cause the algorithm to select a point near the optimum instead of the optimal point. Numeric stability of the qreg computations has been improved.

20. The SEM Builder has the following improvements:

a. The Contextual Toolbar no longer flashes when opening a path diagram file or when using the Undo or Redo buttons.

b. Opening a path diagram and using the Undo and Redo buttons are now much faster. The speed improvement is particularly noticeable on large or complex path diagrams.

c. The Builder window now becomes inoperative during Undo and Redo operations. Previously, it was possible to connect a path to or from a variable that would be removed by the operation, leading to warning messages.

21. sem has the following fixes:

a. sem with a fitted model that was not full rank in e(V) reported a model-versus-saturated chi-squared test that had an inflated p-value because the degrees of freedom for the test was inflated. sem now reports a note indicating this has occurred instead of reporting the test.

b. sem with method(mlmv) in the very rare situation where the saturated model converged, but was not full rank, yielded a log likelihood greater than the fitted log likelihood. In this case, sem reported a zero-valued chi-squared statistic instead of the negative difference in the chi-squared values. sem now reports a note indicating that the saturated model is not full rank instead of reporting the chi-squared test.

22. stcox, stcrreg, and streg with option coeflegend reported the hazard ratios. This behavior was misleading because the legend elements correspond to coefficients when used in expressions. These commands with the option coeflegend now report the estimated coefficient values.

23. sts graph with option risktable() and suboptions rowtitle() and group() did not respect those suboptions when option by() was also specified with multiple variables. This has been fixed.

24. sunflower with option name() and without option binar() unintentionally changed the contents of the default graph named Graph. This has been fixed.

25. svy estimation with svyset poststratification did not account for observations excluded from the estimation sample when running the estimation command. This could result in positively or negatively biased results depending on which poststratum observations were excluded and the value of the corresponding poststratum weights. In either case, standard errors were inflated. svy now reports a note when this occurs and reruns the estimation command with the poststratification adjustment accounting for the new estimation sample.

The following commands did not have this problem: mean, proportion, ratio, total, svy: tabulate oneway, and svy: tabulate twoway.

26. svy bootstrap, svy brr, svy jackknife, and svy sdr incorrectly ignored option eform. This has been fixed.

27. testparm produces a more informative error message when varlist does not identify any testable coefficients.

28. Stata could crash if more than two gigabytes of data were copied to the clipboard from the Data Editor and then subsequently pasted back to the Data Editor. Now the Data Editor limits the amount of data that can be copied to the clipboard to two gigabytes.

29. In Stata/MP with more than 50 processors or cores and in rare cases that depended on the number of observations being sorted, sort could stop with error message "failed to select pivot values". This has been fixed.

30. (Mac) The Do-file Editor's Find bar would sometimes not appear when activated, and an empty line would appear in its place. This has been fixed.

31. (Mac) When using a do-file to create graphs, closing nontabbed Graph windows that were in fullscreen mode left empty spaces in Mission Control. This has been fixed.

32. (Mac) Stata would open in the root directory, "/", when launched directly from the Finder on Mac OS X 10.10 (Yosemite). This has been fixed.

33. (Mac) Stata for Mac (GUI) would set the graphics setting to off when launched from the command line with the -q flag. Setting graphics off prevents graphs from rendering to the screen. The -q flag no longer sets the graphics setting to off.

34. (Mac) The console version of Stata would search only in /Applications/Stata for the Stata directory. It now searches for the Stata directory in /Applications/Stata13.1, /Applications/Stata13.0, /Applications/Stata13, and /Applications/Stata, in that order.

35. (Unix GUI) The import excel dialog box, when lower was selected from the "Variable case" combo box, did not issue the command with option case(lower). This has been fixed.

-------- update 06may2014 -----------------------------------------------------

1. estimates for with estat ic incorrectly failed with an error message. estimates for now works with estat ic.

2. etregress with option vce(robust) reported a likelihood-ratio test for the test of independent equations (null is no endogeneity). This test did not use the robust VCE and thus would over reject the alternative of endogenous treatment in the usual case where the robust SEs are larger than the standard SEs. So the test was conservative with regard to endogeneity. etregress now reports a robust Wald test of independent equations.

3. Function rnormal(m,s) returned 0 in the unusual case where s was 0, even when m was specified to be nonzero. rnormal() now returns m in such cases.

4. gsem, when the observed endogenous variable was part of a predictor term using factor-variables notation, failed to catch unsupported nonrecursive systems. This has been fixed.

For example, the following model specifications will now cause a syntax error because y1 and y2 form a nonrecursive system that is not supported by gsem.

gsem (y1 <- y2 x1, logit) (y2 <- i.y1 x2)

gsem (y1 <- y2 x1, ologit) (y2 <- i.y1 x2)

gsem (y1 <- y2 x1) (y2 <- c.y1##c.x2)

5. import excel refused to open some Excel files created by Apple Numbers for Mac. This has been fixed.

6. Functions colnumb(), el(), and rownumb(), when used in a scalar expression, reported an error message if the supplied matrix name was an unambiguous abbreviation of 2 or more variables, even when a matrix of that exact name existed. This has been fixed.

7. merge with option keepusing(), where variables in the using dataset shared names with variables in the master dataset and some were omitted from the keepusing() varlist, sometimes returned nonmissing values for unmatched observations when it should have returned missing values. This has been fixed.

8. odbc sqlfile stopped execution of the SQL file if an SQL command in the file did not return results. This has been fixed.

9. qreg with probability weights (pweights) or importance weights ( iweights) did not produce correct standard errors. This has been fixed.

10. reshape, when the j variable value label contained only labels for extended missing values, produced error code r(198). This has been fixed.

11. sts graph did not respect option censopts() when option risktable was also specified. This has been fixed.

12. teffects psmatch and teffects nnmatch with option nneighbors(#) did not enforce that # matches were found for each observation. Estimation continued so long as one match could be found. teffects now strictly enforces that # matches are found for each observation and provides diagnostics to help you identify observations with too few matches.

13. (Mac) The Open dialog for the Viewer now allows .log files to be selected.

14. (Mac) The 31mar2014 update introduced a bug where printing the Results or Viewer window would print only the first page. This has been fixed.

15. (Mac) Class files were not enabled in the Open dialog for the Do-file Editor. In rare cases, a Do-file Editor would not be correctly sized in its window when there were multiple tabs in the window. These problems have been fixed.

16. (Mac) The Variables window was not updating after a Stata command was executed from AppleScript. This has been fixed.

17. (Mac) Opening log and SMCL files from the Finder now opens the files in their own tab in the Viewer.

-------- update 24apr2014 -----------------------------------------------------

1. estat eqtest after sem with prefix command svy now uses survey-adjusted F tests for the equation-level coefficients when design degrees of freedom is present. Option nosvyadjust was added to estat eqtest to provide unadjusted F tests.

2. expoisson using fweights with more than one covariate gave the wrong coefficient estimate for all covariates beyond the first. This has been fixed.

3. ivregress has the following fixes:

a. ivregress, when used with a certain configuration of factor variables, could produce an error message. This happened when the terms of a factorial interaction were split between the exogenous regressors and the instrumental-variables list. This has been fixed.

b. ivregress, in the unusual cases where each exogenous regressor was perfectly collinear with the endogenous regressor or constant-valued, produced an error message. ivregress now drops these variables and performs the estimation.

4. marginsplot, when used correctly with the results from margins after a constant-only model, reported error message "variable _cons not found". This has been fixed.

5. predict after gsem and meglm now produces a more informative error message when empirical Bayes estimates are to be computed but the data have changed since estimation.

6. sem has the following fixes:

a. sem with option group() specified with prefix command bootstrap, jackknife, or svy incorrectly exited with error message "label already defined" when the group variable had value labels attached. This has been fixed.

b. sem exited with a Mata conformability trace error message when specified with exactly one endogenous latent variable with no predictors. This has been fixed.

7. svy sdr with option subpop() specified with command mean, proportion, ratio, or total and option over() incorrectly reported an invalid syntax error message after the replications were complete. This has been fixed.

8. _coef_table with option dfmatrix() required a nonmissing value stored in e(df_r); otherwise, it failed to use t in the column labels for the test statistic and p-values. This has been fixed.

-------- update 31mar2014 -----------------------------------------------------

1. dtaversion and dtaverify are new, undocumented commands in Stata. "Undocumented" means the commands are not documented in the manuals but are documented online.

dtaversion reports an internal .dta file-format version number and the corresponding Stata version number. The command is probably of little interest outside StataCorp.

dtaverify confirms the consistency of a .dta file; that is, it checks whether the .dta file is corrupt. For most of us, this command is useful when we want to determine the file is not damaged from an accident, say, coffee spilling on the computer. Programmers writing programs to write Stata .dta files will find dtaverify even more useful because they can use it to debug their programs. dtaverify provides a lot of useful information when the .dta file has errors.

2. Online help and the search index have been brought up to date for Stata Journal 14(1).

3. Calling exit, clear STATA from a do-file more than one execution level deep would cause Stata to hang. This has been fixed.

4. forecast exited with an error message when an arima model had covariates. This has been fixed.

5. fp could produce an error message when used with an estimation command that created new variables. This has been fixed.

6. Function invnFtail(df1,df2,np,p) could return incorrect results when p was very small. This has been fixed.

7. Function regexr() was much slower in Stata 13 compared with Stata 12. The speed of the function in Stata 13 is now improved to be comparable with the speed in Stata 12 on fixed-length strings. The speed is also improved on the new long strings.

8. glm with family(binomial) and nonzero fractional responses would incorrectly exit with the error message "no observations", r(2000), and a note that the outcome does not vary. This has been fixed.

9. graph twoway with multiple labels specified in option tlabel() produced bolding effects on the labels. This has been fixed.

10. gsem has the following fixes:

a. gsem failed to report a syntax error when weights were specified. This has been fixed.

b. gsem with certain constraints could try to fit a Gaussian outcome variable with a starting value of zero for an error variance, causing gsem to exit with the error message "initial values are not feasible". This has been fixed.

11. import excel failed to open some Excel files created by Apple Numbers for Mac. This has been fixed.

12. meglm ignored covariance constraints specified with the constraints() option. This has been fixed.

