Archive for the ‘statistics’ Category

What might be coming next

Wednesday, April 20th, 2016

Python 2 is reaching its End Of Life (EOL) in 2020 so sometime before then I will want to shift SOFA to Python 3. I much prefer Python 3 but the main thing will be the libraries SOFA relies on to operate – especially on Windows and Mac.

Speaking of Mac, I am finding it very time-consuming supporting the platform. Not to enable SOFA’s core functionality to work but for the image processing libraries (esp convert and gs). Along the way I have spent countless weekends compiling using homebrew etc. Slow, tricky, and often fruitless. And it is difficult to test. I only have access to a Snow Leopard machine (virtualised to allow revert to snapshot) and that is no longer very relevant to what people need for newer versions of OS X. Some very kind people have offered to help with testing (thanks!) but the problem seems to be the packaging steps. Maybe what I need to do is ask the people who volunteered if any of them are able to compile convert and gs for me on their machines. I can then just include those versions in my packages and hopefully everything works.

A final work item is to add a Fisher’s LSD test. A friend is helping with this.

1.4.6 Adds basic time series

Friday, January 1st, 2016

SOFA line charts and area charts now treat dates as dates in the x-axis which makes it easier to look at time series data.

New option added to interface

New option added to interface

X-axis date aware

Time series selected – X-axis date aware

X-axis not date-aware

Time series not selected – X-axis not date-aware

Example time series chart

Example time series chart

Additional improvements include:

  • Better error message when not enough values in group to run analysis e.g. ANOVA.
  • Better handling of precision in p-value results displayed.
  • Better handling of dates pre-1900.
  • Better messages to user about potentially excessive categories in charts.
  • Add support for float years as date values for time series.
  • Add support for specifying port connecting to postgresql.
  • Allows boxplots when fewer values to display.

And there were two other changes:

  • Removed broken google docs integration – just as easy to manually download and import normally.
  • Removed two pop-ups – no longer needed.

There are also a number of bug fixes:

  • No longer a missing legend in multiseries scatterplots just because the first scatterplot only had one series of data.
  • Fixed bug with saving database connection details when a number involved (port).
  • Fixed PostgreSQL bug when saving connection without password – now succeeds rather than failing silently.
  • Fixed MySQL bug with adding rows.
  • Fixed bug in Windows with checkboxes not enabling/disabling properly unless panels refreshed.

1.4.5 More box plot options

Saturday, August 15th, 2015

There are several ways of doing box plots. Some show outliers, some don’t. Some set the whiskers at the min and max values, some don’t. Until now, SOFA kept it simple by only allowing one approach. But sometimes a little more flexibility is needed. So now users can choose between three options:

Option 1) This is the default. Outliers are displayed. Lower whiskers are 1.5 times the Inter-Quartile Range below the lower quartile, or the minimum value, whichever is closest to the middle. Upper whiskers are calculated using the same approach.

boxplot_whiskers_1_5

Option 2) Outliers hidden. Lower whiskers are 1.5 times the Inter-Quartile Range below the lower quartile, or the minimum value, whichever is closest to the middle. Upper whiskers are calculated using the same approach.

boxplot_hide_outliers

Option 3) Whiskers are at the minimum and maximum values.

boxplot_whiskers_min_max

And SOFA displays a small note at the bottom of each box plot so it is clear what approach has been used.

Additionally version 1.4.5 adds:

  • ODS importing can now cope with repeated column names.
  • Better error message when unable to get regression line details because of limited variability.

And there are numerous bug fixes as well:

  • Fix bug when problem with imported data.
  • Fix to title and date concatenation code so doesn’t break when title has non-ascii characters e.g. in Spanish (affecting Windows and Mac)
  • Reordered regression line plotting in js so appears on top of dots.
  • Added zero division error trap to spearman’s test error output.
  • Fixed bug which prevented ods_reader from importing repeating rows at the end of a spreadsheet. Only repeated empty rows at the end are considered the end of the data.
  • Properly handle all read operations on internal-use text files e.g. proj, css etc. Can cope with a utf-8 BOM (only) to cope with Windows Notepad editing. Breaks if other encodings used which is fair enough.

We hope you like the latest version.

