Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
st: Stata 13 ships June 24
"William Gould, StataCorp LP" <firstname.lastname@example.org>
st: Stata 13 ships June 24
Sun, 09 Jun 2013 17:03:36 -0500
Stata 13 ships June 24.
You can order it starting now.
It is a tradition that Statalist members are the first to know when we
have a new release.
Well, we have one. I suspect the following new features will be of
the greatest interest:
1. Longer strings, even BLOBs (Binary Large OBjects)
The maximum length of string variables increases from 244 to
Yes, I'm referring to the string variables that we all have in
our .dta datasets.
It works like this:
str1, str2, ..., str244
(working just as previously)
str245, ..., str2045
(new and working just as you would expect)
(new, beyond 2,045, and working just like str#)
strLs (pronounced sturls) can contain ascii or binary.
strLs are coalesced, meaning the same copy of a strL is shared
across observations to save memory.
You work with strLs the same as you work with str# variables.
All of Stata functions and commands work with them. This
includes -substr()-, -generate-, -replace-, -by-, -sort-,
-tabulate-, etc. The exception is that you cannot use a
strL variable as a key variable in a -merge-.
Programmers: strLs can be longer than local and global macros,
which have a maximum length of (only) 1,081,511 characters.
That means that all of Stata's string functions now work with
2. Treatment-effects estimators.
Treatment-effects estimators measure the causal effect of
treatment on an outcome in observational data.
A new suite of features allows you to estimate average treatment
effects (ATE), average treatment effects on the treated (ATET),
and potential-outcome means (POMs). Binary, multilevel, and
multivalued treatments are supported. You can model outcomes
that are continuous, binary, count, or nonnegative.
Different treatment-effects estimators are provided for
When you know the determinants of participation (but not the
determinants of outcome), inverse probability weights (IPW)
and propensity-score matching are provided.
When you know the determinants of outcome (but not the
determinants of participation), regression adjustment and
covariate matching are provided.
When you know the determinants of both, the doubly robust
methods augmented IPW and IPW with regression adjustment are
provided. These methods are doubly robust because you need to
to be right about either the specification of outcome or the
specification of participation, but not both.
Treatment effects are the subject of the all-new _Stata
Treatment-Effects Reference Manual_.
3. Endogenous treatment-effect estimators
As I said, treatment effects measure the causal effect of
treatment on outcome. Sometimes we do not have conditional
independence, which is to say, unobserved variables affect both
treatment and outcome.
The new endogenous treatment estimators address such cases.
-etregress- handles continuous outcome variables.
-etpoisson- handles count outcomes.
(-etregress- is an updated form of old command -treatreg-; -
etpoisson- is new.)
4. Multilevel mixed-effects and generalized linear
Existing command -sem- fits linear SEMs.
New command -gsem- joins -sem- and fits generalized SEMs.
Generalized SEMs is a term we have coined to mean
generalized linear response functions and to mean nested and
crossed effects, which can be used together or separately.
Generalized linear response functions include linear regression,
naturally enough, and they include probit, logit, complementary
log-log, Poisson, negative binomial. multinomial logit, ordered
probit, ordered logit, and more.
Nested and crossed effects means latent variables at different
levels of the data. 2 levels. 3 levels. More levels.
-gllamm- users: There is a lot of overlap in the models that
-gllamm- and -gsem- can fit. When there is overlap, -gsem- is
5. More multilevel mixed-effects models.
Stata already had multilevel mixed-effects linear, logistic, and
Now we also have probit, complementary log-log, ordered
logistic, ordered probit, negative binomial, and generalized
And all the commands -- even the existing ones -- now allow
constraints on variance components and can provide robust and
cluster-robust standard errors.
And the new commands are not only faster, they are bordering on
Mixed-effects regression now has its own manual.
The new -forecast- command lets you combine results from
multiple Stata estimation commands and/or other sources to
produce dynamic or static forecasts and forecast intervals.
Specify models. Specify identities. Obtain baseline forecast.
Specify alternative paths. Obtain forecast. That means
forecasts under alternative scenarios and ability to explore
impacts of differing policies. Especially useful for
7. Power and sample size
Solve for power, sample size, minimum detectable effect, or
Comparisons of means (t tests), proportions, variances, correlations.
Matched case-control studies, cohort studies, cross-sectional studies.
Standard and customizable tables and graphs.
And its own manual.
8. New and extended random-effects panel-data estimators.
Ordered probit and ordered logistic join the existing
random-effects panel-data estimators linear regression,
interval-data regression, tobit, probit, logistic, complementary
log-log, and Poisson.
Robust standard errors to relax distributional assumptions.
Cluster-robust standard errors for correlated data.
9. Effects sizes.
Results the way behavioral scientists and especially
psychologists want to see them.
Comparison of means: Cohen's d, Hedges's g, Glass's Delta,
point/biserial correlation. Estimated from data or from
published summary statistics.
Variances explained by regression and ANOVA: Eta-squared and
partial eta-squared, omega-squared and partial omega-squared.
Partial statistics estimated from data. Overall statistics from
data or from published summary statistics.
10. Project Manager.
Organize any kind of file (do-files, ado-files, datasets, raw
files, etc.) into hierarchical list for quick access.
Manage hundreds, even thousands, of files per project.
Manage multiple projects.
Create groups in project to categorize files.
Click to open datasets, display saved graphs, open do-files in the
Rename file. Filter on filenames. Search for file using keywords.
You have to try it to appreciate it, but in the meantime, you
can find pictures at www.stata.com.
11. Java plugins.
Call Java methods directly from Stata. Interact with Stata's
datasets, matrices, macros, etc. Take advantage of existing
Java libraries, or write your own code.
12. And more
I should mention the improved help-file searching, and that
Stata now supports secure HTTP and FTP, and fast PDF manual
navigation, and ordered probit with Heckman-style sample
selection, and the new way of estimating ML models without
writing an evaluator program, and the new fractional-polynomial
prefix command, and that quantile regression can now produce
robust estimates of standard errors, and that factor variables
now support value labels for labeling output, and the new way
to import data from Haver Analytics, and automatic
business-calendar creation, and the new import commands that
make reading data really easy, and how you can create Word and
Excel files from Stata, and solve arbitrary nonlinear systems,
and a lot of other things.
I could go on, but instead I'll mention that we have finally implemented the
feature that is the most requested at user meetings around the world:
You can now type -cls- to clear the Results Window.
* For searches and help try: