发帖
雷达卡
链接：https://pan.baidu.com/s/1XEUyQ_9zg6wGoTbOPrwUBw
提取码：vknq
-------- update 02jul2019 -----------------------------------------------------------------------------------------------------------------------------------------------------------------------
1. Online help and the search index have been brought up to date for Stata Journal 19(2).
2. bayes: mlogit, when labels of the outcome variable contained white space, did not recognize the base outcome level and exited with a Mata error message. This has been fixed.
3. bayesmh and bayesstats summary, both with option batch(), incorrectly reported batch standard deviation instead of the usual standard deviation. This has been fixed.
4. binreg has the following fixes:
a. binreg with option iter(0) incorrectly exited with an uninformative error message instead of calculating the deviance without iterating. This has been fixed.
b. binreg, when maximization options search and from() were specified without option ml, did not exit with an error message. This has been fixed.
5. eivreg has the following fixes:
a. eivreg used the wrong formula to compute standard errors, resulting in standard errors that were too small. This has been fixed.
b. eivreg postestimation commands forecast, linktest, and predictnl are no longer available after eivreg.
c. predict after eivreg no longer allows options residuals, stdp, stdf, pr(), e(), or ystar().
d. margins after eivreg no longer allows option predict(statistic) for the following statistics: residuals, stdp, stdf, pr(), e(), or ystar().
6. fmm with set emptycells drop in effect exited with an unhelpful error message, even when the model was specified correctly. This has been fixed.
7. glm has the following fixes:
a. glm with options irls and iter(0) and any combination of options family() and link() except family(gaussian) with link(identity) incorrectly exited with an uninformative error
message instead of calculating the deviance without iterating. This has been fixed.
b. glm, when maximization options search and from() were specified with option irls, did not exit with an error message. This has been fixed.
8. import delimited, when option case() was not specified or when case(lower) was specified, and when variable names in the file to be imported were Stata reserved words (for example,
if or in) also containing capital letters (for example, If or IN), failed with an uninformative error message. This has been fixed.
9. import fred exited with error message "I/O error" when the FRED server returned an HTTP error code 429 (too many requests). This occurs when the number of requests to the FRED
server exceeds limits imposed by FRED. Now Stata reports the note "Too many requests reported by FRED server; please wait." and then proceeds to wait for a short time before
automatically trying the request again.
10. ir and stir, both with option ird, ignored options estandard, istandard, and standard(). An error message is now produced.
11. irt, when an item variable contained all missing values, exited with a noninformative error message. irt now exits with error message "item ... contains only missing values".
12. lincom unintentionally left behind matrix rtable in memory. This has been fixed.
13. menl, when fitting a linear mixed-effect model containing factor variables with more than two levels, ignored the base specification, thereby producing an overparameterized model.
This has been fixed.
14. mi estimate has the following fixes:
a. mi estimate with estimation command clogit would not display notes about multiple positive outcomes within groups or how many groups were dropped during estimation. Now the
appropriate notes are displayed.
b. mi estimate with option cmdok and estimation command mestreg incorrectly exited with error message "command prefix mestreg <cmdline> not allowed". Now mi estimate, cmdok:
mestreg executes without error.
c. mi estimate, when estimation_command contained an if expression with a single bar, "|", inside parentheses, (ifexp1|ifexp2), exited with a syntax error message. This has been
fixed.
15. mi impute chained has the following fixes:
a. mi impute chained with functions of imputed variables did not generate these variables in double precision. This could affect the precision of imputed values when, for example,
the original imputed variable involved in a function was an integer and the regression imputation method was used to impute it. This has been fixed.
b. mi impute chained, when the expression in option include() contained the same operation sign multiple times, such as include(x1 + x2 + x3), produced an uninformative error
message. This has been fixed.
c. mi impute chained with option savetrace(), when there were multiple imputed variables with the same prefix names, produced an uninformative error message. This has been fixed.
16. pkequiv has the following improvement and fixes:
a. pkequiv with option limit() accepted equivalence limits only from 0.1 to 0.99, inclusive. Now it accepts equivalence limits from 0.01 to 0.99, inclusive.
b. pkequiv, with unbalanced data (different numbers of subjects in each sequence) or missing values in any variable, produced incorrect results (it used the formulas for balanced
data). pkequiv now properly accounts for unbalanced data in the computation.
c. pkequiv did not respect if exp or in range. This has been fixed.
d. pkequiv, when some ids did not receive both treatments, miscalculated the degrees of freedom, resulting in confidence intervals that were slightly too narrow. This has been
fixed.
e. pkequiv with option tost miscalculated p-values for the two one-sided tests (TOST), reporting p-values that were twice as large as they should have been. This has been fixed.
