Stata 15 help for whatsnew

Title

Additions to Stata since release 15.0

Description

Update history:

Stata 15.0 base 06jun2017 Stata 15.1 base 06nov2017 updated to 20feb2019

This file records the additions and fixes made to Stata since the release of version 15.0. The end of this file provides links for earlier additions and fixes.

Updates are available for free over the Internet. Click here to obtain the latest update, or see updates for detailed instructions.

The most recent changes are listed first.

-------- update 20feb2019 -----------------------------------------------------

1. Functions tin() and twithin() now support dates in business calendar (%tb) format.

2. describe now stores the dataset label in r(datalabel).

3. areg specified with a factor variable having a single chosen level, when that factor level was not observed in one or more of the absorption groups and the model contained subsequent interaction terms containing a continuous variable, produced incorrect coefficients and standard errors. This has been fixed.

4. ereturn post and ereturn repost with option esample() failed to update variable indices in the existing Mata views. For Mata views that were created using variables with an index greater than that of the variable specified in option esample(), those columns were referencing variables with a higher index than they should have. This has been fixed.

5. fracreg with frequency weights reported the wrong number of observations. This has been fixed. The reported coefficients, standard errors, test statistics, and confidence intervals were correct.

6. gsem no longer requires option nocapslatent when option lclass() is specified and any of the observed variable names in the model begins with a capital letter.

7. svy: gsem, when svyset weights contained zero values, sometimes exited with an unhelpful error message. svy now properly marks the estimation sample for gsem in this case.

8. gsem now exits with an informative error message when factor-variable operators are specified on a latent variable not identified in option lclass(). The old behavior, where gsem ignored factor-variable operators specified on latent variables not identified in option lclass(), is not preserved under version control.

9. After the 15 Oct 2018 update, import excel, when importing an .xlsx file that was already open in Excel, exited with an uninformative error message. This has been fixed.

10. makecns now allows the number of constraints to be equal to the number of parameters. This affects estimation commands such as cnsreg and any estimation command that uses ml.

11. Mata functions fputmatrix() and fgetmatrix() were unable to save Mata structure matrices and vectors. This has been fixed.

12. Mata function regexr(), when a) the regular expression to match was a single byte, b) the match occurred at the end of the source string, and c) the replacement string was length 0 or 1, failed to replace the matching substring.

For example,

mata: regexr("ABCDa","a","") mata: regexr("ABCDa","a","c")

both wrongly returned "ABCDa" instead of "ABCD" and "ABCDc". This has been fixed.

13. ml, when fitting a model with option svy and survey weights were not svyset, posted the name of a temporary variable in macros e(wvar) and e(wexp). This has been fixed.

14. predict, when specified with a new variable name that matches a variable in the current estimation results, now exits with an error message even when that variable does not exist in the current dataset. The old behavior, which provided invalid predicted values, is not preserved under version control.

15. predict after asmixlogit, when specified with option scores and a specified variable type, retained the specified variable type for only the first score variable. This has been fixed.

16. predict after asmixlogit and with option scores, when asmixlogit was specified with a full-factorial interaction among variables with fixed coefficients and was specified with option random(varlist, correlated) where varlist consisted of only a single variable, produced scores that were incorrectly set to 0. This has been fixed.

17. predict after gsem, when specified with option classposteriorpr and when the fitted model contained two or more multinomial outcomes with missing-value patterns that would cause one or more observations to be dropped from the estimation sample, exited with an unhelpful error message. This has been fixed.

18. predict after ologit and oprobit ignored option nooffset. This has been fixed.

19. predict after ologit and oprobit, when specified with option scores, failed to exit with an error message when option nooffset was specified. This has been fixed.

20. sem now exits with an informative error message when factor-variable operators are specified on a latent variable. The old behavior, where sem ignored factor-variable operators specified on latent variables, is not preserved under version control.

21. sem with observed exogenous variables and two or more latent exogenous variables sometimes had difficulty fitting an otherwise-identified model depending on the order of the specified paths. This has been fixed.

22. The SEM Builder has the following fixes:

a. The SEM Builder, when the standard error was included in the customized appearance for a selected connection, would report the standardized estimate instead of the standard error estimate. This has been fixed.

b. The SEM Builder, when attempting to change a path's arrow size, resulted in no change. This has been fixed.

c. The SEM Builder, when changing a path's arrow barb size to 0, resulted in the arrow being replaced by a line that ran off the canvas. This has been fixed.

23. svy: total, for data with a stratum having a single sampling unit and data that have been svyset with option singleunit(centered), now centers using the average of the stratum totals for a given stage instead of using the grand mean. The old behavior produced standard errors that were too large and is not preserved under version control.

24. teffects nnmatch using both exact and distance matching ignored the specified caliper for the distance matching. This has been fixed.

25. Stata now creates random names for SVG path elements when exporting graphs to the SVG format. When an HTML document contains multiple Stata SVG graphs that share common SVG path names, a browser can have trouble rendering the SVG graphs; randomizing the path names helps prevent that. If you would like control over how the SVG path names are generated, specify option pathprefix() when exporting the graph. Stata will use your path prefix and an index that is incremented for each path to create stable path names. See [G-3] svg_options for more information.

-------- update 17dec2018 -----------------------------------------------------

1. Online help and the search index have been brought up to date for Stata Journal 18(4).

2. churdle has the following fixes:

a. churdle, when specified with prefix svy and weights with missing values, incorrectly exited with an error message. Now, point estimates are produced.

b. churdle, when specified with prefix svy, produced incorrect standard errors for the auxiliary parameter. This affected only models where heteroskedasticity was not modeled. This has been fixed.

3. cpoisson, when constraints were specified and initial values were not feasible, incorrectly exited with an uninformative error message. This has been fixed.

4. ereturn display, when specified with results from mean and total, ignored option eform() instead of reporting the estimated parameters in exponentiated form. This has been fixed.

5. estat sbknown, when estimated on one panel of a multiple-panel dataset, exited with an uninformative error message. This has been fixed.

6. histogram, when specified with an if expression or in range, incorrectly applied these sample restrictions to additional plots specified in option addplot(). This has been fixed.

However, when option by() is also specified, the if expression or in range sample restrictions will be applied to additional plots specified in option addplot(). In other words, histogram's behavior has not changed when option by() is specified.

7. mestreg did not check for the presence of delayed entries or gaps. mestreg now returns error message "delayed entries or gaps not allowed".

8. predict with option pr after gsem, meologit, meoprobit, xtologit, and xtoprobit did not allow you to predict probabilities for negative ordinal outcomes. This has been fixed.

9. predict with option scores after tpoisson, which was not left-truncated at zero, produced incorrect scores. This has been fixed.

10. stteffects ipwra with survival-time outcome model weibull, exponential, or lnormal sometimes failed to converge. Convergence has been improved.

-------- update 15oct2018 -----------------------------------------------------

1. icd10cm and icd10pcs have been updated for the 2019 fiscal year. Type icd10cm query or icd10pcs query to see information about the changes.

2. areg, when specified with factor variables and weights, sometimes failed to omit a factor level that was constant within the absorb() variable. When this happened, the standard error for the coefficient on the associated factor level was typically astronomical compared with the other standard error estimates. This has been fixed.

3. estat lcmean and estat lcprob, when fmm or gsem fit a model with a pointmass outcome in the base level of the latent class, incorrectly exited with error message "prediction is a function of possibly stochastic quantities other than e(b)". This has been fixed.

4. estimates table, when set emptycells drop was in effect and an interaction term was specified in option keep() or option drop(), exited with error message "invalid syntax". This has been fixed.

5. fredsearch with option tags(), when the number of tags specified was greater than 5, returned an uninformative error message. fredsearch now returns error message "at most 5 tags can be specified in option tags()".

6. When exporting a graph to EPS or PDF format, Stata now hides lines with a line width of none or 0.

7. gsem, when specified with a truncated Poisson outcome, sometimes caused Stata/MP to crash. This has been fixed.

8. The 02 April 2018 update caused import excel with option sheetreplace to remove all workbook cell ranges. This has been fixed.

9. logit, when fitting a model where a large number of the levels of a factor variable are perfect predictors and when set emptycells drop is in effect, could cause Stata to crash. This has been fixed.

10. margins after npregress with option noderivatives always used default kernels epanechnikov and racine even if the user specified another kernel during estimation. This has been fixed.

11. margins after npregress with option noderivatives, when margins was called more than once, removed the auxiliary variable used to store the mean function prediction. This has been fixed. Marginal effects and predictions were not affected by this issue.

12. marginsplot failed to plot margins values larger than c(maxfloat). This has been fixed.

