What is CBC/HB?
CBC/HB is a system for hierarchical Bayes estimation of individual level utilities using data collected with CBC. You may e-mail firstname.lastname@example.org if you have questions relating to information on this web page. We strongly suggest using the latest version of CBC/HB.
Version 5 History
11 January 2018
- This version fixes an issue with respondent filtering. This can occur if the locale set in the software used to read CSV files is different than the machine’s locale. Decimal values are not converted to numbers correctly.
15 December 2017
- If a project contained an user-specified attribute with only the two values of 0 and 1, the .hbu file would show the attribute contained a single level where it should show zero.
- Under Windows 10 the manual would not launch unless a specific PDF reader program such as Adobe Reader were installed. Windows 10 by default uses the Edge browser to display PDF files.
21 April 2016
This version includes some minor bug fixes, and the default prior variance has been updated to 1.0 (previously 2.0) based on recent research.
03 September 2014
- Iteration limits have been increased by a factor of 10: the limit for burn-in iterations is now 10,000,000; the limit for the number of saved draws is now 10,000,000; the limit for the skip draw factor is now 1000; the limit for the graphing skip factor is now 10,000.
- Utilities were not properly constrained in the case where dummy coding was used and the constrained attribute had only two levels.
16 August 2013
- Bug: When running a project that includes a none, if a user chose not to estimate the none option the estimation would fail with a message that the number of parameters was incorrect. This was introduced in v5.4.
- Bug: When using a .chs file containing a mix of tasks with and without none options, the build process would produce an error. This was introduced in v5.4.
- Bug: The Tie Draws now displays correctly when using Windows is using a DPI value higher than 96.
12 July 2013
- Bug: When using Dual-CSV files, and the design file does not contain the none option and the response file has constant sum with a none, the responses would not be read correctly. This issue is present in all releases of version 5.
- Bug: Tie Draws would read values from a legacy draw file incorrectly, resulting in abnormally large values in the results. This was introduced in v5.5.
- Bug: Tie Draws would result in an error if using a new CSV draw file when two levels of different attributes would have the same text.
- Bug: Pasting attribute or level text that contains spaces would result in only the first word of the text being pasted.
11 June 2013
- Feature: A new option to export the coded design is available from the Attributes tab under the Other Tasks button.
- Change: The RLH is now on a standard 0-1 scale within the CSV output files, instead of a 0-1000 scale. The .hbu files keep the 0-1000 scale in order to be compatible with SMRT.
- Change: Draw files are now exported as CSV files with a header.
- Change: Tie Draws no longer requires an .hbu file for attribute information. Tie Draws also can operate on the new CSV draw file as well as the previous .dra files.
- Bug: Rebuilding the attribute table from file would not refresh the information on the Attributes tab until the project was reloaded.
- Bug: A .cho file with an extra newline at the end of the file would result in a read error.
- Bug: Reading previous studies could result in an error.
07 May 2013
- Bug: Previous versions of v5.4 had an issue with the Tie Draws routine that would cause the software to crash.
- Bug: Occasionally there was an errant warning about the demographics file being open in another application.
- Feature: The software now recognizes Academic Research licenses.
16 January 2013
- Bug: v5.4.5 introduced a problem estimating MaxDiff data.
07 January 2013
- Bug: v5.4.0 introduced a problem with detecting a none option in single CSV files.
29 November 2012
- Bug: Single CSV data could cause a slowdown in the interface.
28 November 2012
- Bug: v5.4.0 introduced a problem with coding covariates from demographic variables. Runs with covariates should be rerun with this version; runs without covariates are not affected.
- Bug: When updating a project from a version prior to v5.4, the project was incorrectly identified as coming from CBC/HB v4.
- Bug: In the log file, the section describing the number of covariates occurred twice.
23 October 2012
- Feature: CBC/HB now checks for circular logic in constraints.
- Change: Small modifications to the XML layout file used by the Online Simulator.
- Bug: An error would occur running utility calibration (introduced in v5.4).
- Bug: When checking a .cho or .chs file for errors, the file would remain locked after finishing.
- Bug: An error would occur when removing a blank respondent filter from the end of the filter list.
- Bug: When viewing a CSV data file within the software, the progress bar would remain after reading the file.
- Bug: An error would occur reading a demographic file with no labels (introduced in v5.4).
28 September 2012
- Note: The following issues are only present in v5.4, and are the result of the extensive changes made in the addition of being able to estimate best/worst data.
- Bug: MaxDiff data in a single CSV file could sometimes errantly detect a none option.
- Bug: Importing a .con file containing constraints would give an error.
- Bug: When using Best/Worst data, estimating worst tasks only with a dual-response none resulted in incorrect coding of the design matrix.
- Bug: When importing MaxDiff data, the system would compare generic attribute labels to those found in the .val file.
- Bug: When estimating MaxDiff data, if the attribute codings were (incorrectly) set to part worth, an 'Out of bounds' error occurs.
- Bug: When estimating Best/Worst data, dual-response none information would code as missing if best- or worst-only tasks were selected.
