If you discover a new bug please do let us know - go to our bug report form.

Fix bug where the window was closed prematurely if the user clicked "include all" when adding a term to the model

Fix a crash that occurs when the number of parameters in the model was close to the current limit

Fix logic error when storing MCMC models if the model fit had not yet been started

Increase default maximum number of model parameters allowed when the software is first started to 500

Match offered variables in customised prediction plot to outputs from prediction, instead of based on column name

Fix synchronisation of columns names when customised prediction is run

Improve keyboard selection and autocompletion in column selection drop-down boxes

Allow custom bar widths to be entered when creating histograms

Fix display of current MCMC iteration when running a model via macros

Correctly recreate the necessary structures to continue running models with orthogonal parameterization when a worksheet is loaded

Improve resizing behaviour of various windows

Add option to plot coverage intervals in customised prediction plots

Fix grid visibility in the variance function window

Add MLDA command to allow creation of a column containing the deviation from a group average

Fix extra parameters being removed when deleting model terms

Fix removing common coefficients to not remove similar separate coefficients at the same time

Avoid creating an extra column for the power of one term in polynomials

Fix crash if removing parameters with the NEXP command

Reduce unnecessary recalculations of the missing value indicator column

Correct calculation of level subscripts in the equations window to take into account missing values

Fix error when removing values or centiles in the customised predictions window

Open new windows on the current screen in multiple-monitor setups

Resize graph titles in the trajectories window to fit into the available space

Fix potential crash if a column is renamed after the response variable windows has been closed

Add option for automatically filling in range options in customised predictions window

Changed coverage interval variable labelling to be more consistent between customised predictions specification and plot windows

Improve the speed when opening hierarchy window by only autosizing the grid after it has been filled

Base available columns in customised prediction plots on the columns associated with the original prediction, rather than column names

Correct submatrix vector-multiply when there is only one row

Display macro editor within main window when it is opened

Display model comparison window within main window when it is opened

Changed the default value for MISR from one to zero for discrete models (excluding multinomial).

Check whether common coefficients have been specified when using MCMC estimation

Check that the model type specified in the MCMC command matches the currently set up model

Check the denominator is valid when running probit models with Gibbs estimation

Speed up changes to the columns displayed in the names window

Add missing line of prior for negative-binomial models

Always use "." as the decimal separator when taking numeric inputs

Fix crash if negative-binomial responses is chosen for multiple-membership models

Allow setting an initial value for negative-binomial overdispersion parameter via the equation window

Only offer columns with data when adding/removing variables via the estimates window

Increase default window size for the data window

Automatically resize columns in model comparison window when it is opened

Fix selecting the column via typing for the prediction, variance and graph highlight windows

Clarify that the values generated by the tail areas window are the upper tail

Fix columns being reversed after an undo operation from data manipulation windows

Update the back-end as the prediction window is filled in to prevent it been reset in certain cases

Use alpha instead of v for the overdispersion parameter in negative-binomial models

Fix crash when setting up a model where cross-classification is turned on an no level-1 has been specified

Fix error (introduced in version 2.33) in the DIC calculation for binary probit models using Gibbs sampling

Fix incorrect likelihood displayed during burn-in period for binary probit models using Gibbs sampling

Add commands to calculate factorial, logfactorial, normal-pdf, poisson-pdf and random draws from a negative-binomial distribution

Fix error in generated command for "distributed random numbers"

Add option for negative-binomial to "generated random numbers" window

Update compiler used for back-end to Visual C++ 2017. Note, the standard random distribution generators have changed in this version, so results may differ. If you need consistent results between versions set RNGV to zero.

New Windows GUI built on the .Net framework

Worksheet columns are now stored in double precision

A 64-bit GUI version of the software is now available

Allow multiple graph windows to be open at once

Allow multiple MCMC diagnostic windows to be open at once

Increase maximum value accepted by EXPO function

Fix display of confidence/credible intervals for model comparison window

Remove display of confidence intervals for random part variables in model comparison window when (R)IGLS estimation is used

Speed up sorting when the data is already in the desired order

Add row for z-ratio to contrasts window

Allow XSS command to be recognised again

Allow loading worksheets where the uncompressed size of the worksheet is great than can be stored in 32-bits

Change "z-score" label in model comparison window to "z-ratio"

Fix bug where it was not possible to select maximum levels minus one as the desired number of levels

Always display z-ratio and p-values to three decimal places in the intervals and tests window