13. As of Stata 15, matrix subscripting did not recognize /: as a shortcut for retrieving all the free parameters from the specified matrix. This problem was noticed when running community-contributed command esttab with option unstack. This has been fixed.

14. mixed, meqrlogit, and meqrpoisson now store the number of iterations in e(ic).

15. ml display, when working with estimation results containing free-parameter notation in the column stripe of e(b), now interprets each of those free parameters as a separate equation when counting how many parameters to exponentiate when an eform_option is specified. The old behavior, grouping the elements with free-parameter notation into a single equation, is not preserved.

16. mleval, mlvecsum, mlmatsum, and mlmatbysum exited with error message "equation 1000 out of range" for estimation commands that specified more than 999 equations. This restriction has been removed. Now the number of equations allowed by these estimation commands is only limited by set matsize.

17. program lines containing multiple adjacent spaces enclosed by compound double quotes, `""', would have those spaces compressed to a single space. This has been fixed.

18. stcoxkm, which is not an estimation command, incorrectly modified existing stored estimation results or incorrectly created new estimation results when there were none. This has been fixed.

19. Stata would crash at the end of the following sequence of events: a. Create a Mata view from a varlist with time-series operators or factor variables. b. Call discard. c. Run an estimation command that posts e(sample). d. Create a new Mata view. This has been fixed.

20. (Linux) c(username) could return a truncated copy of the user ID from the operating system. The truncation would occur after 8 characters or after a nonalphanumeric character. This has been fixed.

-------- update 20sep2018 -----------------------------------------------------

1. Online help and the search index have been brought up to date for Stata Journal 18(3).

2. bootstrap, when used with panel-data commands and variables having time-series operators, produced incorrect standard errors. More lagged observations than necessary were excluded from the bootstrap replication samples. Under standard assumptions, the reported standard-error estimates were conservative. Stata's xt estimation commands that allow both time-series operators and the vce(bootstrap) option were also affected. This has been fixed.

3. levelsof, when used with strL variables and with only one observation in the dataset, produced an error message. This has been fixed.

4. margins has the following fixes:

a. margins, after fitting a model with an offset, when options expression() and at() were both specified, in rare cases computed margins from the expression using only the first observation instead of the entire estimation sample. Specifically, only the first observation was used when the expression included more than a single prediction, the prediction relied on the offset, and each variable included in option at() was set to a constant value. This has been fixed.

b. margins with option nose, option dydx(), and multiple predict() options reported marginal effects using only the first predict() option instead of computing the marginal effect for all specified predictions. This has been fixed.

5. mixed with option dfmethod(kroger) or option dfmethod(satterthwaite), when group identifiers were string variables, exited with an uninformative error message. This has been fixed.

6. predict after menl now estimates predictions for observations with a missing dependent variable.

7. Beginning with the 07 August 2018 update, reshape wide with option string and option j(varname), where varname was a string variable containing spaces, did not correctly identify unique values of varname if the characters before the first space were identical for two or more categories of varname. reshape wide now exits with an error message indicating that string j variables are not allowed to contain spaces.

To use a string variable that contains spaces with reshape wide, you can first create a corresponding numeric variable using encode and then specify the resulting variable in reshape's option j().

-------- update 07aug2018 -----------------------------------------------------

1. The following new functions are added to Stata and Mata:

a. expm1(x) returns e^x - 1 with higher precision than exp(x)-1 for small values of |x|.

b. ln1p(x) and log1p(x) return the natural logarithm of 1+x with higher precision than ln(1+x) for small values of |x|.

c. ln1m(x) and log1m(x) return the natural logarithm of 1-x with higher precision than ln(1-x) for small values of |x|.

2. export excel has new option keepcellfmt that preserves the cell style and format of an existing worksheet when writing data.

3. estat wcorrelation could generate an incorrect correlation or covariance matrix when used after menl. This occurred only when option rescovariance(unstructured, index(indexvar)) or option rescorrelation(unstructured, index(indexvar)) was specified with menl and the user dropped observations from the dataset after fitting the model such that all observations from one or more levels of indexvar were omitted. This has been fixed.

4. irf create with option order() did not produce the correct order when more than two variables changed position in the new order. This has been fixed.

5. metobit and meintreg incorrectly omitted the variance component for the outcome variable, resulting in e(k_rs) being one less than it should have been. This has been fixed.

6. mlogit and mprobit, when the specified depvar had long value labels that started with a number, exited with error message "invalid name". This has been fixed.

7. predict could generate incorrect predictions after menl. This occurred only when option rescovariance(unstructured, index(indexvar)) or option rescorrelation(unstructured, index(indexvar)) was specified with menl, and the requested prediction was a function of the random effects, and the user dropped observations from the dataset after fitting the model such that all observations from one or more levels of indexvar were omitted. This has been fixed.

8. proportion, when specified without the svy prefix, without option vce(cluster ...), and without pweights, now computes the standard error of the estimated proportion using the formula

sqrt(phat(1-phat)/n)

as originally documented in Methods and formulas. The old behavior, using a similar formula that has n-1 in the denominator instead of n, is not preserved under version control.

9. The 06 June 2018 update caused putexcel to remove a cell fill-color format when any font format was written to the same cell. This has been fixed.

10. reshape wide with option string and option j(varname), where varname is a string variable containing spaces, failed to generate the variables corresponding with the observations where the string variable contained spaces. This has been fixed.

11. svy: mean with option over() identifying more than 100 groups and stage-level sampling weights exited with error message "varname not byte". This has been fixed.

12. xtile, when an if condition was specified after an expression that ended with a parenthesis, incorrectly produced an error message. This has been fixed.

13. Stata-Java API Specification, when called with version less than 14, incorrectly encoded native strings from Stata to Java using either iso-8859-1 or MacRoman, depending on the platform. This was incorrect because all strings in Stata are now represented with the UTF-8 encoding. This is now fixed so that the UTF-8 encoding is always used when converting Stata native strings to Java.

14. When using the Stata-Java API Specification, the method ValueLabel.removeVarValueLabel(int index) failed with return code r(198) when a valid variable index was specified. This has been fixed.

15. (Mac) When printing from the Viewer or Results window, changing the font size in the print dialog caused Stata to crash. This has been fixed.

16. (Mac) When printing from the Viewer or Results window, changing the scale in the print dialog could result in blank pages being printed. This has been fixed.

-------- update 27jun2018 -----------------------------------------------------

1. Online help and the search index have been brought up to date for Stata Journal 18(2).

2. graph bar with option percentages, when using a scheme that sets a nonzero number of minor ticks, used the wrong scale for the bar heights when the maximum value among the variables in yvars was greater than the maximum computed percentage. This has been fixed.

3. gsem using the default integration method, intmethod(mvaghermite), and with prefix svy, in the rare situation where predict failed to compute the adaptive parameters within the default maximum number of adaptive iterations, reported standard errors that were too large, usually by a multiple of 3 or more. This situation will now cause svy to exit with an error message stating that the scores could not be computed and also suggesting that the model be refit using a different integration method.

Other commands affected by this change are meglm, metobit, meintreg, melogit, meprobit, meologit, meoprobit, mepoisson, menbreg, and mestreg.

4. meglm, metobit, meintreg, melogit, meprobit, meologit, meoprobit, mepoisson, menbreg, and mestreg, when the fixed-effects equation contained factor variables, returned an incorrect number of parameters in e(k_f). This has been fixed.

5. sem has improved starting values logic for models where some latent variables have only latent indicators, such as in hierarchical CFA models.

-------- update 06jun2018 -----------------------------------------------------

1. menl has the following improvements:

a. menl now uses standard maximization options such as iterate() and tolerance() to control maximization of the NLS algorithm. It was previously controlled by pnlsopts()'s suboptions. pnlsopts() is no longer allowed with the NLS algorithm.

b. menl with nonlinear least-squares (NLS) models -- models without random effects and with i.i.d. errors -- used a default of only five maximum number of iterations. Now it uses the current value of set maxiter, which is iterate(16000) by default. The increased default maximum number of iterations may result in convergence of the NLS models that previously failed to converge in five iterations.

2. putdocx has the following improvement and fix:

a. putdocx table has new argument bwidth in option border(bordername [, bpattern [, bcolor [, bwidth]]]), which is used to set the width of the specified border.

b. putdocx table tablename(i,j) = (exp), when exp contained unbalanced parenthesis ( or ), incorrectly exited with error message "expression must be enclosed in ()". This has been fixed.

