General Comment: In Datab, the maximum potential number of cells in a table is 16,777,216. There is a another version with a limit of 2.1 billion cells available. It is included with the UNIX versions, (labeled Databd) and can be obtained for the PC version.
PC Version 2.0 Released July 1996, Version 2.11 is now available for free to registered users. Version 2.11 includes the following changes:
- The implementation of the Moolgavkar-Venzon algorithm used by the BOUNDS command was improved and problems that could lead to improper reports of convergence for some problems in which there were user-defined range restrictions on the parameter space were fixed. When a bound cannot be found, the output was modified to provide a clearer and more concise summary of the tail area.
- The TCHECK command was modified to correct problems with the output for stratified models.
- In Gmbo, Pecan, Peanuts and Amfit, sometimes when the bounds command is used to compute a likelihood bound for a parameter with a user-specified range, the user-defined bound will erroneously be reported as the likelihood bound. This problem is fixed in 2.07 Nov 1997. The problem occurs in all earlier versions. In most situations, the problem can be avoided by using the PROFILE command instead of the BOUNDS command or by removing the user-defined range restriction.
- In Gmbo, Peanuts, and Amfit prior weights were not handled correctly in some cases.
- In all programs, if the total number of categories in a table produced by the FREQ command is greater than 1500, the counts are incorrect.
- In Gmbo, stratified logistic regression models are not handled properly resulting in a failure to converge and incorrect standard errors if the model includes both strata and model parameters. There is no workaround for earlier versions.
- In Gmbo, in stratified models where there are strata in which all of the records are cases (i.e. P=1 and the stratum parameter is infinity), the program sets P=0. This inflated the deviance by a constant but did not affect the parameter estimates or their standard errors. However, the Pearson Chi-square statistic was inflated by a huge amount. The contribution for records in the affected non-informative strata should be 0 but was (n - ~0)/~0. i.e. essentially infinity (In reality this is typically 10**18).
- In Gmbo and Amfit, the RESIDUAL command must be given (once) before parameter-specific delta beta values can be computed. ( This will be fixed in release 2.07)
- In Pecan, incorrect records are read from a BSF file created from a Pecan data set in which some of the potential case-control sets are non-informative, i.e. they had no cases or no controls.
- In Pecan, the count of the number of records used printed in the model summary is incorrect for data sets in which there are non-informative case-control sets. All other parts of the parameter summary are correct.
- In Peanuts, there is a problem with the KM command when both the TO and BY subcommands are given. In this case, the file that is written will only contain data for the first BY group. The workaround is to use the STRATA command to define the BY groups and then use the KM command without the BY but with the TO subcommand.
- Peanuts, Case-cohort models: If you are fitting a model to case-cohort data and are using the sampling fraction, the estimated survival curve is not adjusted for the sampling fraction. This feature does work properly in Version 1.8, but not in 2.0.
- Peanuts, Case-cohort models: Under some circumstances, the program will crash when carrying out the case-cohort variance adjustment when there are a large number of tied failure times.
- The Datab program is case-sensitive with respect to the names of rate file. Thus, Myrates.oxr and myrates.oxr are recognized as different files.
- In Datab, the resolution value is ignored for the first time period in tables that involve a duration scale (i.e. those using the DURATION command). You can work around this problem by defining the length of the first duration category to be less than or equal to the desired resolution. This problem does not occur for tables defined in terms of date (i.e. those using the CALENDAR or TIME commands).
- In Datab, the DROP command does not function properly if there are more than 4 items in the drop list. The workaround is to drop variables one at a time.
- In Datab, the SUMMARY command to create default summary variables in collapsed tables does not handle FCOUNT type variables properly. All values of the FCOUNT variables are zero in the collapsed table. The workaround is to use TRAN to create a new variable that is a copy of the FCOUNT variable and to use this new variable to create a SUM type summary variable in the new table.
New Features Available:
- A new table LOOKUP function can be used in transformations in Datab or any of the other Epicure programs. The function uses specially formatted n-way tables. Complete documentation of the new function, the table format, and utilities to convert the existing rate tables to the new format is included in the update.
- The PSAVE command has been updated to automatically include the parameter covariance matrix as a part of its output. With the new format, the information for each parameter is followed by the data for the row of the lower triangle on the covariance matrix for that parameter. If the parameter is fixed or aliased the elements in the corresponding row (column) are all 0.
- The syntax and output format for the TCHECK command (used to print summary information about risk sets and time-dependent covariates in Peanuts) has been modified. The new syntax is:
TCHECK varlist [; [TO fname] [HEADER] [FROM lotime] [UNTIL hitime]] @
where varlist is a list of up to 50 covariates (increased from 10) to be written; fname is the name of a file that will contain the ouput. The file will include an Epicure command header if you specify the HEADER option. If output is written to a file it will not be written to the screen. (This will save considerable time.) The FROM and UNTIL subcommands are used to limit the output to those risk sets with failure times in the specified range.
When the data are written to a file, the format is designed to make it easy to input the data into other programs, e.g. Pecan for exact analyses of tied failure times. The output includes: the riskset number (SETNO), a case control indicator (CC), and the current time (%time) as well as the data that you choose to write.
If you would like to receive version 2.10, please send us email us at firstname.lastname@example.org
Last updated October 12, 1999
UNIX Version 2.10 for the Sun and Dec Alpha
Version 2.10 was released for Sun and Dec Alpha workstations on October 8, 1998.
Version 2.04 is now available for the HP.
- In Gmbo, in stratified models where there are strata in which all of the records are cases (i.e. P=1 and the stratum parameter is infinity), the program sets P=0. This inflates the deviance by a constant but does not affect the parameter estimates or their standard errors. However, the Pearson Chi-square statistic is inflated by a huge amount. The contribution for records in the affected non-informative strata should be 0 but is (n - ~0)/~0. i.e. essentially infinity (In reality this is typically 10**18). IF YOU SUSPECT THAT YOUR DATA MAY FIT THIS CASE, please send us email at email@example.com
- In Amfit and Gmbo, the RESIDUAL command must be given (once) before parameter-specific delta beta values can be computed.
- The effective limit on the number of parameters is 50. If there are more than 50 parameters, the parameter boundaries may be set to strange values leading to incorrect fits.
Problem on the Dec Alpha only
- When the workspace size is set to anything over 15,000,000, DATAB will not correctly compute the sums in the first cell of the collapsed table. If you need a workspace this large, you should use the double precision version of DATAB. Otherwise, please use a smaller workspace.
Return to HiroSoft's Home Page
Back to epicure overview