Rename the MISS calculate function to NMIS to avoid conflicts

Correct MCMC iteration display when thinning is used

Fix "execute selection" button in the macro editor to work as expected

Add option to the RTYPe command for computing unadjusted comparative variances

Add command corresponding to the "subscripts as names" selection box

Remove option for using worksheet space to store temporary matrices

Correct missing values output when saving Stata dta files as double precision

Change the default random number generator from Wichmann-Hill to Mersenne Twister

Add window for specifying 3D graphs

Fix crash when attempting to fit a multiple-membership model with multivariate models

Remove option to set the worksheet size as this is now redundant

Fix bug where the last value of the response variable for ordered multinomial models was incorrect when exported to BUGS

Add menus for saving WinBUGS/OpenBUGS script files

Set default version of BUGS files exported to 1.4

Update BUGS interface to allow calling WinBUGS/OpenBUGS directly from MLwiN

Fix crash if a command shorter than four characters is issued

Increase maximum number of displayed iterations in the trajectories window to 50000

Update version of BLAS/LAPACK used to version 3.7.0

Update SPSS I/O library to version corresponding to SPSS 24

Ensure that command generated to change dataset label quotes the new label

Improve commands generated through the GUI

When importing ASCII data only flag up integer values that are too long to be used as IDs

Fix swap command to allow turning off polynomial status for variables

Store the state of the new random number generator when saving the worksheet

All rows to be pasted can now be previewed, rather than just the first 10, via the Edit->Paste Menu

Updated included zlib library to version 1.2.11

Updated included shapelib library to version 1.4.0

Updated included SPSS I/O library to version 24

Fix default selected categories for ordered multinomial in customised predictions screen

Add facility to set up some basic structures to the "intervals and tests" window

Add message to "intervals and tests" window to indicate that z-ratios and p-values assume normality

Fix crash in ASSI command if one column is specified and only text data is provided

Fix pasting a single text variable via the names window

Add window for viewing current box values

Add window for viewing current string values

Add option for MLREcode to multilevel data manipulations window

Add "clear display" button to customised graphs window

Fix pairwise residual plots with subset of parameters

Update labels in the predictions window to attempt to make them clearer

Swap drop-down locations for plus and minus error bars

Scale down graph margin titles when the text is too wide to fit

Modify column display from PRINT command to allow it to work better with the "copy as table" functionality

Display status messages during model storage

Remove limit of twelve graphs when initially opening the trajectories window

Add unit information to the equation window

Save MCMC settings to the worksheet, even if the model estimation has not yet been started

Allow MCMC diagnostics to also be triggered by clicking the graph title

RNGV with no parameters will now report the version currently in use

Add DTOC command to allow creating a categorical variable from unique values of a continuous variable, using the previous values as labels

Fix ADDM command to not add the created indicator columns into the fixed part of the model

Allow more control over whether to display variances and/or correlations in model comparisons

Update "add term" window to better match the capabilities of the corresponding command

Add window to allow specification of default graph export options for newly drawn graphs

Reset column colour in the names window if it becomes unmarked

Fix use of MQL2 in mixed response models

Fix crash if hierarchical centring is turned on and one of the X variables has a zero at the end

Fix crash if elements of the prior matrix are greater than 32768

Avoid some unnecessary calulations when working out coverage intervals in customised predictions (can avoid some errors)

Correct column number check when loading very old worksheet versions.

Mark system columns to prevent accidental overwriting

Fix crash in variance function for discrete models where explanatory variables are added prior to changing the response type

Correct positioning of pop-up dialogue boxes in the equations window

Improve output when running MCMC models via the scripting version

Fix crash when attempting to use the TAN function in the CALC command

Add optional display of p-values, z-scores and confidence intervals in the model comparison window

Provide finer control over which values are displayed in the model comparison window (see MMOD command)

Update ESS calculation used for model comparisons

Keep bayesian-p values for factor loadings in stored models in the worksheet when saving

Add more functions to the CALC command

Add option for t-distribution to the "Tail Areas" window

Use the global number formatting options in TABUlate command

Add iteration information to model comparison table

Remove newline and carriage returns when trimming strings

Add command for setting location of missing indicator columns

Remember used burnin value after model has run

Store actual used burnin and thinning values in model comparisons

Store output column location in the worksheet

Use length of deviance column in DIC calculation

Check a response has been specified before checking ID columns

Avoid crash when reading the data label from SAS transport files

Improve synchronisation of graph, X, and Y variable settings with back-end

Fix potential text corruption when sending link names

Provide more customisation in the "Manage Stored Models" window

Avoid crash when pasting if the clipboard does not contain text

Fetch log-formulation setting from the backend to ensure that it remains consistent

