__Additions made to Stata during version 9__

This file records the additions and fixes made to Stata during the 9.0,
9.1, and 9.2 releases:

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

-------- **update 20jul2007** -----------------------------------------------------

__Ado-files__

1. **adoupdate** did not set the saved result **r(pkglist)** correctly in all
cases. **adoupdate** should save in **r(pkglist)** a space separated list
of package names that need updating (option **update** not specified) or
that were updated (option **update** specified). In fact, **adoupdate** did
this except when a package list was specified, and then it returned
the package list specified. This has been fixed.

2. **_diparm** reported a conformability error when one of the equations in
a multiple equation specification contained predictors. This has
been fixed.

3. **fcast** produced a syntax error when applied to a VAR with only one
lag. This has been fixed.

4. **irf cgraph** reported an unbalanced-quotes error when there was a
space at the end of a quoted subtitle. This has been fixed.

5. **irf cgraph** produced a syntax error when **level()** was specified for
one of the graphs to be combined. This has been fixed.

6. **nestreg**, when used with **stcox** or **streg**, incorrectly bound standalone
variables into blocks with multiple variables when the first block
was bound in parentheses. For example,

**. nestreg: stcox (x1 x2) x3 x4 x5**

was incorrectly interpreted as

**. nestreg: stcox (x1 x2) (x3 x4 x5)**

This has been fixed.

7. **nlcom** and **testnl** returned an error when used after **svy jackknife**
with replicate weights. This has been fixed.

8. **nlogit** failed to omit observations with missing values in the
dependent variable or the level alternative variables. This has
been fixed.

9. **nlogittree** now allows **if** and **in** qualifiers.

10. **predict** was unable to generate scores after **heckprobit** when the
fitted model did not have a dependent variable specified in the
selection equation. This has been fixed.

11. **stepwise**, when used with **stcox** or **streg**, incorrectly bound
standalone variables into terms with multiple variables when the
first term was bound in parentheses. For example,

**. stepwise, pr(.4): stcox (x1 x2) x3 x4 x5**

was incorrectly interpreted as

**. stepwise, pr(.4): stcox (x1 x2) (x3 x4 x5)**

This has been fixed.

12. **sts graph** with the **hazard** option would return the error message
"variable __ub not found" when variable abbreviations were turned
off. This has been fixed.

13. **suest** complained about missing score values when used after **heckman**
and the dependent variable contained missing values. This has been
fixed.

14. **sunflower** failed to draw the final segment of a hexagon's outline.
This was only noticeable on Mac computers and has been fixed.

15. **svy: tabulate** with the **subpop()** option miscounted omitted strata
when the row or column variable contained missing values for all of
the subpopulation observations. This has been fixed.

16. **testnl** reported incorrect standard errors when the estimated
parameter values were close to the boundary of the support for the
functions specified in the user defined constraints. For example,

**. testnl sqrt(_b[x1]^2 - _b[x2])**

resulted in incorrect standard errors if the squared value of **_b[x1]**
was very close to the value of **_b[x2]**, leading to the square root of
a negative number. This has been fixed.

17. The **vce(bootstrap)** and **vce(jackknife)** options were not aware of the
**noconstant** option when presented with indicator variables that were
collinear with the constant. This resulted in an incorrectly
dropped indicator variable. This has been fixed.

18. **xi** with the **noomit** option failed to keep track of one of its
generated indicator variables. On subsequent **xi** calls with the
**noomit** option, **xi** failed to remove one of the previously generated
indicator variables. This has been fixed.

19. **xi** with the **noomit** option generated variables without the default **_I**
prefix when the **prefix()** option was not specified. This has been
fixed.

__Stata executable, all platforms__

20. The Data Editor could not display more than 20 variables at a time.
This limitation affected users with very high-resolution displays.
This has been fixed.

21. **fcast** after **var** produced an error when the VAR contained only one
endogenous variable. This has been fixed.

22. Using a Stata matrix function that returns a scalar within an **if**
clause of a **regress** command caused **e(b)** and **e(V)** to be saved
incorrectly, thereby causing postestimation commands not to work
after **regress**. This has been fixed.

23. Inverse probability functions **invbinomial()**, **invFtail()**, **invibeta()**,
**invnFtail()**, and **invnibeta()** could enter an endless loop when the
unique solution could not be found due to the limit of machine
precision. Now, as with other inverse functions, they return
missing in such cases. Also **nibeta()** now returns 1 if *x* > 1 and
noncentrality parameter *L* __<__ 1.

24. **infile** (fixed format) now accepts a **%**[**#**]**S** *infmt*. **%**[**#**]**S** is similar
to **%**[**#**]**s**, but **%**[**#**]**S** preserves leading and trailing spaces; **%**[**#**]**s**
does not.

25. **infile** (fixed format) treated the ASCII character 255 (hexidecimal
0xFF) as an end of file marker and would stop reading the file if it
encountered it. This character is now treated just as any other
ASCII character.

26. Mata miscompiled the expression **y[p] = y**. The notable feature of
this expression is that **y[p]** appears on the left-hand side of the
expression and **y**, the same variable, appears on the right. This has
been fixed.

27. Mata's **bufput()** function ignored the *offset* argument when it copied
elements into the buffer. This has been fixed.

28. **odbc load** incorrectly imported real data as missing values for
values that were within 27 of the upper boundary of integer data
types. This has been fixed.

29. Returned values from r-class programs could mistakenly include **r()**
values from the last command the ado-file ran under certain rare
circumstances. This has been fixed.

30. **sreturn list** mistakenly listed matrices returned in **r()**. This has
been fixed.

31. **svy**, when used with regression commands such as **regress**, incorrectly
reported missing standard errors when the **subpop()** option excluded
one or more strata with a single sampling unit. Here, **svy**
incorrectly counted the out-of-subpopulation singleton strata even
though they were dropped from the estimation sample. This has been
fixed.

32. **unabcmd** now recognizes the abbreviations for the following formerly
built-in commands: __cnr__**eg**, __logi__**t**, __mlog__**it**, __olog__**it**, __oprob__**it**, __prob__**it**,
__reg__**ress**, and __tob__**it**.

__Stata executable, Windows__

33. In certain circumstances double buffering could prevent some Stata
windows from rendering correctly when used with multiple monitors.
This has been fixed.

34. In the Do-file Editor, the Search -- Find Next menu item and its
corresponding F3 keyboard shortcut were not working. This has been
fixed.

35. The Stata Data Editor displayed an empty system menu under Windows
Vista. This has been fixed.

36. Two scroll bars sometimes appeared on the Variables window under
Windows Vista. This has been fixed.

__Stata executable, Mac__

37. The Output Settings dialog is no longer displayed when translating a
SMCL file to another format.

38. The scroll bar thumb would not scroll past the 32,767th line. This
has been fixed.

39. Clicking multiple times on the Apply button in the Variable
Properties window of the Data Editor could crash Stata on early
versions of Mac OS X. This has been fixed.

40. Graphs copied to the Clipboard could possibly contain two filled
rectangles as the background instead of just one. This has been
fixed.

__Stata executable, Unix__

41. Remotely launching Stata (for example, **ssh mymachine xstata-se**)
would cause Stata to believe it was running in batch mode. This has
been fixed.

-------- **update 22may2007** -----------------------------------------------------

__Ado-files__

1. **binreg** reported the error "option rrr not allowed" when the **rr**
option was used with **vce(bootstrap)** or **vce(jackknife)**. This has
been fixed.

2. **ivprobit** did not apply **if** or **in** restrictions specified by the user
when checking for perfect predictors. Also **ivprobit** would exit with
an error if the two-step estimator was used and the reduced-form
model had perfect predictors. These issues have been fixed.

3. **jackknife** reported an error when weights were specified with
commands that have multiple equation specifications using the
parenthetical notation. This has been fixed.

4. A note has been added to the **ksmirnov** output to warn users when ties
are found in the variable whose distribution is being tested.

5. **nlogit** with **vce(robust)** neglected to cluster on the **group()** variable
when computing standard errors. This has been fixed.

**nlogit** with **robust** did not have this problem.

6. **roccomp** could produce different chi-squared tests when monotonic
transformations were applied to the ratings. This has been fixed.

7. **rreg** reported the error "option noheader not allowed" when used with
the **bootstrap** prefix. This has been fixed.

8. **svyintreg** reported the error "initial values not feasible" for
constant-only models. This has been fixed. **svy: intreg** did not
have this problem.

9. **svy: tabulate** did not always post the correct **e(sample)** when one or
more strata were omitted because they contained no subpopulation
members; some observations from the omitted strata may have been
left in the estimation sample identified by **e(sample)** when they
should not have been. This has been fixed.

10. **svy: tabulate** failed to report on omitted strata when options **ci** or
**se** were not specified. This has been fixed.

11. **xtgee** reported the error "option level() incorrectly specified" when
used with **vce(jackknife)**. This has been fixed.

12. **xtlogit**, when reporting the number and size of groups, ignored
groups with more than 32,740 observations. This has been fixed.

13. **xtreg** exited with an error message that the "independent variables
are collinear with the panel variable" if the scale of the panel
variable was too large. This has been fixed.

-------- **update 07mar2007** -----------------------------------------------------

__Ado-files__

1. **ctset** with the option **clear** did not clear the ct setting. This has
been fixed.

2. **estimates table** reported a conformability error when used after
**mlogit** when one or more equations contained a space. This has been
fixed.

3. **icd9** and **icd9p** were updated to use codes through the V24 update
released on October 1, 2006. The descriptions were also updated so
that the description of invalid codes that were valid in the past
end in a hash mark (#).

4. The **matrix input** dialog did not create a symmetric matrix correctly
when negative values were specified. This has been fixed.

-------- **update 16feb2007** -----------------------------------------------------

__Ado-files__

1. Online help and the search index have been brought up to date for
Stata Journal 7(1).

2. **cttost** exited with an error message if the total number of failures
and censored observations in a certain period was equal to the total
number of individuals in the study, and if the number of failures
was not smaller than the number of individuals in the study two
periods before. Now, in this setting, **cttost** produces the correct
output.

3. **svy jackknife** reported an incorrect subpopulation size and number of
subpopulation observations when the prefixed estimation command
dropped observations due to missing values. This has been fixed.

-------- **update 06feb2007** -----------------------------------------------------

__Ado-files__

1. **lookfor** can now search for phrases enclosed in double quotes.

2. **predict, scores** after **ologit** and **oprobit** incorrectly returned
missing values for models having one cutpoint. The score variable
for the cutpoint contained missing values in observations where the
dependent variable equals the second ordered value. This has been
fixed.

3. **svy: ologit** and **svy: oprobit** produced incorrect standard errors when
used to fit models where the dependent variable had only two values.
This was caused by a problem with predicted equation-scores for
these models, which has been fixed.

__Stata executable, Windows__

4. (Windows Vista only) Pressing Ctrl-C to copy text from the Command
window caused Stata to crash. This has been fixed.

5. Printing the Do-file Editor from the File-Print menu or the main
Stata toolbar would print spurious pages. This has been fixed.

-------- **update 24jan2007** -----------------------------------------------------

__Ado-files__

1. **hausman** failed to correctly recognize equation names in **mlogit**'s
estimation results when one or more equation names contained a
space. This has been fixed.

2. **hausman** now exits with an error message if either model was fitted
with **cluster()** or with **pweight**s. The performance of the test can be
forced by using **force**.

3. **lrtest** exited with an error message if the dependent variable of one
of the models being compared contained time-series operators and the
restricted model was specified before the unrestricted model in the
call to **lrtest**. This has been fixed.

4. **statsby** did not respect the** level()** option. This has been fixed.

5. **svy: tabulate** failed to report standard errors or confidence
intervals if they were requested when the BRR variance estimation
method was used. This has been fixed.

6. **xtpoisson, re normal** applied the **noskip** option by default; this has
been fixed.

-------- **update 12jan2007** -----------------------------------------------------

__Ado-files__

1. **adjust** has been modified to perform intermediate computations by
using double precision. The resulting gain in precision is usually
evident in the fourth or fifth decimal place.

2. **glm**'s Newey-West standard-error calculation, when used with the
Anderson kernel function, failed to account for this kernel's being
nonstandard. The Anderson kernel is nonstandard because (1) it
permits a noninteger lag, that is, a bandwidth and (2) the kernel
weights are nonzero outside the bandwidth/lag range. **glm** now
accounts for both of these nonstandard behaviors.

3. **gprobit** computed incorrect weights for the GLS stage of the
computations. As a result, the coefficient estimates were
inefficient. This has been fixed.

4. **graph** now better supports variable labels that are by default used
to label axes and legend keys. Labels with unbalanced parentheses
and brackets are now supported, as well as labels with separately
double-quoted strings. As with text appearing in graph options,
each double-quoted section of text now appears on its own line in
the title or label. Results for options **xtitle()**, **ytitle()**,
**legend(label())**, and **legend(order())** are similarly improved.

5. **pac** exited with an error when used with the option **yw** if **varabbrev**
was set to **off**. This has been fixed.

6. **stcox** and **streg** now exit with an error when option **vce(bootstrap)** or
**vce(jackknife)** is specified when weights were **stset**. Prior to this
update, weights corresponding to the resampled observations were
used as is for computing the replicates.

**bootstrap** does not allow weights.

**jackknife** allows weights but currently cannot handle them correctly
for **stcox** and **streg**.

7. **suest** allowed survey estimation results that used poststratification
to be mixed with estimation results that did not. This has been
fixed.

