ADO-files What's New

Stata Ado-files — November 1998

Install this update if, when you type help whatsnew after entering Stata, the update history does not contain the line

        Update, November  1998 (from STB-46)

Summary of changes from “help whatsnew”

Note:   the updates are cumulative. (STB-46) denotes files that have been changed since the previous update.

• (STB-46) svymlog, svyolog, svyoprob
  The options deff, deft, meff, meft, prob, and ci did not work when specified at runtime (they did, however, work on redisplay). They now work.
• (STB-45) canon
  The reported model chi-squared was (slightly) incorrect; it is now canon previously issued an error message when either equation contained only one variable.
• (STB-45) reshape
  reshape long would confuse variables with names like v8 and v80 when v8 was mentioned but not exist.
• (STB-45) svytab
  New command svytab produces two-way tabulations with tests for independence. It is for use with complex survey data or other clustered data.
• (STB-45) xttab
  xttab did not calculate the within-percentage correctly in unbalanced data.
• (STB-44) heckman
  The reported model chi-squared was (slightly) incorrect; it is now calculated correctly.
• (STB-44) reshape
  reshape would produce the odd message "___ values would be changed" and then proceed to give correct results. The message is now suppressed.
• (STB-44) sts
  Specifying graph, twoway's ylog option with sts graph produced an error.
• (STB-44) table
  Standard deviations are now reported correctly when aweights are specified.
• (STB-44) xtgls
  xtgls, when computing AR(1) correlations by panel for a large number of panels, did not place all the correlations in the saved result S_E_rho. There were no errors in reported results.
• (STB-44) xtreg
  xtreg, fe now uses better logic for dropping of constant-within-panel variables. Previously, a constant variable might not be dropped and then results looked odd (but were correct).
• (STB-43) heckman
  heckman now has the capability to estimate models with variables that perfectly predict selection. Previously heckman would simply drop such variables from the selection equation, which is inappropriate in most cases.
• (STB-43) kap
  kap has been modified to deal with a large number ratings. Previously, it would issue the error message "too many ratings".
• (STB-43) kdensity
  kdensity now works correctly with aweights.
• (STB-43) stack
  describe, run after stack, no longer shows that the data is saved under a temporary filename.
• (STB-43) sts
  The new atrisk for use with sts graph shows the number of subjects at risk at the beginning of each interval; see help sts.
• stata52. (STB-42) sts
  sts graph now plots survival curves, etc., from time t=0 rather than from the first observed time of failure or censoring. The new noorigin option restores the original behavior.
• (STB-42) bstrap, bsample
  These commands did not work with the idcluster() option; they now do.
• (STB-42) kap
  One of the new syntax features introduced for kap in STB-41 — the ability to specify more than two variables — did not work; doing that returned a syntax error. All new features work now.
• (STB-41) fracpoly, fracplot, fracgen
  fracplot now produces a component+residual plot for Cox regression using martingale residuals. Previously, no residuals were used. (You can recover the old type of plot by using the undocumented nopts option.) In addition, the following changes to the output have been made:
  
  
  1. P-values from F-tests presented when option compare is specified have been corrected: previously they did not correctly account for the degrees of freedom associated with estimation of the FP powers. The change is very slight in most cases.
  2. The compare table has been reformatted slightly: a column for df has been added and the column previously labelled P is now labelled P(term).
  3. fracgen reports creation of new variable(s) followed by the phrase "if _sample", stressing the use of an estimation sample.
• (STB-41) lincom
  The new hr option allows the output to be labeled as Hazard Ratios, which is useful when lincom is used after survival estimators.
• stata51 (STB-41) mcc, mcci
  mcc and mcci now report the exact McNemar test in addition to what was previously reported. There is no change in command syntax. (Programmers: the new significance value is stored in the global macro $S_24.)
• (STB-41) ologitp, oprobitp
  These commands now work after the svyolog and svyoprob commands introduced in STB-40.
• (STB-41) reshape
  reshape now handles blanks in filenames correctly.
• (STB-41) svyolog, svyoprob, svymlog
  Previously these commands — introduced in STB-40 — did not handle zero weights properly; now they do.
• (STB-41) weibull
  weibull now converges more rapidly.
• stata49 (STB-40) kap
  A new absolute option is added for dealing with unobserved outcomes and a syntax addition allows kap to be used as an alternative to kappa.
• stata50 (STB-40) ttest, ttesti, sdtest, sdtesti
  ttest and ttesti now display standard deviations. In addition, string variables are allowed with by().

  sdtest and sdtesti previously reported incorrect one-sided p-values — the degrees of freedom where interchanged — the two-sided p-value was corrrect. This is fixed. In addition, sdtest and sdtesti now display standard deviations and string variables are allowed with by().