Store (R)IGLS and linearisation method in model comparisons

Fix drawing ^ symbol on parameter names in the variance function window

Add beta 64-bit version of MLwiN

Correct bug that was causing models with hierachical centring to give incorrect results (introduced in 2.33)

Fix crash if multivariate MH was chosen as the MCMC method for models with hierarchical centring

Fix crash when reading the denominator if there is missing data (introduced in 2.33)

Provide warning if the clipboard is inaccessible, rather than crash

Fix crash when loading multinomial MCMC models

Correct MCMC estimation when whole records are missing (bug introduced in version 2.33)

Correctly synchronize model structures when using the FPAR command

Add option to choose whether to apply polynomial or power first if both are specified for a variable

Recognise that model has changed after NEXP command

Improve compatibility with old macros by adding a dummy WAIT command

Allow variables added with SETX and ADDM to be removed again later

Add check that the first argument of various toggling commands is numeric

Fix display of higher level priors for multivariate probit models in MCMC

Fix crash when calculating weights when the data has missing values

Various MCMC speedups

Fix crash when storing models with elements of the covariance matrix removed

Add button to select all categories to customised prediction window

Keep command, output and macros windows if the worksheet is reinitialised

Make shapelib dependency optional

Fix some bugs that caused MLwiN to sometimes crash when loading MCMC information

Fix crash if error bar length did not match other data being plotted

Correct DIC calculation for binary-probit models with Gibbs

Add initial support for negative-binomial models with MCMC

Fix reading variable names for imputed data

Add p-values to "intervals and tests" output

Add initial support for reading Stata 14 .dta files

Make graph point identification more consistent for multiple membership models

Fix display of residual plots when a subset is selected

Allow informative priors to be specified correctly for one-level models

Take thinning into account when pre-allocating space to hold MCMC chains

Display Bayesian-p value for factor loadings in model comparision table

Avoid picking up constraints columns when allocating columns for a new group

Check for size zero when allocating matrices

Avoid crashing when calculating percentiles at zero or one hundred percent

Prevent crash when clearing a partially set up MCMC model

Add buttons for deciles and quantiles in the customised prediction set-up

Add command to turn off saving MCMC chains, or redirect them to a file instead

Allow calculating standardised residals to be optional

Update residuals window to reflect the estimation method used

Avoid crash if loading a worksheet saved in 2.31 that had historical trellis-graph information

Check all input columns have the same length when running the TABUlate command

Fix calculation of the standard deviation for group and grand totals in tabulations

Fix potential crash in (R)IGLS if whole higher-level units were missing

Fix crash in MCMC factor models when units were missing

Correct miscalculation of missing index for MCMC if whole block were missing

Fix loading imputed data from Realcom-Impute

Fix potential crash when resizing the customised graph window

Correct potential crash when saving Stata/SAS data files in January

Check for non-finite values when generated MCMC proposal distributions

Fix bug when resizing C1096/C1097 in the ADDM command

Provide error message if ID columns contain missing values when calling WTCOl, rather than failing silently

Fix crash when calculating predictions if the model had not recently been run with (R)IGLS

Allow setting model starting values via the equations window

Increase maximum worksheet values selectable via the GUI

Display zero in equations window for missing estimates in C1096-C1099

Fix reversing the columns when undoing a data manipulation operation

Switched to the POLE library (https://github.com/catlan/pole) to allow cross-platform saving/loading of Minitab files

Updated engine to allow building under Linux/Mac OS X

Updated screenshots in manuals

Prevented group labels overlapping with the last point

Turn off legend for customized prediction graphs with no grouping

Turn off graph margin labels for customised prediction graphs with no trellising

Removed nonfunctional old command references

Fixed position of graph series labels when error bars have been added

Allow variable number of variance function rows to be specified

Fixed potential matrix overwrite when calculating residuals

Allow temporary matrices to increase in size if necessary

Improve check for IDs in multivariate multiple membership models

Fixed occasional crash if removing several variables from the model at once (for example categorical dummies)

Increase precision in various calculations to double

Fixed reading MCMC information from worksheets saved in versions prior to version 2.26