**suest** can be used only with survey estimation results that are
consistent with the poststratification information in the current
**svyset**tings.

**suest** will now verify that the poststratification information is
consistent between **svyset** and the specified estimation results.

**suest** will exit with an error message when non-**svy** estimation
results are specified and poststratification variables have been
**svyset**.

8. **svy**, in the 21nov2006 update, was changed to use **encode** on
string-valued stratum and sampling unit variables. This change
restricted the number of sampling units to no more than 65,536 if
the variable that identifies sampling units was a string variable.
This upper-limit restriction has been removed.

To decrease the amount of overhead, consider replacing string
variables that identify the strata and sampling units with encoded
numeric variables.

9. **svy: tabulate** failed to drop categories of the row and column
variables that were exclusively defined in omitted strata. This has
been fixed.

10. **xtrc** exited with an error when used with more than 1,000 panels.
This has been fixed.

__Stata executable, all platforms__

11. **in** now accepts **F** as a synonym for **f** as shorthand for the beginning
of the **in** range. **L** has been added as a synonym for **l** as shorthand
for the end of the **in** range.

12. Mata has the following new features and fixes:

A. New function **_diag()** replaces the principal diagonal of a matrix
with a specified vector or scalar.

B. New (undocumented) functions **trace_AVBV()** and **trace_ABBAV()**
compute the trace of special matrices used for spatial
statistical calculations.

C. Appending rows to a matrix (**[M-2] op_join**) in Mata is now
faster.

D. Matrix multiplication, *A***B*, is now faster when *A* contains zeros.

E. **cross****(***X***, ***Z***)** is now faster when *X* contains zeros.

F. **select****(***X***, ***v***)** exited with a conformability error when *X* was a row
vector and *v* was a scalar; this has been fixed.

G. **sizeof()** could not be used; including **sizeof()** in code resulted
in the error "'sizeof' found where almost anything else
expected" during compilation. This has been fixed.

13. Two new functions have been added for use with Stata **plugins**:
**sstore()** and **sdata()**. **sstore()** stores string data in Stata and
**sdata()** reads string data from Stata; see
http://www.stata.com/plugins/.

14. The random-effects panel-data estimators **xtlogit, re**; **xtprobit, re**;
**xtcloglog, re**; **xtpoisson, re normal**; **xtintreg**; and **xttobit** could
ignore large panels when fitting the model. This was caused by an
internal numerical underflow, although no warning message was
presented, and this is now fixed. What follows is a description of
the bug and how to determine whether the bug affected any models you
have fitted.

The above commands use adaptive Gaussian quadrature to estimate
panel-level likelihoods, where each is the integral of some function
(call it G) times the normal density. The function G is the
panel-level likelihood conditional on the random effects and thus is
itself a product of observation-level densities of the observations
within that panel. Typically, but not necessarily, densities are
between 0 and 1, and in fact they are between 0 and 1 for
binary-data models. In any case, the product of these densities
over a large panel (a panel with many observations) can be small.
If the values got too small, they were treated as if they were
exactly zero. The net effect was that such panels were ignored in
the estimation of the parameters.

This problem arose in datasets with large panels, that is, datasets
with many observations per panel. How large a panel needed to be in
order to be affected depended on the average observation-level
density for the panel. If this density (probability in a binary
model) averaged 10%, underflow would occur in panels containing 300
or more observations. If this density averaged 50%, underflow would
occur in panels containing 1,000 or more observations. At 90%, the
underflow would occur with panels containing 6,500 or more
observations. Since most users use these models to estimate for
large N, small T data, this large-T problem did not affect many
users.

Also, regardless of panel size, calculations for determining the
optimal Gaussian quadrature points were also subject to a similar
underflow problem. This has also been fixed. The net result of
this fix, however, is merely that quadrature points are now more
precisely estimated, resulting in better-approximated
log-likelihoods. The gain in precision is slight, usually being in
the fourth or fifth decimal place of the estimated regression
parameters.

15. Stata/MP with more than two processors had a bug in **tobit** and **cnreg**.
Let N_PROCS be the number of processors with N_PROCS>2, and let
N_OBS be the number of observations for the model. When N_OBS/100>1
and N_OBS/100<N_PROCS, Stata/MP made a mistake in determining the
actual number of processors to use. The mistake could crash Stata
or produce wrong results. This is fixed.

16. When exporting a graph with the Stata logo as a PostScript file,
Stata would specify the wrong font name for the Stata logo. This
caused applications such as Adobe Illustrator or Distiller to
complain that a font wasn't found. This bug is not serious since
those applications would substitute a font for the incorrectly named
font. This has been fixed.

Stata also now outputs characters outside the range of 7-bit ASCII
characters in PostScript's preferred octal format. Users should see
no difference when their PostScript or EPS files are rendered or
printed.

__Stata executable, Windows__

17. A process spawned by **winexec** would inherit and hold open any file
handles previously opened by Stata (such as an open log file) even
if Stata later closed those file handles. Stata now does not allow
a process spawned by **winexec** to inherit Stata's file handles.

__Stata executable, Mac__

18. You may now change the font in the Results, Viewer, Command, Data
Editor, Do-file Editor, Review, and Variables windows. You may
select a fixed-width font only in the Results, Viewer, Data Editor,
Do-file Editor, and Variables windows.

Stata continues to allow you to change the font in the Graph window.

19. Stata previously output text from graphs and SMCL exported to
PostScript by using the incorrect font encoding. This caused
characters outside the 7-bit ASCII range to not translate properly
when using an external application's PostScript interpreter. This
has been fixed.

-------- **update 30nov2006** -----------------------------------------------------

__Ado-files__

1. Online help and the search index have been brought up to date for
Stata Journal 6(4).

__Stata/MP executable, 64-bit Itanium Windows__

2. The 64-bit Itanium Windows Stata/MP executable could have problems
completing an **update executable** command; this has been fixed.

-------- **update 21nov2006** -----------------------------------------------------

__Ado-files__

1. **bstat** exited with an error indicating a name conflict when the
**stat()** option contained a matrix with equation names. This has been
fixed.

2. **cnreg**, **logit**, **mlogit**, **ologit**, **oprobit**, **probit**, **regress**, and **tobit**
are now noticeably faster (up to five times faster) when used within
large loops, such as **statsby** with datasets that have many small
*by*-groups.

3. **histogram** with an *if* condition that excluded one of the categories
in the **by()** option resulted in an extra empty plot when option
**addlabels** or **kdensity** was specified or resulted in an error when
option **normal** was specified. This has been fixed.

4. **nestreg** reported confidence intervals for the default confidence
level even when a different confidence level was specified in the
**level()** option of the prefixed command. This has been fixed.

5. **quadchk** exited with an error when used to check models using the
**offset()** option. This has been fixed. **quadchk** also ignored weights
and the **collinear** and **constraints()** options. This has been fixed.

6. **svy** now identifies the rare case when a sampling unit bridges
adjacent strata in multistage designs. **svy** treats all sampling
units as nested within strata and units from prior stages.

7. **svy: heckman** excluded from the estimation sample observations with a
missing value in the dependent variable when it should not have.
This has been fixed.

8. **svy jackknife** reported incorrect values for the sample size,
population size, and number of strata when some of the strata were
omitted. This has been fixed.

9. **svy** linearization is now two to five times faster. **svy: mean**, **svy:**
**proportion**, **svy: ratio**, and **svy: total** are considerably faster when
the **over()** option identifies many subpopulations.

10. **svy: mean**, **svy: proportion**, **svy: ratio**, and **svy: total** reported
missing standard errors due to strata with one sampling unit even
when a **subpop()** option excluded the offending strata. This has been
fixed.

11. **svy: mean**, **svy: proportion**, **svy: ratio**, and **svy: total** failed to
omit strata that contain no subpopulation members. This has been
fixed.

12. **svy: tabulate twoway**, when option **row** or **column** is specified, uses
the ratio estimator to compute cell proportions relative to the row
or column margins. For direct standardization, option **stdize()**, a
better estimator is available: the standardized ratio estimator.
This standardized ratio estimator is now used in place of the simple
ratio estimator when **stdize()** is specified.

13. **svy: tabulate** incorrectly reported zero standard errors when used
with survey data that had strata with one sampling unit. Standard
errors and confidence intervals are now suppressed with a note
explaining why when the **se** or **ci** option is specified in such cases.

14. **xtcloglog** and **xtpoisson** with the **re** and **normal** options would ignore
constraints on **lnsig2u**. This has been fixed.

15. **xtintreg**, **xtlogit**, **xtprobit**, **xtpoisson** with the **re** and **normal**
options, and **xttobit** all accepted the **collinear** option but exited
with an error when used with a constraint to define a collinear
model. This is now fixed.

16. **xtintreg** and **xttobit** ignored constraints on **sigma_u** and **sigma_e**.
This has been fixed.

17. **xtline** has new **i()** and **t()** options. The **i()** option allows the
panels (groups) to be identified using a string variable. The **t()**
option allows a variable with noninteger values and repeated values
within group.

The important effect of this is that **xtline** can now be used to
easily create two-way graphs with separate plots for each level of a
categorical variable. For example,

**. xtline mpg, t(displacement) i(rep78) overlay**

creates five line plots of **mpg** versus **displacement**, one for each of
the five levels of **rep78**.

These plots can be rendered as other **twoway** plottypes by adding a
**recast()** option. For example, adding **recast(scatter)** to the command
above will produce five scatterplots rather than five line plots.

__Stata executable, all platforms__

18. In Mata's **ghk()** function, when the covariance matrix is not positive
definite, the **ghk()** Cholesky routine will pivot out the offending
row and column of the covariance matrix. The **ghk()** function failed
to permute the corresponding upper bound of integration. This has
been fixed.

__Stata executable, Windows__

19. Stata for Windows now supports Microsoft Windows Vista.

__Stata executable, Mac__

20. For efficiency, when resizing a graph window, Stata draws a preview
of the graph rather than completely redrawing the graph. However,
the preview could be drawn at the wrong size if the window is
quickly resized by jerking the mouse. This has been fixed.

21. When there is a --**more**-- condition in the Results window, Stata will
continuously poll for a key press to clear the --**more**-- condition,
which causes the processor usage to be very high. Stata now does
not poll as often when there is a --**more**-- condition, causing the
processor usage to be very low.

22. Some graph markers and objects would not be drawn with an outline in
the Graph window when using the Quartz 2D graphics engine. This has
been fixed. This bug did not affect EPS and PS files.

23. The console version of Stata/MP for Intel-based Macs could not
correctly determine whether an executable update was available.
This has been fixed.

24. The Do-file Editor will now treat a tab character as multiple spaces
when printing a do-file. The Do-file Editor previously treated a
tab character as one space when printing.

25. Stata now creates the Stata license file as stata.lic instead of
Stata.Lic after license initialization. This distinction allows Mac
OS X systems using the UFS file system to read the license file
because the UFS file system is case sensitive. The default file
system for Mac OS X is HFS+ and is not case sensitive, and it will
accept both stata.lic and Stata.Lic. Current Stata users who are
using the UFS file system can simply rename the file Stata.Lic to
stata.lic. Current Stata users who are using the HFS+ file system
do not need to rename Stata.Lic to stata.lic.

__Stata executable, Unix (GUI)__

26. Clicking a link in the Results window to open a help file in the
Viewer could cause Stata to crash. This has been fixed.

-------- **update 27oct2006** -----------------------------------------------------

__Ado-files__

1. For **bootstrap** and **jackknife**, if a user specified a **cluster()** option
before the colon and a conflicting **cluster()** option after the colon,
the second **cluster()** option was being ignored. An error message is
now issued.

2. If a user-written e-class command did not set **e(cmd)**, **bootstrap**,
**jackknife**, **svy brr**, and **svy jackknife** exited with an error message.
This has been fixed.

3. If a URL longer than 44 characters was used to open a dataset,
**codebook**'s **problems** option reported an error. It no longer does so.

4. If a value label had unbalanced quotes, **codebook** stopped with an
error message. This has been fixed.

5. **ivprobit** did not set **e(depvar)** when the **twostep** option was used.
This has been fixed.

6. **screeplot**'s **addplot()** option failed with multiple plots. This has
been fixed.

7. **svy: regress** when typed in the abbreviated forms **svy: regr**, **svy:**
**regre**, and **svy: regres** (with one s), failed to adjust standard
errors for mean squared error of the regression. This has been
fixed.

There was no problem when you typed **svy: regress** or its most common
abbreviation, **svy: reg**.

8. Previously, **xtgee** could produce model results in which the estimated
working correlation matrix was not positive definite, as a proper
correlation matrix must be. Now, when **xtgee** reaches a solution for
which the working correlation matrix is not positive definite, the
command issues an error message stating that convergence has not
been achieved.

Also, for exchangeable correlation matrix structures, the estimation
algorithm resets the correlation parameter to its minimum bound
implied by the maximum panel size when the correlation parameter is
infeasible. This reset will help models that can converge to do so,
although if the data are incompatible with an exchangeable
correlation structure, then convergence cannot be achieved.

-------- **update 06oct2006** -----------------------------------------------------

__Ado-files__

1. **areg** with option **vce(bootstrap)** or **vce(jackknife)** resampled
observations instead of the groups identified by the **absorb()**
variable. This has been fixed.

2. **estat mfx** after **asmprobit**, in extremely rare cases, gave incorrect
standard errors for the marginal effects of discrete case-specific
covariates. These standard errors were incorrect only when
different individuals faced different alternatives and when the
subsample used to compute marginal effects was missing the first
alternative. This has been fixed.

