dm0112_1: Estimating text regressions using txtreg_train. C. Schwarz. Stata Journal 23: 799–812.
This update to the txtreg commands fixes a bug that prevented the running of the commands in some versions of the scikit-learn Python package. The txtreg package also now provides a “requirements” file that allows the installation of all required Python packages by using the following terminal command:
conda env create --name [env_name] --file txtreg_requirements.yml
gr0092_1: Speaking Stata: Automatic axis labels: Nice numbers and transformed scales. N. J. Cox. Stata Journal 22: 975–995.
An update to the mylabels command fixes a bug that could bite if the options myscale() and clean were specified together. The fix provided an opportunity to simplify the code.
st0534_3: mstatecox: A package for simulating transition probabilities from semiparametric multistate survival models. S. K. Metzger and B. T. Jones. Stata Journal 21: 847; 19: 748–751; 18: 533–563.
We have automated nearly seven years’ worth of unit checks. Coverage is not 100% but is moderately high for all six commands. In relative terms, mstsample’s coverage is poorest because we use R’s survival and mstate packages to verify certain parts of our output, and mstatecox covers some scenarios that these R packages either do not cover easily or do not cover outright.
This update mostly addresses odds and ends uncovered by the automated unit checks, some more trivial than others. There is no new functionality. The help files have also been updated, where needed.
Notable changes include the following:
mstsample now produces identical output for gap versus clock time when both a) those two metrics are equivalent (for example, only one stage has exiting transitions) and b) stime() is not equal to 0. See https://github.com/hputter/mstate/issues/23 for more details.
mstsample no longer issues an error message when the Cox model in memory is nonparametric.
mstsample’s msfit subroutine has been rewritten to address a quirky edge case in the presence of collapsed transitions with different “from” stages. The uncommon situation could produce incorrect msfit quantities for certain subsets of t-values.
mstsample now properly includes any offset in its internal calculations. Its msfit subroutine also demeans any offset when it calculates the transition-specific baseline cumulative hazard. This demeaning behavior mirrors the subroutine’s existing behavior for any covariates in the main or tvc (if present) equations.
mstutil now requires you to specify the from() and to() options if the Cox model in memory has strata.
All the package’s non-r-class commands now respect information in r-class memory.
mstphtest’s r(global_
#
) matrices now have the same column order as those from r(phtest_
#
). That column order is χ
2, d.f., and p-value.
msttvc made calls to Stata’s random-number generator, which is behavior the user may not have expected. It no longer does so, meaning c(rngstate) is now guaranteed to be the same before and after calling msttvc.
You can view a detailed breakdown of all the changes via mstatecox’s GitHub repository: https://github.com/MetzgerSK/mstatecox/compare/st0534_2…st0534_3.
st0615_1: Multistate life tables using Stata. J. O. Muniz. 2020. Stata Journal 20: 721–745.
Ben Jann’s moremata package is no longer needed to run mslt. The obsolete Mata function mm_colrunsum() that was part of mslt has been replaced by Stata’s official runningsum() Mata function, eliminating the program’s previous code dependency.
A dialog box (mslt.dlg) is now available for implementing mslt. Type db mslt in the Command window, or after running the command once after Stata is invoked, select User ⊳ Statistics ⊳ Multistate lifetables (mslt).
The help file now informs the user how to access the mslt dialog box from the menu or by using the Command window.
A link to ilt (Muniz 2023) has been added to the Also see menu in the mslt help file.
st0675_1: Testing for time-varying Granger causality. C. F. Baum, S. Hurn, and J. Otero. Stata Journal 22: 355–378.
The Stata version required by this routine is Stata 14 because of its reliance on the runiformint() function introduced in that version.
st0696_1: Conditional evaluation of predictive models: The cspa command. J. Li, Z. Liao, R. Quaedvlieg, and W. Zhou. Stata Journal 22: 924–940.
We recently noticed a small typo in the previous code such that the command reports an error when the number of basis terms is set to 1 (m = 1). However, this bug does not impact any results presented in the article, because the number of basis terms in typical applications of the conditional superior predictive ability test is always greater than 1. Nonetheless, the case where m = 1 is of particular interest, because the conditional superior predictive ability test then simplifies to the unconditional superior predictive ability test (Hansen 2005). Therefore, we have corrected this bug and updated the package to aid empirical researchers in implementing both the conditional superior predictive ability and unconditional superior predictive ability tests using the same package.