f. pkequiv with option tost or anderson and with option limit() ignored the specified limits when computing the TOST or Anderson and Hauck test. It incorrectly used the default
equivalence limit of 0.2. This has been fixed.
g. pkequiv without options noboot, symmetric, or fieller used an unstratified bootstrap to calculate the probability that the confidence interval would lie within the equivalence
limits if the experiment were repeated. This has been corrected to stratify the bootstrap samples on the sequence in which treatments were received.
h. pkequiv, when the treatment variable took values other than 1 and 2, produced incorrect results. This has been fixed.
i. pkequiv with option compare(), with any value of compare() except compare(1 2), produced incorrect results. This has been fixed.
j. pkequiv with option fieller miscalculated the confidence interval and mislabeled it as equivalence limits, which were not reported. It now correctly reports the equivalence
limits and confidence interval for the ratio.
k. pkequiv with option symmetric mislabeled the confidence interval for the test formulation as equivalence limits, which were not reported. It now correctly reports the
equivalence limits and confidence interval for the difference in means.
l. pkequiv without options fieller or symmetric displayed the confidence interval for the difference in means under the heading "test limits". It now uses the heading "confidence
interval".
m. pkequiv with options fieller or symmetric, when the sequence variable contained an odd number of observations with sequence equal to 1 or 2, incorrectly exited with error message
"option df() incorrectly specified". This has been fixed and no longer causes an error.
17. pkshape has the following improvements and fixes:
a. pkshape now creates variable labels for the new variables created.
b. pkshape now uses the information about treatments provided by users to label values of the sequence, treatment, and carryover variables.
c. pkshape now supports Unicode characters in variable names and treatment labels.
d. pkshape, when more than nine treatments were used, produced incorrect results. In those cases, pkshape would mislabel one or more treatments, which would lead to incorrect
results. This has been fixed.
e. pkshape, when numbers 1-9 were used as treatment codes, would produce erroneous treatment and carryover values. This has been fixed.
f. pkshape with option order() and a string input sequence variable did not produce an error message as it should have and ignored the treatment sequence provided by the input
sequence variable. This could produce incorrect treatment assignments when the specified ordering in order() was not alphabetical. pkshape now produces an error message in this
case.
g. pkshape exited with an error message without preserving the original data when any of the following variables existed and the corresponding option was not specified with a new
variable name: variable outcome and option outcome(), variable treat and option treatment(), variable carry and option carryover(), variable sequence and option sequence(), and
variable period and option period(). Now data are preserved.
h. pkshape now saves outcome variables as type "double" if the input variables were saved as type "double".
i. pkshape, when period variables except the first two contained nonnumeric values, exited with a misleading error message. It now produces an appropriate error message.
j. pkshape, when replicate combinations of id and sequence were found, exited with a misleading error message. It now produces an appropriate error message.
18. predict after menl, when specified with a linear combination containing an interaction between a continuous variable and factor variable, produced an uninformative error message.
This has been fixed.
19. qreg with prefix by ran the model k x k times instead of k times, where k is the number of by-groups. This has been fixed.
20. stir without option strata() produced error message "missing by() option" when any of the following options were specified: ird, estandard, istandard, standard(), pool, nocrude, and
nohom. It now produces error message "missing strata() option".
-------- update 21mar2019 -----------------------------------------------------------------------------------------------------------------------------------------------------------------------
1. Online help and the search index have been brought up to date for Stata Journal 19(1).
2. asclogit with option vce(cluster clustvar) using a string clustvar variable, when the case variable was not nested in clusters, incorrectly reported results instead of producing an
error message. This has been fixed.
3. churdle with frequency weights reported the wrong number of observations. This has been fixed. The reported coefficients, test statistics, and confidence intervals were correct.
4. compare, in cases with extreme double-precision values outside the range of float precision (such as c(mindouble) or c(maxdouble)), could show missing values in the output table.
This has been fixed.
5. estat sd after mixed now allows option coeflegend when option post is also specified. coeflegend requests that estat sd display a legend that identifies how posted coefficients,
standard deviations, and correlations should be specified in expressions.
6. estat sd after meglm, melogit, meprobit, meologit, meoprobit, mepoisson, menbreg, meqrlogit, and meqrpoisson no longer allows option coeflegend unless option post is also specified.
The old behavior is not preserved under version control.
7. After the 20feb2019 update, estimates table specified with results containing two-part named free parameters, such as /mills:lambda of heckman, twostep, exited with error message
/mills not found. This has been fixed.
8. nlogit with option vce(cluster clustvar), when the case variable was not nested in clusters, incorrectly reported results instead of producing an error message. This has been fixed.
提升卡
置顶卡
沉默卡
变色卡
抢沙发
千斤顶
显身卡
京公网安备 11010802022788号