Standard errors of marginal effects for both discrete and continuous
alternative-specific covariates, as well as those for all continuous
covariates, were always correct.

3. Bayesian information criterion (BIC) calculations within **glm**, with
**family(binomial)** and an explicit binomial denominator, were based on
an incorrect total number of observations. This has been fixed.

4. **quadchk** used with version control for versions before 9 exited with
an error. This has been fixed.

5. **rolling** returned an error message when used with panel data and an
**if** qualifier that disqualified an entire panel. This has been
fixed.

6. **svyset** with the **noclear** option resulted in one sampling stage even
when the original settings had more than one stage. This has been
fixed.

7. **xtgee**, when used with **family(binomial)**, issued an "estimates
diverging" error when used with fractional dependent variables
containing values equal to one. This has been fixed.

8. **xtset**, when both panel and time variables were specified,
misreported unbalanced panels as either strongly balanced or weakly
balanced. This has been fixed.

__Stata executable, all platforms__

9. **pctile** and **xtile**, when dividing the data into more than 1,001
quantiles, would occasionally put data into the wrong quantile
because of small rounding errors. This has been fixed.

10. **svy** reported missing standard errors when there were strata with one
certainty unit. This has been fixed.

__Stata executable, Windows__

11. Pressing Page Up/Down in the Command window no longer sounds a
system beep.

12. The Value labels list box from the **label define** dialog could obscure
part of the Define label names list box on systems where the window
appearance setting was changed. This has been fixed.

__Stata executable, Mac__

13. Stata prevented windows from being dragged above the menu bar on
Macs with multiple monitors. This has been fixed.

-------- **update 28aug2006** -----------------------------------------------------

__Ado-files__

1. Online help and the search index have been brought up to date for
Stata Journal 6(3).

2. **clogit** standard errors were not correct if **pweight**s were used
without option **cluster()**. This problem has been fixed.

3. **cloglog** and **xtcloglog** reported error messages when the **collinear**
option was specified even when constraints were specified that
properly identified the model. This problem has been fixed.

4. **graph bar**, in rare cases, produced bars based on incorrect values
when a variable was repeated for a statistic and followed by another
statistic and variable. This problem has been fixed.

5. **histogram**, **kdensity**, **twoway histogram**, **twoway kdensity**, **twoway lfit**,
and **twoway qfit** produced unusable graphs when the **by()** option
identified one or more groups that resulted in an empty graph. This
problem has been fixed.

6. **linktest**, when run after **glm**, used a Stata 6 version of **glm** rather
than the modern one. This problem has been fixed.

7. **ml** commands using **fweight**s with options **technique(bhhh)** or **vce(opg)**
produced incorrect standard error estimates. This problem has been
fixed.

8. **ml** exited with an unhelpful error message when checking for
convergence and the likelihood evaluator resulted in an unstable or
asymmetric Hessian matrix. This problem has been fixed.

9. **nestreg** returned an error message when **svy** omitted strata due to
zero weights or subpopulation estimation. This problem has been
fixed.

10. The function evaluator version of **nl** exited with an error if the
user tried to pass any options to the function evaluator program.
This problem has been fixed.

11. **nlogit**'s robust standard-error calculation failed to take into
account the within-choice-set correlation inherent to **nlogit** models.
This problem has been fixed.

12. **pcamat**, **factormat**, **corr2data**, and **drawnorm** now allow singular
correlation (or covariance) structures. **pcamat**, **factormat**,
**corr2data**, and **drawnorm** have new option **forcepsd** that maps a matrix
intended to be correlations (covariances) into a
positive-semidefinite singular correlation (covariance) matrix.

13. **ranksum** now allows **by()** to contain a string variable.

14. **_rmdcoll**'s **nocollin** option has been renamed to **normcoll**. The
**nocollin** option will continue to work as a synonym for **normcoll** but
will become undocumented.

15. **stepwise** dropped variables collinear with the constant even when the
**noconstant** option was specified in the prefixed command. This
problem has been fixed.

16. **svy jackknife** sometimes reported the number of subpopulation
observations and subpopulation size even when the **subpop()** option
was not specified. This problem has been fixed.

17. **table**'s **format()** option was not applied to the frequency column when
the frequency was greater than nine digits. This problem has been
fixed.

18. **testnl** with an expression that included a comma produced an error
message. Expressions that are bound in parentheses now allow
commas. For example, **testnl _b[x]=M[1,3]** is still not allowed,
whereas **testnl (_b[x]=M[1,3])** is now allowed.

19. **tsset** and **xtset** now indicate whether the panels are balanced.

__Stata executable, all platforms__

20. **asmprobit** sometimes incorrectly marked the estimation sample when
invalid observations were dropped. This problem has been fixed.

21. **graph twoway** plot types **pcarrow**, **pcbarrow**, and **pcarrowi** may now be
drawn on plot regions with log scales or reversed scales.

22. Once **logit**, **logistic**, or **probit** reported an error 902 (no room to
add more variables due to width), the error would continue being
reported on subsequent attempts even after the memory had been set
to a value large enough to perform the estimation. This problem has
been fixed.

23. Mata function **findexternal()** is updated to allow returning pointers
to functions as well as matrices, vectors, and scalars.

24. **svy brr**, **svy jackknife**, **svy: mean**, **svy: proportion**, **svy: ratio**, and
**svy: total** did not omit strata for which every observation contained
a zero sampling weight. This problem has been fixed.

__Stata executable, Windows and Mac__

25. When Stata checked if an update was available from Stata's website
but no updating was performed, it displayed a note that an update
was available after Stata was launched. In rare cases, Stata's
update status was not synchronized with Stata's website. This
problem has been fixed.

-------- **update 31jul2006** -----------------------------------------------------

__Ado-files__

1. Estimation commands called with options **vce(bootstrap)** or
**vce(jackknife)** now report a note when a variable is dropped due to
collinearity.

2. **asmprobit** when used with **vce(bootstrap)** or **vce(jackknife)** now
requires that **basealternative()** and **scalealternative()** options be
specified.

3. **heckman****,** **twostep**'s dialog incorrectly displayed a button for
launching the **svyset** dialog. This problem has been fixed.

4. **mlogit**, **mprobit**, and **slogit** when used with **vce(bootstrap)** or
**vce(jackknife)** now require that the **baseoutcome()** option be
specified.

5. **nlogit** issued a "too many values" error if the number of groups
exceeded **tabulate**'s row limit; see limits. This problem has been
fixed.

6. **nlogit** reported incorrect degrees of freedom for the LR test of
homoskedasticity if constraints were applied to the regression
parameters. This problem has been fixed.

7. **ologit** sometimes reported a missing model F statistic when it should
have reported a missing model LR statistic. This problem has been
fixed.

8. **rolling** failed to compute results with commands that generate new
variables. This problem has been fixed.

9. **svy** now omits strata that have zero-valued sampling weights in every
observation.

10. **tabstat** with the **by()** option truncated the word Total when the **by()**
variable's name and value label were fewer than 5 characters. This
problem has been fixed.

__Stata executable, all platforms__

11. The following commands now have the **collinear** option: **arch**, **arima**,
**asmprobit**, **biprobit**, **clogit**, **cloglog**, **frontier**, **gnbreg**, **heckman**,
**heckprob**, **hetprob**, **intreg**, **ivprobit**, **ivtobit**, **mlogit**, **mprobit**,
**nbreg**, **nlogit**, **poisson**, **_rmcoll**, **_rmdcoll**, **slogit**, **streg**, **treatreg**,
**truncreg**, **xtcloglog**, **xtfrontier**, **xtintreg**, **xtlogit**, **xtnbreg**,
**xtpoisson**, **xtprobit**, **xttobit**, **zinb**, **zip**, **ztnb**, and **ztp**.

12. **_parse expand** and **_parse canonicalize** now have the **gweight** option
for parsing global weights.

13. **datasignature** did not always calculate the correct checksum when
using the **fast** option on Power PC-based Mac, AIX, Solaris, HP/9000,
HP/Itanium, and Irix computers. This problem has been fixed.

14. Mata's **allof()** and **anyof()** functions did not allow pointer or
structure arguments and, in fact, could crash when such arguments
were specified. This problem has been fixed, and the functions may
be used with pointer or structure arguments.

__Stata executable, Windows__

15. The ability to submit multiple commands to the Command window has
been reinstated.

16. Review window selections are now retained.

17. Under rare conditions, programmable dialogs could show up with a
black background. This problem has been fixed.

__Stata executable, Unix (GUI)__

18. The previous update of Stata introduced a problem where printing a
Viewer window would cause the Command edit field to not accept any
keyboard input. This problem has been fixed.

19. When building a list of fonts for display in the Graph Preferences
dialog, Stata queries the font server for a list of fonts and
expects the fonts to be specified in the X Logical Font Description
(XLFD) format. If a font was not in the XLFD format, Stata crashed.
Stata now ignores fonts that are not returned in the correct format.

-------- **update 13jul2006** -----------------------------------------------------

__Stata executable, Windows__

1. Certain windowing layouts could cause Stata to incorrectly display
windows upon starting up. For example, a pinned Variable pane could
cause a portion of Stata's background to not be drawn properly.
This problem has been fixed.

-------- **update 11jul2006** -----------------------------------------------------

__Ado-files__

1. **clogit** now accepts **pweight**s.

2. **tetrachoric**'s default algorithm for computing tetrachoric
correlations has been changed from the Edwards and Edwards estimator
to a maximum likelihood estimator. The Edwards and Edwards
estimator performed poorly with skewed data. **tetrachoric**'s features
are now similar to those of **spearman** and **ktau**. Standard errors and
two-sided significance tests are now included. A frequency
adjustment when one cell has a zero count is now available with the
**zeroadjust** option. This change has been made without version
control; to get the old behavior, the **edwards** option can be used.

3. **xtmixed** has new option **collinear**, which tells **xtmixed** to retain
collinear variables within random-effects equations.

-------- **update 06jul2006** -----------------------------------------------------

__Ado-files__

1. **clear** has been updated to clear the new timers described below for
profiling ado and Mata code.

2. **heckman**'s dialog incorrectly displayed a button for launching the
**svyset** dialog. This problem has been fixed.

3. **lowess** ignored certain suboptions of the **by()** option. This problem
has been fixed.

4. **predict, scores** after **streg, distribution(exponential)** computed
scores incorrectly when *t0* was not 0. This problem has been fixed.

5. **survey** estimation commands that allow the **technique()** option
reported incorrect standard errors when **technique(bhhh)** was
specified. This problem has been fixed.

6. **svy: tabulate oneway** reported an error message when specified with a
constant variable. This problem has been fixed.

7. **tsline** misread or would not accept certain *datelist* specifications
in the **tlabel()** and **tline()** options. This problem has been fixed.

8. **twoway fpfitci** did not draw the confidence interval when the option
**estcmd()** was used to specify some estimation commands. This problem
has been fixed.

__Stata executable, all platforms__

9. Stata/MP, the parallel-processing flavor of Stata, is now available
for 64-bit Solaris on x86-64 hardware. The supported platforms for
Stata/MP are Windows (32-bit, 64-bit Itanium, and 64-bit x86-64),
Linux (32-bit, 64-bit Itanium, and 64-bit x86-64), and Solaris
(64-bit SPARC and 64-bit x86-64).

10. The internal utilities for **gllamm** now support inverse-probability
weighting schemes.

11. Mata has the following new functions:

A. Mata functions **minindex()** and **maxindex()** return the indices of
minimums and maximums (including tied values) of a vector, *v*.

B. Mata function **timer()** is a new undocumented function for
profiling the execution time of Mata code.

12. Mata had the following fixes:

A. Mata functions **crossdev()** and **quadcrossdev()** aborted with error
if the centering variables contained missing values. Now the
functions return an all-zero result.

B. Mata functions **rowshape()** and **colshape()** would not let you
change certain types of null matrices to other types. For
instance, you could not change a 1 *x* 0 matrix into a 0 *x* 0.
This problem is fixed. A side effect of the fix is that you
must recompile any programs that use **colshape()**. You do not
need to recompile programs that use **rowshape()**, however.

C. Mata function **st_strscalar()** did not return results from Stata's
**c()**, such as **st_strscalar("c(byteorder)")**. This problem has
been fixed. Users are reminded that another way to obtain
c-class values is via Mata's **c()** function: **c("byteorder")**.

13. Inserting data via **odbc insert** to an existing database could,
depending on the underlying SQL code generated, cause SQL syntax
errors. This problem has been fixed.

14. The regular expression functions **regexm()**, **regexs()**, and **regexr()**
could cause Stata/MP to crash; this problem has been fixed.

15. To prevent output from wrapping, **return list**, **ereturn list**, and
**sreturn list** now display only the first line of long string results.
The result looks better and is more easily read. If a report is
truncated, two dots are added to the end of the part shown.

16. In **tabulate twoway**, cells with zero counts contributed missing (**.**)
instead of zero to the value of the likelihood-ratio chi-squared
test. This problem has been fixed.

17. **timer** is a new undocumented command for profiling the execution time
of ado-code.

18. Small Stata (Windows and Mac) could crash when the extended macro
function **properties** was used in a program. This problem has been
fixed.

__Stata executable, Windows__

19. The Command window now has keyboard shortcuts reinstated from Stata
8. Using Ctrl-Backspace to delete a whole word is an example.