13. merge could issue the error message ".dta file corrupt", r(688), when merging datasets that both contained strL variables of the same name, and in a different, but special, order. The error message was unlikely and incorrect; none of the .dta files were corrupt. The invalid error message produced by merge has been fixed.

14. mi impute logit, mi impute ologit, and mi impute mlogit now issue a more informative error message when any of the unsupported maximize_options are used in combination with the augment option. The documentation now explicitly lists all such options.

15. mi impute intreg, bootstrap ignored all censored observations when obtaining a bootstrap sample during the estimation step of the imputation procedure. The problem would also arise when the intreg method was used within mi impute chained and the bootstrap option was specified either as a global option -- mi impute chained (intreg), bootstrap -- or as a method-specific option of the intreg method -- mi impute chained (intreg, bootstrap). This has been fixed.

16. mlogit now reports the number of completely determined observations in a footnote after the coefficient table.

17. oprobit with fweights neglected to count the number of observations with completely determined outcomes. This has been fixed.

18. predict after gsem with poisson or binomial outcomes and with missing values within the estimation sample exited with the error message "numerical overflow occurred". This has been fixed.

19. sem has the following fixes:

a. sem with more than 40 model variables and more than just a few groups would sometimes incorrectly report the error message "insufficient memory" even when the model parameters would fit within a coefficient vector with c(matsize) elements. This has been fixed.

b. sem with method(mlmv) and pweights was producing invalid and highly inflated standard errors. The inflation factor was on the order of the average of the sampling weights. This has been fixed.

20. The SEM Builder has the following fixes:

a. When performing group analysis, you could save a path diagram that could not be restored or that led to Builder errors when restored and edited. This problem occurred only if the final estimation before saving the path diagram was run while the Group control in the Toolbar was not set to the first group, and even then, only rarely. This has been fixed.

b. If a Builder file was saved after estimation with method mlmv and you requested that required covariances among observed variables not be added to the diagram, those covariances would be shown on the path diagram when the file was later used. This has been fixed.

c. Opening path diagrams when the diagram's filename contained special characters could put the Builder in a bad state. One common symptom was loss of the Contextual toolbar. This has been fixed.

d. The SEM Builder no longer allows you to create objects that are of 0 width or 0 height.

21. svy reported negative F statistics instead of reporting a missing value for fitted models where the design degrees of freedom was smaller than the number of model coefficients. This has been fixed.

22. svy brr with sem would produce an unintended error message "operator invalid" after looping through the replicates. This has been fixed.

23. teffects has the following fixes:

a. The nearest-neighbor algorithm used by teffects nnmatch and teffects psmatch failed to check for a user-initiated execution break. This has been fixed.

b. teffects overlap overwrote the estimation result e(cmdline). This has been fixed.

c. teffects produced an error message when control() was specified with a value label that had spaces. This has been fixed.

24. (Windows) Function acosh(x) could return incorrect results when x was very large. This has been fixed.

25. (Windows) Function asinh(x) could return incorrect results when |x| was very large. This has been fixed.

26. (Mac) You may now specify the ODBC driver manager by using set odbcmgr {iodbc|unixodbc}.

27. (Mac) The Do-file Editor has the following fixes:

a. (Mac) Multiple uses of Find could cause a crash when using the keyboard to navigate or dismiss the Find control. This has been fixed.

b. (Mac) Stata help files could not be opened in the Do-file Editor. This has been fixed.

c. (Mac) The print dialog for the Do-file Editor would not show an up-to-date preview of what would be printed based on the print line number preference. This has been fixed.

28. (Mac) The search-has-wrapped status window no longer appears when the cursor has been moved past the last found item.

-------- update 15jan2014 -----------------------------------------------------

1. Online help and the search index have been brought up to date for Stata Journal 13(4).

2. forecast now solves models more quickly. Forecast models that include estimation results from arima will show significantly faster solve times, and nearly all forecast models will show at least slight speed improvements.

3. Commands that utilize substitutable expressions (gmm, nl, nlsur, and mlexp) now benefit from the multiple-core features of Stata/MP and exhibit nearly perfect parallelization.

4. generate and replace, when the right-hand-side expression included references to elements of Stata matrices, did not benefit from parallelization. These commands now exhibit nearly perfect parallelization when the matrix subscripts are specified as numeric scalars.

5. asmprobit failed to display the coefficient table when there were no case-specific variables and no constant terms in the model. This has been fixed.

6. estat recovariance after mixed failed to report the covariance of the factor variable varname for a random-effects equation specified with R.varname and option covariance(exchangeable). This has been fixed.

7. estat wcorrelation after mixed, when models included 1) multiple random-effects specifications with the same level variable or 2) a random-effects equation specified with R.varname and option covariance(exchangeable), produced incorrect results. The affected results were the within-cluster correlation and covariance matrices, the variance-covariance matrix of random effects r(G), and the model-based design matrix for random effects r(Z). This has been fixed.

8. estimates table after sem with option group() reported "(empty)" for all intercepts, means, and variance components. This has been fixed.

9. export delimited failed to respect double quotes embedded in string data. export delimited now correctly respects double quotes embedded in string data by exporting two double quotes for each real double quote.