3. graph export has the following fixes:

a. graph export, when exporting to a PDF file a twoway scatter graph that contained multiple overlayed scatterplots where the symbols in each plot specified transparency settings on the same color, failed to display symbols with transparency. This has been fixed.

b. graph export could crash Stata when exporting to SVG format a graph that contained many graph items. This has been fixed.

4. When Mata function LA_DGEHRD() was called within a timing code section using timer 1, the reported time was wrong. This has been fixed.

5. When using Mata, views and subviews that were created using variable lists containing factor-variables notation or time-series operators used more memory than necessary. Although calculations were correct, Stata typically became noticeably slower. This has been fixed.

6. matrix list and matlist, when used to display matrices posted by contrast and pwcompare, failed to include the coding or contrast operators in the output. This has been fixed.

7. putpdf table with option width(), when the widths for each column in the table were specified as a group of numbers and the sum of those numbers equaled 100 as required, sometimes incorrectly exited with error message "failed to set table width". This has been fixed.

8. regress with aweights or iweights and option vce(hc2) or option vce(hc3) did not use properly weighted leverage values when computing the variance estimates. This typically resulted in standard-error values that differed in the third significant digit. This has been fixed.

9. (Windows) The 18 April 2018 update caused the exporting of Stata graphs to WMF format using translate to fail. This has been fixed. However, WMF is a legacy metafile format and does not support modern graphics features such as transparency and smooth lines. Users are encouraged to switch to the EMF format, which is also supported by older versions of Stata.

-------- update 08may2018 -----------------------------------------------------

1. The extended regression model commands now allow greater flexibility in modeling treatment effects.

a. New suboptions povariance and pocorrelation are now available within options entreat() and extreat() where the treatment model is specified.

Different variance parameters can be estimated for each potential outcome (for each treatment level) by specifying the suboption povariance. povariance may be specified only when fitting a model with eregress or eintreg.

Different correlation parameters can be estimated for each potential outcome (for each treatment level) by specifying the suboption pocorrelation.

b. New suboptions povariance and pocorrelation are now available within option endogenous() when the suboption probit or oprobit has also been specified.

Different variance parameters can be estimated for each level of the endogenous covariate by specifying the suboption povariance. povariance may be specified only when fitting a model with eregress or eintreg.

Different correlation parameters can be estimated for each level of the endogenous covariate by specifying the suboption pocorrelation.

2. meglm, metobit, meintreg, melogit, meprobit, meologit, meoprobit, mepoisson, menbreg, and mestreg, when integration method mvaghermite, mcaghermite, or ghermite is specified, have improved performance for crossed models specified using the R. operator.

3. areg, when specified with factor variables and option vce(cluster clustvar), exited with an error message if any of the independent variables contained _cons in their names. This has been fixed.

4. bootstrap with option bca and a variable named _merge in the dataset incorrectly exited with error message "variable _merge already defined". This has been fixed.

5. jackknife with option keep and a variable named _merge in the dataset incorrectly exited with error message "variable _merge already defined". This has been fixed.

6. margins after npregress has the following fixes:

a. margins with option dydx(), when value labels contained spaces, exited with an error message. This has been fixed.

b. margins with option at(varname = generate(exp)), when exp contained spaces, exited with an error message. This has been fixed.

7. predict after arch, when arch was specified with both options archmlags() and archmexp(), produced an error message. This has been fixed.

8. predict after meglm, metobit, meintreg, melogit, meprobit, meologit, meoprobit, mepoisson, menbreg, and mestreg now produces an error message if option ebmeans or ebmodes is specified without option reffects. Previously, predictions of the mean response or another specified statistic that used the empirical Bayes means or empirical Bayes modes were produced. That behavior was not documented and is not preserved under version control.

9. pwcorr, when specified with two variables and option obs, reported the wrong number of observations for the individual variables when some of the observations contained missing values in one variable but not the other. This has been fixed.

-------- update 18apr2018 -----------------------------------------------------

1. The Java Runtime Environment that is redistributed with Stata is now updated to version 8 update 162.

2. graph twoway with option by(), when using a scheme that sets a nonzero number of minor ticks, sometimes rescaled the axes differently between the individual graphs. This usually happened to the y axes but could also happen to the x axes. This has been fixed.

3. import fred with proxy settings enabled exited with error message "connection timed out". This has been fixed.

4. predict with option score after eteffects, when one score was requested rather than one score for each equation, exited with an error message. This has been fixed.

5. (Mac) Stata could randomly crash when printing from the Results or Viewer window. This has been fixed.

-------- update 02apr2018 -----------------------------------------------------

1. Online help and the search index have been brought up to date for Stata Journal 18(1).

2. npregress has the following new features and fix:

a. npregress with new option citype() allows you to specify normal-based and bias-corrected bootstrap confidence intervals in addition to the default percentile confidence interval.

b. margins after npregress with new option citype() allows you to specify normal-based and bias-corrected bootstrap confidence intervals in addition to the default percentile confidence interval.

c. margins after npregress now allows contrast operators.

d. npregress produced an error message when a specific level of a factor variable was specified as a regressor. This has been fixed.

3. levelsof has new option hexadecimal for numeric variables, which formats the levels in hexadecimal format. This is useful when the variable's levels are noninteger or extremely large integers (absolute value >= 1e19). This option guarantees that the values in the macro that levelsof creates are exactly numerically equal to their values in the variable. This option is unnecessary for integers whose absolute values are less than 1e19.

4. bayes prefix for multilevel models (for example, bayes: melogit) sometimes would not allow crossed effects, issuing an error message that the crossed latent variables were in conflict. This has been fixed.

5. bayes: mvreg with option gibbs performed incorrect full Gibbs sampling that affected the joint distribution of regression coefficients from different equations. This has been fixed. Note that the marginal sampling distributions of regression coefficients were unaffected and correct.

6. ci proportions exited with an error message when varlist included a Stata wildcard character (~, ?, -, or *). This has been fixed.

7. predict after glm, when glm was used with a user-written link function that required specification of a variable, exited with error message "invalid syntax". This has been fixed.

8. svy: tabulate, when the specified variable(s) or the value label of the specified variable(s) contained a single left quote (`), exited with error message "too few quotes". This has been fixed.

9. (Windows) When the current working directory was a drive letter (for example, "c:"), dyndoc, dyntext, and markdown constructed the output file path using a relative path notation "c:outputfile" instead of absolute path notation "c:/outputfile". In this case, clicking on the link to the file failed to open it in the web browser. This has been fixed.

-------- update 08mar2018 -----------------------------------------------------

1. estat teffects, when specified after commands eregress, eintreg, eprobit, or eoprobit with prefix svy:, incorrectly exited with error message "invalid subcommand teffects". This has been fixed.

2. graph twoway with option by(), when using a scheme that turned off y-axis grid lines, produced graphs with y-axis grid lines anyway. This has been fixed.

3. levelsof did not correctly format large integers (> 1e7). This has been fixed.

4. margins with option vce(unconditional), when used after nl, produced incorrect standard errors that were, on average, the square of what they should have been. This means that if the previously reported standard error was larger than 1, then it was too large, and if the previously reported standard error was smaller than 1, then it was too small. This has been fixed.

5. moptimize(), when moptimize_init_indepvars() was specified using a Mata subview that was created by selecting a subset of the columns of a Mata view containing time-series operators or factor variables, sometimes caused Stata to crash. This has been fixed.

6. putdocx table with options title() and headerrow(), when the table continued on more than one page, only created the table header on the first page and did not repeat the table header on subsequent pages. This has been fixed.

7. putexcel ul_cell = formula(formula), when double quotes were specified in formula, created an invalid workbook. When the workbook was opened in Excel, a stopbox was issued. This has been fixed.

8. svy, when specified using the prefix syntax to replay survey estimation results from meglm, metobit, meintreg, melogit, meprobit, meologit, meoprobit, mepoisson, menbreg, and mestreg, such as

svy: meglm

exited with error message "last estimates not found". This has been fixed.

9. svy: nl, when the specified model contained a constant term, produced standard errors that were slightly too large by the factor f = (df_t - df_m) / (df_t - df_m - 1). Here, df_t is the total degrees of freedom, and df_m is the model degrees of freedom. This has been fixed.

10. teffects nnmatch with option biasadj() and frequency weights produced incorrect standard errors for the estimated average treatment effect (ATE) and the average treatment effect on the treated (ATET). This has been fixed.

11. Stata ignored the last line of some Stata text files, such as do-files and dictionaries, if the last line did not contain an end-of-line delimiter. This has been fixed.

12. (Windows) Stata displayed an empty confirmation dialog box if you opened a read-only file in the Do-file Editor and attempted to edit or execute the file. Stata no longer displays the empty dialog box and now requires you to use a different filename if you try to save the file.