__Stata executable, Mac__

20. The contextual menu item Copy Review Contents to Do-file Editor for
the Review window was enabled even when the Review window was empty.
This problem has been fixed.

21. The Font Size submenu from the Prefs menu was not enabled when the
Variables window was selected. This problem has been fixed.

22. Using the Quartz drawing engine for drawing graphs could render some
non-ASCII characters as outlined boxes. To avoid this limitation of
the Quartz text renderer, you may use ATSUI (Apple Type Services for
Unicode Imaging) to render text in graphs by using **set**
**use_atsui_graph on**.

__Stata executable, Unix (GUI)__

23. Closing a Viewer window while a modal dialog was up would cause
Stata to crash. Stata now prevents you from closing the Viewer
until the warning dialog is closed.

24. A GTK warning message could appear from the console Stata was
started from if a dialog had a dropdown list combo box. This
problem has been fixed.

25. After the 17may2006 executable update, the **setposition** member
program did not work correctly for certain types of programmable
dialog controls. This problem has been fixed.

-------- **update 20jun2006** -----------------------------------------------------

__Ado-files__

1. Online help and the search index have been brought up to date for
Stata Journal 6(2).

2. **areg** exited with an error when **if** or **in** was properly supplied after
the **absorb()** option. This problem has been fixed.

3. **egen**'s **median()** function failed when presented with expressions
containing double quotes. This problem has been fixed.

4. **estat hettest** has new options **iid** and **fstat**. By default, **estat**
**hettest** performs a score test for heteroskedasticity that assumes
that the regression disturbances are normally distributed. **iid**
causes **estat hettest** to perform a score test for heteroskedasticity
that drops the normality assumption. **fstat** causes **estat hettest** to
perform an F test for heteroskedasticity that drops the normality
assumption.

5. **estat** failed to work after **stcox, shared()**. This problem has been
fixed.

6. Mata functions **mean()**, **variance()**, **quadvariance()**, **meanvariance()**,
**quadmeanvariance()**, **correlation()**, and **quadcorrelation()** previously
required two arguments, the second being *w*, the weight. The second
argument is now optional and, if omitted, unweighted estimates
(equivalent *w*=1) are returned.

7. Mata functions **variance(X,w)**, **quadvariance(X,w)**, **meanvariance(X,w)**,
and **quadmeanvariance(X,w)** aborted with an error if *X* or *w* contained
all missing values. Now the functions return an all-missing result.

8. **nl** exited with an error if a parameter's name was a substring of
another parameter's name and the parameter whose name was longer was
declared first. This problem has been fixed.

9. **nlogit** did not use weights when computing the initial estimates and
null likelihoods used to test the regression coefficients and
homoskedasticity. This problem has been fixed.

10. **roctab**, with the **detail** option, reported incorrect values of
sensitivity when the classification probabilities corresponding to
different levels of the classification variable were equal. This
problem has been fixed.

11. **svy: mlogit** exited with an invalid-syntax error when the dependent
variable had value labels with spaces in them. This problem has
been fixed.

12. **treatreg** now accepts the **from()** option.

13. **tsline** reported an invalid-number error when the time variable had a
**%d** format and a time axis option was specified using a date string.
This problem has been fixed.

-------- **update 17may2006** -----------------------------------------------------

__Ado-files__

1. **estat vif** after **regress** has a new option, **uncentered**, that computes
uncentered variance inflation factors.

2. **estat vif** after **regress** reported invalid results with options
**vce(bootstrap)** or **vce(jackknife)** or with the **svy** prefix. **estat vif**
now reports an error message.

3. **estimates table** did not label estimates properly when equation names
contained spaces. This problem has been fixed.

4. **glm** would exit with an error message if an HAC covariance matrix was
requested and the list of regressors included time-series operators.
This problem has been fixed.

5. **glm** reported incorrect deviances and deviance diagnostics when
fitted to proportion data with **family(binomial 1)**. This problem has
been fixed.

6. **graph twoway** now allows more than 100 variables to be plotted even
when those variables are from the same observations. Previously the
only way to plot more than 100 variables was to ensure that some of
the variables were plotted over a different range of observations.
One plot is still limited to 100 or fewer variables.

7. The dialog box for **graph twoway** did not generate the correct options
for affecting the rendition of lines when used with **twoway lfitci**,
**twoway qfitci**, or **twoway fpfitci**. This problem has been fixed.

8. **ivprobit** and **ivtobit** exited with an error message when regressors
were dropped because of collinearity, when the list of regressors
exceeded about 500 characters, or when time-series operators were
used. These problems have been fixed.

9. **nlogit** now allows the **robust** option for models that nest deeper than
three levels.

10. **qreg**, and similar quantile regression commands, sometimes dropped
the dependent variable when it equaled a linear function of the
covariates. This problem has been fixed.

11. **svyset** now has the **fay(***#***)** option.

12. **svy: tabulate twoway** reported a one in the marginals when the **column**
and **row** options were specified, even when given data that resulted
in a zero in the marginals. This problem has been fixed.

13. **table** now issues an improved error message when the **sd** statistic is
used with pweights.

14. **xi** omitted a group when called with the **noomit** option and syntax
**i.***varname1*******varname2* or **i.***varname1***|***varname2*. This problem has been
fixed.

15. **xtreg, fe cluster()** did not report the proper degrees of freedom or
p-value for the model F test. This problem has been fixed.

16. **xtset** is a new command for declaring panel data. This command is
meant to eventually replace **iis** and **tis**. For panel data with a time
variable, **xtset** and **tsset** are equivalent. View the **xtset** command as
a new convenience command, something you do not really need but
might like to use anyway.

__Stata executable, all platforms__

17. **datasignature** is a new command that many users will find useful.
**datasignature** reports a short string (a signature) that you can use
to detect whether your data have changed. See "*Examples of*
*interactive use*" under *Remarks* in the help file.

18. **insheet**, when given a file with more than the maximum allowed number
of variables, could cause Stata to crash. This problem has been
fixed.

19. **irf create** now estimates dynamic-multiplier (DM) functions and
cumulative DM functions, as well as their standard errors. The **irf**
reporting commands were modified to display the new DM estimates.

20. Mata functions **select()** and **st_select()** are new. They select rows
or columns of a matrix on the basis of a criterion.

21. Mata incorrectly executed statements of the form "*v***[***i***]** **=** *name*" and
"*v***[***i***,***j***]** **=** *name*" when *v* and *name* were structures. Rather than
assigning a copy of structure *name* to *v*, Mata assigned *name* itself,
so that subsequent changes in *name* changed *v***[***i***]** and *v***[***i***,***j***]**. This
problem has been fixed. You do not need to recompile Mata code.

22. Mata inconsistently classified **""** as being greater than nonempty
strings (such as **"a"**), whereas Stata classifies them as being less
than nonempty strings. Mata has been modified so that it agrees
with Stata: **""** < **"a"**.

23. Mata's **>=** operator returned an incorrect result when used with
strings and the string to the left of the operator was greater than
the string to the right of the operator. This problem has been
fixed.

24. **signestimationsample** and **checkestimationsample** are new programmer
commands for coordinating estimation and postestimation ado-files.

25. Classes can now define an **.oncopy** member program to perform
operations when a copy of an object is created. This is an advanced
concept; see classman section 4.8 Advanced initialization, .oncopy.

26. For Stata/MP, Stata/SE, and Intercooled Stata, the limit on the
number of dyadic operators in an expression has been increased from
500 to 800, and the limit on the number of numeric literals in an
expression has been increased from 150 to 300. See limits.

27. New functions **_byn1()** and **_byn2()** are available within a
**byable(recall)** program. They return the beginning and ending
observation numbers of the by-group currently being executed. See
byable for more information.

__Stata executable, Windows__

28. The Data editor on Windows ME could freeze Stata. This problem has
been fixed.

29. The dropdown print menu from the toolbar print button could get
corrupted when displaying multiple graphs. This problem has been
fixed.

30. Stata could crash if scheduled to run at a preset time when there
was no user logged in to Windows. This problem has been fixed.

31. When invoking a dataset via the recent files menu in Stata that
happened to also be on a shared network drive, an extra slash was
prepended to the path. This problem has been fixed.

32. Stata/MP on Windows XP loaded a wrong version of the common controls
DLL. Various abnormal GUI behaviors, such as the dialog backgrounds
becoming black, could occur if XP Themes were used. This problem
has been fixed.

__Stata executable, Mac__

33. **xmlsave** now uses only double quotes as tag attributes when saving to
the Excel doctype. This change circumvents a bug in Excel for Mac
when reading XML data files generated from Stata.

__Stata executable, Unix (GUI)__

34. When a programmable dialog has the keyboard focus, pressing Enter
simulates clicking the OK button if one exists, pressing Escape
simulates clicking the Cancel button if one exists, and pressing
Shift and Return simulates clicking the Submit button if one exists.

35. The Command window is no longer resized if the main Stata window is
resized vertically. The Review and Variables windows are no longer
resized if the main Stata window is resized horizontally. You may
still drag the splitter controls to resize the windows.

36. Sometimes, the Command window lost the keyboard focus, preventing
users from inputting any commands unless they clicked on the body of
the main Stata window. This problem has been fixed.

37. Exporting a graph as TIFF would result in a graph with the width and
height swapped. This problem has been fixed.

38. Resizing a Graph window could cause its graph to not draw at the
final window size if the graph could not finish drawing while it was
being resized. This problem has been fixed.

39. Resizing the Data Editor also caused the Data Editor's toolbar and
edit field to resize vertically. This problem has been fixed.

40. Right-clicking in a Viewer window and selecting Find in Viewer...
would not display the Find bar. After selecting File > Log >
View... to display the Choose File to View dialog, clicking the
Browse... button and selecting a filename would not be entered into
the File or URL: edit field. Both problems have been fixed.

41. Selecting Prefs > Manage Preferences > Load Preferences > Open...
could crash Stata. This problem has been fixed.

42. If any modal dialog caused Stata's Results window to scroll before
the dialog could close completely, the first line to be obscured by
the top of the dialog would not be updated after it scrolled. This
problem has been fixed.

43. **shell** under Linux could fail if more than half of the memory
available on the system was allocated to Stata. This problem has
been fixed.

-------- **update 14apr2006** -----------------------------------------------------

__What's new in release 9.2 (compared with release 9.1)__

Highlights for release 9.2 include the following:

**o** Stata/MP is a new parallel-processing version of Stata that can
perform calculations in parallel on computers with 2, 4, 8, 16,
and even 32 processors or cores. Stata/MP has all the
capabilities of Stata/SE but can perform many analyses much faster
by taking advantage of special code to perform symmetric
multiprocessing (SMP). See Stata/MP and
http://www.stata.com/statamp/ for more information.

**o** Mata now has structures, which will be of special interest to
those writing large, complicated systems. See item 14 A below.

**o** Mata now engages in more thorough type checking, and produces
better code, for those who explicitly declare arguments and
variables. See item 14 B below.

**o** **hsearch** is a new command that searches help files. It is an
alternative to the **search** command. You will want to try **hsearch**.

We recommend that you set version to **version 9.2** at the top of new
do-files and ado-files. See help **version**.

__Ado-files__

1. **bootstrap** and **jackknife** exited with an error when an appropriate
*eform_option* was specified. This problem has been fixed.

2. **bsample** returned incorrect frequency weight values (in the **weight()**
option) when called with a resample size less than the selected
sample size. This problem has been fixed.

3. **cloglog** and **xtcloglog** have new option **eform**.

4. **corr2data** was inadvertently setting the random-number generator seed
to the current value of **sortseed**. This problem has been fixed.

5. **glm** and **xtgee** labeled generic exponentiated coefficients as "ExpB"
and "e^coef". For consistency, both labels have been changed to
"exp(b)". This change applies only to models where exponentiated
coefficients do not already have meaningful names, such as hazard
ratios.

6. **hsearch** is a new command for searching the Stata help files on your
computer.

7. **ml** complained that the variable **ML_subv** already exists when called
twice interactively with the **subpop()** option. This problem has been
fixed.

8. **ml display** has new option **showeqns**.

9. **sts graph** did not respect value labels when the **by()** and **separate**
options were combined with **hazard**. This problem has been fixed.

10. **svy jackknife** with the **mse** option after **svyset**ing **iweight**s with
**strata()** resulted in incorrect standard errors. This problem has
been fixed.

11. **xthtaylor** no longer requires the user to specify both endogenous
time-invariant variables and endogenous time-varying variables;
users may now specify endogenous time-invariant variables and/or
endogenous time-varying variables.

__Stata executable, all platforms__

12. **asmprobit** with unbalanced data and the BHHH optimization technique,
**tech(bhhh)**, did not work properly and typically did not converge.
This problem has been fixed.

13. Date values in Stata should be integers. Given noninteger values,
however, date functions and date formats rounded toward 1jan1960.
Now they always round down (toward the past).

14. Mata has the following new features:

A. Mata now has structures (see **[M-2] struct**), which will be of
special interest to those writing large, complicated systems.
Because of the change, the format of **.mlib** libraries is
different. There is no backward-compatibility issue -- you do
not need to recompile old code -- you *NEVER* need to recompile
old code. However, because of the format change, you will not
be able to add new members to old libraries. Libraries cannot
contain a mix of old and new code. To add new members, you must
first rebuild the old library.

B. Mata now engages in much more thorough type checking, and
produces better code, for those who explicitly declare arguments
and variables. As before, declaration is optional. However, if
you make explicit declarations, old code may not compile because
Mata detects problems it previously could not. As an example,
assigning a string result to a variable explicitly declared as
**real** is no longer allowed. If you have old code with such
problems, either change the variable's declaration to
**transmorphic** or introduce a new variable to fix the problem
properly.