• svy6 (STB-40) svymlog, svyolog, svyoprob
  Three new commands, equivalent to mlogit, ologit, and oprobit, are introduced for estimating multinomial logistic and ordered logit and probit models using complex survey data.
• (STB-40) boxcox
  In rare cases, boxcox issued an error message when option mean was specified.
• (STB-40) cttost
  cttost issued the error message "invalid syntax" instead of "option invalid" when an invalid option was specified.
• (STB-40) ir
  ir ignored irrelevant options rather than issuing the error "option invalid".

• (STB-40) stack
  The new group() option provides an easier way to use stack. In addition, stack's into() option now understands variable ranges.
• (STB-40) stereg, ereg
  If score(newvar) was specified and the model was estimated in the log-time metric, stereg and ereg created newvar with the sign reversed. This is fixed.
• (STB-40) sureg
  Previously degrees of freedom were calculated as N-round(mean(k1,k2,...))*e, where N is the number of observations, k1, the number of estimated parameters less the intercept in the first equation, k2 the same thing for the second equation, etc., and e the number of equations. This has been changed to N-round(mean(k1,k2,...)).
• (Stata website) xtgee, xtpois, xtprobit
  Estimation of the exchangeable working correlation parameter was incorrect in the presence of singletons — groups containing single observations. In the presence of singletons, the estimated working correlation was attenuated; if the proportion of singletons was small, the effect was negligible. (This same bug affected S-plus and SAS's PROC GENMOD.)

  In addition, these commands are updated to change the calculation of the scale parameter. Formerly, N was used as the divisor; now N-k is used. Either will work in that any consistent estimate of phi is adequate, but most other implementations use N-k.

  There are two parts to this update: the updated xtgee.ado ado-file and an updated executable. Your Stata executable needs to be dated at least 14 Oct 1997 (type query born to the Stata prompt). If it is older than that, obtain the latest executable.