10. export excel would not execute quietly if you specified option replace and a path to a file that contained a space. This has been fixed.

11. fcast compute after vec has the following fixes:

a. fcast compute after vec failed to clear the global macro S_vecfcast_cr after execution and thus would accumulate a long string in this macro when called repeatedly. When the resulting string exceeded the maximum allowed length, the command would produce error message "macro substitution results in line that is too long". This has been fixed.

b. On error, fcast compute after vec dropped all variables created by previous calls to fcast compute. This has been fixed.

12. forecast solve exited with an error message when the forecast model included an estimation result in which equation names were specified (such as from reg3). This has been fixed.

13. gmm exited with an error message if the user supplied analytic derivatives and specified option nocommonesample but did not specify option xtinstruments(), even if the model could be fit. This has been fixed.

14. gsem has the following fixes:

a. gsem failed to report a variance component in the coefficient table if the preceding row was the base outcome from a multinomial response variable. This has been fixed.

b. gsem with multiple observed endogenous variables each having a path from an interaction containing a latent variable was not applying the automatic anchoring logic correctly, causing coefficients on those paths to be constrained to 1. This has been fixed.

c. gsem was not using the variables specified in the censoring suboptions allowed in family(gaussian) when computing the signed estimation results. This has been fixed.

15. import delimited has the following new features and fixes:

a. A new option has been added to import delimited. charset("charset") sets the character set used for importing ASCII text data. By default, latin1 (ISO-8859-1 encoding) will be used; however, mac (Mac OS Roman encoding) may be specified where appropriate.

b. import delimited ignored option rowrange() when specified with a row range greater than 10,000,000. This has been fixed.

c. The dialog box for import delimited failed to use compound double quotes with custom delimiters that contained actual quotes, resulting in an error in the generated command. This has been fixed.

16. ivregress with options first and vce(robust) could produce an error message when the endogenous variable list contained a factor variable that included an omitted category. This has been fixed.

17. margins has the following fixes:

a. margins no longer allows option coeflegend if the post option is not also specified.

b. margins after gsem with multinomial outcomes reported all margins as "(not estimable)". This has been fixed.

18. Matrix Kronecker products with time-series operators or factor-variable notation in the row or column names were not yielding the expected equation names in the resulting matrix. This has been fixed.

19. mean, proportion, ratio, and total with options vce(bootstrap, bca) and over() would incorrectly report error message "option sovar() not allowed". This has been fixed.

20. mi estimate has the following fixes:

a. mi estimate, when used with mean, proportion, ratio, or total, could produce missing multiple-imputation standard errors when some of the imputation-specific standard errors were legitimate zeros. For example, the mean command reports a standard error of zero for a constant variable. mi estimate now reports the estimates instead of missing values.

b. mi estimate, when used with mean, proportion, ratio, or total, now produces missing multiple-imputation standard errors for any parameter that was estimated using a sample of only one observation in any of the imputations, that is, for any parameter for which the corresponding element of the e(_N) matrix is 1 in any of the imputations.

21. mi impute chained has the following fixes:

a. mi impute chained could produce error message "missing imputed values are produced" during the initial conditional imputation when option include() contained expressions of imputation variables and the names of those variables happened to be abbreviations of other variables in the dataset. This has been fixed.

b. mi impute chained mistakenly produced error message "missing imputed values are produced" when option noimputed was specified and option include() contained a categorical or binary imputation variable with a factor specification other than i.varname. This has been fixed.

22. During conditional imputation, mi impute chained and mi impute monotone failed to check the order in which the conditional and conditioning variables were imputed when the conditional variable contained the same number of missing observations as one of the conditioning variables. The commands now produce an error message if a conditional variable is listed before its conditioning variables in the specification of an imputation model.

23. mi reshape has the following fixes:

a. mi reshape allowed @ in the specification of stubs when it should not have. If @ is specified, the proper error message is now produced.

b. All variables corresponding to the same stub must be registered of the same mi type: imputed, passive, or regular. mi reshape did not return the proper error message if these variables were of mixed mi types. This has been fixed.

24. nlcom and lincom, when estimating an expression involving a comma (like _b[/cov(e.x,e.y)]), produced an error message. This has been fixed.

25. nlogit could drop too many alternative-specific variables if one alternative-specific variable was collinear with a variable specified in the last alternative varlist after it was interacted with indicators of the alternatives. This has been fixed.

26. power twomeans incorrectly defined the effect size (delta) as the ratio of the difference between the experimental-group mean and the control-group mean to the standard error of the difference between two sample means. This led to the definition of an effect size that involves sample sizes. The effect size is now defined to be the difference between the experimental-group mean and the control-group mean. This change does not affect any previously obtained estimates of sample size or power, only the reported and computed effect sizes.

27. sem has improved starting values logic for models that use latent variables to model error covariances between observed endogenous variables.

28. sqreg computations could fail in qreg when computing robust standard errors unnecessarily. sqreg bootstraps the standard errors. This has been fixed.

29. stptime and strate, when the jackknife method was used, could report jackknife confidence intervals that did not include the estimated rate for survival datasets that contained no failures, only one failure, or only one cluster with failures. They now report missing confidence intervals and a note with explanation in these situations.

