Software Updates

gr42 8: Quantile plots, generalized. N. J. Cox. Stata Journal 16: 813; Stata Journal 12: 167; Stata Journal 10: 691; Stata Journal 6: 597; Stata Journal 5: 470; Stata Journal 4: 97; Stata Technical Bulletin 61: 10–11 (reprinted in Stata Technical Bulletin Reprints, vol. 10, pp. 55–56); Stata Technical Bulletin 51: 16–18 (reprinted in Stata Technical Bulletin Reprints, vol. 9, pp. 113–116).

Some graph defaults have been tweaked. Default y-axis titles are now shorter: previously, they could be noticeably long. Horizontal y-axis labels are now the default.

The help file has been extended with more detailed discussion, more examples, and more references.

gr0053 1: Speaking Stata: Axis practice, or what goes where on a graph. N. J. Cox. Stata Journal 12: 549–561.

The help file for multqplot to draw multiple quantile plots has been expanded with

more details on use of the combine() option;

more comments on the relation of quantile plots to box plots;

an extended example showing use of the command to consider whether logarithmic transformations should be used; and

more details on control of axis labels and titles.

gr0061 3: Speaking Stata: Design plots for graphical summary of a response given factors. N. J. Cox. Stata Journal 17: 779; Stata Journal 15: 605–606; Stata Journal 14: 975–990.

Any attempt to use the missing option of graph dot, graph hbar, or graph bar is now ignored and advice on what to do instead is shown. Otherwise, changes are all to the help file, including a fix to an external URL no longer reachable; more examples of code for specific needs; and some other minor corrections and improvements.

st0534 1: mstatecox: A package for simulating transition probabilities from semiparametric multistate survival models. S. K. Metzger and B. T. Jones. Stata Journal 18: 533–563.

This nontrivial update makes two major changes to mstsample, which may affect previously estimated results.

mstsample now does its hazard calculations differently for semiparametric models to increase the estimates’ consistency regardless of sample size or the number of time points. The inconsistency affected the transition probability estimates from these models in some situations.

Models with time-varying covariates (TVCs) are affected above and beyond what item 1 above implies because of some other TVC-specific calculation changes. We recommend you install this update immediately. The relevant help files have also been updated to reflect the changes, where needed.

Additionally, there are tweaks to several of mstatecox‘s commands:

an error that would stop the command from continuing in certain situations (mstsample), and

better-displaying output (all other commands listed below).

More specifically:

Overall, in addition to the moremata package, mstatecox now requires you to install the ftools and gtools community-contributed packages. The two new required packages produce sizable speed gains for mstsample (five to six times faster).

mstcovar

• – Previously, some of mstcovar‘s output would display, even if mstcovar was wrapped in a quietly prefix or called from another command that suppresses output (for example, simulate). This is no longer the case.

• – mstcovar now also quietly posts the estimation sample’s mean for each covariate to help with mstsample‘s baseline estimates. It impacts nothing for the user.

mstdraw

• – For the transinfo option, the output for the transition summary list and transition matrix now displays. Previously, the output was suppressed.

• – You may now specify if for prgraph. Doing so will graph the data for the if condition you specify. It makes the most sense to use this functionality for the stub_Rslt_* variables, for example, to subset the results for a particular time range. See help file’s second example, last line.

mstsample

• – A major change: mstsample now calculates S(t) via H(t) for semiparametric models. Previously, it used the product-limit estimate. Using the product-limit estimate for S(t) led to smaller transition probability estimates in some situations, most notably, when there were relatively few time points. By calculating S(t) via H(t), transition probability estimates in these small-time-point situations will be more consistent. In situations with more time points, the difference between the two approaches is minimal. (mstsample has always calculated S(t) via H(t) for nonparametric models.)

• – A major change: For semiparametric models with TVCs, mstsample obtains H(t) via the baseline hazard contributions (Kalbfleisch and Prentice 2002, 114). The previous calculations were based on an assumption that was sometimes met, but often not. The transition probability estimates are now much more consistent across the board.

• – A major change: In addition to the default ftools and gtools optimization noted below, mstsample now has an additional speed option. With this option, the command is faster than the conventional calculations and produces identical final estimates. The speed gains come from how the final simulation output is processed, not from how the simulations themselves are executed. However, mstsample with speed specified will no longer save any datasets with the slicetrigger option, regardless of how small or large your overall simulation run is. In addition, you cannot use the path option with speed, and you also cannot recover the simulation results for every sim draw-subject (gen() variables with “SIMS: *” labels). You may get only the final processed simulation results, which you can still save (gen() variables with “RESULTS: *” labels).

• – For models without TVCs, mstsample now quietly reruns your Cox model with demeaned covariates. The coefficient estimates are unaffected, but the demeaning hugely impacts the accuracy and general stability of the baseline (cumulative hazard) estimate. It does not affect the dataset in memory, nor does it affect the initial stcox call in ereturn memory.

• – The code will now run faster because it now uses various functions from the gtools package. One command from the ftools package (join) is also possibly used, depending on how many observations are in memory. This is default behavior—there is nothing additional you have to do or specify other than ensuring ftools and gtools are installed.

• – mstsample would issue an error message and exit if you fit a Cox model a) with a partial tie correction (exactp) or marginal tie correction (exactm) and b) manually included TVCs using msttvc.

• – If you ran immediate back-to-back mstsample calls for a Cox model with tvc() specified, the transition probabilities from the second mstsample call would be off because of unanticipated behavior from stjoin. This has been fixed.

msttvc

• – Previously, the “!! - CAUTION” message would display, even if msttvc was wrapped in a quietly prefix or called from another command that suppresses output (for example, simulate). This is no longer the case.

• – The formatting of the “CAUTION” message, as well as the subsequent success message, is more readable.

• – The success message is now more explicit about Stata’s ability to report all hazard ratios correctly after msttvc.

mstutil

• – There are added checks for moremata, ftools, and gtools to be installed. If any of the packages are not present, mstutil will display the commands needed to install.

• – mstsample would issue an error message and exit if you used the toStage variable as stcox‘s strata() variable. mstutil now heads off that behavior.

Supplemental Material

Supplemental Material, gr0053_1 - Software Updates

Supplemental Material, gr0053_1 for Software Updates by in The Stata Journal

Supplemental Material

Supplemental Material, gr0061_3 - Software Updates

Supplemental Material, gr0061_3 for Software Updates by in The Stata Journal

Supplemental Material

Supplemental Material, gr42_8 - Software Updates

Supplemental Material, gr42_8 for Software Updates by in The Stata Journal

Supplemental Material

Supplemental Material, st0534_1 - Software Updates

Supplemental Material, st0534_1 for Software Updates by in The Stata Journal