• (STB-39) bs
  Very long (over 600 character) bs commands could result in an "invalid syntax" error. The problem is now fixed.
• (STB-39) kap
  kap now displays tables with missing rows/columns more prettily.
• (STB-39) reshape
  All new syntax is introduced; the new reshape command continues to understand the old syntax.
• (STB-39) sw
  sw now may be used with nbreg.
• (STB-38) stbase
  stbase without the at() option produced incorrect results when either the data had gaps or there were multiple failures per subject. In such cases, stbase lost the observation on the first failure of each subject. stbase produced correct results when the data did not have gaps and there were no multiple failures per subject.
• (STB-38) ttest, ttesti
  ttest and ttesti now produce answers when one of the groups has only one observation.
• (STB-37) areg
  areg refused to produce estimates of the constant-only model when no regressors were specified.
• (STB-37) sts
  The t2title() option did not work; sts graph ignored it.
• (STB-37) svyratio
  svyratio with the by() option did not label the output clearly.
• (STB-37) xtgee
  The noconstant option did not work; xtgee refused to estimate the model.
• (STB-36) bs
  Previously, bs was supposed to have a cluster() option but did not. It now does.
• (STB-36) cs
  cs with the by() option produced incorrect Mantel-Haenzel combined risk ratio when one or more of the strata had infinite risk ratios (no cases in the strata). To produce the incorrect result, (1) you had to specify the by() option, (2) you must not have specified the or or rd options (which would cause cs to report something other othan risk ratios), and (3) there must have been one or more strata with infinite risk ratios (risk ratio reported as missing in the output). The incorrectly calculated result is now fixed.
• (STB-36) dotplot
  Previously, the mean median and nogroup options did not work: if you specified them, you got a syntax error. They now work. In addition, new options vert bounded and incr() have been added.
• (STB-36) lincom
  lincom now allows longer expressions.
• (STB-36) merge
  There was no problem. The on-line help file, however, incorrectly said that the nokeep option omitted observations that would have been marked _merge==1 when it should have said that it omitted observations that would have been marked _merge==2
• (STB-36) matrix
  There was no problem. The on-line help file, however, provided an incorrect syntax diagram for matrix accum
• (STB-36) prtest prtesti
  These commands now report the standard error under the null hypothesis in addition to the statistics previously reported.
• (STB-36) sts
  sts graph with the adjustfor() option could produce titles for graphs that were so long that Stata would then refuse to draw the graph.
• (STB-36) sw
  sw may now be used with clogit
• (STB-36) ttest ttesti
  ttest and ttesti now include a level() option to specify the level for the confidence interval.
• (STB-36) xtgee
  The criteria for detecting divergence of the model was updated.
• (STB-35) bitest, bitesti
  For certain extreme values of the test statistic, bitest and bitesti would report a two-sided p-value of 1 rather than 0.
• crc46 (STB-35) dydx, integ
  These commands have been improved and now fit a cubic spline to the data and use that to produce the derivative and integral.
• (STB-35) joinby
  Under rare circumstances joinby would produce a dataset with no observations when it should have produced something else. Note that if the old joinby produced a dataset that had any observations, it was correct.
• (STB-35) loneway
  When there were weights and missing values of the outcome or group variables, loneway did not produce the correct ANOVA table. This is fixed. In addition, you are warned that there may still be outstanding issues with this command concerning the reliability statistic and the intraclass correlation coefficient; see Gleason (1997), sg65: Computing intraclass correlations and large ANOVAs, STB-35.
• (STB-35) stir
  The note at the top that reports which values corresponded to the exposed and unexposed categories was backwards. Statistics were correct.
• crc45 (STB-35) stset
  Two new options have been added to stset first and force first specifies that events after the first failure event in time-varying data are to be ignored. force drops obsrvations with missing values of the time, outcome, id() t0() or weighting variables. See help stset.
• (STB-35) sw
  sw leaves behind variable _sample for use by other programs. This variable indicates which observations were used in producing results. _sample was incorrectly not set to 0 for observations containing missing values of the dependent variable even though such observations were not used. Note that the stepwise results reported were correct.
  
  In addition, two more changes were made: sw regress now allows the beta option. sw logistic now allows the robust and cluster() options.
• (STB-35) xtgee, xtprobit, xtpois
  Three changes are made. First, the divergence criterion for probit and logit models is relaxed, making the "estimates diverging" error message less likely. Second, the noconstant option may now be spelled out as the manaul indicates. Before the change, it had to be abbreviated. Third, the time variable t() is now required in the case of corr(fixed)
• bsample, bstrap, bs
  Due to an error in bsample, the cluster() option on all three of these commands did not give a proper bootstrap sample.
• egen ... pctile()
  egen's pctile() function now allows noninteger percentiles. Previously, only integer percentiles were allowed; noninteger percentiles resulted in an error message.
• (STB-34) mvdecode
  mvdecode has been improved to handle float variables containing decimal-encoded missing values.
• mvreg
  Degrees of freedom were reported incorrectly.
• (STB-34) prtest, sdtest, ttest
  These commands produced obviously incorrect output when the grouping These commands produced obviously incorrect output when the grouping variable was labeled and the label contained blanks.
  
  In addition, confidence intervals were labeled 95% even when other than 95% was requested (the requested confidence interval was, however, shown).
• sts graph
  sts graph with the lost option produced incorrect Kaplan-Meier plots (without the lost option results were correct).
  
  In addition, sts graph has been modified to display the Kaplan-Meier curve out to the last censoring or death. Previously, the curve was displayed out to the last death only.
• (STB-34) sw
  sw refused to allow pweights or iweights.
• (STB-34) xtgee, xtprobit
  When faced with certain inestimable models, xtgee and xtprobit would report "conformability error". The error message is improved. In addition, the checks for diverging models were improved.

  In addition, sts graph has been modified to display the Kaplan-Meier curve out to the last censoring or death. Previously, the curve was displayed out to the last death only.