30. svy jackknife with svyset data that did not have a specified pweight variable always reported error message "insufficient observations to compute jackknife standard errors no results will be saved". This has been fixed.

31. svy tabulate, when used with a string variable, reported a temporary variable name in the output instead of the specified variable. This has been fixed.

32. teffects psmatch with options atet, vce(iid), and caliper() unnecessarily required valid matches for control subjects and would exit with an error if this requirement was not met. With these three options, teffects psmatch would unnecessarily exit with an error when it could not find a treated observation that was within caliper() distance of a control observation. This has been fixed.

33. (Windows) As of the 30oct2013 update, unconnected lines such as grid lines and axes now use butt line caps rather than round line caps. With this change, the behavior for Windows is the same as Mac and Unix platforms. The new behavior matches the results produced by graph export when producing EPS, PS, and PDF files.

34. (Windows) Pasting multiple commands into Stata's Command window and then minimizing Stata before the commands completed could result in one character being removed from a subsequent command, resulting in an error message. This has been fixed.

35. (Mac) Graphs could not be exported to EPS or PS while being edited in the Graph Editor. This has been fixed.

36. (Mac) Stata could crash if multiple graphs were drawn while the Do-file Editor was in full-screen mode. This has been fixed.

37. (Mac) Generating a complex graph and immediately exporting the graph to a PDF file could include visual artifacts in the PDF file. This has been fixed.

38. (Mac) When users tried saving a recording in the Graph Editor, Stata would crash if the "Save graph recording as:" edit field was empty and button Browse... was clicked on. This has been fixed.

39. (Mac) Saved preference max_memory would not be read correctly if the setting was large. This has been fixed.

40. (Mac) Selecting Edit > Undo after editing text in the Data Editor and applying the change would cause Stata to crash. This has been fixed.

-------- update 30oct2013 (Stata version 13.1) --------------------------------

1. power has three new methods for power and sample-size analysis of ANOVA models.

power oneway performs power and sample-size analysis for one-way ANOVA models. It computes sample size, power, or effect size given other study parameters. It provides computations for unbalanced designs and for mean contrasts. It also supports multiple specifications of input parameters. For example, to compute power or sample size, you can either specify group means directly or specify their between-group variability.

power twoway performs power and sample-size analysis for two-way ANOVA models. It computes sample size, power, or effect size given other study parameters and supports unbalanced designs. It also supports multiple specifications of input parameters. For example, to compute power or sample size, you can either specify cell means directly or specify the variance of the tested effect.

power repeated performs power and sample-size analysis for one-way and two-way repeated-measures ANOVA models. It computes sample size, power, or effect size given other study parameters and supports unbalanced designs. It also supports multiple specifications of input parameters. For example, to compute power or sample size, you can either specify cell means directly or specify the variance of the tested effect.

As in all other power methods, oneway, twoway, and repeated allow you to specify multiple values of parameters and automatically produce results in a table or on a graph.

2. power oneproportion and power twoproportions have new option continuity, which applies a continuity correction to the normal approximation of the discrete distribution.

3. power now provides facilities for users to add their own methods to power. See power userwritten for details.

4. Six noncentral chi-squared functions are added or improved. The following three functions are added:

New functions Description ---------------------------------------------------------------- nchi2den(df,np,x) probability density of noncentral chi-squared distribution nchi2tail(df,np,x) right-tailed noncentral chi-squared distribution invnchi2tail(df,np,p) inverse of right-tailed noncentral chi-squared distribution ----------------------------------------------------------------

The following three functions are improved (they now support broader domains for degrees of freedom and noncentrality parameter):

Improved functions Description ---------------------------------------------------------------- nchi2(df,np,x) cumulative noncentral chi-squared distribution invnchi2(df,np,p) inverse cumulative noncentral chi-squared distribution npnchi2(df,x,p) noncentrality parameter of noncentral chi-squared distribution ----------------------------------------------------------------

5. gsem's family(gaussian) option now has suboptions for censoring when link(identity) is used. See [SEM] gsem family-and-link options.

6. irf now works after arima and arfima.

7. New command estat aroots checks the eigenvalue stability condition after estimating the parameters of a model with time-dependent disturbances using arima.

8. New command estat acplot estimates autocorrelation and autocovariance functions after estimating the parameters of a model with time-dependent disturbances using arima or arfima.

9. psdensity now works after arima with multiplicative autoregressive moving-average (ARMA) terms.

10. putexcel has new option keepcellformat, which preserves a cell's style and format when putexcel writes a value to a cell.

11. mecloglog, melogit, and meprobit have new option asis for retaining perfect predictors. The new default behavior is to detect and omit perfect predictors. The old default behavior of retaining perfect predictors is preserved under version control.

12. bsample and bootstrap with option idcluster() will now produce a unique identifier for each resampled cluster even when option strata() is specified. The old behavior of indexing resampled clusters starting with 1 from within each stratum is preserved under version control.

13. export excel did not export a variable formatted as %d as a date. This has been fixed.

14. forecast solve exited with an error message if you attempted to forecast a panel-data model and specified suboption saving() of option simulate. This has been fixed.