Attempt to convert multivariate structures when loading old data (for example from MLwiN v1)

Save correct response information when saving to old worksheet formats

Improve saving/loading older worksheet versions

Don't write loop contents to temporary files while running macros

Don't close open macro editor windows when loading data

Allow fixed-width data input to contain spaces again

Fix the RPAT command to work if there is only one response

Fix PUPD to correctly store the previous estimates

Fix potential crash in RAND command

Once again allow graph symbol size to be set

Allow execution of selected lines in the macro editor

Label the x-axis with the stored iteration number in the trajectories window

Increase the maximum number of error bars

Add "rotate" option for graph symbols

Ensure sample data is consistently labelled

Documentation updated

Add standard error column to variance function window

Fixed crash in "set graph highlight" window if any of the defined colours had alpha information

Fixed crash in multivariate MCMC models with some variance starting values

Fixed misalignment in stored model results table if fixed part parameters are added in later models

Correctly handle reading missing data codes for Stata versions prior to 8

Fixed crash if more residual storage columns were needed than the maximum number of variables allowed in the model

Fixed write errors when writing BUGS code version of model

Fixed crash in SETX command

Corrected calculation of inverse normal (may slightly affect Raftery-Lewis and Brooks-Draper diagnostics)

Fixed confusing label in customised predictions plot

Fixed crash when reducing the number of levels defined in a model

Fixed off-by-one error in multiple-membership models

Added check that unit IDs are not duplicated within rows in multiple membership models

Added Bayesian-p values and correlations to stored model results

Fixed crash in the SORT command if the number of columns chosen to sort was greater than half the total columns in the worksheet

The centring flag is now toggled off again after adding variables via the equations window, this was previously causing variables added via other methods to have incorrect centring applied

Attempting to plot empty columns should no longer cause MLwiN to crash

Fixed an error in the calculation of fixed part robust standard errors when the number of levels in the model is greater than one

CAR models can now have random effects mean centred, which allows an intercept to be included in the model

Fix crash if standardised weights are defined at more than one higher level

Allow raw weights to be specified without standardised weights, to match the online help

Ignore Stata value labels with empty column name to work around issue with the R foreign package

Added command (FXMM) to automatically check and fix duplicate column IDs in a row for multiple membership models

Fixed crash if the data had certain patterns of missing

Added command for specifying uniform priors in MCMC

Changed ASCII file input to only give an accuracy message for integer variables

Updated compiler to Visual C++ 2010

Fixed crash when requesting level 2 residuals for a multivariate cross-classified MCMC model

Fixed crash in BDIC command if the deviance has be saved somewhere other than c1091

Allow graph colours to be defined as RGB values

Fixed calculation of robust standard errors in 1 level models

Correct bug where DIAG calc function was generated a column that was too long

Correct case where matrix dimensions could get out of sync with the column length

Display v parameter in equations window for negative-binomial models

Fix import/export dialog labels

Include SETD parameters in RPAR command output

Display names window even if there is a prediction grid loading error

Fixed crash in split record dialogue box (when getting unique column values)

Fixed crash when running a saved multivariate MCMC model due to hierarchy information not being available

Fixed incorrect column overwrite in VMAT, ZMAT, YRES, YMAT and XMAT commands

Fixed incorrect column overwrite in MKBL command

Ensure matrix version of columns remains consistent with total length

Pad with missing values if exporting to SPSS and the columns aren't all the same length

Fixed crash when reading Stata files that report categories, but the number is zero

Fixed crash in MLBOx when a unit only contained one value (all quantiles for the unit will now be set to this value)

Improve space check when saving undo information

Fixed crash in PUPD command (used in bootstrap estimation)

Fixed crash in FDIn command

Add initial support for reading Stata 13 format data files

Fixed incorrect number of units reported in multiple membership models if a zero was included in the first ID column

Improved X axis labelling for categorical variables

Remember robust standard error options

Fix crash when using DINP in mlnscript

Fix second form of MRAN

Fix bug in ascii file input where the choice to make a variable categorical was not always correctly applied

Not selecting any columns when exporting should now save all non-empty columns, rather than crashing

Memory for data is now allocated dynamically. This should make it no longer necessary to set a worksheet size. This will also allow more memory to be used as it no longer has to be contiguous

A label can be be associated with the dataset

Display error is hierarchical centring is defined at an invalid level

The colour palette used for graphing can now be modified

Added command for storing residuals when performing imputations

Fixed potentially incorrect hierarchy information if the predictions window is open when changing a model