13. (Windows) The "Recent files" menu truncated file paths at the end instead of the middle for extremely long file paths. This has been fixed.

14. (Mac) On macOS 10.13 (High Sierra), using the Data Editor to keep or drop observations issued an invalid command. This has been fixed.

15. (Unix) On some Unix operating systems, the Do-file Editor would not remember all syntax highlighting preferences between Stata sessions. This has been fixed.

-------- update 13feb2018 -----------------------------------------------------

1. bayesmh and prefix bayes now allow option initsummary to be specified with option dryrun. This is useful for inspecting the initial parameter values before performing MCMC simulations.

2. irtgraph icc with option blocation, when the style of decimal point was set to comma, did not render the lines for the estimated item difficulties. This has been fixed.

3. margins, when used with estimation results from ologit or oprobit where the last term in the main equation was an interaction involving a continuous variable and that continuous variable was also specified in option dydx(), did not compute the standard error of the marginal effect correctly. The affected standard errors were typically inflated. This has been fixed.

4. After the 11jan2018 update, Mata function st_addvar(), when used to create new variables and when there were estimation results already present in e(), returned variable indices that were off by one position. This typically only affected calls to st_view() that used these indices instead of variable names. This has been fixed.

5. stset incorrectly cleared variable _stack, which was generated by a previous stack command. This has been fixed.

6. The matching logic for estimated coefficients specified in expressions, such as those allowed by lincom, test, and generate, was lacking version control for matrix stripe name elements of e(b) that look like functions. For example, stripe name element cov(y) in an expression was transformed to var(y) before attempting to match against the stripe elements of e(b). The matching logic has been updated to search for the original specification when version is less than 15.0 and to search for both variants when version is 15.0 or higher.

-------- update 30jan2018 -----------------------------------------------------

1. import excel's option allstring now has an optional argument, allstring("format"), that allows you to control the display format of numeric spreadsheet columns being converted to Stata string data.

2. New command lmbuild makes creating Mata function libraries easier than does mata mlib.

3. clonevar, when newvar started with the name or an abbreviated name of the cloned variable, incorrectly exited with error message "ambiguous abbreviation". This has been fixed.

4. collapse (min) varlist, when a variable in varlist contained all missing values and when at least one of those was an extended missing value, could return any missing value for that variable instead of the smallest missing value. This has been fixed.

5. estat lcprob, after fmm or gsem, when the name of the categorical latent variable was a valid abbreviation of a variable in the dataset, reported value labels attached to the variable (in the dataset) instead of the class levels. This has been fixed.

6. estat mindices, when specified after sem with a large number of observed endogenous variables, sometimes reported a missing value for all the modification indices representing an omitted path or covariance. This has been fixed.

Our technical services team previously recommended using the undocumented option slow when estat mindices reported missing values. This option is no longer necessary.

7. gsem, when specified using the variable range syntax but with a space on either side of the dash character, such as

gsem ... fvar -lvar gsem ... fvar- lvar gsem ... fvar - lvar

exited with error message nothing found where name expected. gsem now tolerates spaces in this specification.

8. histogram, when specified with both options horizontal and addlabel, incorrectly placed the bar height labels as if option horizontal had not been specified. The histogram was correctly presented horizontally, but the bar height labels were incorrectly placed as if the histogram were vertical. This has been fixed.

9. logistic, logit, mlogit, ologit, oprobit, and probit now accumulate log-likelihood values using quad precision. These commands now converge to a solution for some extreme datasets that previously caused these models not to converge.

10. logistic, logit, and probit, when specified with an interaction term that had a level that was a perfect predictor and when set emptycells drop was in effect, incorrectly exited with error message "initial vector: extra parameter ... found". This has been fixed.

11. Matrix stripe parsing and matching logic was lacking version control for stripe name elements that look like functions. For example, stripe name element cov(y) was transformed to var(y) without regard for version control. Now the behavior is to transform only when version is set to 15.0 or higher.

12. ml, when specified with option technique(bhhh), sampling weights, and option group(), exited with an uninformative error message even when the weights were constant within group. This has been fixed.

13. When a graph was exported to SVG format, rotated marker symbols would not render properly. This has been fixed.

14. (Mac) The 11 Jan 2018 update improved scrolling performance of tables in the Results window but introduced a bug where horizontal lines would render only in black. This has been fixed.

-------- update 11jan2018 -----------------------------------------------------

1. Online help and the search index have been brought up to date for Stata Journal 17(4).

2. grmap has new options that control the alignment of borders and outlines for polygons and text boxes.

3. Function nt(df,np,t) incorrectly returned missing values in rare cases where the iterative method failed to reach convergence. The iterative method now converges and correct values are returned.

4. Function tden(df,t), in extreme cases where the degrees of freedom df was greater than about 1e+10, returned incorrect values. This has been fixed.

5. gsem, when option capslatent was specified or implied, ignored variables in the model if their first letter was a capital letter. gsem now recognizes those variables as latent variables.

6. hausman, when used with fitted models from sem or gsem with fitted covariances, unintentionally exited with an error message such as "option varname not allowed". This has been fixed.

7. margins has the following fixes:

a. margins with option asbalanced, when working with estimation results from ordinal outcome models like ologit and oprobit and when a factor variable was not among the marginsvars, incorrectly exited with error message "not conformable". This has been fixed.

b. margins with option expression(), when the expression contained a time-series-operated variable and when that same variable specification was also specified in option at(), dydx(), dyex(), eydx(), or eyex(), did not correctly compute the requested margin or marginal effect. This has been fixed.

c. margins after gsem, meglm, melogit, meprobit, metobit, meintreg, meologit, menbreg, or mepoisson with factor variables and random effects, when marginsvars were specified and when set emptycells drop was in effect, exited with error message "could not calculate numerical derivatives". margins now computes the margins as specified.

d. margins with option expression(), when the expression contained a factor-variable indicator and in the rare case when the level values of that factor variable were set to fractional values using option at(), did not correctly compute the requested margin or marginal effect. This has been fixed.

8. marginsplot, when plotting marginal effects of multiple factor variables in options dydx() and eydx(), has improved spacing of the x-axis scale.

9. predict after gsem, when specified with a single new variable name and option pr, predicted the mean of the first outcome variable even when it was not a multinomial, ordinal, or Bernoulli outcome. predict now predicts probabilities for the first outcome that is multinomial, ordinal, or Bernoulli.

10. putdocx image with option width(#[unit]), when # was set to a width larger than the body width of the document, caused the image to be clipped or not to be displayed in the document. This has been fixed. Now if the image width specified is larger than the body width, the image width in the document is set to the body width.

11. putdocx table with option drop(), when the row index i equaled the last row in the table, incorrectly exited with an error message. This has been fixed.

12. putexcel ul_cell=picture(filename) gave an error message if you used compound quotes around filename. This has been fixed.

13. putpdf table with option margin(type, #[unit]), when # was specified as a floating-point number, set the margin to the greatest integer less than or equal to #. This has been fixed.

14. The SEM Builder has the following fixes:

a. The SEM Builder incorrectly drew double-thick lines when rendering the box for a generalized response variable. This has been fixed.

b. In the SEM Builder, when the outline width for any variable was set to a sufficiently thick width, the paths connecting that variable to other variables would have a visible gap from the oval or rectangle outlining the variable. This has been fixed.

15. svy: tabulate in a two-way specification, when there were empty cells and option row or column was specified with option ci, incorrectly exited with error message "conformability error". This has been fixed.

16. xtreg, fe, when used to fit a constant-only model, sometimes posted the total sum of squares in e(rss) instead of the residual sum of squares. This affected the values posted in e(rmse), e(sigma_e), and e(rho). This has been fixed.

17. The previous update (20dec2017) enabled simplified Chinese localization by default when the system language was set to Chinese (zh). Now Chinese localization is only enabled by default if the system language is set to Chinese (zh) and the region is set to China (CN). If you wish to manually enable simplified Chinese in the user interface, type set locale_ui zh_CN.

18. Edit > Copy Table could omit the minus sign from negative numbers if they were in the first cell of a row being copied. This has been fixed.

19. (Windows) Exporting a graph to a file or the Clipboard in the EMF format could result in an image that contained a graph that was incorrectly sized within the image's boundaries. This has been fixed.

20. (Unix GUI) Stata sometimes crashed when using the Data Editor or Data Browser, especially when Stata had previously loaded Java. This has been fixed.

-------- update 20dec2017 -----------------------------------------------------

1. Stata 可选择中文本土化用户界面 (Chinese language support has been added).

Stata's menus, dialogs, and the like can now be displayed in Chinese, as well as Japanese, Spanish, and Swedish. Manuals, help files, and output remain in English.