C. Mata now includes all 29 Stata date functions; see **[M-5] date()**.

15. Mata had the following fixes:

A. The first time Mata was invoked after launching Stata, if the
first thing typed into Mata was a double quote character and
there was no matching close quote, Stata crashed. This problem
has been fixed.

B. Mata function **ascii()** returned negative values for the extended
ASCII characters rather than returning values in the range 128
to 255. This problem has been fixed.

C. Mata functions **fread()** and **_fread()** caused Mata to issue a
stack-overflow error. This problem has been fixed.

D. Mata function **regexr()** did not allow replacement with an empty
string (for deletions) and did not allow substitutions on
strings shorter than 10 characters. Both problems have been
fixed.

E. Mata comparison operator **>=** was treated the same as **>** for string
comparisons. This problem has been fixed.

16. **merge**, specified with match variables, caused Stata to crash when
the master dataset had no observations. This problem has been
fixed.

17. **net install** and **net get** now accept a **force** option, which specifies
that the downloaded files replace existing files if any of the files
already exist, even if Stata thinks all the files are the same.

18. PNG (Portable Network Graphics) files produced by **graph export** on
64-bit systems could contain valid image data but an invalid
checksum on that data. This could cause some third-party
applications to have problems displaying the PNG files. This
problem has been fixed.

19. **use** did not clear the previous sort list when a *varlist* was
specified. This problem has been fixed.

20. **xtcloglog**, **xtintreg**, **xtlogit**, **xtpoisson**, **xtprobit**, and **xttobit** used
single precision instead of double precision storage for the log
likelihood. This problem has been fixed.

21. **xtintreg** and **xttobit** results were not as precise as they could have
been because of inaccuracy in the calculation of the derivatives of
the log likelihood. This problem has been fixed.

__Stata executable, Windows__

22. **odbc load** now imports the ODBC bit data type as Stata byte.
Previously bits were imported as Stata integers. This change
reduces the memory needed to import data.

23. Alt-Tab functionality and focusing of the Command window has been
improved for nondefault layouts.

24. New layouts are now saved properly when issuing the **exit** command.

25. Under Windows 98, closing the Data Editor caused Stata to crash.
This problem has been fixed.

__Stata executable, Mac__

26. You may now paste multiple lines of text into the Command window.
Stata executes each line of text that is terminated by a carriage
return. If a line of text does not end in a carriage return, it is
simply pasted into the Command window. Any errors in the lines of
text that are executed do not prevent subsequent lines from
executing, which can cause unintended results. As such, this
setting is off by default, which means that only the first line of
text is pasted into the Command window and is not executed.

27. Stata now uses very little of the processor while Mata is idling in
interactive mode.

28. Stata now prevents multiple lines of text from being pasted into an
edit control from a programmable dialog. Only the first line of
text is pasted. Pressing the Home or End key in an edit control of
a programmable dialog now moves the selection to the beginning or
end, respectively, of the edit control.

29. Stata's Do-file Editor tiles new Do-file Editor windows 20 pixels to
the right and 20 pixels below the last moved or created Do-file
Editor window. However, a new Do-file Editor window could
potentially be tiled off the screen if the last window is too close
to the right edge or bottom of the screen. Stata now starts over
from the top-left corner of the screen if a new Do-file Editor
window is too close to the right edge or bottom of the screen.
Also, Do-file Editor, Viewer, and Graph windows will now tile
properly on multiple displays. Stata previously always tiled new
windows on the primary display rather than the display in which the
last window was located.

30. Pressing Return in the Data Editor's Edit dialog inserted a carriage
return into the currently active edit control. This problem has
been fixed.

31. The 20jan2006 executable update, when run on a PowerPC-based Mac,
would require a G4 processor or better (this limitation did not
affect Stata when run on an Intel-based Mac). Stata will now run on
any PowerPC computer that supports Mac OS X just like it did prior
to the 20jan2006 update.

__Stata executable, Unix (GUI)__

32. If the main Stata window was resized to smaller than the default, it
still opened at the default window size the next time Stata was
launched. This problem has been fixed.

33. Stata now prevents multiple lines of text from being pasted into an
edit control from a programmable dialog. Only the first line of
text is pasted.

34. The Do-file Editor could crash if there was no carriage return on
the last line of the do-file. This problem has been fixed.

-------- **update 27feb2006** -----------------------------------------------------

__Ado-files__

1. Online help and the search index have been brought up to date for
Stata Journal 6(1).

2. **bstat** always reported 95% confidence intervals even when the **level()**
option was specified. This problem has been fixed.

3. **cs** exited with an error when **by()** contained more than one
by-variable, and option **rd** was specified. This problem has been
fixed.

4. **estat dwatson** could produce erroneous results when used after a
command that performed a regression that included temporary
variables. Now an error message is produced.

5. **gladder** and **qladder** did not apply user-specified schemes to
individual plots. This problem has been fixed.

6. Mata function **cat()** now executes faster on large files.

7. **merge** was documented as allowing **uniqu** as a valid abbreviation for
both **unique** and **uniqusing**. The help file now documents that **uniqu**
is an abbreviation only for **unique**.

8. **mfx, eyex** and **mfx, dyex** would report negative standard errors for
marginal effects if the value of the x variable was negative. This
problem has been fixed.

9. **ml** now saves **e(k_eq_model)**, the number of equations to include in a
model Wald test, even when **lf0()** is specified.

10. **permute** with the **left** option did not count all cases where replicate
values were less than or equal to the observed value of the test
statistic. This problem has been fixed.

11. **pksumm** sometimes incorrectly reported the error 'follow-up times are
different for each patient'. This problem has been fixed.

12. **predict** after **logit** and **probit** reported an error when an independent
variable was dropped because of collinearity. This problem has been
fixed.

13. **predict**, after **xtcloglog**, exited with a file-not-found error. This
problem has been fixed.

14. **predict, xbu** after **xtreg, fe** failed if any of the regressors
contained time-series operators. This problem has been fixed.

15. **predictnl** after **ologit** exited with an error when the special
**predictnl** function **predict()** was specified. This problem has been
fixed.

16. **recode**'s **min** element replaced a variable's value with missing when
the **if** qualifier was specified and no observations met the **if**
criterion. This problem has been fixed.

17. **stptime** exited with a syntax error when an **if** qualifier containing
strings was specified. This problem has been fixed.

18. **svy: tabulate twoway** would report tests of independence when the
table had a zero margin in a column beyond the number of rows. This
problem has been fixed.

19. **xtlogit** and **xtprobit** did not accept constraints on **[lnsig2u]_cons**.
This problem has been fixed.

20. **pwcorr** with perfectly correlated data now returns correct
significance values.

-------- **update 20jan2006** -----------------------------------------------------

__Ado-files__

1. **nestreg** exited with an error when a block of variables was
preliminarily dropped from the model specification. This problem
has been fixed.

2. **stepwise** exited with an error when a term consisting of a single
variable was preliminarily dropped from the model. This problem has
been fixed.

__Stata executable, all platforms__

3. The maximum number of programmable menu items that can be added to
Stata has been increased from 1,000 to 1,250.

4. **asmprobit**'s GHK simulator generated too many uniform numbers in the
simulation. This problem has been fixed.

5. Class member programs were not using the version number specified at
the top of a class-definition file. This problem has been fixed.

6. **confirm format** is a new command that allows formats to be verified.
Optionally, you can confirm that the given format is of a particular
type: **numeric**, **string**, **date**, or **ts**.

7. An expression evaluated by a command issued in a loop with an **`=***exp***'**
on the initial loop command line could return missing value (**.**) even
when it should not. This problem has been fixed.

8. **filefilter** used on files containing embedded NULL characters in
certain positions sometimes caused the filters to be ignored for a
portion of the file following the NULL before eventually resuming
correct filtering. This problem has been fixed.

9. **graph box**, when specified with only a single variable that did not
have a value label, left the y-axis untitled. In such cases, the
y-axis will now be titled with the variable name.

10. **graph twoway** now allows you to create custom axis ticks and labels
that have a different appearance (color, size, tick length, etc.)
from the other ticks and labels on the axis. Such custom ticks can
be used to emphasize points in the scale, such as important dates,
physical constants, or other special values. See the new **custom**
suboption of {**y|x|t**}**label()** in *axis_label_options*.

11. **graph twoway** has new options, **yoverhangs** and **xoverhangs**, that
attempt to adjust the graph region margins to prevent long labels on
the y- or x-axes from extending off the edges of the graph; see
*advanced_options*.

12. **graph twoway** has a new suboption, **norescale**, to the options
{**y|x|t**}**label()**. This suboption specifies that the ticks and labels
specified in the {**y|x|t**}**label()** option be placed directly on the
graph without rescaling the axis or associated plot region for the
new values; see *axis_label_options*.

13. **help**, under rare conditions when accessing help files across a
network, could be slow; this problem has been fixed.

14. **inspect** output has been slightly modified. All output on the
right-hand side has been shifted to the left three spaces so that
the "Nonintegers" heading is placed on one line.

15. **label define**, followed by something that was not a valid label name,
caused a memory leak that could result in Stata's incorrectly
reporting an 'unexpectedly out of memory' error with a return code
of 9710. This problem has been fixed.

16. **lnnormal()** and **lnnormalden()** are new functions that provide the
natural logarithm of the cumulative normal distribution function and
the normal density function, respectively.

17. The new extended macro function **adosubdir** returns the subdirectory
in which Stata would search for a file on the ado-path.

18. Mata has the following new functions:

A. **adosubdir()** returns the subdirectory in which Stata would search
for a file on the ado-path.

B. **ghk()** provides a Geweke-Hajivassiliou-Keane multivariate normal
simulator.

C. **halton()**, **_halton()**, and **ghalton()** compute Halton and Hammersley
sets.

D. **lnnormal()** and **lnnormalden()** find the logarithm of the
cumulative normal and of the density function, respectively.

E. **pathsearchlist()** returns a row vector of full paths/filenames
specifying all the locations Stata will search for a named file
along the official Stata ado-path.

19. Mata also has the following changes:

A. Mata's logical negation (not) operator **!** treated missing values
incorrectly. A missing value is supposed to be treated as
meaning true just like any other nonzero value. Thus if **x==.**,
**!x** is **0**. Instead, **!x** returned missing (**.**). This problem has
been fixed.

B. Mata's **printf()** and **sprintf()** functions left-justified strings
printed with **%#s** formats and right-justified strings with **%-#s**
formats. This effect was the reverse of what was correct and
has been fixed.

C. Mata/Stata interface functions **st_local()**, **st_global()**,
**st_numscalar()**, and **st_strscalar()**, when used to store something
more than 65,532 times in a tight loop, could cause a Mata stack
overflow error. This problem has been fixed.

D. On 64-bit computers, Mata restricted matrix size such that r*c*8
< 2^31 for regular matrices and r*c*16 < 2^31 for complex
matrices even when more than 2 GB of memory was available. This
problem has been fixed, and matrices larger than 2 GB may now be
used in Mata on 64-bit computers.

20. **odbc** commands with extended variable lists did not respect quotes or
compound quotes to allow nonalphanumeric variable names to be
imported. This problem has been fixed.

21. Exporting a graph with the **width()** or **height()** options caused Stata
to crash if the graph window was closed. Stata now prevents you
from exporting a graph when the graph window is closed.

22. **_qreg**, a programmer's routine, skipped observations when *if* or *in*
qualifiers were specified. This problem has been fixed.

23. The **regexr()** function and the Mata **regexr()** function did not allow a
string to be replaced with an empty string. For example:

**. display regexr("mystring", "mystring", "")**

displayed "mystring" instead of "". This problem has been fixed.

24. **tabulate, exact**, for computing Fisher's exact test, now allows
larger tables, and an enumeration log is displayed. New option
**nolog** suppresses the log.

25. **twoway lfit** and **twoway qfit** can now pick up value labels to annotate
the x-axis by using the existing option **mlabel(, valuelabel)**; see
*axis_label_options*.

26. A contextual menu item has been added to the Viewer that closes all
viewers except the current one.

__Stata executable, Windows__

27. Multiple do-file editors may now be opened from a single open
dialog. The first file goes to the currently open do-file editor,
whereas subsequent files open in new editor windows.

28. The most recently used file information was not saved properly when
exiting Stata with the **exit** command. This problem has been fixed.

29. The Variables Property dialog has been modified to set the focus to
the variables edit field and to set its text selection to all text.

30. Specifying a **width()** of less than 115 pixels when exporting a graph
would not result in a bitmap image with the specified width. This
problem has been fixed.

31. When Alt-Tabbing between windows and the main Stata window, focus
will automatically be sent to the Command window. Previously, focus
was sent only after alphanumeric keystrokes were detected.

__Stata executable, Mac__

32. You can select multiple files from the Open dialog and Do dialog.
Selecting multiple do-files from the Open dialog will open all files
in separate Do-file Editor windows. Selecting multiple do-files
from the Do dialog will **do** do-files in the order that they were
selected. Selecting multiple datasets and graphs will open only the
first file selected.

33. Horizontal scroll wheel support has been added to the Data Editor
and Do-file Editor. You may now use an Apple Mighty Mouse scroll
wheel to scroll in any direction.