Version 1.4.4 good for Mac Users

Monday, May 11th, 2015

Mac users can finally export output in PDF format including individual charts and report tables. And depending on the version of OS X Mac users will also be able to export as PNG images. Ensuring image exporting works across multiple versions of OS X is an ongoing project of work which users can help with if interested.

mac_export_output

mac_output

Users needing to produce monochrome output for publication will like the addition of a new monochrome theme.

monochrome

All New Features in 1.4.4:

  • Mac users can export output in PDF format (and PNG depending on version of OS X).
  • Added new monochrome theme.
  • Chi Square proportions output much easier to interpret successfully.
  • The name of the grouping variable is now displayed when running comparisons of groups e.g. Country if comparing Italy and Germany.
  • Exporting to spreadsheet detects if too many fields for xls output and informs user that only csv will be generated. Also truncates table name so worksheet name not too long.
  • Import dialog only displays file types suitable for importing.
  • Added message to let user know spreadsheet creation being skipped if no report tables to export.
  • More user help on need for raw data (not pre-summarised) and long-format vs wide-format data as appropriate.
  • Code reorganisation to make it possible for SOFA to be called in GUI form by external code GUI code.
  • Scripts are now easier to use for standalone purposes.
  • Added note about treatment of datetime data as categorical by SOFA for purposes of statistical tests.
  • When exporting to spreadsheet and csv changes reserved sofa_id field name to was_sofa_id so it is OK to reimport after changes.
  • More informative for larger range of potential problem e.g. database engine not functioning.

Bug Fixes:

  • Fix bug resulting in Pearson’s r being displayed instead of Spearman’s rho.
  • Fixed bug on some systems when saving a worksheet with spaces in name.
  • Prevented numerous bugs related to quoting table names, fully qualified file names etc.
  • Fixed bug with misuse of escape_pre_write on python code rather than normal content.
  • Skew, and normality test now cope with the nan issue better e.g. sqrt of a negative number. Just says unable to calculate instead of displaying nan (not a number). skewtest function now copes with negative number as input to square root.
  • Fixed bug when importing NaN text – now treated as a missing value in a numeric field.
  • Removed bug which sometimes prevented Mac users from being able to successfully change the report name.
  • Stopped making export folder if no output to export into it.

Importing tab-separated files and more

Saturday, March 15th, 2014

SOFA 1.4.3 now lets you import directly from tab-separated/tab-delimited files.

Importing tab-delimited text
Another change is less momentous but should really please people doing lots of row stats reports. As SOFA gained more measures it became increasingly more effort to select individual measures one-by-one, checkbox by checkbox. Now there is a toggle button for Select All/Deselect All. Much better :-).

Select or Deselect All

And the bonus themes are now part of the standard release making it easier than ever to make your charts and tables look good – I hope you like them.

New Themes

One change that most users will never notice is better support for running SOFA via scripts. An exciting automation project is currently under development using this functionality and I hope to have some news to share soon.

Here’s the full list of changes:

  • Can import tab-delimited data.
  • More options for attractive charts and reports. Three new themes available – sky, prestige (screen), and prestige (print).
  • Better support for automation (i.e. headless, running without GUI) esp in international context.
  • Exporting to spreadsheet now relies on more robust code library (xlwt)
  • Easy to select or deselect lots of row stats measures at once.
  • Faster opening in many cases.

And the bug fixes:

  • Minor tweak to PostgreSQL plug-in to handle timestamps without timezone.
  • Resolved bug when SQLite numbers are stored in a non-numeric field and processed for Chi Square test.
  • Importing csvs now copes better when only missing vals in sample of a field. Gives user the choice.
  • Fixed bug when doing a Row Stats table with a rows variable e.g. by Gender and some of the fields can’t be calculated for some of the row categories.
  • Headless importing now works in the event of inconsistent data types in fields.
  • Headless importing now reads entire dataset rather than a sample to avoid need for (human) decisions.
  • Scripts no longer rely on translated arguments. Much safer to use on other machines with different locales.
  • Fixed circular import bugs which only became visible when other bugs occurred.

Using SOFA alongside other statistics packages

Saturday, February 15th, 2014

All statistics packages have their strengths and weaknesses so it is not uncommon for people to want to use more than one – even on the same project. SOFA is focused on making it easy to use some core statistical tests and producing attractive, high definition tabular and charting output. SOFA also makes it easy to link to, or import from, a wide range of formats: xls, xlsx, csv, google docs spreadsheets, MS Access, NySQL, MS SQL Server, PostgreSQL, SQLite, and, more recently, CUBRID.