If your computer language is set to Chinese (zh), Stata will automatically display the user interface in Chinese. To change languages manually using Stata for Windows or Unix, select Edit > Preferences > User-interface language...; to change languages manually using Stata for Mac, select Stata 15 > Preferences > User-interface language.... You can also change the language from the command line; see [P] set locale_ui.

2. bayesstats summary, bayesstats ess, bayestest interval, and bayesgraph, after fitting prefix command bayes and subsequently calling estimates use or estimates restore, failed with error message "file not found" if the simulation results were not saved by bayes with option saving(). This has been fixed.

3. predict with option ystar, ystar(a,b), or e(a,b) after churdle exponential produced incorrect results. This has been fixed.

The difference between the old and the new results is negligible when the values of the dependent variable are large, such as 100 or larger in our experiments. Large values of the dependent variable are typical with exponential mean models. However, in the rare case where the dependent variable is small, differences between old and new predictions may be larger; in our experiments, the change was at most 6%.

4. eregress, when the model included a variable that began with a capital letter, sometimes incorrectly exited with an error message regarding specification of initial values. This has been fixed.

5. esize twosample, when a string variable was specified in option by(), incorrectly exited with an uninformative error message. This has been fixed.

6. mecloglog, meglm, meintreg, melogit, menbreg, meologit, meoprobit, mepoisson, meprobit, mestreg, and metobit ignored user-specified constraints imposed on R.varname terms in random-effects equations. This has been fixed.

7. putdocx table with option title(), when there was only one column in the table, exited with error message "failed to add table title". This has been fixed.

8. pwcorr has the following fixes:

a. pwcorr with no observations and only two variables specified incorrectly stored r(rho) as 0. It now exits with error message "no observations".

b. pwcorr with no observations for the first two variables and more than two variables specified incorrectly stored r(rho) as 0. It now correctly stores r(rho) as missing.

9. stepwise with options lockterm1 and lr and forward selection or forward stepwise selection computed likelihood ratios at the first addition step by comparing each model with an additional term with an empty model rather than with the model that includes the forced first term. Thus, terms could be added to the model that did not satisfy the specified significance level for inclusion. This has been fixed.

-------- update 21nov2017 -----------------------------------------------------

1. areg, when fitting a model with the absorb() variable specified among the indepvars and with option vce(cluster clustvar) where clustvar was a variable other than the absorb() variable, incorrectly exited with error message "not sorted". This has been fixed.

2. In Stata 15.1, drop and keep were modified under version control to store the number of observations dropped in r(N_drop) and the number of variables dropped in r(k_drop). However, they cleared previously set r() results too soon, meaning that an r() result passed to an if condition on drop or keep was treated as missing. This has been fixed.

3. graph commands that specified option msangle() to change the angle of marker symbols in the current plot retained the adjusted marker angle for all plots even when msangle() was not specified with subsequent commands. This has been fixed.

4. graph matrix, when using the default stroke alignment for bounding boxes of plots and diagonal text boxes, did not align correctly. The result was that some borders appeared too thick. This has been fixed.

5. gsem has the following fixes:

a. gsem, when fitting a model with both endogenous and exogenous latent variables and with an interaction term containing a latent variable, exited with an uninformative error message. gsem now fits the model as specified.

b. gsem, when specified with a multinomial logit outcome variable and crossed random effects, incorrectly exited with an uninformative error message. gsem now fits the model as specified.

6. Prefix command mfp used with intreg incorrectly dropped all left- or right-censored observations from the estimation sample. mfp now keeps these observations.

7. ml with method d1debug or method d2debug, when the specified model contained factor variables, exited with error message "could not calculate numerical derivatives". This has been fixed.

8. predict after logit and probit, when the fitted model omitted the base level of a factor variable because it was a perfect predictor, did not properly assign missing values to the predictions according to the information in the matrix e(rules). This has been fixed.

9. stcox, when specified with option shared() and variable names in indepvars that exceeded the default 12-character width for the table, produced a coefficient table with misaligned elements in the row for the theta parameter estimate. This has been fixed.

10. xtile with option altdef failed to use the alternative formula for calculating percentiles and instead calculated them using the default formula. This has been fixed.

11. (Mac) When Stata was run on macOS High Sierra, the autocompletion list box from the Command window remained visible after the user attempted to cancel it. This has been fixed.

-------- update 06nov2017 (Stata version 15.1) --------------------------------

1. bayesmh has new likelihood t() in option likelihood() for fitting Bayesian regression models with Student's t distributed errors, also known as Bayesian Student-t regression models. This option is useful for performing robust linear regression inference in the presence of outliers or when the error terms are not normally distributed. This option specified with the t prior permits bayesmh to fit robust multilevel models in which error terms and random effects follow Student's t distribution.

2. margins, for some nonlinear predictions after certain estimation commands, now runs faster and computes standard errors and marginal effects more accurately. These improvements are a direct result of implementing analytical derivatives of the prediction with respect to each linear form and free parameter in the fitted model.

Affected estimation commands include, but are not limited to: heckman, logit, nbreg, poisson, and probit. See margins derivatives for the complete list.

3. menl has the following new features and fix:

a. menl now allows you to specify unstructured and exchangeable within options rescovariance(rescovspec) and rescorrelation(rescorrspec) and to specify distinct within option resvariance(resvarspec). These new residual covariance structures let you either estimate each variance and covariance parameter separately or estimate equal variances and a common covariance for the residuals.

b. menl has new suboption group() within options rescovariance() and rescorrelation(). This option lets you model residual covariance and correlation structures at the lowest level of your model hierarchy without having to include random effects at that level in your model. This is useful, for instance, when fitting nonlinear marginal or population-averaged models that model dependence of residuals directly, without introducing random effects.

c. menl no longer requires you to include random effects in your model specification. The improved flexibility in model specification, in combination with option resvariance(), rescorrelation(), or rescovariance(), permits fitting nonlinear models without random effects but with heteroskedastic and dependent error terms.

d. menl with option rescovariance(independent), when the data contained a residual panel with more than two observations, displayed the standard errors for the residual variance (or standard deviation) incorrectly. This has been fixed.

e. menl for models with uncorrelated heteroskedastic errors now performs a more numerically efficient scaling computation for the residuals, which may lead to slightly different parameter estimates for small datasets or for ill-identified models.

4. svyset has the following improvements:

a. svyset has new option rake() that provides a specification for calibrating weights using the raking-ratio method.

b. svyset has new option regress() that provides specifications for calibrating weights using generalized regression (GREG) or truncated regression methods.

c. svyset with a list of variables specified in option brrweight(), bsrweight(), jkrweight(), or sdrweight() now reports the weighting variables in a more compact form. Only the first and last variables in the list are shown. For example, typing the following command would produce output of the form below:

. svyset, jkrweight(jk1 jk2 jk3)

... jkrweight: jk1 .. jk3

5. estat esize has the following improvements and change:

a. estat esize has new option epsilon that reports the table of effect sizes using epsilon-squared instead of eta-squared. In comparison with eta-squared, epsilon-squared provides a less biased estimate of the effect size.

b. estat esize with option omega now computes omega-squared using the modern method of Grissom and Kim (2012). Previously, estat esize used the method of Hays (1963).

c. estat esize reported confidence intervals for omega-squared. However, these confidence intervals were found to lack the proper coverage. As such, they are no longer reported, and the lower and upper bounds are no longer stored in the matrix of effect sizes, r(esize). The old behavior is preserved under version control.

6. esizei, when specified using the syntax form for effect sizes for F tests after an ANOVA, has the following changes:

a. esizei has new effect-size estimate epsilon-squared that is reported in the default output. In comparison with eta-squared, epsilon-squared provides a less biased estimate of the effect size.

b. esizei now computes omega-squared using the modern method of Grissom and Kim (2012). Previously, esizei used the method of Hays (1963).

c. esizei reported confidence intervals for omega-squared. However, these confidence intervals were found to lack the proper coverage. As such, they are no longer reported, and the lower and upper bounds are no longer stored in r(lb_omega2) and r(ub_omega2). The old behavior is preserved under version control.

7. estat icc has the following improvements:

a. estat icc is now available after meglm using a gaussian family with an identity link, a bernoulli family with a logit or probit link, and a binomial family with a logit or probit link. This lets you calculate intraclass correlations after fitting a linear, logit, or probit mixed-effects model when using meglm. Previously, you had to use separate commands for each family-link combination.

b. estat icc is now available after estimation with svy estimation results from melogit, meprobit, meglm, metobit, and meintreg. This lets you calculate intraclass correlations for survey data.