34. The menu item Tools > Back and its keyboard shortcut Cmd-[ have been
added to the Viewer to match the keyboard shortcut for Back in
Safari. However, Cmd-[ had been previously reserved for Edit >
Shift Left (to shift text one tab stop to the left in the Do-file
Editor). To avoid any conflicts, the keyboard shortcuts for Shift
Left and Shift Right have now been changed to include the shift key
(press Cmd-Shift-[ to shift text left and Cmd-Shift-] to shift text
right).

35. Stata now displays the current number of columns and rows of the
Results, Viewer, and Do-file Editor windows as they are being
resized.

36. Stata no longer displays window transition effects when it is in the
background.

37. Pressing the keyboard shortcut Cmd-W did not close the Review
window. This problem has been fixed.

38. In some cases, clicking the Back button in the Viewer would result
in the Viewer toolbar's being overwritten with the Viewer's output.
This problem has been fixed.

39. In some cases, a single line in the Results window would appear
chopped after output had scrolled. This problem has been fixed.

40. Typing **exit** from the Command window exited Stata without prompting
to save any unsaved changes in the Do-file Editor. This problem has
been fixed. Type **exit, clear** to exit Stata without prompting.

41. Stata(console) for Mac failed to perform an **update swap** after an
executable had been downloaded. This problem has been fixed.
However, current Stata(console) for Mac users will have to manually
download the updated executable with the fix from
http://www.stata.com/support/mac/.

__Stata executable, Unix (GUI)__

42. Stata now displays the current number of columns and rows of Viewer
windows as they are being resized.

43. Pressing Ctrl-W now closes the Viewer.

-------- **update 12jan2006** -----------------------------------------------------

__Ado-files__

1. **cnsreg** has the new **mse1** option and accepts **iweight**s.

2. Estimation commands with model Wald tests composed of more than just
the first equation now save the number of equations in the model
Wald test in **e(k_eq_model)**.

3. **kwallis** now exits with an error when there are no observations to
perform the calculations or when fewer than two populations are
specified.

4. **ml** now saves the number of equations used to compute the model Wald
test (from the **waldtest(***#***)** option) in **e(k_eq_model)**.

5. **ml display** after **ml**-based user-written Stata 9 estimation commands
that were prefixed with **svy** did not display the coefficient table
header properly. This problem has been fixed.

6. **ml score** has the new **missing** option.

7. **pkexamine** has the following changes:

A. **pkexamine** with an *if* condition that excluded the observation
with zero time and concentration did not use the proper zero
time and concentration in the computations. This problem has
been fixed.

B. **pkexamine** no longer adds an observation with zero time and zero
concentration when the dataset has an observation with nonzero
concentration at time zero.

C. **pkexamine** exited with an error message when **fit()** contained the
total number of observations. This problem has been fixed.

8. **predict** after **ologit** and **oprobit** reported an error when the correct
number of new variable names was specified for computing multiple
equation-level score variables from a constant-only model. This
problem has been fixed.

9. **qreg** did not set **e(wtype)** and **e(wexp)** when weights were used. This
problem has been fixed.

10. **suest** after **intreg** reported an error when there were missing values
in one or both of the dependent variables. This problem has been
fixed.

11. **suest** produced incorrect variance estimates when used with **cnsreg**.
This problem has been fixed.

-------- **update 13dec2005** -----------------------------------------------------

__Ado-files__

1. Errors in a couple of help files have been corrected.

-------- **update 12dec2005** -----------------------------------------------------

__Ado-files__

1. Online help and the search index have been brought up to date for
Stata Journal 5(4).

2. **adoupdate** is a new command for keeping the user-written ado-files
you have installed from *The Stata Journal*, the SSC, and other
sources up to date.

3. **bootstrap** did not pass the **nodrop** option to **jackknife** when both the
**nodrop** and **bca** options were specified. This has been fixed.

4. **estat** exited with an error message after **arima** if the **condition**
option was specified. This has been fixed.

5. **glm** ignored the **exposure()** option. This has been fixed.

6. **histogram** occasionally reported one fewer bin than the number
produced in the graph. This has been fixed.

7. **jackknife** reported a "last estimates not found" error when called as
a prefix to a command without arguments. This has been fixed.

8. **jackknife** reported an "options not allowed" error when called as a
prefix to a command that does not allow options even when no options
were specified. This has been fixed.

9. **lrtest** output inappropriately wrapped to a new line if the names of
the tested models were long. This has been fixed.

10. **nl** now uses the format */param* in listing the parameters in the
coefficient table.

11. **nl**, when used with weights, reported an incorrect sum of weights and
residual deviance. This has been fixed.

12. **recode**'s **max** element treated missing values as the maximum value of
a variable if the variable contained all missing values. This has
been fixed.

13. **stphplot** did not display the legend properly in certain
circumstances. This has been fixed.

14. **suest** now has option **svy** for applying the current **svyset**
characteristics to the variance calculation.

15. The 15sep2005 update changed some low-level graphics calls. In rare
cases, the previous low-level graphics calls were not properly
interpreted by some HP printer drivers, causing the entire graph to
be printed black. We have confirmed with users who had experienced
the problem that the 15sep2005 update allows these drivers to render
the graphs correctly.

This problem also occurred when copying and pasting into Microsoft
Office on the Mac platform. The 15sep2005 update fixed this
problem.

-------- **update 11nov2005** -----------------------------------------------------

__Ado-files__

1. **nestreg** exited with an error when used with a **survey** estimation
command other than **svy: regress**. This has been fixed.

-------- **update 10nov2005** -----------------------------------------------------

__Ado-files__

1. **adjust** after **ivprobit** can now calculate adjusted probabilities when
the maximum likelihood estimator is used.

2. **bootstrap** with the **bca** option exited with error when the user
specified names for bootstrap statistics. This has been fixed.

3. **estat bootstrap** reported an invalid subcommand error when used after
**mean**, **proportion**, **ratio**, and **total** even when **vce(bootstrap)** was
specified. This has been fixed.

4. **ds** sometimes failed when certain values were specified together in
the **indent()** and **skip()** options and the linesize was smaller than
needed. This has been fixed.

5. **glm** saved an invalid **e(power)** when user-defined links used the
**SGLM_p** macro for non-standard purposes (such as holding a variable
name instead of a number). This has been fixed.

6. **ivreg** ignored the **beta** option. This has been fixed.

7. **nestreg** is a new prefix command for reporting nested model
comparison tests (also known as block or hierarchical regression
analysis).

8. The dialog for **stepwise** has been modified so that only the weights
that are accepted for a specific command are presented.

9. **predict** exited with an error after **xtgee** with the **link(nbinomial)**
option. This has been fixed.

10. **xtdes** exited abruptly if the time variable had fewer than two
distinct nonmissing values. Now a more informative error message is
issued.

-------- **update 20oct2005** -----------------------------------------------------

__Ado-files__

1. **merge**, when specified with no **using** files, gave an incorrect error
message. This has been fixed.

2. **merge** failed if any of the **using** files were located on a network
drive. This has been fixed.

3. **nl** has received the following enhancements. The previous behavior
of **nl** is preserved by setting **version** to one earlier than 9.1.

A. **mfx** can be used after **nl** to obtain marginal effects.

B. **nl** now supports the **vce(***vcetype***)** option. Allowed *vcetype*s
include **bootstrap**, **jackknife**, **robust**, and three
heteroskedasticity- and autocorrelation-consistent (HAC)
variance estimators.

C. **predict** after **nl** now allows you to obtain the probability that
the dependent variable lies within a given interval, the
expected value of the dependent variable conditional on it being
censored, and the expected value of the dependent variable
conditional on it being truncated. These predictions assume
that the error term is normally distributed.

D. The iteration log now reports the new sum of squares after each
iteration is complete instead of the old value.

E. The coefficient table now reports each parameter as its own
equation, analogous to how **ml** reports single-parameter
equations.

F. **nl** no longer reports an overall model F test because the test
that all parameters other than the constant are jointly zero may
not be appropriate in arbitrary nonlinear models.

G. The value of the log-likelihood function assuming i.i.d. normal
errors is saved in **e(ll)**, and likelihood-ratio tests can be
performed after estimation using **lrtest**.

4. **qreg**'s **wlsiter(***#***)** option with *#* > 1 did not change the coefficients
compared with when *#* was 1. This has been fixed.

5. **stepwise** incorrectly exited with an error when adding back a term
that was previously removed for backward stepwise regression (when
using options **pr()** and **pe()**). This has been fixed.

6. **stepwise** incorrectly exited with an error when performing forward
hierarchical selection (when using options **pe()** and **hierarchical**).
This has been fixed.

-------- **update 05oct2005** -----------------------------------------------------

__Ado-files__

1. **asmprobit** with the **cluster()** option set **e(wtype)** to **pweight** causing
an error in **estat mfx** after **asmprobit**. This has been fixed.

2. **egen**'s **rowtotal()** function, when passed a variable list containing a
wildcard followed by numbers, such as "***00**", sometimes calculated an
incorrect sum due to the inclusion of a temporary variable. This
has been fixed.

3. **svy brr** with the **fay()** option produced incorrect variance estimates.
This has been fixed.

4. **tabodds** with the **or** and **cornfield** options failed when the *var_case*
(case variable) contained missing values. This has been fixed.

-------- **update 19sep2005** -----------------------------------------------------

__Ado-files__

1. **codebook** has new option **dots** for use with the **compact** option. **dots**
displays a dot for each variable processed before displaying the
compact codebook table. This will interest those working
interactively with large datasets. Additionally, the **compact**
option, introduced in the 15sep2005 update, displayed counts for
string variables before displaying the correct compact codebook
table. The spurious output is now suppressed.

2. **graph export**, when specified with an extremely small pixel size for
the **width()** or **height()** options sometimes caused Stata to crash.
Stata now limits the pixel size range to be between 8 and 16,000.

3. **xtpoisson** sometimes labeled the variables in the output incorrectly
when variables were dropped from the estimation. This has been
fixed.

-------- **update 15sep2005** -----------------------------------------------------

__What's new in release 9.1 (compared with release 9)__

Highlights for release 9.1 include the following:

**o** Multiple **log** files may be opened simultaneously. See item 39
below.

**o** Survey linearization is now 2 to 100 times faster, the larger
multiplier applying to large datasets with many sampling units.
See item 46 below.

**o** The string length limit has been increased to 244 for Intercooled
and Small Stata to match Stata/SE. See item 32 below.

**o** Up to 300 estimation results may be stored. See item 34 below.

**o** Mata has a new permutation function making it easier to program
exact permutation tests. See item 41 D below and the examples in
**help mata cvpermute()**.

**o** Mata has several other new features, including new string, regular
expression, and binary I/O functions. See item 41 below.

**o** Graphs exported to PNG or TIFF bitmap format may now have their
sizes specified, allowing the resolution to be greater than screen
resolution. See item 36 below.

**o** There is a new command for computing marginal effects after an
alternative-specific multinomial probit model. See item 3 B
below.

**o** There is now a console version of Stata/SE for the Mac. See item
63 below.

**o** Collecting statistics across a by list using the **statsby** command
is now allowed in more situations, including with postestimation
commands. See item 21 below.

**o** Fay's adjustment is now allowed with balanced repeated
replications (BRR) for complex survey data. See item 25 A below.

We recommend that you set version to **version 9.1** at the top of do-files
and ado-files. **version 9.1** improves the default behavior for **permute** and
**xtreg**. See help **version**.

__Ado-files__

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

2. Some maximum likelihood estimation commands (such as **arima**, **nbreg**,
**streg**, and **xtnbreg**) ignored constraints when there were no
predictors in the model. This has been fixed.

3. **asmprobit** has the following changes.

A. **asmprobit** has a new default covariance parameterization and a
new option, **structural**. The new default parameterization
estimates the covariance of the alternatives differenced with
the base alternative. For a J-alternative model, the
differenced covariance is a J-1 x J-1 matrix with J(J-1)/2 - 1
estimable parameters, where the variance of the latent errors
for the scale alternative minus the base alternative is fixed to
2.

**structural** specifies that the structural J x J covariance
parameterization be used, which sets J-1 of the correlations to
zero and fixes the base and scale variances to 1.

B. **estat mfx** is a new postestimation command for computing marginal
effects after **asmprobit**. The **Postestimation** / **Marginal effects**
**or elasticities** menu displays the **estat mfx** dialog after an
**asmprobit** has been fitted.

C. When called with **pweight**s, **asmprobit** incorrectly stored **iweight**s
instead of **pweight**s in **e(wtype)**. This has been fixed.

D. **asmprobit** did not allow the constant-only model. Now it does.

E. Computing the likelihood for cases with only two alternatives
had the alternative probabilities reversed. If all cases had
two alternatives, the signs of the estimated coefficients were
reversed. If only some cases had two alternatives, estimates
were inaccurate. This has been fixed.

F. **predict** after **asmprobit** mislabeled the predictions when one or
more alternatives were excluded throughout the prediction
dataset. This has been fixed.

4. **biplot** with the option **separate** by default always places both
subgraphs on common x- and y-scales. Previously, the scales
sometimes differed.

5. **bootstrap** exited with an error, complaining about time-series
operators in the expression when an expression contained a numeric
literal with a decimal. This has been fixed.

6. **bootstrap** and **jackknife** exited with an obscure error message when
used with **xtmixed**. This has been fixed.

7. **codebook** has new option **compact** that produces a compact codebook
with each line showing a variable name, number of observations,
number of unique observations, mean, minimum, maximum, and variable
label.