15. merge and append, if an error occurs, are supposed to leave the original data in memory. If an error occurred, however, they could leave the original data with extra observations at the end. This has been fixed.

16. predict with option variance or option correlation produced incorrect results after mgarch vcc. This has been fixed.

17. Function subinword() could match the ending of a word rather than matching an entire word. This has been fixed.

18. Pasting data into the Data Editor did not update the internal list of sorted variables. Now if you paste data into one of the variables on which the data was sorted, the list is updated.

19. Strings passed from Stata to the Java Runtime Environment containing characters encoded with ASCII values between 128 and 255 might not get passed through correctly. This problem was originally discovered when writing data to an Office Open XML (.docx) file. This has been fixed.

20. (Windows) update failed to install from a local directory when Stata was run from a shared network. This has been fixed.

21. (Windows) Pressing the Break button when executing code with nobreak would cause Stata to issue a message box stating "Break disallowed at this time". Though informative, this behavior prevented programmers from coding specialized break logic. Stata for Windows will now have the same behavior as Stata on other platforms and simply ignore break when code is executed with nobreak.

22. (Mac) The Project Manager now opens SMCL and help files in the Viewer window instead of the Do-file Editor.

23. (Mac) The default width for the Viewer window was less than 80 columns. This has been fixed.

24. (Mac) Dragging and dropping text to the Do-file Editor or Command window did not result in any text inserted. This has been fixed.

25. (Mac) OS X 10.9 (Mavericks) or Java on Mavericks has a bug that could lead to a crash in Stata when Stata accessed Java in the Project Manager, the GUI for the power command, the import delimited command, https or ftp URLs, or the Mata _docx*() functions. This update works around the bug to avoid the possibility of such a crash.

26. (Mac) OS X 10.9 (Mavericks) introduced a change that caused Stata to always start in the root directory "/" instead of the working directory from Stata's previous session. This would only affect Stata if it was launched by double-clicking on it in the Finder, not by double-clicking on a do-file or dataset from the Finder. This has been fixed.

27. (Unix) Double-clicking on a .smcl file from a project in the Project Manager now opens the file in a Viewer.

-------- update 07oct2013 -----------------------------------------------------

1. The logic for detecting perfect predictors built into logit and probit has been added to the following estimation commands. Option asis has also been added.

glm with family binomial xtgee with family binomial xtcloglog, pa xtcloglog, re xtlogit, pa xtlogit, re xtprobit, pa xtprobit, re

2. contrast with survey estimation results posted the unadjusted F statistics and p-values in r(F) and r(p), respectively, even when option nosvyadjust was not specified. This has been fixed.

contrast with option post will continue to post the unadjusted values to e(F) and e(p); they are used to recompute the reported survey-adjusted values on replay.

3. correlate with time-series operators and option means failed to report all the variable names correctly in the means table. This has been fixed.

4. gsem with time-series operators on variables interacted with multilevel latent variables (random covariates) would cause error "not sorted". This has been fixed.

5. gsem and all me commands have the following fixes:

a. gsem and meglm with options dnumerical and vce(robust) or vce(cluster ...) reported a Mata conformability trace error when any kind of constraint was specified or implied. This has been fixed.

b. gsem and meglm will now attempt to fit the specified model even if the fixed-only model fails to converge.

c. gsem and all me commands with option intmethod(mvaghermite) fitting multilevel models with one or more groups that yield a large negative log likelihood ignored missing values when computing and using the adaptive parameters, yielding an incorrectly fit model. This has been fixed.

6. hausman has the following fixes:

a. hausman has improved factor-variables support in its table.

b. hausman with option equations() and interactions present in the specified equations exited with error "operator invalid". This has been fixed.

7. import delimited would exit with an error if the file to be imported contained two variables with the same name. Now import delimited will import the data, renaming any duplicate variables to v#, where # represents the column's position.

8. import excel used an excessively large amount of temporary memory to import string variables. This has been fixed.

9. margins, contrast, and pwcompare, after gsem with some but not all ordinal depvars, incorrectly flagged some margins, contrasts, and pairwise comparisons as "(not estimable)". This has been fixed.

10. margins has the following fixes:

a. margins now has better error messages when a name specified in marginslist does not appear among the list of covariates in the current estimation results.

b. margins with gsem results was mistakenly treating multilevel latent variables as intercepts, which caused it to always fail the estimable functions test. This has been fixed.

c. margins with option over() and expression generate() in option at() would cause a syntax error. This has been fixed.

d. margins was not recognizing a named equation as part of option equation() within a predict() or expression() specification. When this caused a problem, it reported "(not estimable)" for margins that actually were estimable. User-written estimation commands that set e(marginsprop) to contain addcons and allow an equation name to be specified in option equation() of predict were the only commands affected by this.

11. matrix list and matlist misaligned row and column headings that ended in #, *, @, or | but were not part of a multiple-line factor-variables interaction specification. This has been fixed.

12. mi impute mvn, when the check for collinear independent variables failed, produced an error with a Mata traceback log. It now produces an appropriate error message.