But there is no point overcomplicating SOFA so it can do every statistical test that might be needed for a particular project. SOFA users have been routinely surveyed on what features they would like added and it has not consolidated into a clear list of priorities. People need lots of different things depending on their specific projects.

So a sensible goal for SOFA is to make it easy to import and export data, including metadata such as variable and value labels. This strategy has already resulted in the addition to SOFA (version 1.4.2) of built-in export-to-spreadsheet functionality. And it has already been improved for version 1.4.3 (not released yet at the time of this writing).

The question is, what packages should SOFA target as priorities for interoperability? Feel free to fire me off an email at grant@sofastatistics.com.

Confidence Intervals for ANOVA & t-tests in 1.3.3

Friday, April 5th, 2013

95% confidence intervals have now been added to ANOVA and t-tests. And associated output has right justified numbers to make it easier to read.

Confidence Intervals

Version 1.3.3 also lets you sort by category labels in clustered bar charts, line charts, area charts and box plots. Area charts can also be sorted ascending or descending by count/mean/sum.

The series and category are now displayed in tooltips e.g. Italy, 20-29 for clustered bar charts, multi-series line charts, and box plots. This is especially helpful when there are lots of categories and/or series.

Boxplot Improvements

  • Improved statistics output footnotes.
  • Borders on bar-type charts are now optional. This can be useful when bars are very short.
  • Chi square clustered bar charts can cope with higher default limits for number of values.
  • Importing field names with more than 90 characters prohibited at the point of import rather than causing problems later.
  • The group by max number of values is now controlled by a single my_globals setting (making it easier to override).
  • The default settings for some remaining max values have been increased.

There was one minor bug-fix this version – line charts now cope better with lots of categories (increased padding around max label width in overall width calculations).

And a problem with the deb installer was also fixed.

Further improvements in 1.1.4

Friday, February 24th, 2012

The latest version adds a range of improvements:

  • Added lower and upper quartiles to Row Stats report tables.

    Quartiles

  • Box plots now start y-axis from just below the minimum y value of the data unless the content is close enough to the bottom of the graph to make it worth using 0 anyway.
    Y axis adjusted automatically  for box and whisker plots
  • Showing the percent sign in percent columns for report tables is now optional – which is good news for many dissertation students.

    Show (or hide) percentage symbols

  • SOFA now displays value labels sorted by the numerical version of numbers even if stored as text. So no more 1, 11, 2,3 etc in cases where people have stored the number as a Text data type.
  • Added some more valid US date formats using dot dividers.
  • New help button for importing data.
  • New help button to advise on how to make of flexible data filters.
  • English translations are handled better (no more messages about not having US English and using UK English instead etc).

Plus there are some useful bug fixes:

  • Fixed bug where getting observed values e.g. for chi square test, fell over when one field in pair had missing values while the other didn’t.
  • Fixed bug in calculation of upper and lower whiskers in box plots.
  • Single bar charts don’t show a bar title anymore – only needed if multichart.
  • Fixed bug which only changed variable definitions when the extra settings dialog was closed with OK and didn’t ever set it otherwise e.g. when changing the selected project.
  • Now copes with newer versions of matplotlib on Linux.
  • No longer stores empty strings as variable labels if user doesn’t enter a label.

New tutorial videos on SOFA Statistics

Sunday, January 15th, 2012

Check out these two new tutorial videos for SOFA Statistics:

0.9.13 Recode data e.g. age to age group; better stats support

Saturday, July 10th, 2010

SOFA Statistics 0.9.13 has a number of exciting new features.

1) Easy recoding of data e.g. age to age group:

Recoding data e.g. age to age group

Recoding data e.g. age to age group

2) Better support when choosing statistical tests:

When a selection is made SOFA displays some helpful tips to affirm the choice made or suggest alternatives.

Helpful tips

Helpful tips

The dialog also makes it easy to answer questions about your data which will help make a decision.

Answering questions about your data

Answering questions about your data

3) Better support for importing CSV files with different data encodings:

Confirm encoding when importing csv files

Confirm encoding when importing csv files

4) Nicer Windows installer

New Windows installer

5) More flexible installation options. Anything containing ‘sofa’ is OK.

There have also been some important bug fixes:

  • Recoding copes with REMAINING keyword properly and copes with varied field types properly.
  • Generally copes better with system encodings like Chinese Traditional (big5).
  • CSV importing handles non-English much better.
  • Corrected handling of non-English characters when errors are encountered.