Improved speed when running binomial proportion models

When reading starting residuals for MCMC replace missing values with zero instead of skipping them

Changed copy shortcuts (Control-C now does a standard copy and Control-G takes a screenshot of the current window)

Introduced MLOGit command for unordered multinomial logit models

Fixed crash when running a one level model for further MCMC iterations after loading a worksheet where the estimation had already been started

Fixed crash when changing the level identifiers when the cross-classified flag is turned on

Avoid possible file name conflict generating temporary files from more than one instance of MLwiN

Fixed linearisation method changing from 2nd order PQL to 2nd order MQL is the nonlinear setting screen is opened

Fixed bug where settings could be read incorrectly after upgrading Windows

Added a command line option (/reset) to clear saved settings.

Fixed incorrect predictions when using cross-classified MCMC models

Fixed incorrect graph highlighting when using cross-classified MCMC models

Fixed potentially incorrect number of level one units reported for cross-classified models with no missing data

Fixed incorrect copying of random part variance matrix if SETE is called with more than one element at a time (for example when setting to diagonal)

Fixed potential data corruption if the available worksheet space gets too low during a CALC command

Fixed the column to the left of the constraints column being deleted when clearing a model

Fixed PREG command to allow selecting the response and setting coverage intervals

Fixed crash after running OLSE command if not all the fixed effects were chosen

Fixed off-by-one error in DUMM command

Fixed off-by-one error in SETX command

Changed quantile method in MLBO command to be more consistent

Fixed label display when first loading the tail areas window

Changed default memory allocation to not use the worksheet space

Fixed crash when saving a WinBUGS model when hierarchical centring is turned on

Fixed error bars so that they can be specified independently

Fixed some equation window subscript display issues

Fixed display issue when first opening the tail areas window

Fixed occasionally wrong sign when using the DET function in the CALC command

Increased the intermediate precision of the PREDict command

Changed graph component. This should provide a more reliable graphing experience. This has been set up to behave as closely as possible to the old version, however a few features are no longer available (for example variable size points)

Switched to blas/lapack for many matrix operations

Fixed potential data corruption when saving factor chains

Fixed crash if attempting to save an empty worksheet as a Stata dataset

Don't hide values of zero on MCMC diagnostic and trajectory plots

Fixed possible crash when reducing the number of levels in a model

Fixed posible corruption if factors are set up before any models are run

Fixed skipping some data when reading in BUGS coda files

Corrected MCMC multiple membership error messages

Improved reading missing data in Stata versions prior to 8

Fixed crash in equation window where matrices are present at low zoom settings

Allow zero as a valid cross-classification ID for MCMC estimation

Correct list of reserved variable names in Stata

Fixed incorrect heading and error if using TABS with categorical variables

Fixed indicator column generation and data corruption in RECS command

Fixed data corruption and incorrect results in MLBOx

Fixed various errors in generated BUGS output

Fixed updating priors in the equations window when the estimation method is changed via a command

Fixed missing estimates when using simple notation and random intercept isn't switched on

Improved MCMC residual starting values when residuals aren't provided (for example cross-classified models)

Fixed crash in MCMC when using a multiple-membership model where there are more classification units level 1 units

Fixed crash when exporting data if the names window is open

Changed tabulate to not print rows/columns where there are no values

Fixed crash when changing notation settings with the data window open

Fixed a potential crash if using block updating in MCMC

Show cases in use regardless of model type

Improve warning if rounding occurs when importing data

Provide warning if a column used in a customised prediction is going to be overwritten

Save MCMC state if estimation mode is set to MCMC and MCMC macro commands are used

Store factor information in model comparisions if appropriate

Differentiate between levels with the same name in a single model for model comparisons

Fixed a bug in the example MH algorithm provided in the manual

Enable reading Stata 12 data sets

Add command to allow selection of an alternative random number generator

Send correct MCMC command if the response type is changed via a command

No longer provide starting residuals from hierarchical models to MCMC when running cross-classified model

Fix MLLAg for data where blocks start with missing values

Fix expanded WinBUGS output where no residuals are specified

Fix crash when plotting a trellis graph, based on one of row or column codes, where there were more than five unique codes

Fix crash when saving/loading models where measurement error is defined

Fix intervals and tests window in the case where extra random variation is specified

