Version History

Changes

1. Beginning in this version, the HB jump size does not change after burn-in iterations are completed. This change is due to feedback from academics and HB experts.
2. Updated terms of service.

Fixes

1. Includes various minor fixes.

Fixes

1. Fixed display issue with covariates when using high DPI settings.
2. Fixed problem copying models to clipboard.

Changes

1. This version allows users to get licenses by signing into Sawtooth Account.

Fixes

1. This version fixes an issue with respondent weighting in Logit runs. This can occur if the locale set in the software used to read CSV files is different than the machine’s locale. The weight values are not converted to numbers correctly, and result in respondents with extremely large weights. The resulting utilities are of poor quality.

This version has a new feature which allows for the import of model specifications from a CSV file.

This version fixes an issue where sensitivity simulations would fail if a dependent variable had a dependency on an independent variable.

1. When defining a scenario, if an attribute level is left blank and simulated without first filling a value, the blank will be interpreted as N/A. Previously it was interpreted as zero and could lead to an error message regarding extrapolation.
2. When creating an Excel simulator, an error is now displayed if no utility runs are specified.
3. The dialogs used in sensitivity analysis did not resize properly.

Simulations could fail with an error message "The specified file could not be found" even though the utility files were present and valid.

1. Utility archive files would become corrupt if the draws file exceeded 2GB in size. The software will now allow draw files up to 8GB.
2. The estimation time contained in the HB estimation log was incorrect.
3. The Academic Research license was not correctly recognized.
4. Excel simulators created by the software contained now show "Sawtooth Software, Inc." as the author.

Bug: Using MBC in a non-English Windows could result in point estimates files containing zero for the RLH value and an extra column following RLH containing large numeric values.

1. Bug: Projects with many continuous independent variables would cause error checking to become very slow.
2. Bug: Category labels with comma characters could make the utility run report have some odd table cell alignments.
3. Bug: Creating an Excel simulator in a non-English version of Windows could result in abnormally large utilities or an error message "Input string was not in the correct format."
4. Bug: Sensitivity analysis including continuous independent variables could result in an error message "Index was outside of the array."

1. Bug: The "Most Frequent Chosen Combinations" report was wrong in some circumstances. This was introduced in v1.0.1. This error did not affect the 1-way and 2-way counts reports; only the n-way combinatorial counts report. Note: the combinatorial simulations results were not affected by this bug.
2. Bug: The 'View/Edit' option for the data file no longer allows column sorting.
3. Bug: The time remaining shown during HB estimation did not correctly display times greater than 24 hours. This has been fixed.
4. Bug: Creating an Excel-based simulator would fail if there were more than 500 total categories in all independent variables.
5. Bug: Creating an Excel-based simulator would fail if a dependent variable was dependent on another dependent variable that came after it in the data file. Now, dependencies can be based on variables either before or after in the list.
6. Bug: Creating an Excel-based simulator would previously fail if a respondent was present in the utility run but not in the data file (because the user had deleted that respondent from the data after utilities had been run). We now provide an error message in this case.
7. Change: The "Effects of Independent Variables on Dependent Variables" computed probabilities using the conditional dependent variable filter rather than all tasks. Although either way can be correct, viewing probabilities based on all tasks seems more intuitive to users. We've changed this to use all tasks rather than filtered.
8. Change: Performance of the HB algorithm is faster due to better caching for memory. The speed increase can be quite substantial for large datasets.
9. Change: RLH is now on the original 0-1 scale, instead of the 1-1000 transformed scale previously used (we're not multiplying RLH by 1000 any more).
10. Change: The data file and demographic file browse buttons now default to the project folder.
11. Change: The 'Restore Defaults' button on the Filters & Weighting tab now also restores the defaults for respondent weights and task filters.
12. Change: We've improved the way MBC interacts with Windows' alt-tab behavior for switching between programs. Previously, the MBC program tile would not always appear in the list of programs.

1. Bug: Using constraints on a model that involved independent variables coded as 'excluded' could result in an 'index out of range' error.
2. Bug: Under certain circumstances the text 'NaN' could appear in counts reports.
3. Bug: Attempting to save simulation results when the file already existed and was open would result in a crash.

1. Bug: Combinatorial simulations have now been fixed after being broken in v1.0.1. The bug resulted in weighted draws giving the equivalent predictions as unweighted draws. (Weighted and unweighted draws lead to quite similar predictions, though empirical research suggests a slight edge for weighted draws.)
2. Bug: During estimation clicking on the progress bar no longer cancels the estimation. One must click over the 'Cancel' text to cancel.
3. Bug: Users no longer receive an error when removing a blank respondent filter at the end of the filter list.
4. Bug: After viewing the data file within the software, the progress bar now closes as expected.
5. Bug: Demographics files without a labels row now read correctly (this bug was introduced in v1.0.1)
6. Bug: When opening the scenario dialog, the 'Top combinations to display' is now initialized to the last value saved.
7. Change: When creating a scenario, right-clicking on the utility selection view will allow users to select the most recent runs for all dependent variables. (This saves you having to specify the most recent runs for all dependent variables, individually.)
8. Change: The number of covariates is now shown in the HB estimation statistics view. (Helps you verify that covariates are actually being used.)
9. Change: During estimation, each utility in the statistics view now includes the color of the parameter in the estimation graph. (Helps you better determine which color series on the graph refers to which parameter in your model.)
10. Change: When specifying models, an error message no longer appears if an independent variable has no predicting categories chosen when it is specified as being 'Excluded' from analysis.
11. Change: If the user has specified circular logic within the utility constraints area, MBC now reports specifically which constraint is involved in the circular logic. This helps users more quickly de-bug their constraints.

1. Change: Zero is now permitted as a code to indicate missing values in the data file (in addition to a blank field). In continuous independent variables, zero remains a zero in the design matrix.
2. Change: The degrees of freedom used in the non-linearity chi square test has been changed from the number of categories minus one to the number of categories minus two.
3. Change: In netted variables, the offstates are now available for selection.
4. Change: When adding constraints the dialog will inform the user how many constraints have been added.
5. Bug: Combinatorial simulations would generate an error when using a two-stage model where the second stage variables precede the first stage variable in the data file.
6. Bug: When missing data is present or conditional dependent variables are use, duplicate combinations would appear in the Most Frequent Combinations counts report.
7. Bug: Sensitivity simulations did not include first stage models (specified by conditional dependent variables) without including the first stage in a sensitivity table.
8. Bug: The option to build the variable table from file on the Variables tab did not function.
9. Standard errors for netted variables involving Logit utilities could show up as 'Nan' or zero. Standard errors to not apply to shares from Logit utilities.
10. Bug: Occasionally the error "Retrieving a collapsed varialbe as a non-group" would occur.
11. Bug: Weighted draw simulations would generate an error if a respondent had no applicable tasks to use for weighting.
12. Bug: Excel export of counts and simulations would include the same text in multiple columns, instead of displaying as merged cells.

Initial Release.