- Bug: Task fitlers were not being applied to estimation.
17 September 2012
- Feature: CBC/HB now supports Best/Worst CBC data.
- Change: The default starting seed is now 1 rather than 0 (which denotes to use a random starting point). This should reduce confusion as to why utilities are not exactly the same as previous runs.
- Bug: Dual CSV file format had a problem in the way that missing data were read.
- This update contains various fixes and updated components.
29 May 2012
- Bug: Constraints on user-specified parameters were not applied correctly.
- Change: Previously the default task weight for constant-sum files was 5.0. The new default is 1.0, with a warning if greater than 1. This is because the majority of single-CSV files have discrete choices instead of constant-sum.
- This update contains various fixes and updated components.
15 July 2010
- Bug: Version 5.2.4 introduced an issue which causes ACA/HB 3.2 and SSI Web 7 to not load after the update.
- Bug: Numbers were superimposed on the checkmarks of the task list.
- Bug: When using Dual-CSV import, a cryptic message occurs if a version cannot be found in the design file.
12 July 2010
- Bug: Some licenses would cause a 'Value is null' message to be displayed on startup. This message did not affect the software adversely.
- Change: CSV header labels are now being formatted in a more standard fashion.
10 May 2010
- Bug: Using interactions and dummy coding would produce inaccurate results. This error was introduced in v5.
- Bug: Some .val files could not be read correctly.
- Bug: The SetUseConstraints command in the interpreter would not function correctly.
- Bug: Utility calibration would sometimes fail using linear attributes.
- Feature: The list of covariates is now included in the log file.
- Change: Paste operations in the attribute list has been improved.
- Change: The default window size has been enlarged to solve some display issues.
12 January 2010
- Bug: Using constraints would result in being unable to estimate the model with a 'too many constraints' message.
18 December 2009
- Feature: The command interpreter will now import all v3.x text control files (.eff, .val, etc) when a project is created.
- Feature: The command interpreter has a new command (SetCustomPriorCovarianceMatrix) which allows users to specify what the diagonal and off-diagonal elements should be.
- Feature: Automatic detection of MaxDiff data has been improved. The attribute table will automatically recognize MaxDiff data and set the attribute codings accordingly. Also when possible the prior covariance matrix will also be automatically set for MaxDiff data.
- Feature: When adding constraints, there is a new option to enter them using a grid with a layout similar to the older .con files.
- Feature: Copy and Paste buttons are now available to paste attribute and level labels.
- Feature: CSV data files will use the header row to get attribute labels when available.
- Feature: CSV data files now allow alphanumeric respondent numbers. The resulting CSV utility files will also use alphanumerics. Since the .cho, .chs, .hbu, and .dra files are older formats and must remain compatible with other software, they will not be updated to handle alphanumerics. If the data files contains non-numeric respondent numbers, the respondents in these files are renumbered sequentially.
- Change: The memory managment system of CBC/HB has been overhauled to improve performance on larger datasets or when saving draws. Because of that, manual configuration of memory files is no longer necessary and has been removed.
- Change: This version no longer supports Windows 2000.
- Bug: Single CSV data files did not recognize MaxDiff data correctly. This would result in an error regarding a 'none option' which did not exist.
28 September 2009
- Bug: Dual CSV file layouts did not read constant sum responses correctly.
14 September 2009
- Bug: Command interpreter 'SaveAs' command would not save in certain cases.
- Bug: A debug-style message box was shown when changing the file type from .cho to .csv.
- Bug: Email registration failed to include product information.
24 August 2009
- Feature: Estimation with covariates is now available.
- Feature: CSV files (in either a single or dual design/response format) can now be used as input files instead of .cho.
- Feature: Utilities can now be calibrated using the new calibration tool, similar to the old CALIB utility.
- Feature: CBC/HB v5 runs natively on 64-bit machines.
- Feature: Ability to specify prior alpha mean and variance.
- Feature: New command interpreter allows scripting from external applications.
- Change: Output files have now been converted to CSV or text format for easier post-processing.
- Change: Constant sum tasks are no longer 'expanded' in the build process, resulting in increased performance.
Version 4 History
11 November 2008
- Bug: Using two linear attributes in an interaction caused the none option to appear as 0.0 for all respondents. This is present in all versions of CBC/HB.
- Bug: The label in the hbu file for two linear attributes in an interaction was missing.
18 July 2008
- Bug: Using constraints on an interaction when more than one interaction was present could result in incorrect utilities or a memory error message.
- Bug: Saving draws where there is an interaction between two linear terms would result in the draw file missing the last parameter.
30 January 2008
- Feature: Build now includes a cancel option.
- Feature: Batch estimation shows progress information.
- Bug: Memory problems could occur on machines with more than 2GB of memory.
- Bug: An error occurred when reading a VAL file that had different information than the ATT file.
04 September 2007
- Bug: Missing responses when using dual-response none were coded incorrectly. The resulting build file would not run. It did not cause incorrect utilities.
- Bug: Tie Draws could not read .hbu files formatted as UTF-8, which is the default format created by the engine.