8. icd10cm and icd10pcs have been updated for the 2018 fiscal year. Type icd10cm query or icd10pcs query to see information about the changes.

9. putdocx has the following improvements:

a. New command putdocx sectionbreak lets you add a section break to the current document. You may apply document formatting options pagesize() and landscape to the new section. This improvement makes it possible to set different properties for different pages in the document. For example, you can mix the landscape and portrait layouts in a single document.

b. putdocx table now lets you customize the format of all cells in adjacent or noncontiguous ranges. You may apply existing cell formatting options halign(), valign(), border(), and shading() and existing cell text formatting options font(), bold, italic, strikeout, underline(), nformat(), smallcaps, and allcaps. This improvement makes it possible to format many cells at once.

c. putdocx text and putdocx table have new options smallcaps and allcaps to format text in a paragraph or a cell with a mix of larger and smaller capital letters or with all capitalized letters, respectively.

d. putdocx table has new options title() and note() to add a title or a note to a table. Option note() may be repeated to add multiple notes with the same format after a table.

e. putdocx text, putdocx image, and putdocx table with option linebreak(#) now let you add line breaks to a paragraph or a cell. Previously, you needed to specify linebreak multiple times to add more than one line break. You may still specify linebreak multiple times as you did before, but we recommend using the new, more concise syntax.

10. putexcel has the following improvements:

a. putexcel set has new option open. Specifying open opens the specified filename in memory and leaves the file in memory for all future putexcel changes.

b. New command putexcel close closes the workbook and saves all changes to disk. You must use putexcel close when you are done editing the workbook if you have opened the workbook with putexcel set and option open.

c. putexcel has new output type etable, which lets you more easily add the reported coefficient table(s) from the most recent estimation command to an Excel file.

d. putexcel now supports existing option overwritefmt when using the syntax for formatting cells. This lets you write Excel cell formats more efficiently.

11. putpdf has the following improvements and fixes:

a. New command putpdf sectionbreak lets you add a section break to the current document. You may apply document formatting options pagesize(), landscape, font(), halign(), margin(), and bgcolor() to the new section. This improvement makes it possible to set different properties for different pages in the document. For example, you can mix the landscape and portrait layouts in a single document.

b. putpdf table now lets you customize the format of all cells in adjacent or noncontiguous ranges. You may apply existing cell formatting options halign(), valign(), border(), bgcolor(), and margin() and existing cell text formatting options font(), bold, italic, strikeout, underline, and nformat(). This improvement makes it possible to format many cells at once.

c. putpdf text and putpdf table have new option allcaps to format text in a paragraph or a cell with all capitalized letters.

d. putpdf table has new options title() and note() to add a title or a note to a table. Option note() may be repeated to add multiple notes with the same format after a table.

e. putpdf table has new option linebreak(#) to add line breaks in a cell.

f. putpdf text and putpdf image with option linebreak(#) now let you add line breaks to a paragraph. Previously, you needed to specify linebreak multiple times to add more than one line break. You may still specify linebreak multiple times as you did before, but we recommend using the new, more concise syntax.

g. putpdf table now places the cell text 1 point higher within the cell when the content of the cell is string. The old behavior is preserved under version control.

h. putpdf table with option margin(), when the margin was set to less than 2 points, did not set the top or bottom margins. This has been fixed.

12. levelsof has the following improvements:

a. levelsof is now faster. Speed is improved when there are many observations (>10,000) and many distinct values (>100) or when the data are integers.

b. levelsof has new option matrow() for saving the levels of the variable to a matrix. Previously, you also had to run tabulate to obtain the levels as a matrix.

c. levelsof has new option matcell() for saving the frequency of each level to a matrix. Previously, you also had to run tabulate to obtain the frequencies as a matrix.

13. drop and keep now store the number of observations dropped in r(N_drop) and the number of variables dropped in r(k_drop). This change means that these commands now clear results previously stored in r() by other commands. The old behavior is preserved under version control.

14. Mata function uniqrows() is now faster, especially when the number of observations is large (>10,000) and the data are integers.

15. Mata class xl() has 28 new member functions for writing Excel cell formats and font formats more efficiently.

16. zipfile and unzipfile now produce and extract compressed files larger than 2 GBs.

17. Community-contributed command dataex, which generates properly formatted data examples for Statalist, is now being distributed with official Stata for the convenience of anyone who wishes to post data on Statalist.

18. estat lcgof, after gsem with option lclass() was used to fit a model where the observed variables predicted latent class outcomes, reported the likelihood-ratio test of the fitted model versus the saturated model even though the methods currently implemented for computing the likelihood-ratio test do not support this situation. estat lcgof no longer reports the likelihood-ratio test in this case.

19. estat sd, after fitting a model with mixed where the residual structure was one of ar, ma, unstructured, banded, toeplitz, or exponential, labeled the residual structure coefficients in r(b) and r(V) in an incorrect order. This has been fixed.

20. etregress has the following fixes:

a. etregress with option lrmodel incorrectly included the coefficients from the treatment equation in the likelihood-ratio model test. This has been fixed.

b. etregress with options lrmodel and hazard(newvar) stored in newvar the hazards that corresponded to the treatment model with only the constant term (intercept). This has been fixed.

21. gsem with option lclass() to specify models with categorical latent variables no longer implies option listwise. The old behavior is preserved under version control.

22. graph export, when exporting a graph with a Greek letter in the italic style to a PDF file, did not display the Greek letter as italicized. This has been fixed.

23. import excel, when the Excel file was specified as a path and filename without an extension, exited with error message "file not found", even when the file existed. This has been fixed.

24. margins has the following fixes:

a. margins after stintreg, when option expression() contained anything other than a single predict() statement, exited with error message "expression is a function of possibly stochastic quantities other than e(b)". This has been fixed.

b. margins with option pwcompare, when used to compute margins for two or more interaction terms that involved the same number of factor variables, incorrectly computed the values of the margins. This has been fixed.

25. melogit and mepoisson with option intmethod(laplace), when used to fit a two-level model, may not have converged when the model could have been identified. This has been fixed.

26. merge with options keep() and assert() did not always verify the required match results before keeping the requested observations. This could result in merge not reporting an error when it should have. This has been fixed.

27. mi impute chained, when option include() specified an expression longer than 80 characters, exited with error message "too few ')' or ']'" instead of producing imputations. This has been fixed.

28. spset, when the data had been xtset with option delta(timevar), reset the data with the periodicity information removed. This has been fixed.

29. svy jackknife, when option mse was specified at estimation or set by svyset, ignored the finite population correction if option fpc() was specified in the first sampling stage. This has been fixed.

30. (Windows) When exporting a Stata graph to the Clipboard or to an EMF file, Stata now includes image data in both the EMF+ and the EMF formats. Stata previously included image data only in the EMF format. The EMF+ format is an extension to the EMF format that supports features such as color transparency. Exporting both formats together allows modern applications that support the EMF+ format to display color transparency in Stata's graphs while still supporting legacy applications that can only display the EMF format.

31. (Windows) graph export, when exporting to a bitmap format such as PNG, did not respect option width() or height() when set graphics was set to off. This has been fixed.

32. (Windows) window fopen always displayed all files instead of the filtered list of files. This has been fixed.

33. (Mac) macOS 10.13 introduced a change that caused Stata's Data Editor to crash when used with an if expression. This has been fixed.

34. (Mac) macOS 10.13 introduced a bug that caused Stata's Do-file Editor to be unresponsive if Stata was installed on an SMB share. This has been fixed.

35. (Unix GUI) The Save preferences dialog did not open to the correct directory. This has been fixed.

-------- update 25sep2017 -----------------------------------------------------

1. Online help and the search index have been brought up to date for Stata Journal 17(3).

2. zip, zinb, and zioprobit no longer support option vuong, which specifies that the Vuong test for nonnested models be reported. This test was used to compare zero-inflated and noninflated models. However, recent work has shown that testing for zero inflation using the Vuong test is inappropriate. If you wish to proceed with the test anyway, use the zero-inflated commands with undocumented option forcevuong.

-------- update 05sep2017 -----------------------------------------------------

1. teffects ra, teffects aipw, teffects ipw, teffects ipwra, stteffects ra, stteffects wra, stteffects ipw, and stteffects ipwra now allow you to specify option vce(cluster clustvar) to compute variance robust to intragroup correlation.

2. asclogit, asmprobit, asroprobit, and nlogit, when option case() specified a variable with values exceeding c(maxlong) = 2,147,483,620, grouped all the associated observations into a single case. This has been fixed.