8. The collapse dialog has been improved.

9. **estimates table** with a long *namelist* labeled only the first few
columns of estimation results. This has been fixed.

10. **fracpoly** and **mfp**, when used with **glm**, ran an older version of **glm**.
This has been fixed.

11. **graph** when the range of values in the y- or x-dimension was very
close to machine precision, but not exactly 0, sometimes entered an
infinite loop, requiring that processing be manually broken. This
has been fixed.

12. **graph combine** with options **ycommon** and **xcommon** in rare cases did not
produce combined graphs with the intended common axes. This has
been fixed.

13. **graph set** did not recognize the PICT format for setting graphics
exporting options. This has been fixed.

14. **hausman** now supports the existing **sigmaless** and new **sigmamore**
options after **xtreg**. These options are particularly useful when
comparing fixed- and random-effects regressions because they improve
results in small to moderate samples by ensuring that the difference
covariance matrix will be positive definite.

15. **svy jackknife** did not correctly count the number of observations in
the estimation sample when the sampling weight variable contained
zeros. This has been fixed.

16. **logit**, **probit**, and **dprobit** allowed **aweight**s, even though they were
not documented and the resulting calculation had no meaningful
interpretation. When **version** is set higher than 9.0, these commands
now issue an error message when used with aweights.

17. The matrix input dialog now allows users to enter symmetric matrices
by filling in the lower triangle.

18. **mean**, **proportion**, **ratio**, and **total** exited with a syntax error when
supplied with the **over()** option with more than 400 subpopulations.
This has been fixed.

19. **mprobit** has new option **probitparam** that scales the coefficients in
the same way as **probit**. Also, **mprobit** coefficients had reversed
signs when there were two alternatives (a probit model). This has
been fixed.

20. **permute** now uses two random uniform variables to generate Monte
Carlo permutations of the values of the permute variable. **permute**
must be called under **version** control in order to reproduce old
results.

21. **statsby** has several enhancements.

A. **statsby** now works with postestimation commands.

B. **statsby** now works with commands that support neither *if* nor *in*
(although it will be slower with such commands).

C. **statsby** has a new option **forcedrop** that forces all data to be
dropped except the group samples. This allows **statsby** to work
with user-written commands that completely ignore *if* and *in*
qualifiers.

D. **statsby** no longer clears estimation results when the command
being run is not an estimation command.

E. **statsby** is now faster when used with most time-series datasets.

22. **stci** exited with an error message when one or more *twoway_options*
were supplied, even though the **emean** and **graph** options were
specified. This has been fixed.

23. **stcurve** exited with an error message when options **at()**, **at1()**, and
**at10()** were used together. This has been fixed.

24. **stepwise** produced a "variable not found" error message when a
variable name was an abbreviation of another variable name found
earlier in the variable list. This has been fixed. Also, **sw** with
the Stata 8 syntax sometimes incorrectly exited with an error
message if the estimation command dropped a variable due to
nonestimability. This has been fixed.

25. **svy** has several improvements.

A. **svy brr** now allows the **fay()** option when using replication
weight variables. **svy brr** also saves the value of the Fay's
adjustment in **e(fay)**.

B. **svy: tabulate** now accepts string variables.

C. **svy** did not correctly set **e(sample)** when there were missing
values in the independent variables. This has been fixed.

D. **svy: mlogit**, when supplied with a dependent variable with value
labels containing dots (**.**) or colons (**:**), incorrectly exited
with an error message. This has been fixed.

E. **svy: proportion**, when used with variables having value labels
with the right quote character ("**'**"), incorrectly exited with an
error message. This has been fixed.

F. **svy: tabulate** exited with an unhelpful error message when **mata**
did not have enough memory to perform the necessary
calculations. A better error message is now displayed.

G. **svy: tabulate oneway**, when used with the **missing** option, dropped
the correct "Total" margin and labeled the row for the missing
value as the "Total". This has been fixed.

H. Version 8 survey estimation commands (**svyregress**, **svylogit**, ...)
now provide a better error message when used with data that are
**svyset** under version 9.

I. **svyset** did not return the name of the sampling weight variable
in **r(wvar)** as documented in **[SVY] svyset**. This has been fixed.

26. **tsline** sometimes incorrectly exited with an error when the **tlabel()**
option was supplied with integers but no date strings. This has
been fixed.

27. **varwle** displayed missing values for some tests when the lags were
not consecutive. This has been fixed.

28. **xi** ignored the **noomit** option when it was called with *terms* after the
comma. This has been fixed.

29. **xtmixed**, when used with the maximize option **iterate()**, did not issue
the appropriate warning at the bottom of the output if the model
failed to converge. This has been fixed.

30. **xtreg** has the following improvements:

A. **xtreg, fe** adjusted the VCE for the within transform when
**cluster()** was specified. The cluster-robust VCE no longer
adjusts for the within transform unless the new **dfadj** option is
specified.

B. **xtreg, fe** and **xtreg, re** produced cluster-robust VCEs when the
panels were not nested within the clusters. In some cases this
VCE is consistent, and in others it is not. You must now
specify the new **nonest** option to get a cluster-robust VCE when
the panels are not nested within the clusters.

C. **xtreg, re** reported different between R-squared values from one
run to another on identical datasets. This has been fixed.

31. **xtregar, fe** mislabeled variables when variables from the middle of
the list of *indepvars* were dropped due to collinearity. This has
been fixed.

__Stata executable, all platforms__

32. Intercooled and Small Stata now have the same maximum string length
(244) as Stata/SE. See limits.

33. **encode** encoded in reverse order for some platforms. This has been
fixed.

34. **estimates store** and **_estimates hold** now allow up to 300 estimation
results to be stored. The previous limit was 20.

35. **gettoken**'s **bind** option did not work properly if the first character
of the string to be parsed was "**(**" or "**[**". This has been fixed.

36. **graph export** has new options **width()** and **height()** for specifying the
width and height of a graph in pixels when exporting to PNG and TIFF
bitmap formats. If only the width or height is specified, Stata
exports the graph with the requested dimension while maintaining the
correct aspect ratio.

37. **include** is a new command similar to **do** and **run** for executing
do-files, but it differs in that any local macros, changed settings,
etc., created by executing the file are not dropped or reset when
execution concludes. Rather, results are just as if the commands
appeared in the session or file that included the file.

38. **itrim()** is a new string function that replaces multiple,
consecutive, internal spaces with a single space in strings.

39. **log** can now start multiple log files simultaneously. See **log** for
full details.

40. **logistic**, **logit**, and **probit**, when used with the **svy** prefix or with
options **vce(bootstrap)** or **vce(jackknife)**, had the following fixes:

A. They did not display a note indicating when there were
completely determined successes or failures. This has been
fixed.

B. They did not display a note indicating the presence of perfect
predictors. This has been fixed.

C. They did not correctly report on variables that were dropped due
to collinearity. This has been fixed.

41. Mata has several new functions:

A. **bufput()** and **bufget()** assist in performing binary-formatted I/O.

B. **byteorder()** is equivalent to Stata's **byteorder()** function.

C. Existing function **cat()** now takes optional second and third
arguments specifying the beginning and ending lines of the file
to read.

D. **cvpermute()** returns all permutations of the values of a column
vector *V*, one at a time. If *V*=(1\2\3), there are six
permutations: (1\2\3), (1\3\2), (2\1\3), (2\3\1), (3\1\2), and
(3\2\1). If *V*=(1\2\1), there are three permutations: (1\1\2),
(1\2\1), and (2\1\1).

E. **regexm()**, **regexr()**, and **regexs()** are regular expression
functions corresponding to the Stata functions with the same
name.

F. **stataversion()** returns the version of Stata that you are
running. **statasetversion()** allows setting the version.

G. **stritrim()** replaces multiple, consecutive, internal spaces with
a single space in strings.

H. **strtoreal()** converts strings to numeric values.

42. Mata function **cat()** produced an extra empty line in the rare case
when the input stream had "**\r\n**" and the "**\r**" landed at the end of
the internal buffer. This has been fixed.

43. **ologit** and **oprobit**, when used with the **svy** prefix or with options
**vce(bootstrap)** or **vce(jackknife)**, did not display a note indicating
when there were completely determined observations. This has been
fixed.

44. **print @Graph** with no Graph window open sometimes caused Stata to
crash. This has been fixed.

45. **print** *filename***.gph**, where *filename***.gph** was not a Stata format graph
file or was a live-format Stata graph file, sometimes caused Stata
to crash. This has been fixed.

46. **svy** linearization is now faster, especially for large datasets with
many sampling units.

47. **xmluse** with the **firstrow** option, when reading Excel datasets where
the second row did not have sufficient type information, resulted in
an error message saying that a variable couldn't be created because
it already existed. This has been fixed.

48. **xtabond** produced consistent but inefficient estimates when there
were gaps in the time series. This has been fixed.

49. The overlaid twoway graph dialog has a new tab **Y-Axis (2)** for
affecting the rendition of a second y-axis.

50. Axis tabs on all graph dialogs produced incorrect option arguments
for grid-line patterns. This has been fixed.

__Stata executable, Windows__

51. **set doublebuffer** is a new command that enables or disables
double-buffering in the Results, Viewer, and Data Editor windows.
Disabling double-buffering may be necessary if output from the
Results window appears very slowly.

52. (64-bit versions of Windows with a 64-bit version of Stata) **odbc**
loaded only a subset of the requested data. This has been fixed.

53. **permute**, when called with the **using** option with an absolute path in
Windows (as in C:\mystuff\outperm.dta), issued an error message
instead of replaying the results stored in the supplied dataset.
This has been fixed.

54. If **profile.do** was located on a UNC path (that is,
\\computername\path...), Stata prepended an extra backslash (that
is, \\\computername\path...) and caused an error. This has been
fixed.

55. On certain Windows versions, most notably Windows 98, opening the
Viewer could crash Stata. This has been fixed.

56. Tooltips have been added to some items in the Preferences dialog to
aid in readability.

57. The Variables Properties dialog from the Data Editor sometimes did
not select the correct value label for the current variable. This
has been fixed.

58. The Do-file Editor was incorrectly staggered from the correct
position when it was the initial editor opened. This has been
fixed.

59. A runtime exception sometimes occurred when the Search dialog was
opened. The exception was never critical and always allowed for
execution to continue. The cause of the exception has been fixed.

60. The initial dimensions of the Results window were sometimes too
small when using the initial, default settings. The Results window
was not fully utilizing the available space. This has been fixed.

61. Runtime assertion errors could occur when creating Viewers, Graphs,
or Do-File editors. These have been resolved.

62. When a variable control in a dialog has the focus, subsequent clicks
inside the Variables window will send variables to the dialog
control, as was the case in version 8.

__Stata executable, Mac__

63. There is now a console version of Stata/SE for the Mac. See
http://www.stata.com/support/mac/ for more information.

64. Stata now attempts to resolve aliases to the Stata directory and the
Stata executable upon startup.

65. The mouse pointer did not change to a hand when the pointer was over
a link. This has been fixed.

66. If you were running Stata from Mac OS 10.4 (Tiger), some lines would
appear bold when you made a selection from the Results window or
Viewer. This has been fixed.

67. A keyboard shortcut for** File > Save As...** has been added.

-------- **update 05jul2005** -----------------------------------------------------

__Ado-files__

1. **aorder**, when used on a variable list containing variable names with
certain combinations of the letters '**c**', '**d**', '**e**', or '**f**', could
exit with a 'system limit exceeded' error. This has been fixed.

2. **icd9** and **icd9p** have been updated to use the V22 codes; V21, V19,
V18, and V16 codes were previously used. V16, V18, V19, V21, and
V22 codes have been merged so that **icd9** and **icd9p** work equally well
with old and new datasets. See help **icd9** for a description of **icd9**
and **icd9p**; type "**icd9 query**" and "**icd9p query**" for a complete
description of the changes to the codes used.

3. **ml display** now behaves as it did in Stata 8 when called under
version control.

4. **suest** allowed estimation results from **ivreg** when it should have
produced an error message. This has been fixed.

5. **split**, when passed **parse("")** as an option, failed with an error.
**split** now understands that **parse("")** is equivalent to **parse()**.

6. **svy** sometimes identified a larger estimation sample in **e(sample)**
than was actually used. This has been fixed.

7. **xtgee, family(binomial)** is now equivalent to **xtgee, family(binomial**
**1)** for nonbinary dependent variables as documented in the manual.

__Stata executable, all platforms__

8. Find in the Viewer has been improved. Find is no longer performed
through a modal dialog but instead is performed from a Find bar that
is located above the Viewer's status bar. In addition to finding a
search term and selecting it, the Find bar can also highlight all
instances of the search term. Every Viewer window has its own Find
bar. To show or hide a Viewer's Find bar, click its Find button,
press Ctrl-F (Mac users press Cmd-F), or press F3.

Stata always starts a find operation from the current selection. If
there is no selection, Stata starts at the beginning of the
document. To perform a find at the beginning of a document when
there is text selected, click on the Viewer content area to clear
the selection before performing a find, or press Ctrl-F3. If the
Find bar's edit field has the keyboard focus, you may also press
Ctrl-Enter to do the same.

The following keyboard shortcuts are available when the Find bar is
visible.