13. ml has the following fixes:

a. ml with option subpop() reported an incorrect sample size when some of the variables contained missing values outside the subpopulation sample. It was also possible for an entire sampling unit to be dropped because of these missing values, yielding subpopulation variance estimates based on an incorrect sample size. This has been fixed.

b. Estimation commands using ml to fit models with empty cells in interaction terms when set emptycells drop was in effect exited with a Mata conformability error. This has been fixed.

c. ml subroutines mlvecsum, mlmatsum, and mlmatbysum were not working properly when set emptycells drop was in effect; they failed to ignore empty cells in interaction terms. This has been fixed.

14. nestreg with prefix svy and svy options exited with a syntax error when the prefixed estimation command contained parenthesis-bound variables. This has been fixed.

15. odbc load crashed when using the Postgres Unicode ODBC driver to load strings longer than 280. This has been fixed.

16. predict after gsem with structural paths between Gaussian depvars used an inverse projection of the structural coefficients instead of a direct linear combination to compute the linear predictions.

For the example model

Y = B * Y + G * X + e

where Y is the matrix of depvars, X is the matrix of (exogenous) indepvars, e is the matrix of endogenous errors, B is the matrix of structural coefficients, and G is the matrix of exogenous coefficients, predict computed linear predictions as

Yhat = (I - Bhat)^-1 * Ghat * X

when it should have been using

Yhat = Bhat * Y + Ghat * X

This has been fixed.

17. qreg, in rare situations, could cause Stata to crash. This has been fixed.

18. sem and gsem now respect an unconstrained constant path in all cases when option noconstant is specified. For example,

sem (mpg <- _cons) (mpg <- turn) , noconstant

would fit with a zero-constrained constant, while

sem (mpg <- turn) (mpg <- _cons) , noconstant

would fit with an unconstrained constant. Now both specifications will fit with an unconstrained constant.

19. The SEM Builder now adds an object at the default size if the user mistakenly adds an object of 0 width or height.

20. set cformat, set pformat, and set sformat commands have the following fixes:

a. They now report an error when the specified format width is too large. The maximum format width for cformat is 9. The maximum format width for pformat is 5. The maximum format width for sformat is 8.

b. They reported error "invalid numeric format" on an empty format specification when option permanently was specified with a space before the comma. This has been fixed.

21. set obs, when a variable was sorted and when the last value of that variable contained an extended missing value, did not mark that variable as no longer being sorted. This has been fixed.

22. simulate with a user-written command that has an option minimally abbreviated as l(arg) parsed this option as if it were option level(#), causing a syntax error when arg was not a value in the limited range allowed by level(#). simulate no longer parses level(#).

23. svy was not always preserving the r() results normally posted by ereturn display. This has been fixed.

24. svy: tabulate twoway with option wald and using bootstrap variance estimation reported an unhelpful error. This has been fixed so that svy: tabulate twoway reports the unadjusted Wald test using a chi-squared statistic instead of the usual F statistic.

25. testparm ignored the level restrictions on factor variables in interactions when at least one factor variable was specified without a level restriction. The result was a test that included more level combinations of the interaction than was specified. This has been fixed.

26. (Mac) The Do-file Editor has the following fixes:

a. Scrolling when using a scroll wheel from a third-party mouse would be very slow. This has been fixed.

b. Using Find on a long document when running on Mac OS X 10.6 could cause severe performance problems. This has been fixed.

c. Menu File > Do... would open the selected do-file in the Do-file Editor instead of executing it. It now executes the selected do-file.

27. (Mac) The Project Manager did not restore a project's expanded or collapsed state after the project was opened. This has been fixed.

28. (Mac) Commands are now added to the Review window before they finish executing.

29. (Windows) Double-clicking on a *.smcl file in the Project Manager now opens the file in the Viewer.

30. (Unix Console and Mac Console) Pressing Ctrl+C could cause Stata to terminate. The problem would occur only after the Java Runtime Environment was loaded by Stata. This has been fixed.

-------- update 19sep2013 -----------------------------------------------------

1. Online help and the search index have been brought up to date for Stata Journal 13(3).

2. New postestimation command estat wcorrelation after mixed computes and displays the within-cluster correlation matrix for a given cluster calculated on the basis of the design of the random effects and their assumed covariance and the correlation structure of the residuals. This allows for a comparison of different multilevel models in terms of the ultimate within-cluster correlation matrix that each model implies.

3. The icd9 databases have been updated to use the latest version (V31) of the ICD-9 codes.

4. predict after mgarch ccc, mgarch dcc, and mgarch vcc has new option correlation for producing conditional correlations.

-------- update 12aug2013 -----------------------------------------------------

1. Filename completion is now available in the Command window by pressing the Tab key. The filename must be preceded by a double-quote (") character. Otherwise, Stata will assume you wish to do variable name completion.

2. forecast identity has new option generate to generate the left-hand-side variable on the fly, which is useful when fitting equations that include transformations of endogenous variables as covariates.

3. gsem has new option dnumerical to compute the gradient and Hessian using numerical techniques instead of analytical formulas. This option may be helpful when using multilevel latent variables with large groups and when gsem reports "starting values not feasible" for a variety of user-specified starting values.

4. stem has new option truncate(#), which truncates data to the specified value.