Fix crash when turning off exclusion in the hierarchy viewer

Give error message if starting residuals for MCMC are wrong length

Fixed rank2.txt macro to use 2.5% quantiles

Don't truncate numeric values passed to SJOIn

Mark macro as modified if on delete or cut

Fix incorrect "Line too long" message when out of worksheet space

Fix crash if maximum number of explanatory variables is set lower than the maximum number of levels

Fix crash if the maximum number of levels is set lower than three

Store and display pD for model comparisons

Allow variables with the category "c" to be removed from a model via the interface

Don't show extended MCMC information by default when doing model comparison

Fix potential crash when displaying prevously saved graphs where a data set label is defined

Fixed TOLE command so that it no longer sets MAXI to the same value

Fixed crash in SURV command if it referenced the first column of data

Fixed crash when making attempting to make an empty column categorical

Added commands to import/export data into a worksheet

Improved efficiency of UNVEctorise command

The hierarchy viewer will no longer display as cross-classified if the estimation mode is switched back from MCMC

Fixed SIMU crash if no parameters are given

Fixed UNVE crash if no parameters are given

Improved efficiency of COMBine command

Fixed crash in residuals screen if no model has been set up

Fixed crash in customised predictions with multinomial/multivariate models

Fixed MVIE to handle missing values

Fixed CORO command

Fixed DVAR command

It is now possible to set full/diagonal matrix for the response covariance matrix by clicking "cov"

Added option in the "responses" window to set whether responses added to the respose covariance matrix are entered as full or diagonal matrices

Fixed crash when invalid or empty values were entered in the factor setup screen

Fixed crash in the residuals window when no model was set up

Quotes in command parameters must now match

Fixed incorrect number of residuals in multiple membership models where not all the possible IDs appeared in the first ID column

Added initial support for SAS transport files

Fixed crash in DAMI command if no missing data had been imputed

Fix crash if the response is typed in rather than selected

Allow MLwiN to still start up with a default worksheet size if there isn't enough memory to provide the requested size

Don't set a reference category in the case where the variable only contains one category

Fix crash if you attempt to click a point on a histogram to identify it

Correctly calculate interactions involving centred terms

Remove incorrect warning in customised predictions where a group centred variable had previously been in the model, but is now removed

Fix constraints being cleared when running an imputation

Fix a crash in VECTorise if an invalid output column is specified

Fix possible crash in the WTCOlumn command

Fixed offset display

Fixed crash when saving worksheets containing an MCMC run that had been loaded from a previous version

Fixed display error in the NonLinear dialogue box

MCMC refresh rate is now remembered and saved in the worksheet

Graph autosorting is now saved in the worksheet

Iteration pause setting is now saved in the worksheet

No longer remove categories from last column pasted in

Fixed possible data corruption when using the PUPN and PUPD commands and the lengths of 1096..1099 do not match the input values

Fixed crash when opening or creating a macro and the current folder is set to a UNC path

Fixed constraints being ignored on certain computer configurations

Fixed crash in the SETX command

Fixed crash in the XOMIt command

Fixed bug in MOVE command with non-default column names

Fixed macros not resuming if equation settings changed while a macro is paused

Variables and levels in model comparisons are now sorted based on the order of models given to the MSTO command

MCMC seed menu option now applies to macros too

Fixed occasional crash when an estimate went to zero

FTESt and RTESt commands now use the MLwiN numeric precision setting

MCOM now makes use of it's parameters again

Improved pasting model comparison tables into packages such as Word

Customised predictions no longer crash if the number of predicted cases is greater than the length of the data

A warning is now given if a customised prediction is made on a variable that is centred around a group mean

Fixed crash that occurred if you attempted to read in more than one text data file via READ (This also fixes a crash in the ISTA command)

Give a warning rather than an error if data contains missing values for multinomial models

Fix crash if plotting a graph using a worksheet from a previous MLwiN version

Fixed problem with loading category labels from SPSS data files

Fixed problem with loading text columns from Stata data files

Fixed problem with pasting columns containing text into the worksheet

Fixed a potential crash when changing the equation screen layout via commands in a macro

Including/Excluding points from the model via the graphs options screen should now work for models with more than 2 levels

The last point in the data set should now be correctly identified if selected, even after it has been excluded from the model

Fixed crash when trying to display category labels longer than 254 characters

Attempting a customised prediction on a model containing orthogonal polynomials no longer causes a crash

Allowed copy/paste of categories between columns

Fixed potential problem when adding/removing categories from a column

Fixed error in macros for discrete negative binomial models

Fixed LOAD/SAVE problems with some file or directory names

Fixed crash with MOVE command if groups are defined

Fixed forcing PRE and POST on when loading a worksheet

Added an error message if the user asks for a constraint but then does not specify it

Fixed a problem in the BVAR macro that prevented negative binomial models from running

Please note that there is a bug in this release: The problem is that certain file names are being misinterpreted as if they are command arguments. Examples of file names where this happens are ones **containing a space followed by a number, a "C", "G", "B" or "K"**. When this happens MLwiN is unable to open or save the file, giving a "Wrong Params" error. The complete file name including the path could be misinterpreted.