17 July 2007
- Change: Updated to .NET Framework 2.0 for Windows Vista.
- Change: Without a license CBC/HB will now work in demonstration mode.
- Bug: Files using constant sum and none options were coded incorrectly.
09 February 2007
- Bug: Saving draws could occasionally lead to an 'Unknown error' message during estimation.
- Bug: Build would not work correctly with missing dual-response none data.
19 July 2006
- Feature: New options to turn off writing of .hbu, .log, and .res files. These options are located in Edit | Preferences.
- Feature: New option to graph either population averages or population exponential moving averages. This option is located in Edit | Preferences.
- Bug: Interaction terms between two attributes coded as part worth (categorical) were being coded incorrectly. This bug resulted in writing out the interaction coefficients in the wrong order, which also led to their incorrect application within the SMRT market simulator.
- Bug: The 'Tie Draws' procedure would crash if an attribute was excluded.
15 March 2006
- Change: Management of User Ids has changed to allow import of new .userid files.
- Change: The way memory is managed during estimation has been updated. Very large datasets under some systems would run slowly.
- Change: Error checking while reading .cho files has improved.
- Change: During estimation warning messages now appear to inform the user that performance is lagging because either there is not enough memory or they are working from a network drive.
- Bug: During import, .eff files would import Excluded attributes as Part Worth.
- Bug: Some fatal errors during estimation would close the estimation window without being reported.
- Bug: Better handling of writing hbu files during estimation; an error occurred if the hbu was open in another editor, and some system settings can cause the hbu to not be written after long runs.
04 October 2005
- Feature: New update feature will prompt users when a newer version is available. This can be turned off in Edit|Preferences.
- Bug: Tying draws with user-specified attributes could cause the software to terminate abruptly.
- Bug: The Total Task Weight and Prior Variance fields were not correctly formatted for countries where the decimal point is not a period ('.').
- Bug: The build process would encounter an error when coding linear variables in an alternative-specific design.
- Bug: Creating a new project did not check for a previous project of the same name.
- Bug: Respondent filters were not being imported from QAL files correctly.
- Bug: Although constraints and prior covariance matrices were imported correctly, they were not enabled by default.
14 September 2005
- Feature: Eliminating of all text-only control files.
- Feature: New estimation screen displays a visual graph of the parameters to help more easily assess convergence.
- Feature: Batch mode processing of multiple projects.
- Feature: Support for "Dual-Response None" data.
- Feature: Ability to add notations for the project.
- Feature: Ability to constrain parameters for user-specified or linear variables to be less than or greater than another user-specified or linear variable.
- Feature: Tie Draws program now supports interaction terms.
Version 3 History
08 November 2004
- Feature: A CSV output file is now created along with the HBU when tying draws.
- Change: The samples folder has been moved from the user's 'My Documents' folder to the 'Program Files\Sawtooth Software\CBC HB\Samples' folder. This addresses problems with administrative installs.
- Bug: During estimation, if the time remaining was greater than 10 hours, when the number fell below 10, the '1' would be left on the screen. Similarly, if the time started out greater than 100 hours and fell below, etc.
- Bug: If the CHO file was malformed such that a -1 would be read as the number of concepts for a task, the software would lock up or crash.
- Bug: If the EFF file specified an interaction but not a main effect for either attribute in the interaction, estimation would crash.
- Bug: If specifying linear coding (pound notation) in the CHO file, an invalid code would crash the build process.
29 March 2004
- Feature: Proper prior covariance matrix values have been introduced when calculating main effects only and using dummy coding. When dummy coding is used with interactions, the matrix will revert to the identity matrix.
- Feature: Users may now specify their own prior covariance matrix by using a .mtrx file.
- Feature: When using effects coding, constraints may now be specified on attributes involved in interactions.
02 January 2004
- Bug: CHO response range checking was flagging 0 (not answered) as invalid.
29 December 2003
- Bug: When doing the build step, .CHO files did not check to see if the response for a task is within the range for that task.
- Bug: Tying draws was not reading the constraints file correctly.
03 October 2003
- Bug: Converting a .CHO to a .CHS would insert a blank line after the first line of a respondent when no extra variables ("second line" information) are included.
- Bug: Converting a .CHO to a .CHS would incorrectly round decimal values to integers if using pound notation in the .EFF file.
- Bug: If a 'None' option was included in the data file but the researcher decides not to include the 'None' option in the estimation, the build program did not interpret the data correctly. (NOTE: This bug is present in all versions of HB-Sum, which was replaced by this product.)
- Bug: When using dummy coding and specifying monotonicity constraints involving the omitted level, the results were incorrect.
- Bug: Values in the prior covariance matrix were incorrect for 1) a non-linear attributes' main effects and 2) an interaction with a linear attribute.
17 September 2003
- Bug: Constraints for linear attributes were not being read correctly if the first level of the constraint was negative.
- Bug: Pressing the F1 key would ask the user to browse to a help file (which doesn’t exist). Now it will launch the manual.
09 September 2003