3. bayesmh and prefix bayes with option initrandom, when the Cauchy or Laplace distribution was used as the prior for some model parameters, incorrectly exited with error message "not implemented". This has been fixed.

4. bootstrap and jackknife, when used to prefix logistic specified with an invalid eform_option, now exits with an error message before entering the replication loop.

5. estimates table with option varlabel, when not all variables in the estimation results were present in the dataset, exited with error message "variable varname not found". estimates table now uses the variable name in the table instead.

6. Expressions using system variables _b and _se, when used to access coefficients and standard errors after mlogit with a value label attached to depvar, did not recognize equations specified using the level values of depvar. This has been fixed.

7. forecast solve, when computing dynamic forecasts with estimates from a model fit using arima with options arima(0,1,1) and sarima(0,1,1,12), incorrectly exited with error message "missing values encountered". This has been fixed.

8. graph ignored legend option ring() if the scheme placed the legend within the plot region. This has been fixed.

9. gsem with more than one outcome variable and with multilevel latent variables, when missing-value patterns in the observed variables caused the omission of one or more of the multilevel groups, incorrectly exited with an error. This has been fixed.

10. heckman with options twostep and vce(bootstrap) incorrectly exited with error message "name conflict". This has been fixed.

11. margins with weighted estimation results, with the first observation in the dataset containing a missing weight value, and with all the independent variables fixed to a constant value incorrectly estimated all margins as zero valued with the "(omitted)" note. This has been fixed.

12. predict, predictnl, and estat wcorrelation in the unusual case that out-of-sample predictions were obtained after menl with option rescovariance(independent, index(varname)) have the following fixes:

a. predict with statistic rstandard or with another statistic but without option fixedonly, when the values of the index varname in the new data did not exactly match the set of values from the estimation sample, gave predictions based on an incorrect residual covariance matrix. predict now correctly computes the predictions for values of varname that match the estimation sample and returns missing values for those that do not.

b. predictnl without option fixedonly to obtain nonlinear out-of-sample predictions after menl when the values of the index varname in the new data did not exactly match the set of values from the estimation sample, gave predictions based on an incorrect residual covariance matrix. predict now correctly computes the predictions for values of varname that match the estimation sample and returns missing values for those that do not.

c. estat wcorrelation, when used to obtain a correlation or covariance matrix for a cluster outside the estimation sample, returned matrices based on an incorrect residual covariance matrix. This has been fixed.

13. predict, predictnl, and estat wcorrelation in the unusual case that out-of-sample predictions were obtained after menl with option rescovariance() or rescorrelation() and suboption by(varname) or with option resvariance() and suboption strata(varname) have the following fixes:

a. predict without option fixedonly, when the by or strata variable in the new dataset contained values that were different from their values in the estimation sample, computed predictions when it should have returned missing values. This has been fixed.

b. predictnl without option fixedonly, when the by or strata variable in the new dataset contained values that were different from their values in the estimation sample, computed nonlinear predictions when it should have returned missing values. This has been fixed.

c. estat wcorrelation, when the by or strata variable in the new dataset contained values that were different from their values in the estimation sample, could produce a correlation or covariance matrix when it should have exited with error message "invalid level specification: <levelspec>". This has been fixed.

14. _ms_extract_varlist, when used after mlogit with a value label attached to depvar, did not recognize equations specified using the level values of depvar. This has been fixed.

15. putdocx and putpdf have the following improvements and fix:

a. putdocx table and putpdf table with output type etable() now support the export of results after estimates table. This makes it easier to export formatted estimation results and export results from multiple models to a single table.

b. putdocx table and putpdf table with output type etable() now support the export of results after margins. This makes it easier to create tables of marginal effects.

c. putdocx describe and putpdf describe, when tablename is specified, now store the number of rows in r(nrows) and the number of columns in r(ncols) for the specified table.

d. putdocx table, when exporting data in memory to the document using output type data() and when there was no space between the equal sign = and data(), incorrectly exited with error message "ddata: invalid table specification". This has been fixed.

16. The SEM Builder, when working with a path model with estimation results from Stata 14, incorrectly exited with an error message when the model was fit again using Stata 15. This has been fixed.

17. streg and stintreg with option distribution(ggamma), distribution(loglogistic), or distribution(lognormal), when set varabbrev off was in effect, incorrectly exited with error message "[/] not found". This has been fixed.

18. tebalance box, when used after teffects with a treatment variable using value labels with spaces, incorrectly exited with error message "invalid syntax". This has been fixed.

19. (Windows) For system administrators, update did not preserve batch mode during its restart sequence, which prevented the use of silent updates. This has been fixed.

20. (Mac) graph export, when multiple graphs were exported and the Graph window preference "Keep list of graphs visible in window" was enabled, could export some images as blank. This has been fixed.

21. (Mac) When using an input source to enter characters into the Command window from a language other than that of the keyboard, pressing Enter to choose from a list of characters from the input source instead submitted the current contents of the Command window. This has been fixed.

-------- update 20jul2017 -----------------------------------------------------

1. Prefix bayes and command bayesmh support three new prior distributions in option prior().

a. t(mu,sigma2,df) specifies the location-scale t distribution with mean mu, squared scale sigma2, and degrees of freedom df.

b. laplace(mu,beta) specifies the Laplace distribution with mean mu and scale beta.

c. cauchy(loc,beta) specifies the Cauchy distribution with location loc and scale beta.

2. New postestimation command estat gofplot after stintreg allows you to visually assess the goodness of fit of the model by plotting the Cox-Snell residuals versus the estimated cumulative hazard function corresponding to these residuals.

3. estat lcmean after fitting a model with gsem or prefix fmm: can now handle up to 256 class-specific outcome means. The previous limit was 70.

4. putdocx has the following improvements and fixes:

a. New command putdocx clear lets you clear the current document in memory without saving.

b. putdocx table has new output type etable. The etable output type lets you more easily add the reported coefficient table(s) from the most-recent estimation command to the active document.

c. putdocx table has new option linebreak to add a line break to a cell.

d. putdocx table now lets you customize the format of all cells in a row or column. You may apply existing cell formatting options halign(), valign(), border(), and shading() and existing cell text formatting options font(), bold, italic, strikeout, and underline(). Previously, you could only edit formatting for a single cell. This improvement makes it possible to format many cells at once.

e. putdocx table now lets you apply option nformat() to the numeric values of all cells in a row, the values of all cells in a column, or the value of a single cell when editing a table. Previously, you could only specify this option when exporting results from a Stata or Mata matrix.

f. putdocx table now lets you apply text formatting option script() to individual cells when editing a table.

g. putdocx table, when option append was specified for a cell, appended the new text or image to the start of the next line rather than the end of the current text or image. This has been fixed.

h. putdocx table, when exporting the data in memory as a table to the document, incorrectly removed the leading and trailing whitespace from string variables. This has been fixed.

5. putpdf has the following improvements:

a. New command putpdf clear lets you clear the current document in memory without saving.

b. putpdf table has new output type etable. The etable output type lets you more easily add the reported coefficient table(s) from the most-recent estimation command to the active document.

c. putpdf table now lets you customize the format of all cells in a row or column. You may apply existing cell formatting options halign(), valign(), border(), bgcolor(), and margin() and existing cell text formatting options font(), bold, italic, strikeout, and underline(). Previously, you could only edit formatting for a single cell. This improvement makes it possible to format many cells at once.

d. putpdf table now lets you apply option nformat() to the numeric values of all cells in a row, the values of all cells in a column, or the value of a single cell when editing a table. Previously, you could only specify this option when exporting results from a Stata or Mata matrix.

e. putpdf table now lets you apply text formatting option script() to individual cells when editing a table.

6. syntax can now parse up to 256 options. The previous maximum was 70.

7. estat bootstrap, after a command that estimates multiple free (auxiliary) parameters and with option vce(bootstrap) or prefix bootstrap specified at estimation, exited with error message "estimation command error: e(k_eq) does not equal the number of equations". Examples of affected commands include meglm, sem, and gsem. This has been fixed.

8. Expressions using system variables _b and _se to access coefficients and standard errors, when set varabbrev was on, returned the value from the first element that had a possible match instead of the element with an exact match. This has been fixed.

9. graph export, when exporting to PDF a map drawn by grmap that contained a polygon with no fill color, incorrectly exited with error message "unable to save PDF file". This has been fixed.

10. graph twoway has the following fixes:

a. graph twoway, when an option included a color style that specified a color adjusted for both intensity using the * operator and opacity using the % operator (for example, red*.8%20), did not use the correct color or effect. This has been fixed.

b. graph twoway, when an option included a linepattern formula with consecutive ls, showed a small gap between segments that should have been solid. This has been fixed.