The workaround is to either rename and/or move the file so that it does not match the above criteria, or to open the command interface and click the most recent line starting with "LOAD" or "STOR". The user must then replace the "LOAD" or "STOR" command with one appropriate for the type of worksheet and press enter to run them command. The appropriate replacements are as follows:

LOAD | STOR | |

MLwiN | RETR | SAVE |

MLwiN (compressed) | ZRET | ZSAV |

Stata | RSTA | SSTA |

Minitab | RMTW | SMTW |

SPSS | RSPS | SSPS |

We apologise for any inconvenience this may cause. This bug will be fixed in the next release.

GCLEar <N> no longer removes text labels from all graphs

Sandwich estimators are now used for weights set up by the GUI

MCMC options now correctly displays the currently selected leveling level

The "Copy" button in the "customised predictions" window is now always visible when viewing the prediction grid

Renaming stored models is now possible

The following screens no longer incorrectly execute previous commands: *Listwise, Merge, Join, Sort*

Storing MCMC model results now saves more information

Allow copying results from the "variance" window

Ordered Multinomial models in MCMC only previously worked if the last category e.g. A in alevchem example was chosen as reference category - now works in both directions

Fixed crashes after using COMB and UCOM commands

Ignore black lines when reading text data

Preserve whether a term is in the fixed part when loading old worksheets

Fix bug where clearing the graph "y title" actually cleared the "x title"

Fix DESC command to correctly allow setting more than one column description at a time

Improved the consistency of equation colouring

Allow ADDTerm to re-add terms that have been removed from the equation by FPARt

WIPE is now better at clearing worksheet state

Fixed potential error when reading worksheets containing column descriptions

Re-enabled "standardised weights" option in the weights window

Fixed potential crash when changing the equation scale

Fixed reading in Stata files after there was already data in the worksheet

Fixed issue parsing unquoted single parameters

Take missing values into account for centiles and nested means in customised predictions

Remember graph row and column codes when retrieving worksheets

Fixed a crash when loading a worksheet containing column descriptions where the number of columns is different to the number saved in the worksheet

Fix "recode by value" to work on columns after empty columns in the data

Running multivariate discrete response models from a macro gave different results to running from the GUI when binomial (as opposed to extra-binomial) variation was specified for any response(s) other than the last response. This was because when these models were run from a macro, only the last random parameter was constrained, and so extra-binomial variation would be used for all but the last response regardless of the user's specification. Thus if the user wanted to fit, say, a trivariate binomial response model with all responses having binomial variation, then if the user ran the model from a macro the first two responses would fit extra binomial variation and the third response would estimate binomial variation. If the user instead used the user interface all three responses would be fit with binomial variation as required.

MCMC methodology features new in MLwiN 2.13 ( Further details…)

- parameter expansion
- hierarchical centring
- orthogonal parameterisations
- structured MCMC
- structured MVN

Fixed potential crash running multivariate models from a macro

Fixed to remember "None" reference category in modify term

Fixed to truncate labels that exceed Stata limits, rather than crashing

Fixed to save variable with a default name, rather than skip them when the name is invalid in SPSS

Fixed copy when the number of columns to be copied is two fewer than the number of rows

Fixed incorrect results from running multivariate discrete models from macros

Fixed potential crash loading a worksheet containing a model run in MCMC

Fixed Crash when running single level models from macros, where a column of 1s of the same length as the response variable and called 'cons' is not present in the worksheet

Fixed "highlight styles" window to work again

Fixed worksheet space leak when columns were erased in the opposite order in which they are created

Fixed model timing functionality in the GUI

Fixed potential crash when clearing the worksheet

Fixed potential data corruption when running a model through macros

Fixed crash on startup if the default worksheet size is set too high

Problem with with running models from macro files This bug effects version 2.10 beta 6-11 and release versions 2.10 and 2.11