5. Prefix by, when used with summarize with an if expression containing a string variable, could cause Stata to crash. This has been fixed.

6. estat esize mislabeled interaction terms in the reported effects sizes table by attaching the operator c. to all variables in the interaction term. This has been fixed.

7. export delimited issued an error when trying to save a filename with embedded spaces. This has been fixed.

8. gmm, when option xtinstruments() was specified and the user requested a cluster-robust weight matrix or VCE, ignored the cluster variable and instead clustered based on the panel variable. This has been fixed.

9. heckman, with user-specified constraints, produced a likelihood-ratio test for independent equations that was incorrect. Now a Wald test is reported when a user specifies constraints.

10. import delimited could fail to import a row correctly if the file to be imported had both a line ending with a delimiter and the next line starting with a delimiter and if option delimiters() was specified with suboption collapse. This has been fixed.

11. margins after gsem or meglm with family(gamma) would report the error "prediction is a function of possibly stochastic quantities other than e(b)" even when there were no latent variables in the model specification. This has been fixed.

12. predict after gsem was not computing out-of-sample mean predictions (for option mu or pr) when it should. This has been fixed.

13. predict after meglm was not computing out-of-sample mean predictions (for option mu or pr) when it should. This has been fixed.

14. symmetry with an empty if condition returned the error "tempvar not found". This has been fixed.

15. tabulate, when used with a strL variable and options summarize() and missing, could crash Stata. This has been fixed.

16. var with options small and constraints() would display missing values for the F statistic in the header. This has been fixed.

17. (Mac) Executing commands in Stata from an AppleScript script that creates multiple graphs could cause Stata to crash. This has been fixed.

18. (Windows) The linesize setting was not set when the font was changed in the Results window. This has been fixed.

19. (Unix) On systems using European number formats, Stata commands could fail when reading values with decimals. The problem would occur only after the Java Runtime Environment was loaded by Stata. This has been fixed.

-------- update 23jul2013 -----------------------------------------------------

1. ivregress produced an error when you specified interactions of factors with a large number of categories. This has been fixed.

2. gmm now allows equation names within option parameters() and labels the parameter vector passed to moment-evaluator programs appropriately. That allows you to use matrix score in evaluator programs to compute linear combinations of variables quickly. New option haslfderivatives allows you to code derivatives with respect to those linear combinations, simplifying the use of gmm when implementing other commands.

In addition, gmm has a new option, quickderivatives, that provides an alternative method of computing the numerical derivatives required in the VCE computations when analytic derivatives are not provided. This method is slightly less accurate than the default method but is substantially faster with problems that contain many instruments or residual equations.

3. import excel gave a "file too big" error for .xls files over 40 MB. The restriction was intended for .xlsx files only. This limit has been removed for .xls files.

4. import excel was slower in Stata 13 compared with Stata 12. The performance in Stata 13 has been improved to be comparable with Stata 12.

5. reshape, when the variable specified in option j(varname) contained missing values, might give the wrong error message. This has been fixed.

6. The parameter domains for rnbinomial(), the negative binomial random-number generator, have changed to allow a greater range of parameter combinations.

7. sem with a dataset that has one or more variables with capitalized names was reporting the error "latent variable '...' not referenced in the model" even when option latent() was not used. This has been fixed.

8. (Mac) The notification sound Stata plays when Stata is finished executing a command in the background would not play in Mac OS X 10.7 and 10.8. This has been fixed.

9. (Mac) The main Stata window now has a minimum content height of 320 pixels so that it can be resized to a much shorter height than before.

-------- update 02jul2013 -----------------------------------------------------

1. Online help and the search index have been brought up to date for Stata Journal 13(2).

2. .dta files did not fill in the file position in the "map" for variable labels; this has been fixed. This did not affect Stata's ability to read and write .dta files.

3. merge in Stata/SE and Stata/MP would produce an error if the using dataset contained more than 5,000 variables. This has been fixed.

4. Function subinword(), when the string to be substituted with was a single letter and matched both a word at the end of the string to be substituted for and the first letter of the string to be substituted for, returned an empty string rather than performing the proper substitution. This has been fixed.

5. tabulate, sort with a string variable did not always list the categories in descending order of frequency. This was fixed 21jun2013 for Mac and Windows and has been fixed in this update for all others.

6. (Mac) After updating Stata, attempting to open the Do-file Editor would hang Stata for Mac. This has been fixed.

7. (Mac) Executing a selection from a do-file in the Do-file Editor could cause Stata to write out incomplete lines to be executed or even crash Stata if the do-file contained extended ASCII characters. This has been fixed.

8. (Mac) In the Do-file Editor, Replace & Find would replace the search expression from the beginning of the do-file instead of from the current selection. This has been fixed.

9. (Mac) Displaying the current working directory in the status bar is now much faster.

-------- update 21jun2013 -----------------------------------------------------

1. (Mac) When using the Graph Editor on Mac OS X 10.7 or newer, right-clicking on an object would bring up two different contextual menus. This has been fixed.

2. (Windows) If a new project was created through the Do-file Editor or a new file was added to an open project, open Viewer windows became unresponsive. This has been fixed.

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

See whatsnew12to13.

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


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