11. grmap, when a single variable was specified in suboption variables() within option diagram(), did not render the mid-line in the diagrams. This has been fixed.

12. margins has the following fixes:

a. margins with option dydx() could cause Stata to crash if marginal effects were calculated only for a single level of a factor variable. This has been fixed.

b. margins, when an interaction was specified in marginslist after dfactor, gsem, mecloglog, meglm, meintreg, melogit, menbreg, menl, meologit, meoprobit, mepoisson, meprobit, mestreg, metobit, mgarch, nl, nlsur, sem, spivregress, spregress, spxtregress, sspace, or ucm, produced margins for each factor variable separately instead of for the level combinations specified by the interactions. This has been fixed.

c. margins after eintreg, eoprobit, eprobit, or eregress, when an endogenous categorical variable was specified in at() but not specified in either option fix() or option base(), incorrectly used the observed values instead of the values specified in at(). This affected only computations using the endogenous categorical variable in auxiliary equations. This has been fixed.

d. margins, when syntax op(numlist) was used with one of the contrast operators g, gw, h, hw, j, jw, p, pw, q, or qw to request only a subset of all possible contrasts, reported incorrect values of the contrasts. This has been fixed.

e. margins after clogit, with default prediction pu0, when the sum of the values of the first variable in indepvars that contained no 0 values totaled less than 1e-9 in absolute value, reported missing standard errors. This has been fixed.

f. margins after npregress, when marginslist or option at() was specified, incorrectly included in its computations of the reported point estimates observations that were missing in the estimation sample. This has been fixed.

g. margins after npregress, when option over() was specified and when more than one variable was specified in option at(), more than one variable was specified in option dydx(), or more than one variable was specified in the marginslist, labeled coefficients with the wrong variable names. This has been fixed.

13. marginsplot, after margins where a factor-variable name could also have served as an abbreviation for a continuous variable and both variables were specified in option at(), exited with error message "_marg_save has a problem. Margins not uniquely identified." This has been fixed.

14. odbc insert, when using a Unicode ODBC driver, truncated strL variables to 2,045 characters. This has been fixed.

15. pctile with option genp(newvarp) and more than 1,000 quantiles specified in option nquantiles() did not store the percentages in the observations that held the corresponding percentiles. This has been fixed.

16. tebalance box and tebalance density after teffects, when the working directory contained many .gph files that contained spaces in their filenames, incorrectly exited with error message "invalid syntax". This has been fixed.

17. unicode convertfile with option dstcallback() applied default method stop regardless of the method specified. This has been fixed.

18. (GUI) The "Save as" dialog, when used to save a dataset to a format used by an older version of Stata, specified the incorrect version number when it issued command saveold. This has been fixed.

19. (Windows) graph export, when used to export a graph to the legacy WMF format, did not display the plotted values. This has been fixed.

Stata for Windows users are encouraged to use the EMF format when exporting graphs because it supports modern features, such as transparency, that WMF does not.

20. (Mac) graph export, when the exported graph was saved to a Windows share and when the graph was exported to PDF or PNG format, correctly exported the graphs but the files were not visible from the Finder because of a bug in macOS. This has been fixed.

21. (Unix GUI) When using a MobaXterm Windows client, the Find dialog in the Do-file Editor froze if no matches were found for the text specified in the "Find what:" field. This has been fixed.

-------- update 29jun2017 -----------------------------------------------------

1. Online help and the search index have been brought up to date for Stata Journal 17(2).

2. bayes with multilevel models did not allow string variables to be specified as identifiers of random-effects levels. This has been fixed.

3. predict with option latent() after gsem, when the fitted model had paths between endogenous latent variables, incorrectly exited with an error. This has been fixed.

4. putdocx, when an RBG value was specified in option border(), font(), or shading(), did not apply the appropriate color. This has been fixed.

-------- update 19jun2017 -----------------------------------------------------

1. proportion and svy proportion now support new methods to compute limits for confidence intervals. Option citype() lets you specify that the Wilson, exact (Clopper-Pearson), Jeffreys, or Agresti-Coull method be used in addition to the normal (Wald) and logit methods that were previously supported.

2. estat lcgof has the following fixes:

a. estat lcgof, after fitting a model using gsem with option lclass() and 10 or more outcome variables, exited with error message "too many terms in interaction". This has been fixed.

b. estat lcgof, after fitting a model using gsem with both options lclass() and group(), used the wrong number of degrees of freedom and wrong saturated model for the likelihood-ratio test of the fitted model versus the saturated model. This has been fixed.

3. Expressions that included a numeric literal that was longer than 129 characters could cause Stata to crash. This has been fixed so that Stata can handle numeric literals of arbitrary length.

4. gsem has the following fixes:

a. gsem, with multiple outcome variables, multilevel latent variables, and option group(), incorrectly exited with an uninformative error message. gsem now works as documented.

b. gsem, with multilevel latent variables, multiple outcome variables, and missing values in some of the observed variables, incorrectly exited with error message "missing values not allowed in S". This has been fixed.

5. icd10, icd10cm, and icd10pcs have the following fixes:

a. icd10 generate, icd10cm generate, and icd10pcs generate with option range(codelist), when specified with an invalid codelist that included a range ending with a code using the wildcard character (*), incorrectly created a new variable. For example, a range of B10/B29* is invalid, but icd10 generate still created a new variable. These commands now return error message "invalid range specification".

b. icd10 search, icd10cm search, and icd10pcs search, when the specified codelist included a range that ended with a code using the wildcard character (*), incorrectly returned all ICD-10, ICD-10-CM, or ICD-10-PCS codes after the code at the start of the range. For example, a range of B10/B29* is invalid, but icd10 search still returned a list of codes and their descriptions for all ICD-10 codes from B10 and later. These commands now return error message "invalid range specification".

c. icd10 check, icd10cm check, and icd10pcs check, when option version() was not specified, did not check for codes that were valid for previous versions (error 77) or valid for later versions (error 88) of the ICD data. This has been fixed.

6. import delimited, when used with option stringcols() or option numericcols(), could incorrectly import the first row as variable names. This has been fixed.

7. import excel with option firstrow, when the specified workbook contained no data, exited with an uninformative error message. This has been fixed. import excel now executes without an error message but does not create a Stata dataset.

8. margins has the following fixes:

a. margins, when used after eoprobit with option entreat() and when the treatment was interacted with the cutpoints and when margins's computations were restricted to a single treatment group, incorrectly reported some marginal predictions as not estimable. This has been fixed.

b. margins, when used after gsem with option group() and when margins's computations were restricted to a single group, incorrectly reported some marginal predictions as not estimable. This has been fixed.

9. ml, when option technique() specified a switch between nr and bfgs or between nr and dfp and when constraints were specified, incorrectly exited with an error message. This has been fixed.

10. putpdf table, in the rare case when the specified cell contents were another table and the cell comprised multiple cells that were merged vertically using option rowspan() or option span() and the merged cell crossed a page break, drew the bottom border of the cell incorrectly. This has been fixed.

11. stcrreg, with prefix stepwise and parentheses binding the first predictor, incorrectly exited with error message "option compete() required" even when option compete() was specified. This has been fixed.

12. Prefix svy, when version control was set to 11 through 14.2 and when a command specified an interaction using factor-variable notation, exited with an error message such as "c: operator invalid" or "1b: operator invalid". This has been fixed.

13. (Windows) tempfile, in the rare case when the Stata temporary directory path contained certain rarely used Unicode characters and when Windows could not provide an ASCII-compatible version of that name, produced tempfile names that were invalid for some operations. Affected commands exited with an error message reporting that the path was incorrect or the file could not be opened. This has been fixed.

14. (Windows) The Do-file Editor, when content had not been saved before closing, reissued the prompt "Do you want to save the changes to ...?" when "Don't save" was selected the first time. This has been fixed.

15. (Mac) In the Command window or Do-file Editor, dragging a file onto the window did not insert the file path into the window. This has been fixed.

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

See whatsnew14to15.

+---------------------------------------------------------------+ | Help file Contents Years | |---------------------------------------------------------------| | this file Stata 15.0 and 15.1 2017 to present | | whatsnew14to15 Stata 15.0 new release 2017 | | whatsnew14 Stata 14.0, 14.1, and 14.2 2015 to 2017 | | whatsnew13to14 Stata 14.0 new release 2015 | | whatsnew13 Stata 13.0 and 13.1 2013 to 2015 | | whatsnew12to13 Stata 13.0 new release 2013 | | whatsnew12 Stata 12.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 | +---------------------------------------------------------------+ -------------------------------------------------------------------------------


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