Removed offset information when clearing the model

Fixed ISTA to not only use the first imputed data set

Increased precision of data exported by ISAV

Fixed crash in customised predictions when differences are selected in a multivariate model

Fixed SWAP to maintain whether a term is in the fixed part when swapping a term with itself

Fixed SWAP command to maintain the reference category when swapping a term with itself

Information from the EXCL command is now saved in the worksheet

Fixed potential crash in the MRAN command

Fixed potential error in the display of number of cases when loading one worksheet over another

Fixed potential crash after loading a worksheet containing MCMC information

Fixed potential crash when copying a prediction grid where the specification was incomplete

Fixed incorrect command generated when weights is ticked in averages and corellations

Fixed buffer overflow when copying columns with names greater than 19 characters

Fixed potential problem when pasting in extreme values

Fixed crash in customised predictions where terms had been removed from the model

Fixed problem where sometimes variables could not be removed from the model

Fixed labelling in customised predictions window

Allow specifying offset in negative binomial models

Fixed crash when removing offsets

Fixed crash when category name clashed with a column name

Fixed crash when the last model term was not in the fixed part

Fixed incorrect likelihood calculation

Increased precision in some calculations

Fixed crash in orthogonal polynomial

Fixed crash when importing data back from BUGS

Fixed column only trellis graphs not displaying

Fixed crash when specifying coverage interval in customised predictions

Allow UNIQ command to work if the input and output columns are the same

Fixed crash in cross classifications window if the number of classifications was left empty

Allow copying images of graphs again

Fixed graph scaling issues when switching from a categorical to continuous x axis

Fixed crash if the MLwiN window was resized to be too small

Allow Monitoring Chain Length of zero in MCMC

Customised predictions can now be properly cleared from a previously saved worksheet

Fixed crash when minimising the model comparison window

Allow UNC pathnames to be used in files

UNIQue command should now give feedback for invalid parameters

Fixed memory corruption when the user selects only one of median or mode for a multinomial customised prediction

Fixed bug preventing LOGA and LOGO from opening files

MSTO now records likelihood for normal models estimated in RIGLS

Fixed crash on clearing model with only the response specified

Fixed problem loading stored models

Allow custom x-axis graph scale without custom y-axis scale

Fixed crash in MOVE command

Fixed crash when filling customised prediction grid

Fixed a worksheet corruption caused by the EXPL command

Fixed saving worksheets where a previous worksheet has not been loaded

Likelihood is now calculated and stored in the same way in macros as for the GUI

ADDT command changed to use 10000 as base for orthogonal polynomial instead of 1000

Fixed crash on clearing a model with prediction window open

Fixed crash on generating a prediction grid with no model

Corrected schgend category name in tutorial dataset

Reset Centring information when loading a worksheet

Speedup when running macros that write a lot of text to the output

Fixed window name not being removed from status bar if the title changed

Fixed crash in cross-classified models containing missing values

Fixed Predictions grid display problem

Fixed crash on subsequent formatted text input/output

Fixed crash when attempting to plot greater than 6x6 graphs

Fixed a crash caused by running a macro that does not load a worksheet

Fixed a crash when viewing predictions with a mean or median selected

Allow plotting customised predictions where mean is not specified

Increased maximum number of columns involved in data manipulation from 400 to the number of columns in the worksheet

Fixed toggling general notation causing crash

Discrete macros folder now has the correct CHKRES macro

Fixed GBAR command to work correctly

Added support for SPSS system missing value

Discrete macros folder now has the correct POST macro

MLwiN crashes when editing data. This is caused by the cursor moving to the next line after you finishing editing the data. If there isn't a visible cell to move to the application will crash. *Workaround*: A temporary solution is to resize/scroll the data window so that there is always a visible cell below that which you are editing.

Attempting to save to a read-only device causes MLwiN to crash

MLwiN crashes when Paste window minimised.

If the user selects Edit > paste and then attempts to minimise the resulting window MLwiN will crash with a runtime error.This is caused by window resizing code being able to be run while there is no window visible to resize.

Incomplete line in input file causes problems.

If the user attempts to import delimited data where some of the rows do not have data or delimiters for every column MLwiN will import fewer than expected rows and place data in the wrong column. While it can be argued that such input files are invalid, unfortunately Microsoft Excel produces such data when the user attempt to paste data containing blanks at the end of the last column.