**Shortcut** **Keyboard Focus** **Action**
--------------------------------------------------------------------
Ctrl-F Any part of Viewer Show/hide Find bar
Return Find bar edit field Find next
Shift-Return Find bar edit field Find previous
Ctrl-Return Find bar edit field Find from beginning of
document
Ctrl-Shift-Return Any part of Viewer Find from end of document
F3 Any part of Viewer Find next
Shift-F3 Any part of Viewer Find previous
Ctrl-F3 Any part of Viewer Find from beginning of
document
Ctrl-Shift-F3 Any part of Viewer Find from end of document
Ctrl-G Any part of Viewer Find next
Ctrl-Shift-G Any part of Viewer Find previous
--------------------------------------------------------------------

Note: Mac users should press the Cmd key instead of the Ctrl key.

9. The dialog programming documentation has been updated for Stata 9.
In addition, several bugs that were present in the dialog system
have been fixed.

A. The **stata** command did not echo the command in the Results window
as documented. This has been fixed.

B. Spaces that preceded text in a **put** command were not always
handled properly. This has been fixed.

C. **target(stata hidden)** was not accepted with u-action buttons as
documented. This has been fixed.

D. The **require** command did not work with heavyweight i-actions.
This has been fixed.

E. The **VERSION** statement did not work correctly when multiple
operating systems were specified. This has been fixed.

F. **onchange(***i-action***)** did not work with **BROWSE** controls. This has
been fixed.

10. Do-files that generate a **graph** and later produce an error, only
showed the error message but did not produce the error code. This
has been fixed.

11. Time-series **graph**s were limited to 100 tick marks. This has been
fixed.

12. **ktau** computations are now faster.

13. **Mata**'s object files (**.mo** and **.mlib** files; see **[M-1] how**) are now
stored a little differently. This means that if you have any
personal **.mo** or **.mlib** files, you must rebuild them from your source
code; see **[M-3] mata mosave** and **[M-3] mata mlib**. Because this
change is being made so close to the original release date, it is
not being made under version control.

14. **Mata** could mistakenly issue a subscript-out-of-range error for the
statement **d[i]=x** when **i** was a vector. Mata did this when any
element of **i** exceeded **length(x)** instead of when any element exceeded
**length(d)**. This has been fixed.

15. **Mata** produced a stack-overflow error, r(3998), in a rare case. This
has been fixed.

16. **quadchk** has new option **nofrom**. By default **quadchk** starts from the
previous estimation results. Adaptive quadrature is sensitive to
starting data, so specifying **nofrom** will often get slightly closer
results.

17. After **set varabbrev off** an expression using
**_b[***some_abbreviated_varname***]** worked when it should not have. This
has been fixed.

18. **xtlogit, re**, **xtprobit, re**, **xtcloglog, re**, **xtpoisson, re normal**,
**xtintreg**, and **xttobit** now allow up to 195 integration points in the
**intpoints()** option.

19. **xtlogit**, **xtprobit**, **xtcloglog**, **xtpoisson**, **xtintreg**, and **xttobit** when
specified with the **intmethod(ghermite)** option now allow the
**constraints()** option for specifying constraints.

20. The **set level** extended macro function (documented in Stata 7 and
earlier versions) failed to work in Stata 9. This has been fixed.

__Stata executable, Windows__

21. **cd** to a UNC path (that is, **cd \\computername\path***...*) returned
**c(pwd)** as an empty string rather than the correct path. This has
been fixed.

22. In Stata's various file dialogs, such as **File--Open**, if a user
attempted to open or save a file across a UNC path (that is,
**\\computername\path***...*), Stata would prepend an extra backslash
(that is, **\\\computername\path***...*) and cause an error. This has
been fixed.

23. **haver** failed to load datasets that contained more than 2048
variables. This has been fixed. Now, the maximum is determined by
the maxvar setting.

24. Stata for Windows incorrectly enumerated multiple instances of
Stata. This has been fixed.

25. The main Stata window remained hidden when launched until the
commands in **profile.do** completed. This has been fixed.

26. When a Viewer was launched upon startup of Stata (whether through
shortcuts, double-clicking smcl files, etc.) the Viewer never
appeared. This has been fixed.

27. Graph window positions were not properly loaded from a named
preference set. This has been fixed.

28. Copy (Ctrl-C) and Print (Ctrl-P) shortcuts failed to work in the
Graph window. This has been fixed.

29. Resizing a floating Results window could result in the output not
matching the dimensions of the window. This has been fixed.

30. The Results window, Viewer, and Data Editor now double-buffer their
contents to prevent them from flickering when resized or redrawn.

31. When the Variables or Review windows were clicked and text was sent
to the Command window, the right-click menu did not properly allow
for Cut/Undo/Copy operations. This has been fixed.

__Stata executable, Mac__

32. There is a new option to draw smooth fonts with QuickDraw and the
Quartz graphics engine rather than drawing smooth fonts using Quartz
directly. This allows for slightly faster, yet potentially less
accurate, rendering of text output. To enable or disable the
option, check or uncheck the **Draw text using QuickDraw** checkbox from
the General Preferences dialog or **set use_qd_text** {**on** | **off**}; see
**set use_qd_text**. When this option is enabled, the minimum font size
for smoothing is read from the System Preferences, not Stata's
preferences. This enhancement requires Mac OS X 10.2 or higher.

33. Some of Stata's menus that contain hierarchical menus do not display
an arrow to indicate that a menu item contains a hierarchical menu
while running Stata in Mac OS X 10.4 (Tiger). This is an issue in
Tiger, not in Stata. Apple is aware of this and has indicated that
it will be fixed in a future software update to Tiger.

34. Due to a change in how the default PATH environment variable is
defined in Tiger from previous versions of Mac OS X, Stata now
searches the user's home directory regardless of whether it is
defined in the PATH variable or not. This allows Stata to determine
if a profile.do file exists in the user's home directory when
running Stata on Tiger.

35. In Mac OS X 10.4.x (Tiger), Stata files could not be dropped on
Stata's Dock icon so that they will open in Stata. This has been
fixed.

36. Setting the y resolution of graphs copied to the Clipboard using the
Graph Preferences dialog had no effect. This has been fixed.

-------- **update 21jun2005** -----------------------------------------------------

__Ado-files__

1. Online help and the search index have been brought up to date for
Stata Journal 5(2).

2. **adjust** failed when an **if** condition involved literal strings. This
has been fixed.

3. **ml** now documents that **ml** **display** uses **e(k_eform)** to determine how
many equations to exponentiate when supplied with an *eform_option*.

4. **predict** with the **pr** option, when used after **probit** with the **svy**
prefix or the **vce(bootstrap)** or **vce(jackknife)** options, did not
produce the predicted probability values. This has been fixed.

-------- **update 17jun2005** -----------------------------------------------------

__Ado-files__

1. **areg** failed when attempting to fit a constant-only model. This has
been fixed.

2. **asmprobit** uses either **clogit** or **mlogit** for initial estimates. These
estimates are now scaled by the standard deviation of the extreme
value distribution, **c(pi)/sqrt(6)**.

3. **asmprobit** does not attempt to refine the standard
deviation/correlation estimates if the first call to **ml** (using the
Cholesky factored variance/covariance parameterization) does not
converge. In this case, a second call to **ml** is made to
reparameterize the estimates, but **iterate(0)** is used to prevent
another nonconvergent search.

4. **asmprobit**, **mprobit**, and **slogit** used inappropriate error codes. This
has been fixed.

5. **biplot** now allows plotting of any two dimensions, not just the first
two, with the new **dim()** option. **biplot** also now allows you to plot
negative column (variable) arrows with the new **negcol** option. Graph
options for the negative column arrows are specified with the new
**negcolopts()** option. New options **norow** and **nocolumn** suppress the
row points or column arrows.

6. **nlogit** now allows groups to have different sets of alternatives, so
unbalanced groups are now allowed. **nlogit** now saves **e(alt_min)**,
**e(alt_max)**, and **e(alt_avg)** providing the minimum, maximum, and
average number of alternatives per group.

7. **cluster measures** complained if a *measure* was not provided. It now
defaults to using the **L2** *measure*, as documented.

8. **dotplot**, when supplied with an **if** or **in** restriction, produced graphs
with an incorrect range for the frequency axis. This has been
fixed.

9. **_get_gropts**, a programmer's command, now recognizes the **altshrink**
option as a **graph** **combine** option when the **getcombine** option is
specified.

10. **newey** failed if a model contained variables with time-series
operators and one of those variables was dropped due to
multicollinearity. This has been fixed.

11. **newey**, like **regress**, showed "(dropped)" in the table of coefficients
for a dropped variable. Now, like most other commands, it displays
"note: *variable* dropped due to collinearity" at the top of the
output.

12. **predict** with the **pr** option, when used after **logistic**, **logit**, or
**probit** with the **svy** prefix or the **vce(bootstrap)** or **vce(jackknife)**
options, produced values of the linear prediction instead of the
predicted probability. This has been fixed.

13. **predict** *varlist***, reffects** after **xtmixed** now produces an appropriate
error message and exits with an error code if any of the variables
in *varlist* already exist.

14. **svy** estimation commands, when string variables were used to identify
strata or sampling units, incorrectly counted the corresponding
number of strata or sampling units resulting in incorrect variance
estimates. This has been fixed.

15. **svyopts** did not work as advertised in the Stata Press book, *Maximum*
*Likelihood Estimation with Stata, 2nd edition* (Gould, Pitblado, and
Sribney 2003). This has been fixed.

Programmers interested in getting their estimation commands to work
with the new **svy** prefix command should see the **Remarks** section of
program properties.

16. **svy: regress** and **svy: ivreg** exited with an inappropriate error
message when the sampling weights summed to a number less than 1.
Although this has been fixed, we note that the sum of the sampling
weights should sum to the population size (or at least estimate the
population size) for the DEFF design effect to be valid when using
an FPC in the first stage. DEFT is invariant to the scale of
sampling weights.

-------- **update 25may2005** -----------------------------------------------------

__Ado-files__

1. **asmprobit** help now documents that the options **basealternative()** and
**scalealternative()** may contain a number, a label value, or a string.

2. **biplot** did not label row markers properly when the **mlabel()** option
was specified and a subsample of the data was used. This has been
fixed.

3. **cabiplot** and **caprojection** now allow row and column marker labels to
be specified. This is accomplished with the **mlabel()** suboption of
**rowopts()** and **colopts()**.

4. **estimates replay** failed when called under version 8.2 version
control. This has been fixed.

5. **glm**, **mdslong**, **merge**, **predict** after **mds**, and **rolling** failed when the
temporary file directory path contained spaces. This has been
fixed.

6. **graph bar** and **graph dot**, after the 04may2005 update, displayed an
extra set of bars when the **sort** suboption was specified with the
**over()** option. This has been fixed.

7. **graph twoway** error messages have been improved when too many or too
few variables are specified for a plot.

8. **icd9[p] check** with the **generate()** option did not generate a variable
if all codes were valid. This has been fixed.

9. **mprobit** and **slogit** help now document that the option **baseoutcome()**
may contain a number or a label value.

10. **nl** when used with the pre-version-9 syntax accepted but ignored the
**robust**, **cluster()**, **hc2**, and **hc3** options. Now specifying one of
those options with the old syntax will result in an error message.

11. **predict** failed after **asmprobit** with a string **alternatives()**
variable. This has been fixed.

12. **predict** after **xtreg, fe** produced the linear prediction when options
such as **residuals** or **cooksd** were specified. An error message is now
issued instead.

13. **stepwise** incorrectly exited with an error for backward stepwise
regression (when using options **pr()** and **pe()**). This has been fixed.

14. **xtmixed** sometimes failed when given an **if** condition that used value
labels, for example, **if foreign == "Domestic":origin**. This has been
fixed.

15. **xttobit** erroneously treated observations for which the dependent
variable was missing as being right-censored instead of ignoring
them. This has been fixed.

-------- **update 04may2005** -----------------------------------------------------

__Ado-files__

1. **asmprobit** models with unstructured covariances produced an error
message during the reparameterization phase of the **ml** optimization
if the model definition for the variance/correlation parameters
exceeded 80 characters. This has been fixed.

2. **bootstrap** has a **group()** option, which was not documented in the
manual. See **bootstrap** for the updated online documentation.

3. **clogit** performed an extra, but unnecessary, calculation, which
caused it to run slowly on large datasets. This has been fixed.

4. **clogit** uses **logit** to generate starting values when none are
specified. For some models, these starting values resulted in
nonsensical **clogit** log-likelihoods. **clogit** now checks that starting
values make sense before using them.

5. **graph bar** and **graph dot** no longer require user-provided names when a
variable is repeated with more than one statistic. The dialogs for
these commands have been redesigned to take advantage of the new,
simplified syntax.

6. When managing multiple graphs or renaming graphs when used from
disk, Stata sometimes displayed the wrong graph in a Graph window.
This has been fixed.

7. **pause** executed subsequent commands under version 6 control. It now
executes them under the control of the caller's version.

8. **roctab** produced a different standard error for the area under the
curve when a monotonic transformation was applied to the rating.
This has been fixed.

9. **screeplot** has been enhanced to produce scree plots automatically
after **manova**, **canon**, **ca**, and **camat**.

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

11. **svy** dropped observations from the estimation sample that had zero as
the value of the sampling weight. This has been fixed.

12. **xtreg, be** and **xtreg, fe** reported a missing value for the model F
statistic when the **vce(bootstrap)** option was specified. They now
report the correct model Wald chi-squared statistic.

13. **xtreg, re** reported a missing value for the model chi-squared
statistic when the **vce(jackknife)** option was specified. **xtreg, re**
now reports the correct model F statistic.

-------- **previous updates** -----------------------------------------------------

See whatsnew8to9.

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